PubMed Health. A service of the National Library of Medicine, National Institutes of Health.

Maund E, Craig D, Suekarran S, et al. Management of Frozen Shoulder: A Systematic Review and Cost-Effectiveness Analysis. Southampton (UK): NIHR Journals Library; 2012 Mar. (Health Technology Assessment, No. 16.11.)

3Results

Assessment of clinical effectivenes

Study selection

The searches identified 8883 records, 8341 from the original searches undertaken in March 2010, 521 from the update searches undertaken in MEDLINE and EMBASE in January 2011 and 21 from subsequent reference checking (Figure 1). On the basis of reviewing title and abstracts 8161 papers were excluded; 722 papers were ordered for a more detailed evaluation. Additionally, an author sent in one paper. Of the 723 papers, 67 were published in languages other than English. These included Chinese (22 papers), Russian (19 papers), Italian (6 papers), Japanese (6 papers), Dutch (5 papers), French (2 papers), Croatian (1 paper), German (1 paper), Hebrew (1 paper), Korean (1 paper), Norwegian (1 paper), Slovak (1 paper) and Turkish (1 paper).

FIGURE 1. Flow chart of study selection.

FIGURE 1

Flow chart of study selection.

After a detailed evaluation, 691 papers were excluded from the review, including four for which there was no translator available, and five that could not be obtained through the British Library (see Appendix 5 for a list of excluded studies). The main reasons for excluding studies from the review were that they were not a clinical study (146 studies) or they did not meet the study design criteria (193 studies) or the population criteria (244 studies). Of the studies excluded on the basis of population, in 42 studies it was unclear if the patients had primary frozen shoulder, whereas in 17 studies < 90% of patients in each study had primary frozen shoulder.

There were 32 studies included in the review, one of which was a cost–utility analysis conducted alongside a separately published study of effectiveness. Of the 32 studies, 28 were published in English, 2 in Chinese, 1 in Japanese and 1 in Norwegian.

Studies were grouped by the main treatment intervention of interest: six studies were of steroid injection (with and without physical therapy), three of sodium hyaluronate injection, twelve of physical therapy, three of acupuncture, four of MUA, three of distension and two of capsular release. One study, by Calis et al.,66 fell into more than one category and was reported in all relevant sections (steroid injection, sodium hyaluronate injection and physical therapy). In one of the physical therapy studies there was a ‘supervised neglect’ comparator. The included studies were from a wide range of countries (Turkey, Canada, UK, USA, Italy, Japan, Hong Kong, China, Taiwan, the Netherlands, Singapore, Thailand, Greece, Pakistan, Finland, Australia, Denmark and Norway).

Although a quantitative synthesis was planned this was largely not possible, as few studies could be pooled. The overall quality of the studies was poor and there were few studies of the same comparators. Within the classes of intervention there was considerable variability between studies in various aspects of the intervention such as dose (e.g. of steroid or sodium hyaluronate injection), and duration and intensity of treatment, especially in terms of physical therapy and home exercise programmes. The physical therapy interventions encompassed a range of combinations of ‘active’ and ‘passive’ components. There was variability in the outcomes reported, the tools used to measure individual outcomes, time of follow-up and type of data reported. As a result there were few situations in which it was appropriate to pool and undertake the planned subgroup analyses. It was therefore necessary to primarily use narrative synthesis to present the findings of this review. The main analysis was the narrative synthesis with pair-wise comparisons where appropriate. This is presented first with studies grouped by intervention. This is followed by the MTC. This was a secondary, exploratory analysis because of the small number of studies that connected in the network.

Steroid injection

Study characteristics

Six RCTs investigated steroid injections, with or without the addition of physiotherapy, in the treatment of primary frozen shoulder. A summary of the study characteristics is reported in Table 6, with further details available in Appendix 6. All were full papers of studies conducted in the UK,41,67 Turkey,66,68 Canada35 and the USA42 between 1981 and 2008. There was one two-armed trial,68 one three-armed trial67 and four four-armed trials35,41,42,66

TABLE 6. Studies of steroid injection.

TABLE 6

Studies of steroid injection.

The steroids administered were methylprednisolone acetate and triamcinolone hexacetonide. Two studies used a single injection of 20 mg triamcinolone hexacetonide,41,67 one of which also administered lidocaine,67 and three studies used a single injection of 40 mg methylprednisolone acetate or triamcinolone hexacetonide.35,66,68 The remaining study administered three 40-mg injections of methylprednisolone acetate plus lidocaine, one each week for 3 weeks.42 In one of the studies the injection was guided using fluoroscopy.35 Five of the studies reported administering an intra-articular injection,35,41,42,66,68 one of which compared an anterior and a lateral approach,42 and in one study it was unclear.67

The comparators were steroid injection plus physiotherapy,35,41,67 placebo injection,35,41,42,68 physiotherapy with35,41 and without66,67 placebo injection and home exercise alone.66

Five of the six trials evaluated a physical therapy regimen as part of the intervention.35,41,42,66,67 These varied between studies but all consisted of physiotherapy modalities. They included combinations of exercises, mobilisation techniques, TENS, ultrasound, ice, heat packs and interferential therapy. All of the studies appeared to include some form of mobilisation as part of the physiotherapy, although details were scant in some studies. The duration of physiotherapy ranged from 4 to 11 weeks, although, with the exception of one study, duration was < 6 weeks. One study tailored the physiotherapy depending on whether participants had acute or chronic-like symptoms.35

Five studies reported that participants in all groups were asked to undertake home exercises, although the level of detail reported varied between studies. Bal et al.68 had a substantial home exercise programme: participants were asked to undertake shoulder stretching and stabilising exercises five times per day over 12 weeks, using a heat pack before exercise and a cold pack after. Carette et al.35 asked participants to carry out active and auto-assisted exercises twice daily over 12 weeks. In the study by Calis et al.66 stretching and Codman's exercises were performed. Details of home exercise were not reported in the other two studies.41,42 The study by Dacre et al.67 did not have a home exercise component.

There was also some variability between studies in concomitant treatments. Four studies reported that paracetamol of varying doses was available to participants: 1500 mg per day when needed in one study,68 a supply (dose unspecified) in one study,35 one to two tablets every 4–6 hours (a maximum of eight per day) in another study41 and paracetamol when needed in the remaining study.66 Another study stated that all participants were advised to continue NSAIDs.42 One study did not report whether participants received concomitant treatment.67

The inclusion criteria varied between the six studies, although four of the studies used similar criteria for the extent of restriction of movement.35,41,66,68 Three of the studies included participants with frozen shoulder of < 6 months' duration.35,41,42 Where reported, frozen shoulder was diagnosed through presence of pain and range of movement, clinical diagnosis, presence of symptoms, laboratory tests and radiography. The method of diagnosis was unclear in two studies.42,68 Three studies included a small number of participants with diabetes: 6% in the study by Carette et al.,35 5% in the study by Dacre et al.67 and 6% in the study by Ryans et al.41 None of the studies reported results separately for patients with diabetes. The participants' stage of frozen shoulder at baseline was not reported in any of the studies. Where reported, the mean duration of frozen shoulder amongst the included participants ranged from 1342 to 2135 weeks. None of the studies reported whether participants had received any previous treatment for frozen shoulder. The mean age of participants ranged from 54 to 57 years and the proportion of women ranged from 42% to 63%.

Quality assessment

Quality varied between the included studies. Four studies did not report the method of randomisation;42,6668 therefore, it was unclear whether these were truly randomised studies as stated by the authors. In addition, in two of these studies it was unclear whether the intervention groups were comparable at baseline.67,68 These four studies therefore have a potentially high risk of bias even though they met other criteria such as blinding of outcome assessment. Only one of these studies was adequately powered for at least one outcome.67 Bal et al.68 also reported substantial loss to follow-up (20%, all in the placebo group). Dacre et al.67 and Rizk et al.42 reported considerably less loss to follow-up (6% and 8% respectively), whereas Calis et al.66 reported that there were no dropouts.

The remaining two studies were considered of satisfactory quality although one had some risk of bias.41 Both reported an adequate method of randomisation and Carette et al.35 also reported an adequate method of allocation concealment; it was unclear whether Ryans et al.41 used an adequate method because, although sealed envelopes were used, it was not stated whether these were opaque or sequentially numbered. Both studies met most of the remaining criteria including reported blinding of outcome assessment. Ryans et al.41 reported that participants were blinded to type of injection. Carette et al.35 attempted to blind the injection; however, the method used (covering the syringe with foil) may not have been reliable. Blinding of the physiotherapy treatment was not possible. Carette et al.35 also reported an intention-to-treat analysis. The main limitation of this study was that it may not have been sufficiently powered; 17% were lost to follow-up and there appears to be an imbalance between treatment groups, with greater loss to follow-up in the steroid group. Ryans et al.41 did not report an intention-to-treat analysis and there was also a high proportion lost to follow-up (27%), with a higher number of dropouts in the steroid only and placebo groups. As with the Carette study, the Ryans study may not have been sufficiently powered. Full details of study quality are reported in Appendix 8.

Pain

All six studies assessed pain but Calis et al.66 and Dacre et al.67 reported data in graphical form only (Table 7). Consequently, a SMD could not be calculated and these two studies could not be included in a meta-analysis. Calis et al.66 reported that there was significantly greater improvement in pain in the steroid group than in the placebo group (p = 0.02). Dacre et al.67 evaluated steroid, steroid and physiotherapy, and physiotherapy alone. This study reported that all three groups showed a significant reduction in pain after 6 weeks (p < 0.001) with mean measures improving by 49–66%, and further improvements at 6 months. Improvements were reported to be similar in all three treatment groups.

TABLE 7. Pain outcomes in studies evaluating steroid injection.

TABLE 7

Pain outcomes in studies evaluating steroid injection.

Of the four remaining studies, three reported final mean values and SDs were reported or could be imputed. Both Carette et al.35 and Rizk et al.42 reported pain overall and Ryans et al.41 reported daytime pain at rest (Table 7). Each of the three studies reported pain using a different scale.

The study by Bal et al.68 reported mean change and SDs at 12 weeks for pain at night measured on the VAS 0–100 mm. Although a SMD could be calculated for this study, it was inappropriate to pool this SMD with those SMDs calculated from final value means, as the difference in SD reflects not differences in measurement scale, but differences in the reliability of the measurements.48

Table 8 provides the SMDs and 95% CIs. The outcome data for individual groups in the included trials are available in Appendix 7.

TABLE 8. Pain results from studies of steroid injections (SMD and 95% CI).

TABLE 8

Pain results from studies of steroid injections (SMD and 95% CI).

Steroid versus placebo

Three studies reported pain at short-term follow-up. Two studies reported final value mean pain for short-term follow-up: Carette et al.35 reported data at 6 weeks and 3 months and Ryans et al.41 reported data at 6 weeks (Table 8). As stated in the methods section, to allow comparison of the results of the meta-analysis with the MTC for pain, time points closest to 3 months were pooled, that is, 3-month data for Carette et al.35 were pooled with 6-week data for Ryans et al.41 There was a significant decrease in pain with steroid injection compared with placebo: pooled SMD −1.15, 95% CI −1.62 to −0.67 (Figure 2). When back transformed to a 0–100 mm VAS scale, this equates to a mean difference (MD) of 21 mm (95% CI −29.7 to −12.3 mm). No data could be located on the MCID in pain for individuals with frozen shoulder. However, it is suggested that a difference of 14 mm on a 100-mm VAS scale is the MCID for individuals with rotator cuff disease, a shoulder disorder also characterised by pain and restricted range of movement.50 Given that the between-group MCID is thought to approximate 40% of that within individuals,6,53 the MCID between groups for rotator cuff disorder can be estimated to be approximately 5.6 mm. Using this as a proxy for between-group MCID for frozen shoulder, this suggests that there is a clinically significant decrease in pain with steroid compared with placebo. There was no statistical heterogeneity between the studies (I2 = 0%). When 6-week data for Carette et al.35 and Ryans et al.41 were pooled the results were similar (pooled SMD −1.30, 95% CI −1.78 to −0.81; MD 23.8 mm, 95% CI 32.6 to 14.8 mm).

FIGURE 2. Standardised mean difference at up to 3 months: steroid vs placebo.

FIGURE 2

Standardised mean difference at up to 3 months: steroid vs placebo.

Bal et al.68 reported mean change in pain at 12 weeks. There was no significant difference in pain with steroid compared with placebo (Table 8).

Steroid versus physiotherapy combined with placebo

Two studies reported data at short-term follow-up: Carette et al.35 reported data at 6 weeks and 3 months and Ryans et al.41 reported data at 6 weeks (Table 8). In both studies physiotherapy was combined with placebo. When 3-month data from Carette et al.35 were pooled with 6-week data from Ryans et al.41 there was no significant difference in pain between steroid and physiotherapy (Figure 3). When back transformed to a 0–100 mm VAS scale, this equates to a MD of −4 mm (95% CI −11.9 to 3.7 mm), which suggests that there may be no clinically significant difference between treatments. There was no statistical heterogeneity between studies (I2 = 0%). When 6-week data for Carette et al.35 and Ryans et al.41 were pooled there was a marginally significant decrease in pain with steroid compared with physiotherapy combined with placebo (pooled SMD −0.51, 95% CI −0.94 to −0.07). However, when back transformed to a 0–100 mm VAS scale, this equates to a MD of −9.3 mm (95%–17.2 to −1.2 mm), which, as with the pooled 3-month and 6-week data, suggests that there may be no clinically significant difference between treatments.

FIGURE 3. Standardised mean difference at up to 3 months: steroid vs physiotherapy combined with placebo.

FIGURE 3

Standardised mean difference at up to 3 months: steroid vs physiotherapy combined with placebo.

Steroid combined with physiotherapy versus placebo

Two studies reported pain at short-term follow-up: Carette et al.35 reported data at 6 weeks and 3 months and Ryans et al.41 reported data at 6 weeks (Table 8). When 3-month data from Carette et al.35 were pooled with 6-week data from Ryans et al.41 there was a statistically significant decrease in pain with steroid combined with physiotherapy compared with placebo: pooled SMD −1.88, 95% CI −2.43 to −1.33 (Figure 4). When back transformed to a 0–100 mm VAS scale, this equates to a MD of −34.40 mm (95% CI −44.47 to −24.34 mm). This suggests that there is a clinically significant decrease in pain with steroid combined with physiotherapy compared with placebo. There was no statistical heterogeneity between studies (I2 = 3%). When 6-week data for Carette et al.35 and Ryans et al.41 were pooled the results were similar (pooled SMD −2.31, 95% CI −2.89 to −1.72; MD −42.3 mm, 95% CI −52.9 to −31.5 mm).

FIGURE 4. Standardised mean difference at up to 3 months: steroid combined with physiotherapy vs placebo.

FIGURE 4

Standardised mean difference at up to 3 months: steroid combined with physiotherapy vs placebo.

Steroid combined with physiotherapy versus physical therapy combined with placebo

Three studies reported pain for short-term follow-up. In all studies physiotherapy was combined with a placebo injection. Carette et al.35 reported pain at 6 weeks and 3 months, Rizk et al.42 reported pain at 11 weeks and Ryans et al.41 reported pain at 6 weeks (Table 8).

When 3-month data from Carette et al.35 were pooled with 11-week data from Rizk et al.42 and 6-week data from Ryans et al.41 there was a significant decrease in pain with steroid combined with physiotherapy compared with physiotherapy: pooled SMD −0.57, 95% CI −0.94 to −0.20 (Figure 5). However, there was evidence of substantial statistical heterogeneity (I2 = 79%). When 6-week data for Carette et al.35 were pooled with 11-week data from Rizk et al.42 and 6-week data from Ryans et al.,41 the results were similar (pooled SMD −0.86, 95% CI −1.25 to −0.47, I2 = 89%). Potential sources of heterogeneity may be the type of steroid used, different lengths of follow-up and proportion of patients with diabetes. Furthermore, the studies by Carette et al.35 and Ryans et al.41 were assessed to have a potentially lower risk of bias than the study by Rizk et al.42

FIGURE 5. Standardised mean difference at up to 3 months: steroid combined with physiotherapy vs physiotherapy combined with placebo.

FIGURE 5

Standardised mean difference at up to 3 months: steroid combined with physiotherapy vs physiotherapy combined with placebo.

Given the differences in study quality, a sensitivity analysis was performed by excluding Rizk et al.42 from the meta-analyses (Figure 6). Pooling 3-month data from Carette et al.35 and 6-week data from Ryans et al.41 resulted in a significant decrease in pain with steroid combined with physiotherapy compared with physiotherapy: pooled SMD −0.98, 95% CI −1.43 to −0.52. Furthermore, there was no longer evidence of statistical heterogeneity (I2 = 0%). When back transformed to a 0–100 mm VAS, the pooled SMD equates to a MD of −17.93 mm (95% CI −26.2 to −9.5 mm), which suggests that there is also a clinically significant decrease in pain with steroid combined with physiotherapy compared with physiotherapy. When 6-week data for Carette et al.35 and Ryans et al.41 were pooled the results were similar (pooled SMD −1.51, 95% CI −2.00 to −1.02; MD 27.6 mm, 95% CI 36.6 to 18.7 mm).

FIGURE 6. Standardised mean difference at up to 3 months (sensitivity analysis): steroid combined with physiotherapy vs physiotherapy combined with placebo.

FIGURE 6

Standardised mean difference at up to 3 months (sensitivity analysis): steroid combined with physiotherapy vs physiotherapy combined with placebo.

The study by Rizk et al.42 also reported pain for medium-term follow-up. At 24 weeks there was no significant difference between steroid combined with physiotherapy and physiotherapy (see Table 8).

Steroid combined with physiotherapy versus steroid

Carette et al.35 reported data at 6 weeks and 3 months and Ryans et al.41 reported data at 6 weeks (Table 8). When 3-month data from Carette et al.35 were pooled with 6-week data from Ryans et al.41 there was a significant difference in pain with steroid combined with physiotherapy compared with steroid alone: pooled SMD −0.75, 95% CI −1.20 to −0.29 (Figure 7). When back transformed to a 0–100 mm VAS, this equates to a MD of −13.7 mm (95% CI −21.96 to −5.3 mm), which suggests that there may be no clinically significant difference between treatments. However, there was evidence of substantial statistical heterogeneity (I2 = 70%); therefore, this result should be interpreted with caution. When 6-week data for Carette et al.35 and Ryans et al.41 were pooled, there was a significant decrease in pain with steroid combined with physiotherapy compared with steroid alone (pooled SMD −1.00, 95% CI −1.47 to −0.53). When back transformed to a 0–100 mm VAS, this equates to a MD of −18 mm (95% CI 26.9 to 9.7 mm). This suggests that there may be a clinically significant reduction in pain with steroid combined with physiotherapy compared with steroid alone. Furthermore, there was no evidence of statistical heterogeneity (I2 = 0%).

FIGURE 7. Standardised mean difference at up to 3 months: steroid combined with physiotherapy vs steroid.

FIGURE 7

Standardised mean difference at up to 3 months: steroid combined with physiotherapy vs steroid.

Function and disability

Four of the six studies assessed function and disability; however, only the Shoulder Pain and Disability Index (SPADI) total score and SPADI 8-item disability subscales were common to more than one study and follow-up times varied (Table 9). Results were presented as change from baseline data for all but one of the studies.66 Table 10 provides the between-group difference in means and 95% CIs. The outcome data for the individual groups in the included trials are available in Appendix 7.

TABLE 9. Function and disability outcomes in studies evaluating steroid injection.

TABLE 9

Function and disability outcomes in studies evaluating steroid injection.

TABLE 10. Function and disability results of studies of steroid injections (MD and 95% CI).

TABLE 10

Function and disability results of studies of steroid injections (MD and 95% CI).

Steroid versus placebo

Two studies reported SPADI total score and SPADI 8-item subscale score for steroid versus placebo injection (Table 10).35,68

Carette et al.35 reported SPADI total score at 6 weeks and found a significant effect in favour of the steroid injection group (MD −17.8, 95% CI −31.96 to −3.64).

Both studies reported 3-month follow-up and were pooled at this time point, although it should be noted that the study by Bal et al.68 was at high risk of bias. When the two studies were pooled (Figure 8) statistical heterogeneity was high (I2 = 80%); therefore, the treatment effect should be treated with considerable caution. The study by Carette et al.35 reported a significantly greater decrease in SPADI total score in the steroid group than in the placebo and home exercise group (MD −16.10, 95% CI −30.26 to −1.94), whereas in the Bal et al. study68 the treatment effect was in favour of placebo and home exercise, although the CIs crossed the line of no effect. Only Carette et al.35 reported SPADI total score for medium- and long-term follow-up; there was no significant difference between groups for these later follow-up times (Table 10).

FIGURE 8. Shoulder Pain and Disability Index total score at 3 months: steroid vs placebo.

FIGURE 8

Shoulder Pain and Disability Index total score at 3 months: steroid vs placebo.

Carette et al.35 reported SPADI 8-item disability subscale score at 6 weeks and found a significant effect in favour of the steroid injection group (MD −13.80, 95% CI −27.53 to −0.07).

As with SPADI total score, the same two studies reported SPADI 8-item disability subscale score and were available for meta-analysis at 3 months.35,68 Again, the pooled treatment effect should be treated with considerable caution because of high heterogeneity (Figure 9). Carette et al.35 reported a significantly greater decrease in SPADI 8-item subscale score in the steroid group than with placebo and home exercise (MD −14.30, 95% CI −28.03 to −0.57), whereas in the Bal et al. study68 the treatment effect was in favour of placebo and home exercise, although the CIs crossed the line of no effect.

FIGURE 9. Shoulder Pain and Disability Index 8-item subscale score at 3 months: steroid vs placebo.

FIGURE 9

Shoulder Pain and Disability Index 8-item subscale score at 3 months: steroid vs placebo.

Only Carette et al.35 reported SPADI 8-item subscale score for medium- and long-term follow-up; there was no significant difference between groups for this outcome (Table 10).

Ryans et al.41 reported results of the Shoulder Disability Questionnaire (SDQ) and global function at 6 and 16 weeks. There was no significant difference between the steroid group and the placebo group at either time point for either outcome (Table 10). Bal et al.68 reported University of California Los Angeles (UCLA) Shoulder score with steroid injection versus placebo and both groups were similar at the 3-month follow-up [steroid group median 32.5, interquartile range (IQR) 6.2; placebo group median 31.5, IQR 7.7; no p-value reported].

Steroid versus home exercise

Calis et al.66 reported Constant score for short-term follow-up; steroid injection was associated with a significantly better outcome than home exercise alone (MD 9.10, 95% CI 2.98 to 15.22).

Steroid versus physiotherapy

Calis et al.66 compared steroid with physiotherapy (without placebo injection) using the Constant score. At 3 months there was no significant difference between the two treatment groups (MD −5.80, 95% CI −11.80 to 0.20).

Steroid versus physiotherapy combined with placebo

Two studies compared steroid with physiotherapy plus placebo injection.35,41 One study reported SPADI total score35 and the other the SDQ and a global function score.41

There was a significantly greater improvement in SPADI total score in the steroid group than in the physiotherapy group at 6 weeks in the study by Carette et al.35 (MD −14.50, 95% CI −28.25 to −0.75), but not at 3 months or for medium- and long-term follow-up or in the SPADI 8-item disability subscale score between treatment groups at any follow-up time in the same study (Table 10).35 There was also no significant difference between treatment groups in either SDQ or global function score at 6 or 16 weeks in the study by Ryans et al.41

Steroid combined with physiotherapy versus placebo

Two studies investigated steroid combined with physiotherapy versus placebo.35,41 One study reported SPADI total score35 and the other the SDQ and a global function score.41

There was a significant improvement in total SPADI score with steroid in combination with physiotherapy compared with placebo in the study by Carette et al.35 at 6 weeks (MD −27.60, 95% CI −42.03 to −13.17) and 3 months (MD −21.00, 95% CI −35.43 to −6.57), although not at medium- (MD −14.10, 95% CI −28.53 to 0.33) or long-term follow-up (MD −1.10, 95% CI −15.53 to 13.33) (Table 10).35 There was also a significant improvement in SPADI 8-item disability subscale score with steroid in combination with physiotherapy compared with placebo in the same study at 6 weeks (MD −23.90, 95% CI −37.89 to −9.91) and 3 months (MD −20.00, 95% CI −33.99 to −6.01) but not at medium- and long-term follow-up.

There was a significant improvement in both SDQ (MD −4.70, 95% CI −7.85 to −1.55) and global function score (MD −21.00, 95% CI −34.54 to −7.46) with steroid in combination with physiotherapy compared with placebo at 6 weeks in the study by Ryans et al.41 At 16 weeks there was no significant difference between groups in this study (Table 10).

Steroid combined with physiotherapy versus physiotherapy combined with placebo

Two studies investigated the effect of steroid combined with physiotherapy versus physiotherapy (plus placebo injection) on function and disability using SPADI total score35 and the SDQ and a global function score.41

There was a significant improvement in total SPADI score with steroid in combination with physiotherapy compared with physiotherapy in the study by Carette et al.35 at 6 weeks (MD −24.30, 95% CI −38.33 to −10.27), although not at 3 months (MD −12.50, 95% CI −26.53 to 1.53) or for medium- and long-term follow-up (Table 10).35 Similarly, there was a significant improvement in SPADI 8-item disability subscale score with steroid in combination with physiotherapy compared with physiotherapy in the same study at 6 weeks (MD −21.60, 95% CI −35.34 to −7.86) but not at any of the other follow-up points.35

Ryans et al.41 also showed a significant improvement in SDQ score with steroid combined with physiotherapy compared with physiotherapy at 6 weeks (MD −4.30, 95% CI −7.76 to −0.84) but not at 16 weeks, and there was no significant difference in global function score at either follow-up time.

Steroid combined with physiotherapy versus steroid

Two studies compared steroid injection followed by physiotherapy with steroid alone. One study reported a SPADI total score and SPADI 8-item disability subscale score;35 another study reported SDQ and global function score.41

There was no significant difference between treatment groups in SPADI total or SPADI 8-item disability subscale score at short-, medium- or long-term follow-up in the study by Carette et al.,35 or between treatment groups in either SDQ or global function for short- or medium-term follow-up in the study by Ryans et al. (Table 10).41

Range of movement

The range of movement measurements of interest in the review were external rotation, internal rotation and abduction (both passive and active); if internal rotation was not available but hand behind back was reported, this was used as a proxy measure. Five studies reported at least one of these measures (Table 11). Change from baseline data was available from three of these studies35,41,68 and final value data was available from one study.66 A fifth study, by Rizk et al.,42 reported only means with no measures of variance or p-values with which to impute SDs. Table 12 provides the between-group differences in means and 95% CIs. The outcome data for the individual groups in the included trials are available in Appendix 7.

TABLE 11. Range of movement (ROM) outcomes in studies evaluating steroid injection.

TABLE 11

Range of movement (ROM) outcomes in studies evaluating steroid injection.

TABLE 12. Range of movement results of studies of steroid injections (MD and 95% CI).

TABLE 12

Range of movement results of studies of steroid injections (MD and 95% CI).

Steroid versus placebo

One study by Bal et al.68 reported passive internal rotation at 12 weeks. There was a significantly greater improvement in the placebo group compared with steroid injection (MD −28.50°, 95% CI −41.56° to −15.44°). This result contradicts the original analysis in the paper, which reported that there was no statistically significant difference between the two groups, using the Mann–Whitney U-test.

Three studies reported passive external rotation at short-term follow-up, two at 6 weeks35,41 and two at 3 months.35,68 Improvement in passive external rotation was significantly greater with steroid injection than with placebo injection at 6 weeks (pooled MD 9.48°, 95% CI 2.76° to 16.19°; two RCTs) (Figure 10). There was no significant heterogeneity between these studies (I2 = 0%)

FIGURE 10. Passive external rotation at 6 weeks: steroid vs placebo.

FIGURE 10

Passive external rotation at 6 weeks: steroid vs placebo.

Meta-analysis of the two studies that reported passive external rotation at 3 months showed no significant difference between the steroid and the placebo groups (Figure 11)35,68 This pooled result was associated with substantial heterogeneity (F = 80%) and should therefore be treated with caution. The study by Carette et al.35 showed a significant improvement in passive external rotation in the steroid group compared with placebo whereas the study by Bal et al.68 showed no significant difference between groups (Table 11). There were several possible sources of heterogeneity, in particular differences in the risk of bias.

FIGURE 11. Passive external rotation at 3 months: steroid vs placebo.

FIGURE 11

Passive external rotation at 3 months: steroid vs placebo.

Two RCTs reported passive external rotation for medium-term follow-up: 16 weeks41 and 6 months.35,41 There was no significant difference between steroid and placebo groups in passive external rotation (MD 4.67°, 95% CI −2.21° to 11.55°) (Figure 12). There was some evidence of moderate statistical heterogeneity (I2 = 43%). Both studies appeared clinically similar although the follow-up time points differed. Carette et al.35 was the only study to examine passive external rotation at long-term follow-up and found no significant difference between steroid and placebo groups (Table 12).35

FIGURE 12. Passive external rotation at up to 6 months: steroid vs placebo.

FIGURE 12

Passive external rotation at up to 6 months: steroid vs placebo.

Carette et al35 reported passive abduction and hand behind back at 6 weeks and 3, 6 and 12 months. There was no significant difference between groups at any of the time points for either outcome (Table 11).

Steroid versus home exercise

Calis et al.66 reported passive abduction and passive external rotation at 3 months. There was a significantly greater improvement in the steroid group compared with home exercise alone in both passive abduction (MD 16.80°, 95% CI 6.60° to 27.00°) and passive external rotation (MD 8.00°, 95% CI 2.48° to 13.52°).

Steroid versus physiotherapy

Calis et al.66 reported passive external rotation and passive abduction at 3 months. There was significantly greater passive external rotation at 3 months with physiotherapy than with steroid (MD −10.80°, 95% CI −16.94° to −4.66°). There was no significant difference between treatment groups for passive abduction (Table 12).

Steroid versus physiotherapy combined with placebo

Three studies reported passive external rotation at time points up to 3 months.35,41,66 Two of the studies involved physiotherapy sessions over 12 weeks and in both physiotherapy groups a placebo injection was also given.35,41 Both studies reported data at 6 weeks and one35 reported data at 3 months; 6-week data were pooled. There was no significant difference in passive external rotation between groups at 6 weeks (Figure 13). There was substantial statistical heterogeneity between the studies (I2 = 64%).

FIGURE 13. Passive external rotation at 6 weeks: steroid vs physiotherapy combined with placebo.

FIGURE 13

Passive external rotation at 6 weeks: steroid vs physiotherapy combined with placebo.

Carette et al.35 reported passive external rotation, and passive abduction at 3 months. There was no significant difference between treatment groups for both outcomes (Table 12).

Medium-term follow-up data were available for the studies by Carette et al.35 and Ryans et al.41 Carette et al.35 reported passive external rotation at 6 months and Ryans et al.41 at 16 weeks. There was no significant difference between groups (Figure 14). There was no significant difference between groups in passive external rotation at 12 months in the Carette et al. study.35 Similarly, the same study showed no significant difference between treatment groups in passive abduction or hand behind back at medium- or long-term follow-up (Table 12).

FIGURE 14. Passive external rotation at up to 6 months: steroid vs physiotherapy combined with placebo.

FIGURE 14

Passive external rotation at up to 6 months: steroid vs physiotherapy combined with placebo.

Steroid combined with physiotherapy versus placebo

Two studies reported passive external rotation at time points up to 3 months.35,41 Both studies reported data at 6 weeks and one35 reported data at 3 months. Data at 6 weeks were pooled. The meta-analysis (Figure 15) indicated that passive external rotation was significantly greater with steroid injection plus physiotherapy than with placebo injection at 6 weeks (pooled MD 16.99°, 95% CI 10.01° to 23.97°).

FIGURE 15. Passive external rotation at 6 weeks: steroid combined with physiotherapy vs placebo.

FIGURE 15

Passive external rotation at 6 weeks: steroid combined with physiotherapy vs placebo.

Passive external rotation was significantly greater with steroid injection plus physiotherapy than with placebo injection at 3 months (MD 17.60°, 95% CI 7.90° to 27.30°) in the study by Carette et al.35

Medium-term follow-up data were also available up to 6 months: Carette et al.35 reported passive external rotation at 6 months and Ryans et al.41 at 16 weeks. The pooled estimate showed that passive external rotation was significantly greater with steroid injection plus physiotherapy than with placebo injection at medium-term follow-up (pooled MD 8.14°, 95% CI 0.18° to 16.09°); however, there was substantial statistical heterogeneity (I2 = 70%) (Figure 16). There were a number of possible sources of heterogeneity including the dose of steroid used, different length of follow-up and the proportion of participants with diabetes. There was no significant difference between groups at 12 months in the single study reporting long-term follow-up (Table 12).35

FIGURE 16. Passive external rotation at up to 6 months: steroid combined with physiotherapy vs placebo.

FIGURE 16

Passive external rotation at up to 6 months: steroid combined with physiotherapy vs placebo.

Carette et al.35 reported passive abduction at 6 weeks and 3, 6 and 12 months. Improvement of passive abduction was significantly greater in the steroid plus physiotherapy group than with placebo injection at 6 weeks (MD 15.30°, 95% CI 7.14° to 23.46°), 3 months (MD 16.30°, 95% CI 8.14° to 24.46°) and 6 months (MD 13.00°, 95% CI 4.84° to 21.16°). There was no significant difference between treatment groups at 12 months (Table 12).

In the same study there was a significantly greater improvement in hand behind back for the steroid with physiotherapy group than with placebo injection at 6 weeks (MD −9.00 cm, 95% CI −15.80 cm to −2.20 cm) and 3 months (MD −7.10 cm, 95% CI −13.90 cm to −0.30 cm), but not at any of the other time points (Table 12).

Steroid combined with physiotherapy versus physiotherapy combined with placebo

Three studies reported passive external rotation at time points up to 3 months.35,41,42 All studies reported data at 6 weeks35,41,42 and two also reported data at 3 months.35,42 Because of SDs not being available for one study,42 data were pooled from two of the studies for 6-week follow-up.35,41 The increase in passive external rotation was significantly greater with steroid injection plus physiotherapy than with physiotherapy (plus placebo injection) at 6 weeks (pooled MD 10.80°, 95% CI 4.02° to 17.58°) (Figure 17). There was moderate to high heterogeneity between the studies (I2 = 70%); the results of this analysis should therefore be considered with caution.

FIGURE 17. Passive external rotation at 6 weeks: steroid combined with physiotherapy vs physiotherapy combined with placebo.

FIGURE 17

Passive external rotation at 6 weeks: steroid combined with physiotherapy vs physiotherapy combined with placebo.

The study by Carette et al.35 also showed that the increase in passive external rotation was significantly greater with steroid injection plus physiotherapy than with physiotherapy at 3 months (MD 13.00°, 95% CI 3.56° to 22.44°).

Medium-term data were also available from Carette et al.35 and Ryans et al.41 The pooled estimate (Figure 18) showed that passive external rotation was significantly greater with steroid injection plus physiotherapy than with physiotherapy at up to 6 months (pooled MD 7.36°, 95% CI 0.04° to 14.68°). The study by Carette et al.35 showed no significant difference between groups at 12 months (Table 12).

FIGURE 18. Passive external rotation at up to 6 months: steroid combined with physiotherapy versus physical therapy combined with placebo.

FIGURE 18

Passive external rotation at up to 6 months: steroid combined with physiotherapy versus physical therapy combined with placebo.

Carette et al.35 reported passive abduction at 6 weeks and 3, 6 and 12 months. Improvement of passive abduction was significantly greater in the steroid plus physiotherapy group than with physiotherapy at 6 weeks (MD 11.30°, 95% CI 3.40° to 19.20°), 3 months (MD 10.40°, 95% CI 2.50° to 18.30°) and 6 months (MD 7.90°, 95% CI −0.00° to 15.80°). There was no significant difference between treatment groups at 12 months (Table 12).

Steroid versus steroid and physiotherapy

Two studies reported passive external rotation for short-term follow-up;35,41 both studies reported data at 6 weeks and one35 reported data at 3 months; the 6-week data were pooled. Passive external rotation was significantly greater with steroid injection plus physiotherapy than with steroid alone at 6 weeks (pooled MD −7.52°, 95% CI −14.70° to −0.34°). This analysis was not associated with statistical heterogeneity (I2 = 0%) (Figure 19). The study by Carette et al.35 showed no significant difference between treatment groups at 3 months (Table 12).

FIGURE 19. Passive external rotation at 6 weeks: steroid versus steroid and physiotherapy.

FIGURE 19

Passive external rotation at 6 weeks: steroid versus steroid and physiotherapy.

Data were also available for medium-term follow-up for the same two studies. The pooled analysis showed no significant difference between groups and the analysis was not associated with statistical heterogeneity (I2 = 0%) (Figure 20). The study by Carette et al.35 showed no significant difference between treatment groups at long-term follow-up (Table 12).

FIGURE 20. Passive external rotation at up to 6 months: steroid versus steroid and physiotherapy.

FIGURE 20

Passive external rotation at up to 6 months: steroid versus steroid and physiotherapy.

Carette et al.35 reported passive abduction at 6 weeks and 3, 6 and 12 months. At 6 weeks improvement of passive abduction was significantly greater in the steroid plus physiotherapy group than with steroid alone (MD −8.10, 95% CI −16.26 to 0.06); there were no significant differences between treatment groups at later follow-up times (Table 12).

Similarly, for hand behind back, there was a significantly better improvement in the steroid with physiotherapy group than with steroid alone at 6 weeks (MD 8.50, 95% CI 1.70 to 15.30) but not at later follow-up (Table 12).

Quality of life

Only Carette et al.35 reported quality of life data following treatment (Table 13).35 Table 14 provides between-group differences in means and associated 95% CIs. Further outcome data can be found in Appendix 7.

TABLE 13. Quality of life outcomes in studies that evaluated steroid injection.

TABLE 13

Quality of life outcomes in studies that evaluated steroid injection.

TABLE 14. Quality of life results of studies of steroid injections (MD and 95% CI).

TABLE 14

Quality of life results of studies of steroid injections (MD and 95% CI).

There was a statistically significant improvement in the physical component of the SF-36 with steroid plus physiotherapy compared with placebo at 6 months (MD 7.10, 95% CI 0.86 to 13.34). There was no statistically significant difference between these two groups at any of the other time points evaluated.

There was a statistically significant improvement with steroid plus physiotherapy compared with physiotherapy (plus placebo injection) in the physical component at 6 weeks (MD 5.30, 95% CI 0.03 to 10.57) and in the mental component at 6 months (MD 8.20, 95% CI 1.98 to 14.42) and 12 months (MD 6.90, 95% CI 0.68 to 13.12). There was no statistically significant difference between these two treatments groups at any of the other time points evaluated.

For the comparisons steroid compared with placebo, steroid compared with physiotherapy (plus placebo injection) and steroid compared with physiotherapy plus steroid, the study by Carette et al.35 showed no statistically significant difference between treatment groups at any of the time points evaluated.

Adverse events

Three of the studies that investigated steroids reported adverse events.42,67,68 There was a limited amount of information reported regarding adverse events in the included studies (Table 15). None of the studies reported any details regarding how adverse event data were collected or recorded; therefore, the data should be treated with caution. Two reported that no adverse effects were reported in any of the treatment groups67,68 and one reported that there were no withdrawals due to adverse effects.42

TABLE 15. Adverse events in studies evaluating steroid injections.

TABLE 15

Adverse events in studies evaluating steroid injections.

Summary

The six included steroid injection studies were diverse in terms of the intervention: two used a single steroid injection of 20 mg, three a single injection of 40 mg and one used three injections of 40 mg administered over 3 weeks. Five of the studies reported using an intra-articular injection. There was also some variation between studies in the physiotherapy provided, although most of the studies used a programme of < 6 weeks' duration that included mobilisation and home exercise for all groups. Five of the studies were multi-armed trials. Steroid injection was compared with placebo in three studies, home exercise in one study and physiotherapy alone in four studies (with or without placebo). The effect of steroid injection followed by physiotherapy was also assessed. This combined intervention was compared with placebo, steroid alone and physiotherapy (with placebo) alone.

Information about previous treatments that participants had received and stage of frozen shoulder was limited. Outcomes were not reported by stage of frozen shoulder or by whether or not participants had diabetes. As a result, it was not possible to explore in the synthesis the effect of these variables on outcome. It was unclear whether four of the studies were truly randomised and therefore these studies had a potentially high risk of bias.42,6668 The fifth study was of adequate quality,35 whereas the sixth was of reasonable quality but may have a risk of bias.41

For most of the comparisons, data were available from only one or two studies, which may have been underpowered to detect an effect. There were sufficient data to pool the two best-quality studies for some of the outcomes, although in some instances statistical heterogeneity was high. In addition, although both studies used a single intra-articular injection, one evaluated a guided injection of 40 mg of steroid35 and the other used a 20-mg dose and the injection appeared to be unguided.41 For most outcomes and comparisons it was necessary to undertake a narrative synthesis.

Steroid alone

There was evidence of short-term benefit with steroid injection compared with placebo for pain, function and disability, and range of movement but not quality of life. The evidence on function and disability was mixed and this may reflect differences in study quality and/or clinical differences. There was also evidence of short-term benefit with steroid injection compared with home exercise alone for function and disability and two measures of range of movement from one study with a potentially high risk of bias. There was no consistent evidence across outcomes for a benefit with steroid injection compared with physical therapy.

Steroid followed by physiotherapy

There was consistent evidence across all outcomes (pain, function and disability, range of movement and quality of life) of a short-term benefit with steroid injection followed by physiotherapy compared with placebo or physiotherapy; however, this was based on a small number of studies. There was also evidence of better quality of life with the combined treatment in the medium term. When steroid injection plus physiotherapy was compared with steroid alone there was a short-term beneficial reduction in pain and a better range of movement at the 6-week follow-up with the combined intervention but not at 3 months or later and there was no benefit for any other outcomes.

In conclusion, a small number of diverse studies were identified, four of which had a potentially high risk of bias. Based on the best available evidence, there was a benefit, mainly short term, with a single steroid. There was an added benefit with providing physiotherapy in addition to steroid injection compared with home exercise alone and physiotherapy alone for several outcomes. There was also benefit with the combined intervention compared with steroid alone for pain and for range of movement at 6 weeks but not the other outcomes. There was insufficient evidence to conclude with reasonable certainty in what clinical situations steroid injection, with or without physiotherapy, is most likely to be effective for primary frozen shoulder.

Sodium hyaluronate

Study characteristice

Three RCTs investigated sodium hyaluronate in the treatment of primary frozen shoulder. A summary of the study characteristics is reported in Table 16. All were full papers of studies conducted in Turkey,66 Italy69 and Japan70 between 1996 and 2006. There were two two-armed trials69,70 and one four-armed trial.66

TABLE 16. Studies of sodium hyaluronate with or without physical therapy.

TABLE 16

Studies of sodium hyaluronate with or without physical therapy.

Calis et al.66 used a 30-mg injection once weekly for 2 weeks, Takagishi et al.70 used 2 mg injected once weekly for 5 weeks and Rovetta et al.69 used 20 mg of sodium hyaluronate coadministered with 20 mg of steroid (triamcinolone acetonide) injected at 15-day intervals for the first month then monthly for 6 months and physical therapy (mobilisation or exercises) for 4–12 weeks. The comparators were steroid injection,66,70 steroid injection and physiotherapy,69 physiotherapy66 and no intervention except for home exercise.66

Two of the three studies included physiotherapy.66,69 Both studies differed in the duration and content of the regimen, but each contained some form of active exercise or mobilisation. Calis et al.66 had a regimen of 10 daily sessions consisting of a heat pack applied for 20 minutes, ultrasonic therapy for 5 minutes (1.5 W/cm2 intensity), TENS for 20 minutes at the patient's level of tolerance, and stretching exercises. In comparison, Rovetta et al.69 had a 4- to 12-week regimen of the most appropriate therapy for the individual (passive mobilisation, active exercises and facilitation exercises), which was chosen by a physiotherapist. Only Calis et al.66 reported that participants had a home exercise programme, which consisted of stretching and Codman exercises, but did not provide any further details.

Two studies reported that patients received concomitant treatments.69,70 Rovetta et al.69 allowed paracetamol (dose unspecified) to be taken if necessary, whereas Takagishi et al.70 stated that participants who were using topical NSAIDs prior to commencing the study were allowed to continue to do so. Calis et al.66 allowed paracetamol to be taken if necessary.

The inclusion criteria varied between the three studies. Rovetta et al.69 reported that none of the participants had diabetes, whereas Calis et al.66 and Takagishi et al.69 did not report whether or not any of the participants had diabetes. Frozen shoulder was diagnosed through physical examination, laboratory tests, subacromial impingement test, clinical history, radiography or ultrasound when described. The method of diagnosis was unclear in one study.70 The participants' stage of frozen shoulder at baseline was not reported in any of the studies and it was unclear whether participants had received any previous treatments. The mean age of participants ranged from 48 years to 65.8 years. The proportion of women ranged from 63% to 75%.

Quality assessment

None of the studies reported the method of randomisation or allocation concealment, and only one stated that blinded outcome assessment was performed; therefore, all three studies have a potentially high risk of bias. The study by Calis et al.66 was of higher quality and fulfilled five criteria: the number of participants randomised, treatment groups that were comparable at baseline, blinded outcome assessors, use of intention-to-treat analysis and no unexpected imbalances in dropouts. Rovetta et al.69 fulfilled two criteria, stating the number of participants randomised and also having treatment groups that were comparable at baseline, whereas Takagishi et al.70 fulfilled only one criterion, which was stating the number of patients randomised. Full details of study quality are reported in Appendix 8.

Pain

All three studies evaluating sodium hyaluronate assessed pain; however, the study by Calis et al.66 reported pain in graphical form and therefore could not be included in the analysis. Calis et al.66 did, however, report that there was significant improvement in pain severity at 3 months within each of the treatment groups, including the no intervention group (p < 0.001). Rovetta et al.69 and Takagishi et al.70 both used a VAS but assessed different types of pain at different time points (Table 17); therefore, it was inappropriate to pool the two studies. Table 18 provides the SMDs and 95% CIs. The outcome data for the individual groups in the included trials are available in Appendix 7.

TABLE 17. Pain outcomes in studies evaluating sodium hyaluronate.

TABLE 17

Pain outcomes in studies evaluating sodium hyaluronate.

TABLE 18. Pain results of studies of sodium hyaluronate (SMD and 95% CI).

TABLE 18

Pain results of studies of sodium hyaluronate (SMD and 95% CI).

Sodium hyaluronate versus steroid

Takagishi et al.70 found no significant difference in pain at short-term follow-up with sodium hyaluronate compared with steroid (see Table 18).

Sodium hyaluronate combined with steroid and physiotherapy versus steroid combined with physiotherapy

Rovetta et al.69 found a significant decrease in pain at 6 months with sodium hyaluronate combined with steroid and physiotherapy compared with steroid combined with physiotherapy (SMD −0.78, 95% CI −1.50 to −0.06).

Function and disability

Calis et al.66 and Takagishi et al.70 both assessed function and disability. Calis et al. assessed function and disability using the Constant score (Table 19), whereas Takagishi et al.70 used a Japanese-specific activities of daily living questionnaire. However, Takagishi et al. reported baseline values and a p-value for the within-group change from baseline only and therefore could not be included in the analysis. Table 20 provides the MD and 95% CIs. The outcome data for the individual groups in the included trial are available in Appendix 7.

TABLE 19. Function and disability outcomes in studies evaluating sodium hyaluronate.

TABLE 19

Function and disability outcomes in studies evaluating sodium hyaluronate.

TABLE 20. Function and disability results of sodium hyaluronate (MD and 95% CI).

TABLE 20

Function and disability results of sodium hyaluronate (MD and 95% CI).

At 3 months Calis et al.66 found a significantly greater improvement in the Constant score with sodium hyaluronate compared with no treatment (MD 8.90, 95% CI 2.62 to 15.18), but no significant difference in Constant score between sodium hyaluronate and physiotherapy (MD −6.00, 95% CI −12.16 to 0.16) or between sodium hyaluronate and steroid injection (MD −0.20, 95% CI −5.86 to 5.46).

Range of movement

The range of movement measurements of interest were external rotation, internal rotation and abduction (both passive and active); if internal rotation was not available but hand behind back was reported, this was used as a proxy measure of active internal rotation. All three studies that evaluated sodium hyaluronate reported at least one of these measurements (Table 21). Table 22 provides the between-group difference in means and 95% CIs. The outcome data for the individual groups in the included trials are available in Appendix 7.

TABLE 21. Range of movement (ROM) outcomes in studies evaluating sodium hyaluronate injections.

TABLE 21

Range of movement (ROM) outcomes in studies evaluating sodium hyaluronate injections.

TABLE 22. Range of movement results of studies of sodium hyaluronate injections (MD and 95% CI).

TABLE 22

Range of movement results of studies of sodium hyaluronate injections (MD and 95% CI).

Sodium hyaluronate versus no treatment

Calis et al.66 reported passive external rotation and passive abduction at 3 months. There was a significantly greater improvement in the sodium hyaluronate group than in the no treatment group in both passive external rotation (MD 8.30°, 95% CI 2.52° to 14.08°) and passive abduction (MD 12.40°, 95% CI 0.14° to 24.66°).

Sodium hyaluronate versus physiotherapy

Calis et al.66 reported passive external rotation and passive abduction at 3 months. There was significantly greater improvement in the physiotherapy group than in the sodium hyaluronate group in both passive external rotation (MD −10.50°, 95% CI −16.87° to −4.13°) and passive abduction (MD −12.50°, 95% CI −23.98° to −1.02°).

Sodium hyaluronate versus steroid

Calis et al.66 reported passive external rotation and passive abduction at 3 months. There was no significant difference in passive external rotation or passive abduction between groups (see Table 22). Takagishi et al.70 reported external rotation at 5 weeks. There was no significant difference between groups (see Table 22).

Sodium hyaluronate with steroid and physiotherapy versus steroid with physiotherapy

One study, by Rovetta et al.,69 reported internal rotation, external rotation and abduction at 6 months only. There was a significantly greater improvement in internal rotation with sodium hyaluronate combined with steroid and physiotherapy than with steroid combined with physiotherapy (MD 23.70°, 95% CI 13.44° to 33.96°). There was no significant difference in mean external rotation or mean abduction between groups (see Table 22).

Quality of life

None of the three studies that evaluated sodium hyaluronate assessed quality of life.

Adverse events

None of the three studies evaluating sodium hyaluronate reported whether or not participants experienced any adverse events.

Summary

The three included studies investigating sodium hyaluronate used variable doses and in one study sodium hyaluronate was combined with steroid injection and physiotherapy. Because of variations in outcomes reported and comparators used it was not possible to pool data in a metaanalysis. For most outcome measures data were available from a single study only. The studies did not report participants' stage of frozen shoulder or previous treatments received.

All of the studies had a potentially high risk of bias; none reported the method of randomisation and therefore it was unclear whether they were truly randomised. Because it was unclear whether all studies had enough participants to detect a statistically significant treatment effect, it should be kept in mind that where studies did not report a statistically significant benefit (i.e. the CI crossed the line of no difference) this should not be interpreted as evidence of no difference between the groups.

One study reported a medium-term benefit in pain outcome when sodium hyaluronate was added to steroid and physiotherapy. The same study also reported a significant benefit at medium-term follow-up for internal rotation but not for external rotation or abduction. Another study reported a benefit for function and disability with sodium hyaluronate compared with home exercise but not with physiotherapy or steroid. The same study reported significantly better short-term passive external rotation and abduction with sodium hyaluronate than with home exercise and physiotherapy but not steroid injection. A second study reported no evidence of a benefit for external rotation with sodium hyaluronate compared with steroid.

In conclusion, a small number of diverse studies of sodium hyaluronate were identified, all of which may have had a high risk of bias. There was insufficient evidence to make conclusions with any certainty about the effectiveness of sodium hyaluronate for primary frozen shoulder and in what situations it is likely to be effective.

Physical therapy

Study characteristics

Eleven RCTs and one controlled trial investigated various types of physical therapy in the treatment of primary frozen shoulder. A summary of the study characteristics is reported in Table 23, with further details available in Appendix 6. Eleven were full papers and one71 was available in abstract form only. One was a Chinese-language paper.72 The studies were published in the Netherlands,40,73 Turkey,51,66,74 Hong Kong,75 Singapore,76 Thailand,77 Greece,16 Taiwan,78 China72 and the UK71 between 1994 and 2009. There were nine two-armed trials,16,40,51,7274,7678 two three-armed trials71,75 and one four-armed trial.66

TABLE 23. Studies including physical therapies.

TABLE 23

Studies including physical therapies.

A variety of physical therapy regimens were used, most of which were physiotherapy modalities. These included various combinations of exercises (up to and beyond the pain threshold), stretching, manipulation of the glenohumeral joint, ultrasound, superficial heat, SWD, laser therapy, Codman's exercises, wall-climbing exercises, continuous passive motion, manual therapy, dumb-bell gymnastics and massage. For the purposes of the synthesis, the interventions were grouped based on whether or not at least one component involved mobilisation (of any type) under the supervision of a therapist. Therapies that involved some form of mobilisation (with or without other physical therapies) are described as active therapies. Studies using interventions such as laser therapy or TENS without any supervised mobilisation or exercise were classified as therapy without mobilisation.

Seven compared at least two forms of physical therapy.40,71,72,7476,78 One of the studies had what was described as a ‘supervised neglect’ intervention.73 This study by Diercks and Stevens was the only study identified that explicitly used a supervised neglect intervention; therefore, this study was included although it was not a RCT.73 Undertaken in the Netherlands, a physiotherapy intervention of unspecified duration was used that involved a standardised protocol, carried out by a therapist, of active exercises up to and beyond the pain threshold. The supervised neglect comparator involved providing patients with an explanation of the natural course of the disease and instructions not to exercise in excess of their pain threshold, to undertake pendulum exercises and active exercises within the painless range and to resume all activities that were tolerated. In the Calis et al. study, physiotherapy was compared with sodium hyaluronate and steroid injection;66 and in the Pajareya et al. study, physiotherapy was compared with no intervention.77

A single study did not have an arm with an active mobilisation component. This study by Stergioulas16 compared laser therapy with placebo laser (with home exercise). The remaining studies compared multiple active physical therapies. One compared ultrasound and physiotherapy with sham ultrasound and physiotherapy;51 one compared 1-hour sessions of continuous passive motion with physiotherapy;74 one compared osteopathy with physiotherapy and with breathing exercises, massage and range of movement exercises;71 one compared end-range mobilisation (ERM) and mid-range mobilisation (MRM) with mobilisation with movement (MWM) and MRM;78 one compared passive joint mobilisations and exercises with exercises only;76 one compared SWD plus stretching with heat pack plus stretching;75 one compared dumb-bell gymnastics with barehanded exercises;72 and one compared high-grade mobilisation techniques (HGMT) with low-grade mobilisation techniques (LGMT) (both followed by Codman pendular exercises).40 All of the physiotherapy arms included an active mobilisation component. The intervention duration ranged from 2 weeks to 3 months.

Six studies reported that home exercise was used as part of the intervention, although details were limited in some studies. Where specified the home exercise routines included daily Codman and stretching exercises,66 Codman, stretching and active range of movement exercises,51 daily passive range of movement and pendulum exercises,74 pendulum and pain-free exercises16 and daily stretches.75 Home exercise was discouraged in one study.78 Analgesic use also differed between studies where reported. NSAIDs or other analgesics were prescribed when necessary in one study,73 paracetamol was taken when necessary in another,66 paracetamol was allowed to a maximum of 1000 mg daily in another,51 400 mg of ibuprofen was given daily for 3 weeks in one study77 and prescribed and non-prescribed pain medication was allowed in one study.40 No NSAIDs or other analgesics were allowed in one RCT, with a week washout period before the study commenced.74

The inclusion criteria varied between the studies. All included patients with some limitation of movement, although the extent of this varied where reported. Primary frozen shoulder was diagnosed through range of movement, biochemical analyses, radiography and clinical diagnosis where reported. Four studies did not report how frozen shoulder was diagnosed.40,71,77,78 Three included participants with diabetes: 37% in the study by Dogru et al.,51 16% in the study by Vermeulen et al.40 and 33% in the study by Pajareya et al.77 Separate data were not presented for this subgroup. The mean duration of frozen shoulder at baseline ranged from 3 months to 6.8 years. The stage of frozen shoulder included was reported by few studies: stage 1 or 2 in the study by Dundar et al.,74 ‘stiff stage’ in the studies by Yan72 and Leung and Cheing75 and late stage 2 or 3 in the study by Maricar and Chok.76 Outcome was not reported separately by stage of frozen shoulder. Only one study, Vermeulen et al.,40 reported on whether participants had received any previous treatment for frozen shoulder. In this study 81% of participants had previously received physical therapy, 61% steroid injections and 6% surgery. The proportion of female participants ranged from 20% to 86% and the mean age of participants ranged from 50 to 62.5 years.

Quality assessment

The quality of the studies was generally poor. Only one was of satisfactory quality, reporting the number randomised and both an appropriate method of randomisation and adequate allocation concealment.78 This study by Yang et al. also met most of the remaining criteria including comparability at baseline, blinding of outcome assessors and intention-to-treat analysis; however, the study may not be adequately powered to detect significant differences between groups and there was a large number of dropouts (23%). The main limitation of this study was the study design for the purposes of our review; only limited data were suitable for inclusion in our analysis as different sequences of the three same mobilisations were compared at end of treatment (see Appendix 6).78

Leung and Cheing,75 Stergioulas16 and Vermeulen et al.40 reported an appropriate method of randomisation; Leung and Cheing75 and Vermeulen et al.40 did not report whether an adequate method of allocation concealment was used; and Stergioulas16 reported that allocation was performed using opaque envelopes but it was unclear whether the envelopes were numbered sequentially. Although these three studies fulfilled other criteria, such as comparability at baseline and blinding of outcome assessors, there was still a possible risk of bias. Leung and Cheing75 reported that there was no loss to follow-up and loss to follow-up was 15% in the Stergioulas study16 (the majority in the laser group) and 4% in the Vermeulen study40 (with equal numbers from each group). Vermeulen et al.40 also used intention-to-treat analysis and was adequately powered for one outcome.

Calis et al.,66 Dogru et al.,51 Dundar et al.,74 Pajareya et al.77 and Yan72 reported the number of participants randomised. It was unclear whether the method of assignment was truly random in any of these studies and allocation concealment was not reported; therefore, they potentially have high risk of bias. These studies also fulfilled few of the remaining criteria. Dogru et al.51 and Calis et al.66 reported blinding of outcome assessment. Dropouts were relatively high in the Pajareya study77 (12%, the majority in the control group) and the reporting of satisfaction results was also unclear. There were no or very few dropouts in the Calis et al.,66 Dogru et al.,51 Yan72 and Dundar et al.74 studies. Calis et al.66 also used intention-to-treat analysis.

The number of participants randomised was inconsistently reported in the study by Maricar and Chok.76 Neither Maricar and Chok76 or Wies et al.71 reported the method of randomisation or allocation concealment, so were at high risk of bias. Few of the remaining criteria were met. Dropouts were high in the study by Maricar and Chok (41%)76 and not reported by Wies et al.71

Allocation to groups was not randomised in the study by Diercks and Stevens73 and met none of the quality criteria (including blinding of outcome assessment) except comparability at baseline (there were no significant differences in age, sex or duration of disease). This study was quasiexperimental with a successive cohort as the physical therapy group. This study was therefore at a high risk of bias. The paper reported that there was no loss to follow-up.

Pain

Six studies assessed pain (Table 24). The study by Pajareya et al.77 reported analgesic use at 3 weeks only and was therefore excluded from the synthesis. Calis et al.66 reported data in graphical form only; consequently, the SMD could not be calculated and this study could not be included in a meta-analysis. Calis et al.66 did, however, report that there was significant improvement in pain severity at 3 months within both the physical therapy group and the no intervention group (p < 0.001).

TABLE 24. Pain outcomes in studies evaluating physical therapies.

TABLE 24

Pain outcomes in studies evaluating physical therapies.

The remaining four studies each evaluated different physical therapies using different comparators; therefore, it was not appropriate to pool the studies in a meta-analysis.16,40,51,74

Table 25 provides the SMDs and 95% CIs. The outcome data for individual groups in the included trials are available in Appendix 7.

TABLE 25. Pain results of studies of physical therapy (SMD and 95% CI).

TABLE 25

Pain results of studies of physical therapy (SMD and 95% CI).

Physical therapy versus physical therapy

There was no significant difference in pain between continuous passive motion and conventional physiotherapy at short-term follow-up (12 weeks) (Table 25).74 There was no significant benefit from adding ultrasound to a 20-minute supervised exercise programme of Codman's exercises, wall-climbing and joint stretching for patient's tolerance at short-term follow-up (3 months) (Table 25).51 There was no significant difference in pain between HGMT and LGMT at any time point (Table 25).40

Physical therapy without mobilisation versus control

There was a significant reduction in pain with laser compared with placebo laser at 8 weeks (SMD −1.71, 95% CI −2.29 to −1.12) and 16 weeks (SMD −1.89, 95% −2.49 to −1.29) (Table 25).16

Function and disability

Ten of the twelve studies that assessed physical therapies reported function and disability outcomes (Table 26). SPADI score16,51,71,74,77 and Constant score66,73,74 were common to several studies; however, the data could not be pooled in meta-analysis as the comparisons varied substantially. Short- and medium-term data were available from these studies. Final value data were available from six studies.16,51,66,7375 Pajareya et al.77 included only baseline data, so this study was not included in the analysis. Change from baseline data was available from the study by Vermeulen et al.40 Yang et al.78 reported mean percentage of change only. Table 27 provides between-group differences and 95% CIs. Because of differences between interventions and comparators it was not appropriate to pool these studies. Outcome data for individual studies are available in Appendix 7.

TABLE 26. Function and disability outcomes of studies evaluating physical therapy.

TABLE 26

Function and disability outcomes of studies evaluating physical therapy.

TABLE 27. Function and disability results of studies of physical therapy (MD and 95% CI).

TABLE 27

Function and disability results of studies of physical therapy (MD and 95% CI).

Physical therapy versus control

Diercks and Stevens73 reported a statistically significant improvement in Constant score with supervised neglect compared with physiotherapy at 3 (MD −16.37, 95% CI −21.46 to −11.28), 6 (MD −15.40 95% CI −20.50 to −10.30), 12 (MD −17.74, 95% CI −22.75 to −12.73), 18 (MD −16.13, 95% CI −22.20 to −10.06) and 24 (MD 9.22, 95% CI −15.69 to −2.75) months' follow-up. It is worth noting that there were several factors that could have biased this result. The physiotherapy group were prescribed exercises up to and beyond the pain threshold whereas the supervised neglect group were advised to exercise within the painless range. Also, as a successive cohort was recruited after 2 years (the physiotherapy group), treatment protocols may have changed during the time that elapsed (although the authors did report that a standardised treatment protocol was used). Although the baseline characteristics reported were similar between groups it is possible that there may have been unmeasured differences between groups.

Leung and Cheing75 found significant improvement in American Shoulder and Elbow Surgeons (ASES) score with SWD plus stretching compared with home exercise (MD 17.50, 95% CI 1.76 to 33.24). There was no significant difference between heat pack plus stretching and home exercise (Table 27).

Calis et al.66 reported that Constant score was significantly improved with physiotherapy plus home exercise compared with home exercise alone (MD 14.90, 95% CI 8.32 to 21.48).

Physical therapy versus physical therapy

The study by Dogru et al.,51 which compared ultrasound and physiotherapy with sham ultrasound and physiotherapy, found no significant difference between treatment groups in SPADI total score or the SPADI 8-item disability subscale at 3 months (Table 27).

The intention in the study by Yang et al.78 was to compare two different sequences of three types of mobilisation (each type of mobilisation was delivered over a 3-week period). The two groups received the three types of mobilisation but in a different order over 12 weeks (see Appendix 6). Because at the end of the 12 weeks both groups had received the same treatments, only 6-week data were extracted for ERM and MRM versus MWM and MRM. The change in FLEX-SF score was similar at 6 weeks for ERM and MRM (19.9%, SD 8.1%) and MWM and MRM (17.25%, SD12.2%)

The study by Dundar et al.,74 which compared continuous passive motion with physiotherapy, found no significant difference between treatment groups in SPADI total score or the Constant score (Table 27).

Leung and Cheing75 found no significant difference between SWD and heat pack groups in terms of ASES score (Table 27).

In the study by Vermeulen et al.40 there was no significant difference between HGMT and LGMT in the shoulder rating questionnaire score (Table 27). There was also no significant improvement in SDQ score with HGMT compared with LGMT at any time point (Table 27); however, it should be noted that the 95% CI marginally crossed the line of no effect at 12 months (MD −11.20, 95% CI −22.60 to 0.20). There was no significant difference in the proportions of patients who reported ‘(much) worse or no change’ at 3 months (13% vs 12%), 6 months (13% vs 10%) or 12 months (9% vs 18%) and ‘better or much better’ at 3 months (87% vs 88%) 6 months (87% vs 90%) or 12 months (91% vs 82%).

There was no significant difference in SPADI score with physiotherapy compared with osteopathy in the study by Wies et al.;71 however, it should be noted that the 95% CI marginally crossed the line of no effect (MD −19.90, 95% CI −40.11 to 0.31). There was no significant difference between either the physiotherapy or osteopathy group and the control group (breathing exercises, range of movement exercises and massage) (Table 27).

Physical therapy without mobilisation versus control

Stergioulas16 reported several measures of function and disability. At 8 weeks laser treatment was associated with significantly improved SPADI score (MD −14.11, 95% CI −19.50 to −8.72), Croft score (MD −4.34, 95% CI −6.34 to −2.34) Disabilities of the Arm, Shoulder and Hand (DASH) score (MD −9.24, 95% CI −14.35 to −4.13) and Health Assessment Questionnaire (HAQ) score (MD −0.75, 95% CI −1.09 to −0.41) compared with placebo laser. At 16 weeks laser treatment was associated with significantly improved SPADI score (MD −13.83, 95% CI −18.88 to −8.78), Croft score (MD −7.13, 95% CI −9.18 to −5.08) and DASH score (MD −10.51, 95% CI −15.37 to −5.65). There was no significant difference in HAQ score between groups at 16 weeks (Table 27).

Range of movement

Eleven of the included studies that investigated physical therapy reported a range of movement measure of interest (Table 28). All of the studies except two reported short- or medium-term follow-up. One presented data at 24 months73 and one at 12 months.40 Six studies reported final value data and one77 presented data that were not included in the analysis as follow-up was l < 4 weeks. One study reported change from baseline data.40 Maricar and Chok76 reported only graphs and p-values, presented narratively. Because of the differences between the interventions and comparators meta-analyses were not performed. Table 29 provides between-group differences and 95% CIs. Outcome data for individual studies are available in Appendix 7.

TABLE 28. Range of movement outcomes in studies evaluating physical therapy.

TABLE 28

Range of movement outcomes in studies evaluating physical therapy.

TABLE 29. Range of movement results of studies of physical therapy (MD and 95% CI).

TABLE 29

Range of movement results of studies of physical therapy (MD and 95% CI).

Physical therapy versus control

In the study by Calis et al.66 there was a significant improvement in passive abduction (MD 24.90°, 95% CI 14.59° to 35.21°) and passive external rotation (MD 18.80°, 95% CI 13.11° to 24.49°) in the physiotherapy group compared with home exercise alone.

In the study by Leung and Cheing75 there was a significant improvement in external rotation with arm by side (MD 21.00°, 95% CI 4.95° to 37.05°) with SWD plus stretching compared with home exercise. Hand behind back was also significantly greater in the SWD plus stretching group compared with the home exercise group (MD −7.00 cm, 95% CI −13.14 cm to −0.86 cm). There was no significant difference between either SWD or heat pack (both with stretching exercises) and home exercise in external rotation with arm at 90°, and no significant difference between heat pack and home exercise in external rotation with arm by side or hand behind back (Table 29).

The study by Diercks and Stevens73 reported that at 24 months the physiotherapy group had a median of 8 points for the Constant score for external rotation, compared with 10 points in the supervised neglect group, although no measure of variance was provided. This corresponds to a greater range of motion in the supervised neglect group in this plane. For internal rotation, both groups reported a median of 8 points, corresponding to the position dorsum of the 12th dorsal vertebra.

Physical therapy versus physical therapy

Dogru et al.51 found no significant difference between ultrasound and sham ultrasound in external rotation, internal rotation and abduction (Table 29).

There was also no significant difference in the study by Dundar et al.74 between continuous passive motion and physical therapy in passive external rotation, passive internal rotation and passive abduction (Table 29).

The study by Wies et al.71 showed a significantly greater improvement in active abduction in the osteopathy group (MD 45.20°, 95% CI 16.87° to 73.53°) and physiotherapy groups (MD 38.80°, 95% CI 5.76° to 71.84°) compared with control (breathing exercises, massage and exercise). There was no significant difference between the physiotherapy and osteopathy groups (Table 29).

Yang et al.78 reported that the mean percentage of change in external rotation was similar between ERM plus MRM (mean 36.4%, SD 24.3%) and MWM plus MRM (mean 34.2%, SD 13.3%). However, there was a greater change in internal rotation with MWM + MRM (45.6%, SD 38.5%) compared with ERM and MRM (20.5%, SD 24.4%) (Table 29).

Leung and Cheing75 found that there was a significant improvement in external rotation both with arm by side (MD 29.50°, 95% CI 14.28° to 44.72°) and with arm at 90° abduction (MD 30.10°, 95% CI 13.51° to 46.69°) with SWD compared with heat pack. There was also a significant improvement in hand behind back with SWD compared with heat pack (MD −12.30 cm, 95% CI −18.79 cm to −5.81 cm) (Table 29).

Maricar and Chok76 compared passive joint mobilisations and exercises with exercises only and found no significant difference in external rotation or internal rotation between treatment groups at weeks five, seven or eight.

The study by Vermeulen et al.40 showed a significant improvement in active abduction at 12 months with HGMT compared with LGMT (MD 12.60°, 95% CI 0.06° to 25.14°). There was also a significant improvement with HGMT in passive abduction at 3 months (MD 13.10°, 95% CI 1.62° to 24.58°) and 12 months (MD 12.50°, 95% CI 1.03° to 23.97°), and in passive external rotation at 12 months (MD 6.50°, 95% CI 0.27° to 12.73°). There was no significant difference between groups in active external rotation at any of the time points, or in passive external rotation, active abduction or passive abduction at the remaining time points (Table 29).

Physical therapy without mobilisation versus control

There was no significant difference in external rotation or active abduction at short- or mediumterm follow-up between laser and placebo laser (Table 29).16

Quality of life

Two studies, one by Dogru et al.51 and the other by Vermeulen et al.,40 reported quality of life using the physical and mental components of the SF-36 (Table 30). Data at 3 months' follow-up were available (Table 31). As the interventions varied substantially these studies were not pooled. Outcome data for the individual groups are available in Appendix 7.

TABLE 30. Quality of life outcomes in studies evaluating physical therapy.

TABLE 30

Quality of life outcomes in studies evaluating physical therapy.

TABLE 31. Quality of life results of studies of physical therapy (MD and 95% CI).

TABLE 31

Quality of life results of studies of physical therapy (MD and 95% CI).

Physical therapy versus physical therapy

There was no significant difference between ultrasound and sham ultrasound groups in the SF-36 physical or mental components at 3 months' follow-up in the study by Dogru et al. (Table 31).51

There was no significant difference between HGMT and LGMT groups in the SF-36 physical or mental components at any time point in the study by Vermeulen et al. (Table 31).40

Other

Two studies reported other outcomes of interest (Table 32). Pajareya et al.77 reported satisfaction with treatment and percentage of patients reporting treatment success. Yan72 also reported the percentage of patients describing their rate of improvement as bad, average, good and excellent at 3 months. These data are summarised in Table 33 and further information for the individual studies is available in Appendix 7.

TABLE 32. Other outcomes in studies evaluating physical therapy.

TABLE 32

Other outcomes in studies evaluating physical therapy.

TABLE 33. Other outcome results from studies of physical therapy.

TABLE 33

Other outcome results from studies of physical therapy.

Physical therapy versus control

In the study by Pajareya et al.77 similar proportions of each group reported that they had successful treatment at 6 weeks (61.4% vs 60.8%). At 12 weeks, 76.8% in the physiotherapy compared with 60.8% in the control group reported successful treatment, and at 24 weeks the proportions were 80.4% in the physiotherapy group and 82.4% in the control group.

In the same study the number of patients ‘very satisfied’ at 3 weeks was five in the physical therapy group versus one in the control group; the numbers ‘moderately satisfied’, ‘unsatisfied’ and ‘very unsatisfied’ were 7 versus 1, 24 versus 13 and 23 versus 45 in the physical therapy group versus the control group respectively.77

Physical therapy versus physical therapy

Yan72 compared dumb-bell exercises with barehanded exercises and found that a greater number of patients performing dumb-bell exercises than those performing barehanded exercises reported an excellent rate of improvement at 3 months (24 vs 0).

Adverse events

Two studies reported adverse events.74,77 In the study by Pajareya et al.77 patients in the physiotherapy group were asked whether they had pain that persisted for > 2 hours after treatment or more disability the next morning. There were 10 episodes of pain in the physiotherapy group (in four patients) that persisted for > 2 hours after treatment. Patients were also asked whether the trial drugs and/or treatment programme upset them in any way and were examined for signs of echymosis or burn during range of movement evaluation. Side effects of NSAIDs were reported but it was unclear which treatment groups these applied to (see Appendix 6). Dundar et al.74 stated that no side effects were observed during the study. No further information regarding how adverse events were measured or assessed was reported. The study by Stergioulas16 reported that there were no complications. Adverse events did not appear to have been assessed in the remaining nine studies.

Summary

Twelve studies were included that investigated various types of physical therapy, without use of steroid injection. They were very diverse in the physical therapies evaluated, although most evaluated physiotherapy modalities. The comparators were another physical therapy or control. With the exception of one study,16 at least one component of the intervention involved active mobilisation or exercise. Half the studies reported that participants were advised to undertake home exercise.

One study of satisfactory quality was identified and three studies were of reasonable quality but had some risk of bias. The other studies had a potentially high risk of bias. Because of the considerable variability of the interventions investigated in the studies and different outcomes measures used it was not appropriate to pool any studies in a meta-analysis. The studies did not report outcome by stage of frozen shoulder or the presence of diabetes and information was very limited regarding previous treatments received.

Physical therapy versus control

There was evidence from one study (with some risk of bias) that there was a significant short-term benefit for function and disability with SWD in conjunction with stretching compared with home exercise and for some but not all measures of range of movement. A further study with some risk of bias reported significant short- and medium-term benefit with laser therapy (plus home exercise) compared with home exercise alone for pain and function and disability but not range of movement. Another study with a potentially high risk of bias reported significant short-term benefit with physiotherapy (plus home exercise) compared with home exercise alone for function and disability and two ranges of movement. One study reported significant short-, medium- and long-term benefit for function and disability with supervised neglect compared with physiotherapy; however, this study was at high risk of bias.

Comparing physical therapies

The majority of studies comparing two physical therapies reported no significant difference in outcome between therapies; however, with the exception of three studies,40,75,78 these all had a potentially high risk of bias. A single study with some risk of bias found a significant benefit of SWD compared with a heat pack (both with stretching) for three ranges of movement; however, in this same study there was no significant difference in benefit for function and disability. Another study with some risk of bias found a significant improvement in one measure of function and disability and in three ranges of movement with HGMT compared with LGMT, but no benefit for pain. A single satisfactory study found no benefit from providing ERM followed by MRM compared with MWM followed by MRM. Based on single studies with a high risk of bias there was no significant difference between continuous passive motion and conventional physical therapy, between mobilisation with and without ultrasound or between joint mobilisation and exercise. A single study with a high risk of bias reported that daily exercises using dumb-bells were significantly more effective than exercises without dumb-bells. One study available only as an abstract and at high risk of bias reported a significant benefit of physiotherapy and osteopathy compared with breathing and range of movement exercises combined with massage in one measure of range of movement. There was no significant difference between physiotherapy and osteopathy techniques.

In conclusion, the studies identified were diverse and mainly of poor quality. With the exception of physiotherapy including SWD, only single studies were identified of any one physical approach. Based on the best available evidence there may be benefit from SWD and stretching and from HGMT in patients who have already had physiotherapy or a steroid injection. There was insufficient evidence to make conclusions with any certainty about the best mode of physical therapy for primary frozen shoulder.

Acupuncture

Study characteristics

Three RCTs investigated acupuncture in the treatment of primary frozen shoulder. A summary of study characteristics is reported in Table 34. All were full papers of studies conducted in Hong Kong,79 China80 and Taiwan81 between 2006 and 2008. There was one two-armed trial80 and two three-armed trials.79,81 Sample sizes ranged from 74 to 360.

TABLE 34. Studies of acupuncture with or without physical therapy.

TABLE 34

Studies of acupuncture with or without physical therapy.

One study used acupuncture81 and two studies used electroacupuncture,79,80 The number of acupuncture points needled ranged from three to five. Although needle point prescriptions varied between studies, two studies both included Hegu (LI4) and Jianyu (LI15) needle points.80,81 In the acupuncture study, acupuncture was administered in a course of eight sessions. In both studies of electroacupuncture, the intervention was administered in a course of 10 sessions, with acupuncture points needled to a depth of 15–25 mm and electricity administered when ‘De qi’ sensation was felt. Further details of the electrical stimulation used were reported in one study only79 and consisted of 2–100-Hz electrical stimulation at a pulse of 100–400 microseconds for 20 minutes. The comparators were interferential electrotherapy,79 waiting list control,79 TENS,80 physiotherapy81 and acupuncture with physiotherapy.81

One study included a physiotherapy regimen.81 This consisted of a heat pack for 15 minutes, joint mobilisation for 5–10 minutes and active shoulder exercises for 5–10 minutes, five times per week for 4 weeks. One study reported concomitant treatment of a home exercise programme of mobilisation exercises five times a day over 6 months.79 Two studies did not report whether participants received any concomitant treatment.80,81

The inclusion criteria varied between the three studies. It was unclear in all studies whether they included participants with diabetes. Frozen shoulder was diagnosed by an orthopaedic surgeon, through clinical examination and through clinical history. The participants' stage of frozen shoulder was not reported in two studies.79,81 In the other study participants were reported to have either stage 1 (described by the authors as pre-adhesive) frozen shoulder, that is, they had shoulder pain that increased at night with range of motion normal or only slightly affected, or stage 2 (described by the authors as adhesive) frozen shoulder, that is, a reduction in pain but severely affected range of motion.80 None of the studies reported whether participants had received previous treatment for frozen shoulder. Where reported, the mean age of participants ranged from 51.5 to 56.4 years and the proportion of women from 52% to 55%.

Quality assessment

None of the studies reported the method of randomisation or allocation concealment; therefore, these three studies are potentially at high risk of bias. The study by Cheing et al.79 was of the highest quality and fulfilled five criteria: the number randomised, the comparability of treatment groups at baseline, double blinded, blinding of outcome assessors to treatment allocation and no unexpected imbalances in dropouts between groups. This study, however, did not use intention-to-treat analysis. Fang et al.80 fulfilled two criteria: the number of participants randomised and the blinding of participants. Ma et al.81 also fulfilled two criteria: the number of participants randomised and the comparability of treatment groups at baseline for important prognostic factors.

The studies by Fang et al.80 and Ma et al.81 reported outcome measures at ≤ 4 weeks from baseline and are therefore excluded from the subsequent synthesis. Furthermore, the waiting list control arm for the study by Cheing et al.79 reported outcomes at 4 weeks from baseline only. The following synthesis is therefore based on the study by Cheing et al.79 and compares electroacupuncture with interferential therapy only. This study reported on the outcomes of pain and of function and disability only.

Pain

Cheing et al.79 reported results of ‘pain at the moment’ at approximately 2, 4 and 7 months using a 0–10 cm VAS. There was no significant difference in pain between the electroacupuncture and interferential therapy groups at 2 months (SMD 0.35, 95% CI −0.23 to 0.93), 4 months (SMD 0.21, 95% CI −0.37 to 0.78) or 7 months (SMD 0.21, 95% CI −0.37 to 0.78).

Function and disability

Cheing et al.79 reported results of the Constant score at 2, 4 and 7 months. There was no significant difference in Constant score between the electroacupuncture and interferential therapy groups at 2 months (MD −2.80, 95% CI −5.88 to 0.28), 4 months (MD 3.10, 95% CI −1.53 to 7.73) or 7 months (MD −1.70, 95% CI −4.76 to 1.36).

Adverse events

None of the three studies that evaluated acupuncture or electroacupuncture reported on whether or not participants experienced any adverse events.

Summary

Three studies compared acupuncture with another treatment, although only one provided data beyond 4 weeks' follow-up. This study had a potentially high risk of bias and it was unclear whether it had enough participants to detect a difference between groups. Based on a single study, there was no statistically significant difference between electroacupuncture and inferential electrotherapy in pain or function and disability at short-, medium- or long-term follow-up. There was insufficient evidence to make conclusions with any certainty about the effectiveness of acupuncture for primary frozen shoulder and in what situations it is likely to be effective.

Manipulation under anaesthesia

Study characteristics

Four RCTs investigated MUA in the treatment of primary frozen shoulder. A summary of study characteristics is available in Table 35 and full details are provided in Appendix 6. All were full papers of studies conducted in Pakistan,82 the UK38,83 and Finland39 between 2007 and 2009. All were two-arm trials and sample sizes ranged from 3638 to 125.39

TABLE 35. Studies of MUA.

TABLE 35

Studies of MUA.

The MUA procedure was described in three of the RCTs38,39,83 and was performed by a physician39 or, in the two UK studies, by an orthopaedic surgeon.38,83 In two studies participants received a steroid injection (triamcinolone) in conjunction with MUA, 30 mg38 in one of the UK studies and 80 mg82 in the study from Pakistan.

A single study, by Amir-us-Saqlain et al.,82 reported use of a physiotherapy programme following MUA. The other three studies advised participants in the intervention and control groups to undertake home exercises. Kivimaki et al.39 appeared to have the most intensive home exercise programme, although details were scanty in all the studies; participants in this study received physiotherapy advice in two sessions and written instructions for a daily training programme including pendulum and stretching exercises. Quraishi et al.38 reported a home exercise programme of pendulum exercises and wall-climbing movements; specific details were not provided by Jacobs et al.83

Each of the studies had a different comparator. In the study by Amir-us-Saqlain et al.82 from Pakistan, the intervention being assessed was the effect of keeping the manipulated extremity in abduction and external rotation for 24 hours following MUA. Both groups received MUA, including a steroid injection, followed by physical therapy and the experimental group also had the extremity kept in a fixed position following MUA. In the two UK studies, Jacobs et al.83 compared MUA with distension in combination with steroid injection; and Quraishi et al.38 compared MUA, including a steroid injection, with arthrographic distension. Kivimaki et al.39 compared MUA (plus home exercise) with home exercise alone, which involved physiotherapy advice in two sessions and written instructions for home exercise.

The inclusion criteria varied between studies in terms of extent of restriction of movement.38,39,82 Jacobs et al.83 did not specify the symptomatic inclusion criteria other than that patients with primary frozen shoulder were included. Quraishi et al.38 included only patients with stage 2 frozen shoulder. The stage of frozen shoulder was described as painful and stiff or in the ‘freezing’ phase in two RCTs,39,83 and the majority of participants reported pain and stiffness in one RCT (the remainder had either pain or stiffness).82 At baseline the duration of frozen shoulder ranged from a mean or median of 10 weeks to 40 weeks. Two studies reported on whether participants had received previous treatment for frozen shoulder. Quraishi et al.38 reported that 44% had previously received physical therapy, 61% a steroid injection and 31% physical therapy combined with steroid injections. Kivimaki et al.39 alluded to participants having received previous treatment by reporting that in the 3 months prior to randomisation there was no difference between treatment groups in terms of physical therapy, massage or chiropractic manipulations received. The mean or median age of the participants ranged from 53 to 57 years and the proportion of women ranged from 58% to 68%. It was unclear whether participants with diabetes were included in one study,82 one study reported that there were no participants with diabetes83 and in the remaining two studies the proportions with diabetes were 14%39 and 17%.38

Quality assessment

The included RCTs generally varied in quality. All of the studies reported the number of participants randomised and comparability at baseline, although one reported only comparability for range of movement.82 The Amir-us-Saqlain et al.82 and Jacobs et al.83 studies did not report the method of randomisation and so it was unclear whether these studies were truly randomised, and they did not report allocation concealment or whether the outcome assessors were blinded, although this would not have been possible in the study by Amir-us-Saqlain et al.82 These studies therefore had a potentially high risk of bias. Additionally, Jacobs et al.83 was not adequately powered, loss to follow-up was high (19%) and it was unclear whether imbalances in dropouts were adjusted for. Although the authors stated that intention-to-treat analysis was performed, not all patients were included in the analysis. It was unclear whether the study by Amir-us-Saqlain et al.82 was adequately powered and loss to follow-up was high (23%).

Both Kivimaki et al.39 and Quraishi et al.38 reported appropriate randomisation methods; however, only Kivimaki et al.39 reported allocation concealment and blinding of outcome assessors and was therefore classified as of adequate quality. Loss to follow-up was high in this study (34% at 6 months) and was greater in the MUA group than in the control group; the analysis did not appear to be intention to treat. The study did not appear to be adequately powered at the 6- and 12-month follow-ups. Although Quraishi et al.38 reported that the investigator was blinded, it was not clear whether outcome assessment was blinded or whether there was allocation concealment and therefore this study may be at risk of bias. Loss to follow-up was relatively low in this study (8%) although it was unclear whether the study was adequately powered. Full details of study quality are reported in Appendix 8.

Manipulation under anaesthesia versus home exercise

Kivimaki et al.39 compared MUA plus home exercise with home exercise alone. MDs and 95% CIs were reported by the authors for pain, function and disability, passive abduction, passive internal rotation, external rotation and working ability at the follow-up points of 6 weeks and 3, 6 and 12 months from baseline (Table 36). The outcome data for the individual groups are available in Appendix 7.

TABLE 36. Outcomes assessed in the study of MUA versus home exercise.

TABLE 36

Outcomes assessed in the study of MUA versus home exercise.

Pain

There was no significant difference in pain intensity between MUA and home exercise at 6 weeks or 3, 6 or 12 months (Table 37).

TABLE 37. Results of the study of MUA versus home exercise (MD and 95% CI).

TABLE 37

Results of the study of MUA versus home exercise (MD and 95% CI).

Function and disability

There was no significant difference in Constant score between treatment groups at 6 weeks or 3, 6 or 12 months (Table 37).

Range of movement

There was no significant difference in passive abduction, passive internal rotation or external rotation between treatment groups at 6 weeks or 3, 6 or 12 months (Table 37).

Working ability

There was no significant difference in the change in working ability between treatment groups at 6 weeks or 3, 6 or 12 months (Table 37).

Adverse events

The authors reported that there were no major complications during manipulation.

Manipulation under anaesthesia versus distension

Two studies compared MUA with distension;38,83 however, both MUA and distension differed between studies and so the data for pain and function and disability that were available could not be considered for pooling.

One RCT by Jacobs et al.83 compared MUA with steroid injection in combination with distension treatment. Only limited data were available for pain, function and disability and quality of life, which were presented as mean regression coefficients up to 16 weeks (and 24 months for quality of life) (Table 38). The other RCT by Quraishi et al.38 compared MUA plus steroid injection with arthrographic distension. This study reported means and ranges for pain and function and disability at 2 and 6 months. The percentage of patients who were satisfied with treatment was also reported (Table 38). The outcome data for the individual groups are available in Appendix 7.

TABLE 38. Outcomes assessed in the studies of MUA vs distension.

TABLE 38

Outcomes assessed in the studies of MUA vs distension.

Pain

The study by Jacobs et al.83 showed no significant difference in pain intensity between treatment groups up to 16 weeks. The mean regression coefficient (SE) was −2.77 (0.33) for MUA and −2.75 (0.42) for distension combined with steroid injection (95% CI −1.11 to 1.15).

In the study by Quraishi et al.38 at 2 months' follow-up the mean VAS score was 4.7 (range 0 to 8.5) in the MUA plus steroid arm and 2.4 (range 0 to 8) in the arthrographic distension arm (p-value not reported). At 6 months there was a significantly greater improvement in VAS in the arthrographic distension group than in the MUA group (SMD 1.52, 95% CI 0.75 to 2.30) (Table 39).

TABLE 39. Results of the studies of MUA vs arthrographic distension (SMD and 95% CI).

TABLE 39

Results of the studies of MUA vs arthrographic distension (SMD and 95% CI).

Function and disability

The study by Jacobs et al.83 found no significant difference between treatment groups in the Constant score up to 16 weeks: mean regression coefficient (SE) 3.13 (0.24) for MUA and 3.23 (0.42) for distension plus steroid injection (95% CI −1.09 to 0.81).

In the study by Quraishi et al.38 at 2 months' follow-up the mean Constant score was 58.5 (range 24 to 90) in the MUA plus steroid arm and 57.4 (range 17 to 80) in the arthrographic distension arm. At 6 months the Constant score was significantly improved in the arthrographic distension group compared with the MUA group (–6.40, 95% CI −6.56 to −6.24) (Table 39).

Range of movement

Quraishi et al.38 reported no statistically significant difference between groups in external rotation (p = 0.13), internal rotation (p = 0.48) and abduction at 6 months' follow-up (p = 0.62) (data available only in a graph).

Quality of life

Jacobs et al.83 reported that all components of the SF-36 improved for all patients during the course of treatment, with the greatest improvements shown in the physical role and bodily pain components. There was no statistically significant difference with respect to change in SF-36 scores between the treatment groups (data available only in graph format, no p-values reported).

Satisfaction

Quraishi et al.38 reported that at final follow-up 81% of MUA patients were satisfied or very satisfied compared with 94% of patients who received arthrographic distension. All of the patients with diabetes were satisfied with their outcome (three in both the MUA and arthrographic distension groups).

Adverse events

The study by Jacobs et al.83 reported that no systemic or local complications were noted in either treatment group. Quraishi et al.38 did not report any information on adverse events.

Manipulation under anaesthesia followed by manipulated extremity kept in fixed position

Amir-us-Saqlain et al.82 assessed the effect of keeping the manipulated extremity in abduction and external rotation for 24 hours following MUA compared with MUA not using the fixed position. Both groups also received a steroid injection and physical therapy. Only measures of range of movement were assessed 12 weeks post treatment (Table 40). Between-group differences in means and 95% CIs are presented in Table 41. Outcome data for the individual groups are presented in Appendix 7.

TABLE 40. Outcomes assessed in the study of MUA followed by fixed extremity.

TABLE 40

Outcomes assessed in the study of MUA followed by fixed extremity.

TABLE 41. Results of the study of MUA followed by fixed extremity (MD and 95% CI).

TABLE 41

Results of the study of MUA followed by fixed extremity (MD and 95% CI).

Range of movement

Passive abduction (MD 22.49°, 95% CI 13.14° to 31.84°), passive internal rotation (MD 1.00 cm, 95% CI 0.71 cm to 1.29 cm), active abduction (MD 28.99°, 95% CI 17.07° to 40.91°) and active internal rotation (MD 0.91 cm, 95% CI 0.49 cm to 1.33 cm) were significantly greater at 12 weeks post treatment in the MUA combined with steroid injection plus manipulated extremity and physiotherapy group than in the MUA combined with steroid injection plus physiotherapy group. There was no significant difference in passive or active external rotation between treatment groups.

Summary

The four included studies of MUA were diverse in terms of the intervention and the comparator. Some studies used a steroid injection (triamcinolone) in conjunction with MUA. The studies were not considered clinically similar enough to be pooled in a meta-analysis. Across all of the studies information about previous treatments that participants had received and stage of frozen shoulder was limited. It was not possible to explore variation between those with and without diabetes.

Because it was unclear whether all studies had enough participants to detect a statistically significant treatment effect, it should be kept in mind that where studies did not report a significant benefit (i.e. the CI crossed the line of no difference) this should not be interpreted as evidence of no difference between the groups. A single study of adequate quality reported no statistically significant difference between MUA (and home exercise) and home exercise alone in pain, function and disability, range of movement or working ability at 6 weeks and 3, 6 and 12 months.39 Follow-up at 6 months and beyond in this study should be treated with some caution because of high loss to follow-up, which was somewhat higher in the MUA group, and also the lack of intention-to-treat analysis. Two studies comparing MUA with distension had mixed findings. One, with a potentially high risk of bias, found no significant difference in pain or function and disability between MUA and distension (which included steroid) up to 16 weeks post treatment.83 The second, which had some risk of bias, found a significantly greater improvement in pain and function and disability at 6 months with arthrographic distension than with MUA in conjunction with steroid.38 The fourth study, with a potentially high risk of bias, reported that keeping the manipulated extremity in abduction and external rotation for 24 hours following MUA was superior to MUA without this additional process for passive and active abduction and internal rotation at 12 weeks.82

In conclusion, there was very little evidence available for MUA and most of the studies identified had limitations. The single adequate study found no evidence of benefit of MUA over home exercise alone. Generalisability is somewhat unclear because of the limited information about previous interventions that participants had received and stage of frozen shoulder.

Distension

Study characteristics

Three RCTs investigated distension with steroid injection in the treatment of primary frozen shoulder. A summary of study characteristics is reported in Table 42, with further details available in Appendix 6. All were full papers of studies conducted in Australia,43 Denmark84 and Norway36 and published between 1998 and 2008. All were two-armed trials.

TABLE 42. Studies including distension with steroid injection.

TABLE 42

Studies including distension with steroid injection.

The two most recent studies by Buchbinder et al.43 and Tvetia et al.36 used arthrographic distension. One injected 40 mg of steroid (in 1 ml) and up to 82 ml of saline on a single occasion43 and the other injected 20 mg of steroid (2 ml), 4 ml of local anaesthetic and 10 ml of saline on three occasions at 2-week intervals,36 along with contrast medium in the arthrographic distension groups. Both of these studies stated that the procedure continued until rupture occurred. Buchbinder et al.43 had the additional end points of patient termination of the procedure and a maximum of 90 ml having been injected. The older study by Gam et al.84 stated that the injection was confirmed by ultrasound and participants were injected with 20 mg of steroid with 19 ml of lidocaine once per week for up to 6 weeks. Details were not provided in this study as to whether distension occurred or how this was established. The two studies of arthrographic distension used different comparators: Buchbinder et al.43 used a placebo comparator of arthrogram only, whereas Tveita et al.36 used steroid injection (with local anaesthetic).

Home exercise of pendular exercises and scapular setting were given in the study by Buchbinder et al.43 and no manual treatment (e.g. physiotherapy, massage) or other medical interventions were allowed. In contrast, in the Tvetia et al. study36 patients were allowed to continue with their current physiotherapy programme; no patients were prescribed new physiotherapy programmes during the study. The study of non-arthrographic distension did not report any information regarding home exercise.84

There was some variability in concomitant treatments. Buchbinder et al.43 allowed paracetamol and codeine preparations but NSAIDs were not allowed, Tveita et al.36 allowed pain medication organised by the patients' primary care physicians and Gam et al.84 reported that analgesics were permitted.

The inclusion criteria varied between studies; however, the two most recent studies of arthrographic distension required similar extent of restriction in shoulder movement.36,43 The duration of frozen shoulder at baseline ranged from a median or mean of 114 days (approximately 4 months) to 7 months, although the stage at baseline was not reported by any of the studies. Diagnosis was by measuring passive range of movement to onset of pain,43 clinical examination, blood samples and radiography or ultrasound,84 and clinical history and radiography.36

A substantial proportion of patients (28%) had diabetes in the Buchbinder study;43 the remaining two studies had no patients with diabetes. Buchbinder et al.43 also included a small proportion (2.2%) with secondary frozen shoulder. Buchbinder et al.43 and Tveita et al.36 also reported that patients had undergone treatment for frozen shoulder before the study: in the Buchbinder study 28% of patients had received corticosteroid injections whereas in the Tveita study 17% had undergone physiotherapy. The ages of patients ranged from a mean or median of 47 years to 57.3 years and the proportion of women was either 59% or 80%.

Quality assessment

Buchbinder et al.43 was considered of satisfactory quality as the risk of bias was minimised through computerised randomisation and allocation concealment by a biostatistician who kept the assignment scheme. An intention-to-treat analysis was also performed. Patients were blinded to treatment and outcome assessors were also blinded. The treatment groups were also comparable at baseline. The main limitation of this study is that it was not sufficiently powered after dropouts (9%). The intention-to-treat analysis used in the study was adequately powered based on the authors' power calculation (the sample size to detect a difference in SPADI scores).

Both Gam et al.84 and Tveita et al.36 reported a reliable method of randomisation; however, it was not clear whether an adequate method of allocation concealment had been used. Tveita et al.36 reported that patients were not informed of their actual assignment until the first injection was to be given, and no information regarding allocation concealment was reported by Gam et al.84 Therefore, these studies are potentially at risk of bias. Gam et al.84 reported that outcome assessors were blinded and that the groups were comparable at baseline. In the study by Tveita et al.36 there appeared to be a higher proportion with previous shoulder problems and who were on sick leave in the steroid injection group. Tveita et al.36 also reported using statistical adjustment to control for baseline differences for the function and disability SPADI outcome but not for range of movement. Tveita et al.36 used intention-to-treat analysis and both studies reported the same rate of loss to follow-up (9%), with similar numbers in each arm.

Pain

Two studies, one by Buchbinder et al.43 and one by Gam et al.,84 assessed pain. Buchbinder et al.43 reported change from baseline data at 3, 6 and 12 weeks using a 10-point Likert scale (Table 43). However, the study by Gam et al.84 reported data for VAS pain scores at rest and VAS pain scores on function in graphical form only. This study was described narratively.

TABLE 43. Pain outcomes in studies evaluating distension.

TABLE 43

Pain outcomes in studies evaluating distension.

There was no significant difference in pain between arthrographic distension with steroid and placebo (arthrogram) at 6 weeks (SMD −0.40, 95% CI −0.99 to 0.19) or 12 weeks (SMD −0.15, 95% CI −0.73 to 0.44) in the study by Buchbinder et al.43

In the study by Gam et al.84 there was no significant difference between the steroid and distension groups for VAS pain score on function (p = 0.1) or VAS pain score at rest (p = 0.1) (time points unclear). The authors reported that there was a tendency toward less pain in the distension group. Additionally, the authors reported that analgesic consumption was significantly lower at week 11 in the distension group than in the steroid group (p = 0.008). These results were presented in graphical form only.

Function and disability

Two of the three included studies assessed function and disability.36,43 Both reported SPADI total score and one43 reported a problem elicitation technique score, a patient preference disability measure in which patients identify their own problems related to the disease that they would like to see improve as a result of treatment. Change from baseline data were available from Buchbinder et al.43 and final value data were available from Tveita et al.36 (Table 44). These were not pooled as the comparator arms were different between studies: Buchbinder et al.43 compared arthrographic distension with placebo whereas Tveita et al.36 used steroid injection as a comparator. Table 45 provides the between-group MDs and 95% CIs. The outcome data for individual groups are available in Appendix 7.

TABLE 44. Function and disability outcomes in studies evaluating distension.

TABLE 44

Function and disability outcomes in studies evaluating distension.

TABLE 45. Function and disability results of studies of distension (MD and 95% CI).

TABLE 45

Function and disability results of studies of distension (MD and 95% CI).

Distension versus placebo

There was no significant difference between arthrographic distension and placebo in SPADI total score at 6 or 12 weeks in the study by Buchbinder et al.43 (Table 45). In contrast, there was a significant improvement in the problem elicitation technique score with distension compared with placebo at 6 weeks (MD −46.00, 95% CI −80.99 to −11.01) and 12 weeks (MD 54.40, 95% CI 12.57 to 96.23) in the same study.

Distension versus steroid

There was no significant difference in SPADI total scores between distension and steroid at 6 weeks in the Tveita et al. study36 (Table 45).

Range of movement

The range of movements of interest in this review that were reported by the studies of distension were passive and active abduction, passive and unspecified external rotation, passive internal rotation and hand behind back. Change from baseline data were available from Buchbinder et al.43 and final value data were available from Tveita et al.36 (Table 46). It was considered inappropriate to pool the two studies as one had a placebo comparator and one a steroid injection comparator. The study by Gam et al.84 provided limited data for extraction and these were summarised narratively. Table 47 summarises the between-group MDs and 95% CIs. The outcome data for individual groups are available in Appendix 7.

TABLE 46. Range of movement outcomes in studies evaluating distension.

TABLE 46

Range of movement outcomes in studies evaluating distension.

TABLE 47. Range of movement results of studies of distension (MD and 95% CI).

TABLE 47

Range of movement results of studies of distension (MD and 95% CI).

Distension versus placebo

There was no significant difference between groups in the outcomes active abduction, external rotation or hand behind back at 6 or 12 weeks' follow-up in the study by Buchbinder et al.43 (Table 47).

Distension versus steroid

The study by Tveita et al.36 found that there was a significant improvement in passive abduction with steroid compared with distension at 6 weeks after the last injection (MD −34.00°, 95% CI −40.78° to −27.22°). There was no significant difference between groups for active abduction, external or internal rotation, or passive internal or external rotation (Table 47).

The study by Gam et al.84 reported that there was a significant improvement with distension compared with steroid in external rotation (p = 0.0007) but no statistically significant difference in abduction (time point unclear).

Adverse events

All of the studies investigating distension reported adverse events. These were reported in detail in the study by Buchbinder et al.43 and to a lesser extent by Gam et al.84 and Tveita.36 The adverse events reported by all studies are summarised in Table 48.

TABLE 48. Adverse events in studies investigating distension.

TABLE 48

Adverse events in studies investigating distension.

Buchbinder et al.43 reported that open-ended questions were used to assess the extent of adverse events. More patients in the arthrographic distension group than in the placebo group reported pain associated with the procedure (16% vs 5%) and pain lasting up to 48 hours (12% vs 5%). None of the adverse events was reported as serious and a number of additional adverse events were reported (Table 48).

Gam et al.84 reported that the number and type of side effects were recorded. The only adverse events reported in this study were two cases of unacceptable pain after injection (one in each treatment group); these patients dropped out of the study.

In the study by Tveita et al.36 the patients recorded pain intensity related to the injection procedures. Six patients in the steroid group and five in the distension group felt that the injections were very painful. Flushing and disturbances in heat regulation were also common in both groups and a number of other non-serious adverse events were reported (Table 48).

Summary

Three studies were included that investigated the effects of distension with steroid injection. Two of these studies investigated arthrographic distension36,43 and the other investigated non-arthrographic distension.84 The comparators evaluated were steroid alone36,84 and placebo (arthrogram only).43 Only one study was judged to be of satisfactory quality,43 whereas the remainder were potentially at risk of bias. Because of the variability in the interventions and comparators it was not considered appropriate to pool any of the studies. Stage of frozen shoulder was not reported in any of these studies; however, previous treatments were described in two of the studies.36,43

Distension compared with placebo

One study of satisfactory quality compared distension with placebo and reported benefit with distension in the short term in one of two measures of function and disability. There was no difference in three measures of range of movement or in pain at function at short-term follow-up. None of the adverse events reported was considered serious.

Distension compared with steroid

Two studies that were potentially at risk of bias compared distension with steroid. One study that compared arthrographic distension with steroid reported no difference between groups in one measure of function and disability, but reported a benefit in one range of movement in the short term. Some patients in both groups considered the injections as being very painful and one case of glenohumeral joint infection (septic arthritis) was reported. The study that compared non-arthrographic distension with steroid reported no difference between groups in pain at function or at rest, but a significant benefit with distension in one measure of range of movement but not another at short-term follow-up. One patient from each group dropped out as a result of unacceptable pain.

In conclusion, few studies of distension were identified and only single studies of different comparisons were available. Based on one study of satisfactory quality there is a little evidence of potential benefit with distension compared with placebo. There is insufficient evidence to draw conclusions about the efficacy of distension (arthrographic or non-arthrographic) for frozen shoulder.

Capsular release

Two case series of more than 50 participants were identified that investigated capsular release in the treatment of primary frozen shoulder. A summary of the study characteristics is reported in Table 49, with further details available in Appendix 7. One study was published in Norway in 2007 by Austgulen et al.;85 the other in Taiwan in 2002 by Chen et al.86

TABLE 49. Studies including capsular release.

TABLE 49

Studies including capsular release.

The capsular release procedures varied between the two studies. Only a summary of the procedures are provided here; fuller descriptions are available in the original papers. In the study by Chen et al.,86 the authors stated that the intervention involved distension and debridement, release and manipulation. The degree of joint movement was confirmed arthroscopically (after distension by water and the syringe rebound), a synovectomy was performed using an arthroscopic shaver or vaporisation, capsular scar tissue was described by the authors as being debrided and then gentle manipulation was performed. This involved arm elevation in the scapular plane (which was usually associated with audible popping of the contracted capsule) and external rotation followed by internal rotation at varying degrees of abduction. This was done with gradual pressure and stopped if unyielding resistance was met. The authors stated that repetition of these steps led to tearing of the capsular structures. The arm was kept in abduction–external rotation for 2 days following the procedure.

The Austgulen et al. study85 reported that shoulder arthroscopy was performed using normal technique, and the frontal capsule, glenohumeral ligament and coracohumeral ligament were split from the bicep tendon to 6 o'clock. The authors stated that the subacromial space was inspected and adherences loosened. Where the spaces were narrow subacromial decompression was performed until the shoulder could move at maximal outward rotation and at 180° in abduction.

Both studies included a programme of physiotherapy following the procedure. Austgulen et al.85 specifically stated that the patients received aggressive rehabilitation from the first day after surgery (no further details reported). This study also reported that participants were asked to perform home exercises (including stretches) every day. In addition, concomitant NSAIDs were given as needed and postoperative oxycodone was given occasionally.85 Chen et al.86 did not state whether a home exercise routine was undertaken or whether concomitant medication was given.

The inclusion criteria differed between the studies, with Chen et al.86 reporting limited criteria (i.e. basic criteria for diagnosis of idiopathic frozen shoulder). Additionally, Austgulen et al.85 required that participants had already tried physiotherapy but without a satisfactory result. The nature and intensity of the previous physiotherapy was not reported. The diagnostic criteria also varied between studies. In the Austgulen study85 diagnosis was based around limited range of movement and confirmed under anaesthesia. Chen et al.86 used clinical history, physical examination, radiography and arthrography. Neither study reported the stage of frozen shoulder; the average duration of frozen shoulder at baseline was 13 months (range 3 to 60 months) in the Austgulen study85 and 8 months (range not reported) in the Chen study86 (it was unclear whether this was mean or median). A total of 17% of the patients in the Austgulen study had diabetes85 but this information was not reported in the study by Chen et al.86 In both studies participants had received previous treatment for frozen shoulder. Austgulen et al.85 reported that participants had received physical therapy, whereas Chen et al.86 reported that none of the participants had responded to conservative treatment of at least 12 weeks' duration. The average age (mean or median not specified) of participants was 53 years in the Austgulen study,85 whereas Chen et al.86 reported that ages ranged from 32 years to 79 years. The percentage of women was similar in both studies: 67%85 and 75%.86

Quality assessment

Both studies were case series, which are inherently at high risk of bias because of the lack of a control group. The Bradford Hill criteria (a group of minimal conditions necessary to provide adequate evidence of a causal relationship between an incidence and a consequence) state that there must be a temporal relationship between the incidence of the factor (in this case surgery) and consequence (recovery). For a condition such as frozen shoulder, which for most people will resolve over a 1- to 3-year period, this relationship cannot be ascertained without a control group.87 Austgulen et al.85 was considered better quality as this study reported inclusion criteria, the population appeared representative of the frozen shoulder population, SDs were reported and patients were recruited prospectively, although the extent of this was unclear. The number of dropouts was unclear. Chen et al.86 met only one criterion (reporting of inclusion criteria); the remainder were unclear or not met. There were no dropouts in this study. Further details of quality assessment are available in Appendix 8.

Outcomes available

Data were sparse for all outcomes assessed (Table 50). The effect on short-term pain was described narratively in Chen et al.86 Function and disability and relevant measures of range of movement (external rotation, internal rotation and abduction; it was not specified whether these were active or passive) were reported at baseline and follow-up and as average gain in range of movement by Chen et al. (no further details available).85,86 Additionally, Austgulen et al.85 reported data for ability to work, physical activity, sleep at night and satisfaction. Both studies reported adverse events. Numerical results are summarised in Table 51. The complete outcome data for these studies are available in Appendix 7.

TABLE 50. Outcomes assessed in studies evaluating capsular release.

TABLE 50

Outcomes assessed in studies evaluating capsular release.

TABLE 51. Results of the studies of capsular release.

TABLE 51

Results of the studies of capsular release.

Pain

Only one study reported any data regarding pain. Chen et al.86 reported that two-thirds of patients complained of pain exacerbation after arthroscopic brisement; 5% had pain duration of longer than a few weeks. Half of patients had pain relief within 1 month whether in motion or not. After 3 months all except eight shoulders were pain free in any direction of shoulder movement (data were not reported).

Function and disability

Austgulen et al.85 reported a significant improvement in the Oxford Shoulder score from a baseline mean of 41.0 (SD 7.5) to 18.4 (SD 7.3) postoperatively (p < 0.001), a scale for which the best possible outcome is 12 and worst possible outcome is 60 (Table 51).

Similarly, Chen et al.86 reported that an improvement in the mean modified ASES score from a baseline score of 41 (SD 13) to 87 (SD 11) was significant (p < 0.005) at follow-up (Table 51).

Austgulen et al.85 reported significant improvements from baseline to follow-up (p < 0.001) on a 10-point scale in mean working ability (baseline 2.4, SD 2.6; follow-up 7.4, SD 2.5), mean physical activity (baseline 2.3, SD 2.5; follow-up 7.4, SD 2.4) and sleep at night (baseline 1.7, SD 2.5; follow-up 7.2, SD 2.6). This study also reported high satisfaction postoperatively (8.6, SD 1.6 on a 10-point scale) (Table 51).

Range of movement

Austgulen et al.85 reported that there was a significant improvement in mean external rotation from 3° (SD 5°) to 39° (SD 23°) (p < 0.001) and in mean abduction from 34° (SD 8°) to 154° (SD 37°) (p < 0.001) postoperatively.

Chen et al.86 reported a similar average gain in external rotation of 35°. The gain in internal rotation was 30°.

Adverse events

Austgulen et al.85 reported that two patients had frozen shoulder again and had repeat surgery. No deep infections, nerve damage or other complications were reported. Chen et al.86 found that only one patient experienced complications (superficial wound infection). No further information regarding how adverse events had been assessed was reported in either study.

Summary

Two studies investigated capsular release. Both were case series and therefore inherently at high risk of bias. One of the studies stated that debrisement was a component of the capsular release procedure and both reported that physiotherapy was undertaken post procedure.

One case series reported that the majority of patients were pain free at movement at short-term follow-up. Statistical significance was not reported. There was evidence of benefit for function and disability from both studies. Each study reported a significant improvement in one function and disability scale. Additionally, one study reported significant benefits in working ability, physical activity, sleep at night and satisfaction with capsular release. One study reported evidence of significant benefit in two measures of range of movement with capsular release; the other study reported an improvement in two measures of range of movement, but did not report whether or not these were significant.

In conclusion, although the evidence available suggested potential benefit from capsular release, these studies were at high risk of bias and cannot be used to draw conclusions regarding the efficacy of this treatment for frozen shoulder.

Mixed-treatment comparison results

MTC was performed for pain only. As previously stated in the report, this was because pain was the only outcome for which a network was available.

Analyses were planned for pain recorded at ≤ 3 months (excluding pain recorded at ≤ 4 weeks from the start of treatment), > 3 months and ≤ 6 months, and > 6 months and ≤ 12 months. However, data were only available for ≤ 3 months.

Four networks were evaluated:

  • Network 1: studies of any intervention (i.e. conservative and invasive) and any quality.
  • Network 2: studies of any intervention that were at least of satisfactory quality (i.e. method of randomisation was adequate and outcome assessment was blinded). The overall quality of the available trials for the MTC was poor, with most not reporting on the a priori quality criterion of allocation concealment. So that a network of trials could be formed, method of randomisation rather than allocation concealment was used as a quality criterion in the MTC.
  • Network 3: studies of non-invasive treatments of any quality, excluding sodium hyaluronate (because it is rarely used in the NHS).18
  • Network 4: studies of non-invasive treatments that were at least of satisfactory quality (i.e. method of randomisation was adequate and outcome assessment was blinded), excluding sodium hyaluronate.

Interpretation of standardised mean differences

Reduction in pain recorded as a SMD can be transformed back into the units of one or more of the pain measurement instruments. The decision was made to back transform to a VAS 0–100 mm as this measure had more substantial information than other pain measures. This included one study performed in a shoulder-specific population.50 Transformation from SMD to VAS was made using a representative pooled SD at baseline from one of the included trials.51 No data could be located on the MCID in pain for individuals with frozen shoulder. However, it is suggested that a difference of 14 mm on a 100-mm VAS scale is the MCID for individuals with rotator cuff disease, a shoulder disorder also characterised by pain and restricted range of movement.50 Given that the between-group MCID is thought to approximate 40% of that within individuals6,53 the MCID between groups for rotator cuff disorder can be estimated to be approximately 5.6 mm. This value is therefore used as a proxy for the between-group MCID for frozen shoulder.6 The reduction in pain recorded as a SMD can be interpreted on a VAS 0–100 mm scale using Table 52.

TABLE 52. Standardised mean differences and equivalent reductions on a VAS 0–100 mm scale.

TABLE 52

Standardised mean differences and equivalent reductions on a VAS 0–100 mm scale.

From the 31 included trials, the total number of trials potentially available for analysis was nine. The reasons for trials being excluded from the analysis are presented in Table 53.

TABLE 53. Reasons for trials being excluded from the MTA.

TABLE 53

Reasons for trials being excluded from the MTA.

The nine trials and the 10 interventions they evaluated are shown in Table 54. The placebo used varied between studies and included saline injections, lidocaine injection, placebo laser, sham distension and sham ultrasound. The four studies of physical therapy consisted of physiotherapy with an active mobilisation component. Physical therapies without an active component consisted of laser therapy and interferential therapy. Four studies included steroid injections; these were triamcinolone hexacetonide (40 mg once, 20 mg once), methylprednisolone (40 mg once for 3 weeks) and dexamethasone (2 mg once weekly for 5 weeks). Steroid injections were either given alone (three studies35,41,70) or combined with physiotherapy (three studies35,41,42). Seven of the nine studies included in the review had a home exercise component.16,35,39,4143,51

TABLE 54. Network table.

TABLE 54

Network table.

In these nine trials, a variety of pain outcomes and pain scales were reported (Table 55). If a trial reported more than one pain outcome or used more than one scale, only the pain outcome/scale prioritised for use in the analysis (see Chapter 2, Mixed-treatment comparison) is reported.

TABLE 55. Type of pain reported, pain scale used and quality of studies available for MTC analysis.

TABLE 55

Type of pain reported, pain scale used and quality of studies available for MTC analysis.

To check that the non-informative priors used in the analysis were truly non-informative a sensitivity analysis was performed for each network by changing the range of the uniform prior distribution on the SD from 0–2 to 0–0.8, 0–5, 0–10 and, as a final check for stability for network 1, 0–15. An additional sensitivity analysis was also performed by repeating each analysis using a burn-in of 30,000 iterations.

All four networks were evaluated; however, the lack of evidence available to inform networks 2, 3 and 4 led to a lack of stability in the models, particularly a lack of stability in between-study variance. For completeness these analyses have been presented in Appendix 10 although no conclusions can be drawn from the results presented. Only the results for the network of any intervention and of any quality (network 1) are presented here. The slightly higher number of trials included in this network allows the model to achieve stability.

Network 1: studies of any intervention and of any quality

There were nine trials with data that could be used in the analysis of pain at or close to 3 months. These nine trials formed a connected network of nine interventions compared with placebo. The network is presented in Figure 21.

FIGURE 21. Network diagram of studies of any intervention and any quality.

FIGURE 21

Network diagram of studies of any intervention and any quality. PT, physical therapy.

Each line represents one comparison. For example, there were two studies available for the comparison of steroid with placebo. Where there is no line there were no studies available, for example physical therapy without mobilisation versus arthrographic distension and steroid.

The models were run for 100,000 iterations and were not sensitive to the length of burn-in. Stable estimates of treatment effect, CrIs and between-study heterogeneity were produced by the models using a uniform prior of (0, 5), (0, 10) and (0, 15). Convergence, determined by graphical traces and the Brooks–Gelman–Rubin statistic, was achieved. The results of the uniform prior (0, 5) are presented here. Results for uniform priors (0, 10) and (0, 15) are available in Appendix 10.

The model was a good fit as the residual deviance was close to the number of data points. Table 56 presents the treatment effects compared with placebo in order of mean effectiveness. Steroid combined with physiotherapy showed a beneficial treatment effect compared with placebo with 95% CrIs that did not cross the line of no effect. The 95% CrI of the treatment effect of physical therapy without mobilisation and physiotherapy combined with placebo marginally crossed the line of no effect. The remaining interventions (electroacupuncture, steroid injection, physiotherapy, MUA plus physiotherapy, sodium hyaluronate and arthrographic distension) all crossed the line of no effect. There was no clear difference between the interventions in treatment effects, that is, the CrIs overlapped. These results are also displayed graphically in Figure 22.

TABLE 56. Mean reduction in pain compared with placebo with CrIs in order of mean treatment effectiveness for pain at 3 months, including studies of any quality.

TABLE 56

Mean reduction in pain compared with placebo with CrIs in order of mean treatment effectiveness for pain at 3 months, including studies of any quality.

FIGURE 22. Caterpillar plot of the SMD estimates and their 95% CrIs compared with placebo.

FIGURE 22

Caterpillar plot of the SMD estimates and their 95% CrIs compared with placebo. The treatments associated with each number can be identified from Table 56.

Steroid combined with physiotherapy showed a clinically significant as well as statistically significant reduction in pain, with a 95% CrI that corresponded to a clinically significant reduction in pain only: −28.9 mm (95% CrI −54.1 to −7.7 mm). Although the MD was clinically significant for most of the other comparisons, the 95% CrIs contained values that were clinically insignificant. For example, steroid injection had a MD of −21.0 mm (95% CrI −45.0 to 3.5 mm).

Opportunities for comparison with standard meta-analysis are limited as most of the studies were not pooled in a quantitative synthesis. It was possible to compare with the standard meta-analyses for steroid plus physiotherapy versus placebo and steroid versus placebo. The results of the MTC are consistent with the standard meta-analysis of steroid combined with physical therapy versus placebo presented earlier in this report. This standard meta-analysis contained the same two studies included in the MTC. The results of the meta-analysis also showed that steroid combined with physiotherapy resulted in both a statistically and a clinically significant reduction in pain compared with placebo (pooled SMD −0.98, 95% CI −1.43 to −0.52; back transformed MD 17.93 mm, 95% CI −26.2 to −9.5 mm).

Assumptions

In a MTC there are three key assumptions: homogeneity, consistency and methodological similarity.

Homogeneity

The assumption of homogeneity is that trials are sufficiently homogeneous to be quantitatively combined. In the analysis presented, placebo included sham distension, saline injections and placebo laser. Although we believed that it was appropriate to pool different types of placebo, it could be argued that these are not similar enough because of the differing degree of invasiveness. Furthermore, there were differences in the populations of the included studies.

For all studies of conservative treatments, it was not reported whether participants had received any previous treatment for frozen shoulder. However, for both studies of invasive treatments (MUA combined with physical therapy, and arthrographic distension with steroid) it was reported that at least some participants had undergone previous treatments for frozen shoulder including corticosteroid injections or some form of physical therapy. This would suggest that the populations of the invasive studies had a longer duration and possibly more treatment-resistant forms of frozen shoulder than those in the conservative treatment studies.

Consistency

The assumption of consistency is that had treatment C been included in a trial comparing interventions A and B then the treatment effect dAC would be equivalent to that obtained from a trial of interventions A and C.33 Assuming consistency, the treatment effect dAC is the sum of the treatment effects dAB and dBC. However, this assumption requires that the treatments involved belong to closed loops in the network of evidence, such as that seen in a connected polygonal structure.88 Given that the structure of the network presented here was of a radiating star, with only one loop of evidence, it was not feasible to assess consistency.

Methodological similarity

Of the nine studies included in the analysis, five were of at least satisfactory quality (method of randomisation was adequate and outcome assessment was blinded)16,35,39,41,43 and four were not.42,51,70,79 To assess the effect of quality on the estimates, a network of studies that were at least of satisfactory quality was evaluated (network 2). However, this model was unstable and, although presented in Appendix 10, the reliability of the data included in the analysis presented is uncertain and does not allow a robust assessment of the impact of quality.

Summary

Nine interventions formed part of a connected network with placebo and the evidence was informed by all nine trials. Twenty-two studies had to be excluded from the MTC for a range of reasons; therefore, the network was a small subset of the available evidence. Steroid combined with physiotherapy showed a beneficial treatment effect compared with placebo with 95% CrIs that did not cross the line of no effect. Furthermore, this treatment, when back transformed to a VAS 0–100 mm scale, showed a clinically significant reduction in pain with 95% CrIs that excluded clinically insignificant values. The remaining treatments had 95% CrIs that crossed the line of no effect and which contained clinically insignificant values. Overall, there was no clear difference in the treatment effects of any of the interventions, that is, the CrIs overlapped. Although the model was a good fit and gave stable estimates, given the arguable heterogeneity of the placebos and of study populations, and the inclusion of poor-quality studies, the reliability of the results of the MTC is uncertain and they should be interpreted with caution.

Patients' views of interventions for frozen shoulder

Searches of three databases (CINAHL, MEDLINE and PsycINFO) identified 1067 potentially relevant references (Figure 23). Nine full papers were ordered on the basis of screening titles and abstracts. Of these, two papers could not be obtained within the time frame required; however, it is unlikely that these would have met the inclusion criteria. One paper appeared to focus on the clinical effectiveness of steroid injections89 and the remaining paper appeared to examine the quality of life of patients with frozen shoulder but did not appear to address treatments.90 None of the remaining six papers met the inclusion criteria; therefore, no data were available on patients' views regarding interventions for frozen shoulder.

FIGURE 23. Study selection.

FIGURE 23

Study selection.

Economic analyses

Previous economic evaluations

We identified one full economic evaluation, conducted in the Netherlands, that met our inclusion criteria as outlined in Chapter 2 (see Inclusion and exclusion criteria).91 The included economic evaluation, a cost–utility analysis, was conducted as part of a clinical study40 comparing HGMT with LGMT in a Dutch frozen shoulder population. The clinical study, reported in a separate paper, also met the inclusion criteria for the effectiveness review.40 In summary, the clinical study randomised 100 patients to either HGMT or LGMT. Patients were treated twice weekly for 30 minutes during a period of 12 weeks and were encouraged to attend all treatment sessions. From 6 weeks onward, treatment could be reduced in frequency or stopped if a normal range of movement was noticed by the therapist. The study assessed a number of clinical outcomes including active and passive range of movement, shoulder disability (measured by a shoulder rating questionnaire and the SDQ), pain and quality of life (measured by the SF-36). For a full assessment of the study details and quality see Physical therapy. The study was considered to be of reasonable quality, with some risk of bias.

The aim of the economic evaluation component of the study was to compare the two techniques in terms of costs and QALYs from a societal perspective. In addition, the authors also estimated the impact of frozen shoulder on costs and health. Costs were assessed over the 12-month follow-up period and presented as undiscounted annual costs in euros for the price year 2004. All resource-use information was collected from the patients using quarterly cost questionnaires. Wherever possible Dutch standard prices, which were designed to reflect social costs and standardise economic evaluations, were used. Where standard prices were not available, charges were used. Costs included treatment sessions, alternative medicine sessions, hospitalisations (MUA, acromioplasty), home nursing care, medication, travel costs and non-health-care costs (labour and domestic help). All appropriate costs appear to have been included for a societal perspective.

The measure of benefit used in the analysis was QALYS; these were based on the Short Form questionnaire-6 Dimensions (SF-6D) utility index values, which were estimated using SF-36 data collected alongside the study. The SF-6D provides a means for using the SF-36 data to estimate a preference-based single index measure for health using general population values, thereby allowing the authors to obtain QALYs.

Differences in outcome measures (SF-6D and costs) were tested using double-sided nonparametric bootstrapping. The average estimated QALYs were 0.695 for HGMT and 0.702 for LGMT. The difference of 0.007 in favour of LGMT was reported to be not statistically significant (p = 0.71, 95% CI −0.32 to 0.049). The total reported average annual societal costs were €8809 for HGMT and €6911 for LGMT (a cost difference of €1898 in favour of LGMT). The difference in costs was also reported to be not statistically significant (p = 0.37, 95% CI –€2551 to €5711). The authors concluded that the economic analysis does not allow for evidence-based recommendation regarding the preferred treatment.

A full assessment of the quality of this economic evaluation based on the Drummond checklist and the study data extraction/summary are presented in Appendices 11 and 12 respectively. There were a number of limitations to the study including the lack of an incremental analysis, which would not have resulted in an incremental cost-effectiveness ratio (ICER) but would have informed us that LGMT dominates HGMT (i.e. has higher effectiveness and lower costs). The authors dismiss the difference in QALYs and costs based on statistical inference, which is inappropriate. Further probabilistic sensitivity analysis would have allowed parameter uncertainty to be fully characterised and the expected value of further research to be considered. Overall, despite the limitations the analysis was of reasonable quality and the results, although uncertain, provide us with an indication that LGMT may be a more cost-effective option than HGMT.

Although this one economic evaluation provides us with some information regarding the cost-effectiveness of these two physiotherapy techniques, it does not fully address the question posed. The study was conducted in the Netherlands, was limited to two active physiotherapy techniques, failed to appropriately deal with uncertainty and presented utilities using SF-6D, which cannot easily be compared with the preferred EQ-5D measure because of differences in the descriptive systems and values applied to health states. EQ-5D has become the instrument of choice for many agencies including NICE.64 The use of one instrument allows more comparability across projects undertaken by these agencies. The small amount of evidence that we are able to glean from this analysis is useful but, given the lack of economic evaluations for other relevant interventions, to further develop and inform decision-making in the UK it was felt that it was essential for us to undertake some additional work.

Decision model

Based on the preliminary examination of the clinical effectiveness data it was anticipated that any modelling undertaken was likely to be in the form of a simple decision tree. As a tool for modelling decision trees provide an effective method for structuring a problem and combining data from various sources. Possible treatment pathways are laid out in a linear manner and appropriate costs and effects are assigned. The final outcome of interest, in this instance QALYs, is attached to the end of the tree. The proportion of participants achieving particular end points is calculated to give a measure of effect. Costs can be attached to both the end points and the treatments/events within the tree. Total costs for each intervention are calculated by summing the costs associated with each pathway. At each point at which a choice is available there is a chance node; a probability conditional on the previous event is attached to each chance node. This node determines the proportion of individuals who progress down each unique pathway in the tree. A simple example is shown in Figure 24.

FIGURE 24. Example decision tree.

FIGURE 24

Example decision tree.

We reviewed the literature to identify existing models in a similar population and sequences or order of treatments, to explore whether treatments (or sequences of treatments) were influenced by phase of condition and to identify any potential source data (e.g. unit costs, resources or utilities) to help populate the model.

As outlined earlier, we identified one economic evaluation as part of a clinical study; we did not identify any modelling studies in this area. Further, very few of the trials in the review of effectiveness consider phase of frozen shoulder or previous treatment history of participants. Because of the limited data available, it was not possible in the clinical synthesis to investigate the impact of phase of frozen shoulder on treatment effectiveness. The only potential ‘proxy’ for phase of condition was duration of frozen shoulder at baseline. The uncertainty associated with such an assumption would need to be carefully considered when drawing conclusions as there is a considerable individual variability in the duration of each phase: it is suggested that the painful freezing stage may last from 2 to 9 months, the adhesive phase from 4 to 12 months and the resolution phase from 12 to 42 months. It was felt that the poor reporting in many of the clinical trials made it difficult to support the assumption that mean duration of frozen shoulder at baseline could be used as a proxy for stage of condition.

Supplementary examination of the available literature failed to provide any clear indication as to potential treatment pathways or treatment sequences. The recent survey of UK health-care professionals suggested that there is a fairly constant view that treatments should depend on the phase of the condition and/or that a step-up approach should be adopted in terms of the degree of treatment invasiveness.18 With this in mind the clinical advisory group were presented with two alternative modelling scenarios that were thought to be potentially feasible. These were discussed in depth at a group meeting. Option one involved comparing all possible treatments for each phase of the condition (painful, adhesive and resolution) as mutually exclusive options, ignoring any treatment sequencing and phase of condition. This was considered unrepresentative of clinical practice where treatments would often be given simultaneously or in sequence. The results of this type of model would be uninformative and misleading. Option one was rejected as a viable modelling scenario. Option two involved sequencing treatments dependent on their invasiveness. This would potentially allow treatments such as physiotherapy followed by steroid injections followed by MUA to be evaluated. In essence, all treatments, in all possible sequence order, would be considered. This modelling option required an assumption that the clinical effectiveness of each treatment, as obtained from the included clinical trials, would remain the same regardless of where it was placed in a sequence of treatments or what treatments had been received before it. Again, it was the view of the advisory group that this was not a realistic assumption and in fact the effectiveness of treatments is likely to be linked to the position of the treatment in the sequence. The general consensus within the advisory group surrounding this issue was that the assumptions that would be required were, from a clinical perspective, inappropriate.

Given the lack of relevant data from the trials included in the review, the consensus from the advisory group that the required assumptions were clinically inappropriate and the confirmatory position regarding the lack of consensus on recommended treatment pathways in recently published articles,18 it was decided that it was not feasible or useful in this instance to construct a decision-analytic model. In fact, it was perceived that any attempt to undertake modelling would provide hugely uncertain results, which would not allow any conclusions to be made.

Given the lack of UK data on physiotherapy from the identified economic evaluation91 and the lack of economic evaluations of the other relevant interventions we opted to elicit from our clinicians on the advisory group details of the resource use associated with the modalities identified in the review for a frozen shoulder population. The available evidence was then used to obtain utility valuations, which we present as exploratory cost-effectiveness analyses. Our intention, had a de novo model been developed, was to undertake the analysis from an NHS perspective. This perspective has been maintained for the investigative costing exercise. Resources estimated have been valued using national average unit costs (pounds sterling) at 2009–10 prices where possible. Resource use, unit costs and total cost are presented separately, where possible, for each individual treatment. The estimation of resource use gives us only part of the evidence requirements for an economic analysis. The presentation of cost data alone should allow some conjecture about the economic implications of alternative treatments, but does not provide the information required to inform a decision on the cost-effectiveness of these alternatives. To help inform such decisions, the cost estimates obtained from this exercise will be combined with our derived health-related quality of life (HRQoL) estimates (see Mapping from the pain VAS and SF-36 onto the EQ-5D) and presented as ICERs. An ICER is calculated as the difference in the expected cost of two interventions, divided by the difference in the expected QALYs produced by the two interventions. To inform which therapies are of greatest clinical as well as economic value, cost-effectiveness results are usually compared against an acceptable threshold of cost-effectiveness to determine whether a health-care intervention is cost-effective, and a good use of resources, or one that represents poor value for money. The cost-effectiveness threshold for interventions in the English and Welsh NHS is suggested by NICE to be £30,000 per additional QALY; therefore, we have used this threshold.92

The following sections will outline:

  1. resource-use estimation and costing
  2. mapping from clinical outcomes to EQ-5D
  3. estimation of QALYs and ICERs.

Resource-use estimation

Steroid injections

Six RCTs investigated steroid injections in the treatment of primary frozen shoulder, either alone or in combination with physical therapy.35,41,42,6668 Steroid injections can be delivered by a number of different practitioners using different approaches. We have broken these down into a guided injection delivered by an orthopaedic surgeon, a rheumatologist or a radiologist in a hospital setting; unguided injection delivered by a physiotherapist or a GP in a non-hospital setting; and unguided injection delivered by a physiotherapist, an orthopaedic surgeon or a rheumatologist in a hospital setting. Given the nature of the condition and the individual nature of treatments, it was not possible to be precise about the number of injections that an average individual may receive. It was also suggested by our advisory group that the number of patients receiving guided injection within the NHS was likely to be low.

Unguided injections

The clinical experts suggested that it would be rare for individuals to receive more than three injections and on average they would receive two injections spread over various durations of time. Table 57 provides the unit costs of the resources that are used when delivering a steroid injection outside of a hospital setting. We have not included the cost of consumables as these costs are likely to be negligible. The six trials evaluating injections identified in the review of effectiveness assess the use of dexamethasone alone or in conjunction with lidocaine, or triamcinolone alone or in conjunction with lidocaine. We have elected to present costs for both types of steroid although both of the UK trials identified in the review evaluated triamcinolone.41,67

TABLE 57. Unit costs for unguided steroid injections.

TABLE 57

Unit costs for unguided steroid injections.

Alternative scenarios for the delivery of unguided steroid injections were considered and full details of these are presented in Appendix 13. The base case considered three sessions, an initial session of 20 minutes' duration when an assessment would be undertaken plus two injection sessions of 10–20 minutes' duration. Using an average of 15 minutes for the two injection sessions, and assuming a combined injection of triamcinolone and lidocaine, the total cost if delivered by a physiotherapist is £36.18. The total cost if delivered by a GP is £108.68. The choice of steroid injection for the base case was based on the UK trials identified as part of the effectiveness review.41,67 The use of a premixed ready-to-inject vial of dexamethasone plus lidocaine would increase the total cost to £39.06 if delivered by a physiotherapist and £111.56 if delivered by a GP.

Unguided injections are also delivered within a hospital setting. We further estimated that the costs associated with unguided injections in the hospital setting would be £27.17 if delivered in a hospital physiotherapy unit, £88.01 if delivered in an orthopaedic setting and £138.51 if delivered in a rheumatology setting.

Guided injections

Guided injections were assumed to be delivered in a hospital setting by key personnel including an orthopaedic surgeon, a rheumatologist or a radiologist. We estimated that the same number of sessions would be required, using the same type of steroid. The variation in the cost of the guided injections is driven by who delivers the injection. For the base case we have again used the cost of a combined injection of triamcinolone and lidocaine, but present the cost of the premixed dexamethasone plus lidocaine to allow a range to be expressed. We have also assumed that on average two injections will be given over three visits. Alternative scenarios showing varying resource-use levels (suggested alternative numbers of sessions/injections) are presented in Appendix 13. Unit costs for guided injections are presented in Table 58.

TABLE 58. Unit costs for guided steroid injections.

TABLE 58

Unit costs for guided steroid injections.

The total cost of injections delivered within the orthopaedic setting ranges from £299.68 to £302.56 depending on the steroid injected. Injections delivered in the rheumatology setting have a total cost ranging from £472.68 to £475.56. These alternative delivery settings have been presented as it is believed that these are all viable options depending on available NHS services within a local area.

Steroid injection plus physiotherapy

The resource use associated with the combined treatment of steroid injection plus physiotherapy is again related to the setting in which the treatments are given and the method of steroid delivery. A number of scenarios are presented (see Table 59 for relevant unit costs):

TABLE 59. Unit costs for active physiotherapy.

TABLE 59

Unit costs for active physiotherapy.

  • Scenario one: six physiotherapy sessions with an active mobilisation component and two unguided steroid injections all delivered by a physiotherapist. It has been assumed that no additional time was required and the total cost is equal to the total cost of active physiotherapy plus the steroid injections. The total cost ranges from £130.43 to £133.31 if delivered by a community physiotherapist, or from £121.43 to £126.81 if delivered by a hospital-based physiotherapist, depending on choice of steroid.
  • Scenario two: six physiotherapy sessions with an active mobilisation component delivered by a community-based physiotherapist, plus two unguided steroid injections delivered by a GP over three visits. The total cost of this scenario ranges from £235.75 to £238.75, depending on choice of steroid.
  • Scenario three: six physiotherapy sessions with a mobilisation component delivered by a hospital-based physiotherapist, plus two unguided steroid injections delivered by a GP over three visits. The total cost of this scenario ranges from £226.75 to £238, depending on choice of steroid.
  • Scenario four: six physiotherapy sessions with a mobilisation component delivered by a community-based physiotherapist, plus two guided steroid injections delivered within the orthopaedic setting. The total cost ranges from £420.23 to £423.31, depending on choice of steroid.
  • Scenario five: six physiotherapy sessions with a mobilisation component delivered by a community-based physiotherapist, plus two guided steroid injections delivered within the rheumatology setting. The total cost ranges from £605.43 to £607.31, depending on choice of steroid.
  • Scenario six: six physiotherapy sessions with an active mobilisation component, plus two guided steroid injections delivered by a hospital-based physiotherapist. The total cost ranges from £160.43 to 165.81, depending on choice of steroid.

Alternative scenarios could be estimated by varying the number of injections, delivery method and setting. However, the base-case scenarios presented reflect what we believe are current options in the NHS.

Sodium hyaluronate injection

Three trials evaluating sodium hyaluronate were identified in the effectiveness review.66,69,70 In a recent UK-based survey of treatment recommendations by health-care professionals involved in managing frozen shoulder, sodium hyaluronate injection was recommended by < 1% for treating the early ‘painful’ phase or ‘resolution’ phase of frozen shoulder.18 It was further suggested by the clinical advisory group that one injection would typically be delivered over two sessions in a hospital setting. The injection was likely to be delivered by a consultant. However, the costs were not available for sodium hyaluronate for provision under the NHS. It was felt that this intervention was likely to be delivered in a hospital setting, but we were unable to ascertain any costs associated to its delivery that would be borne by the NHS.

Physical therapy

The broader range of physical therapies are not available through the NHS and therefore have not been costed. Two categories of physiotherapy were costed, physiotherapy with mobilisation (active) and physiotherapy without mobilisation. It was felt by the advisory group that physiotherapy without mobilisation would not be delivered on its own in an NHS setting, but may be used to complement another therapy. Therefore, for the purpose of resource-use estimation, physiotherapy without mobilisation has been considered only as an add-on delivered alongside other therapies.

Active physiotherapy

It was estimated by the advisory group that physiotherapy with a mobilisation component would on average comprise six sessions each of 30 minutes' duration, followed by a final review session of 15 minutes' duration. Full details of all estimates of resource use and any plausible ranges are presented in Appendix 13.

As can be seen from Table 59, the cost of physiotherapy is dependent on the setting in which the treatment is delivered. We have used all variations in hourly cost to calculate a plausible range of total cost for the delivery of active physical therapy. Within a community setting, active physical therapy is estimated to cost a total of £126.75. Within a hospital setting the same treatment is estimated to have an average total cost of £98.75.

Physiotherapy without mobilisation

The review of effectiveness identified one trial that was classified as ‘physiotherapy without mobilisation’.16 The trial evaluated laser versus placebo laser (plus home exercise). Given that it is unlikely that laser therapy would be delivered without some form of mobilisation (active physiotherapy) within the NHS, we have assumed that laser therapy would be provided in conjunction with an active mobilisation, adding an additional 15 minutes to a 20- to 30-minute session. The additional cost reflects what we believe is current practice, that is, the delivery of both active mobilisation and other passive modalities such as heat treatment in the same treatment session. The additional cost would be either £49.38 or £63.38 depending on the setting.

Acupuncture

Three trials evaluating acupuncture were identified in the effectiveness review.7981

It was felt by the clinical advisory group that acupuncture, like other physiotherapies, would in clinical practice rarely be given in isolation of other treatments. We have therefore assumed that acupuncture would be delivered in physiotherapy sessions, in either a hospital or a community setting, over the same duration as active physiotherapy, that is, six sessions of 20–30 minutes' duration at a total cost ranging from £117.75 to £126.75.

Manipulation under anaesthesia

The review identified four RCTs investigating MUA in the treatment of primary frozen shoulder.38,39,82,83 The resource implications of MUA were difficult to firmly establish. The actual cost of MUA has been estimated at £424 (range £191–500) based on the HD24C code, which was obtained from the NHS Reference Costs.95 In addition, there is a need for extensive rehabilitation physiotherapy, which was estimated on average to consist of two 30-minute sessions per week for 4 months, plus a final review, which would take place with the orthopaedic surgeon. Therefore, in total, the estimated cost of MUA is £1446. If steroid injections are also used this cost would increase accordingly.

Arthrographic distension

Three trials evaluating arthrographic distension with steroid injection were identitied in the effectiveness review.36,43,84 Arthrographic distension is assumed to be delivered in the hospital setting by a radiologist or an orthopaedic surgeon and involves the use of radiological imaging (e.g. fluoroscopy) to ensure that the injection is accurately placed. The procedure was assumed to require around 15 minutes based on the clinical literature. Because there are no UK trials in the clinical review it is unclear how many visits for injections might be needed. We have therefore conservatively estimated the cost as involving one hospital outpatient visit with injection. The estimated cost of this procedure has been derived from the NHS Reference Costs95 at £113 (range £78–133) based on code RA16Z, plus the cost of a combined injection of triamcinolone and lidocaine (£1.49 + £0.35), giving a total of £114.845 (range £79.84–134.84).

Capsular release

Two studies evaluating capsular release were identified in the effectiveness review.85,86 Capsular release is performed in the hospital setting and delivered by a surgical team, which is likely to comprise anaesthetist, orthopaedic surgeon, anaesthetic nurse, scrub nurse and assistant and recovery nurse. The procedure is likely to take 30–45 minutes, followed by a further 15 minutes for recovery from anaesthetic. The estimated cost of this procedure has been derived from the NHS Reference Costs95 at £1182 (range £787–1489) based on code HB62C. Like MUA, capsular release requires a follow-up of intensive physiotherapy. We have estimated that the physiotherapy following capsular release would be the same as that following MUA. On average this would consist of two 30-minute sessions per week for 4 months, plus a final review that would take place with the orthopaedic surgeon. Therefore, in total, the estimated cost of capsular release is £2204 (range £1809–2511).

Summary

Average costs were estimated for the interventions included in the review. These ranged from £36.18 for an unguided steroid injection to £2204 for capsular release. Where possible, these estimated average costs will be used to inform tentative cost-effectiveness ratios, which will be presented later in the report. They will be combined with HRQoL data estimated by means of a mapping exercise, which is presented in the following section. The aim is to estimate HRQoL by mapping from three clinical outcomes to EQ-5D using an available data set. The main focus of the mapping is to explore the nature and direction of the relationship between different clinical and HRQoL outcome measures. We will then use the EQ-5D estimates to obtain HRQoL outcomes for a subset of clinical trials identified in the review that have presented data on the three relevant outcomes.

Mapping from the pain VAS and SF-36 onto the EQ-5D

The clinical trials identified in the effectiveness review report outcomes using a variety of measures including pain (overall, at rest, on movement, at night), range of movement (e.g. internal and external rotation, elevation), function and disability, quality of life, time to recovery and return to work and recreation, and adverse events. Of the selection of measures used in the trials, none was designed to generate health state utility values, which are required to calculate QALYs, which can then be used to assess cost-effectiveness. Outcomes from treatments affect two basic components – the quantity and quality of life – both of which are captured in the QALY. There are a variety of standardised ‘off-the-shelf’ preference-based measures available to derive estimates of HRQoL, such as the EQ-5D, SF-6D, Health Utilities Index (HUI) and 15-dimension utility measure (15D).65,96,97,98 These HRQoL values can subsequently be incorporated in economic models for performing cost-effectiveness analyses. They are the preferred benefit measure when undertaking any cost-effectiveness analysis.

An additional search of MEDLINE was conducted to identify health quality of life outcomes that could be used to inform any modelling. The lack of use of generic preference-based measures in existing frozen shoulder clinical studies is a potential current barrier to populating economic models with the best evidence on effectiveness and to thereby allow broader comparisons across different diseases.

The included economic evaluation presented HRQoL in the form of the SF-6D, but it was felt that given the general preference for EQ-5D and the possibility of bias in the clinical study further investigation was warranted. The EQ-5D is the recommended instrument of choice for use in cost-effectiveness analyses by NICE and is the most commonly used instrument.

In the context of this clinical review, it is not clear why HRQoL has not been included in the battery of outcome measures in the clinical trials in frozen shoulder populations to date. Possible reasons why such instruments have not been administered may be because it is not practical or they are too time-consuming for patients to complete in trials in addition to all of the other forms/questionnaires that are used – often across multiple time points – or even because studies were not initially designed to focus on economic questions. It may be that the general perception is that these generic instruments are not sensitive enough.

In a situation such as this, in which such instruments have not been used in clinical studies, a pragmatic alternative is to apply a mapping function to convert non-preference-based health data into one of the generic preference-based measures. Mapping essentially provides a statistical equation that allows the estimation of health utilities and the subsequent calculation of QALYs in clinical studies that do not use any preference-based HRQoL instruments. In the published literature, the EQ-5D was recently reported as being the most popular target measure used for mapping.96

On the other side of the mapping equation, the most widely used generic measures/instruments that have been mapped include the SF-36 and SF-12 health surveys, which offer the option of presenting scores across eight dimensions or two summary measures derived from the eight dimensions. These two summary measures are the physical component and the mental component.99,100 Also, a number of shoulder-specific questionnaires (scores) have been shown to be significantly correlated with the EQ-5D in the literature, in particular the SDQ.55 Of particular importance in the current clinical effectiveness review were patient-assessed pain intensity, function and disability, quality of life and range of movement. Pain was the most common outcome measure in the included clinical studies and, for instance, in the MTC analysis, pain was analysed based on the following preference hierarchy: overall pain, pain at night, pain on activity, any other type of pain (excluding pain at rest) and pain at rest. Furthermore, and within the context of the current clinical effectiveness review, three of the studies reported the SF-36 as an outcome measure. As mentioned earlier in the clinical sections of the report, two studies meeting the full inclusion criteria reported the mean physical and mental component summary scores.31,47 One additional study, which did not meet the analysis criteria, reported the means for the eight dimensions of the SF-36.81 The outcomes were reported over the time horizon of 4 weeks and as a result the study was excluded from further analysis. Therefore, in line with the decision made as part of the clinical review we elected not to map from this evidence base, but rather to utilise the potentially more reliable component summary scores that had been reported in the other two trials.35,51

In a recent review of mapping studies, simple additive models that employed a utility index score as the dependent variable and the main effects of either total or instrument dimension scores as independent variables performed almost as well as more complex models comprising many variables.99 With this in mind, and as the main focus of this analysis was to explore the nature and direction of the relationship between different outcome measures, we chose to follow a simple approach and use mean pain VAS score (as it was a main outcome in the clinical review) and mean SF-36 physical and mental component summary scores (as it is one of the most widely used generic quality-of-life measures in mapping studies). These were important outcomes in the clinical effectiveness review.

To undertake an exploratory mapping analysis of this nature we required a data set in which the two measures being mapped had been administered specifically to a frozen shoulder population. We were unable to obtain a data set of this nature for a frozen shoulder population; however, we obtained a UK data set involving patients with rotator cuff disease from the SAPPHIRE RCT.101 Both measures chosen for the exploratory mapping analyses were already calculated and readily available in the SAPPHIRE data set.102

The selection of outcomes on which to map was a pragmatic decision based on the data available in the SAPPHIRE data set. The relationship is estimated using a range of techniques and statistical specifications. Furthermore, we compare the mapping approaches tested in terms of their predictive performance in the SAPPHIRE trial population. Within the context of the current review, we conducted an analysis to explore whether or not relationships between the pain VAS and SF-36 and the EQ-5D exist and the nature of those relationships. In doing so, we considered that we might be able to demonstrate the need for the collection of utility EQ-5D data in future clinical trials and offer a potential alternative approach for frozen shoulder populations for whom there is currently a lack of HRQoL data.

The SAPPHIRE trial compared the efficacy of corticsosteroid (40 mg/ml triamcinolone acetonide) versus local anaesthetic (1% lidocaine) injection in subjects presenting to GPs with (acute) shoulder pain (in one or both shoulders) persisting no longer than 12 months, and with diagnosis of rotator cuff tendonitis.102 The trial evaluated the impact of giving GPs additional training in diagnosing rotator cuff pathologies and in performing shoulder injections, or no training. Trial participants were followed up at 1 and 3 months, with a final follow-up at 6–12 months post randomisation. The outcome measures were the SF-36, EQ-5D, British SDQ103 and VAS to assess the extent of shoulder pain (at night, in daytime at rest, in daytime on movement).

The population characteristics for the two studies included in the clinical review that reported mean overall pain VAS scores,16,51 using a 0- to 100-mm scale, and the two studies included in the clinical review that reported mean SF-36 physical and mental component summary scores35,51 are reported in Table 60. The population characteristics of the participants in the SAPPHIRE trial102,104 are also shown in Table 60.

TABLE 60. Patient baseline characteristics and mean reported scores.

TABLE 60

Patient baseline characteristics and mean reported scores.

The three studies included in the clinical effectiveness review reporting mean pain VAS scores and/or mean SF-36 physical component summary (PCS) and mental component summary (MCS) scores and the SAPPHIRE trial include populations with broadly similar proportions of men and women and of a similar age and duration of shoulder condition or pain and with similar mean pain VAS and SF-36 PCS and MCS scores.

The mapping presented here is an exploratory exercise, based on a patient population with very similar baseline characteristics as the frozen shoulder population. We examine the relationship between the EQ-5D and pain as measured by a VAS 0–100 mm and the EQ-5D and SF-36 (specifically the SF-36 PCS and MCS) health surveys.

Methods

The following sections present the methods used in estimating HRQoL data to inform our tentative analysis and potentially inform other economic evaluations.

Instruments

Pain visual analogue scale Pain was measured on a scale from 0 to 100 mm in both the studies included in the clinical review (see Table 60) and the SAPPHIRE data set. Values anchored on the VAS at 0 generally represent no pain whereas values anchored at 100 generally represent the worst pain possible. Both studies have data available as mean values. When more than one type of pain was measured in any single study, data were selected for the mapping exercise based on the pain hierarchy as reported earlier in the clinical sections of this report.

SF-36 The SF-36 questionnaire covers different aspects of HRQoL and the items can be pooled in eight different subscales: physical functioning, role – physical, body pain, general health, vitality, social functioning, role – emotional, and mental health.107 Patient scores are rated on a 0 (worst)–100 (best) scale for each subscale. These eight subscales can be pooled in two different clusters: the PCS and the MCS. The calculation of the SF-36 summary scores (PCS and MCS) is based on a published algorithm which uses a standardised scoring system with mean = 50 and SD = 10. The two component summary scores are used for the current regression analyses.

EQ-5D The EQ-5D is a five-item questionnaire encompassing the domains of mobility, self-care, usual activities, pain/discomfort and anxiety/depression. Respondents are asked to indicate whether they have no problems, some problems or major problems in each domain, with responses scored as 1, 2 and 3 respectively. To produce a combined utility health score, in which 1 is full health and 0 is equivalent to being dead, individuals' answers about the extent of their problems in each of the five health domains are then weighted. The weights, based on a large UK study, are used to take into consideration population preferences for each health domain.65 The resulting descriptive system defines 243 health states. The worst-valued health state is −0.594 (worse than death), whereas the best-valued health state is 1.0 (perfect health).

Data set

We used the SAPPHIRE trial data set and measurements at 3 months for the primary (base-case) analysis. The number of participants with complete outcomes data at 3 months was (1) 141 for the EQ-5D and pain VAS and (2) 133 for the EQ-5D and SF-36 PCS and MCS. The time point of 3 months was chosen as it represented a clinically relevant time point in the frozen shoulder trials included in this review. The data sets for (1) and (2) were then each divided into two by taking a random sample of 60% of the data and separating that data from the remaining 40% to provide an estimation data set and a validation data set respectively. In the case of (1), the EQ-5D and pain VAS, 85 measurements were available for the estimation data set and 56 measurements for validation. For (2), the EQ-5D and SF-36 PCS and MCS, 80 measurements were available for the estimation data set and 53 measurements for validation.

It should be noted that the collection of new data to validate a model's predictions is the preferred approach.108 However, in many situations this is neither practical nor possible and an alternative approach is to split the data in hand into two parts. The first part is used to estimate the model coefficients (estimation sample) and the second part is used to assess the prediction accurary of the model (validation sample). Thus, data splitting provides a data set to measure the ‘in-use’ (or within-sample) prediction accuracy of the model and simulates the complete or partial replication of the data set, in this case the SAPPHIRE trial data set.

Statistical analysis

Econometric estimation methods tested Ordinary least squares (OLS) regression was used as the main statistical technique in the analysis. The target outcome (dependent variable) was the overall EQ-5D index score, and the explanatory (independent) variables were (1) the pain VAS and (2) the SF-36 PCS and MCS scores. Based on the theoretical limitations of OLS modelling approaches reported in the empirical econometric literature,100,109,110 the current exploratory investigation estimates the EQ-5D index scores also using TOBIT (an ecometric model named after its developer James Tobin,111 which allows for upper censoring of the dependent variable, EQ-5D, at 1.0; a caveat is that it produces biased estimates in the presence of heteroscedasticity or non-normality) and CLAD (censored least absolute deviations, which produces consistent estimates in the presence of heteroscedasticity or non-normality, whereas also allowing for censoring at either the upper or the lower bound).

Regression analyses All regressions were conducted in Stata version 11 (StataCorp LP, College Station, TX, USA) and specifically included the estimation of the following models and included independent variables. The general regression models for (1) the EQ-5D and pain VAS and (2) the EQ-5D and SF-36 PCS and MCS were defined as:

EQ-5Di=α+β1×painVASi+εi
[Equation 1]

where EQ-5D represents the EQ-5D health state index value, i = 1,2,…, n represents individual respondents, painVAS represents the pain score, α is the intercept and ɛ is the error term, and

EQ-5Di=α+β1×PCSi+β2×MCSi+εi
[Equation 2]

where EQ-5D represents the EQ-5D health state index value, i = 1,2,…, n represents individual respondents, PCS represents the SF-36 PCS score, MCS represents the SF-36 MCS score, α is the intercept and ɛ is the error term. Variations of the first model were explored by including squared terms. Variations of the second model were explored by including squared and interaction terms.

We conducted three sets of analyses using alternative approaches relating to the choice of estimation and validation samples from the main SAPPHIRE data set:

  • Regression models using main effects with and without squared terms and interaction term (using individual-level data at 3 months):

    for (1), EQ-5D and pain VAS, total n = 141, estimation data set = 85 (60%), validation data set = 56 (40%)

    for (2), EQ-5D and SF-36 PCS and MCS, total n = 133, estimation data set = 80 (60%), validation data set = 53 (40%).

  • Regression models using main effects with and without squared terms and interaction term (using individual-level data at 1, 3 and 12 months):

    for (1), EQ-5D and pain VAS, total n = 491, estimation data set = 295 (60%), validation data set = 196

    for (2), EQ-5D and SF-36 PCS and MCS, total n = 467, estimation data set = 280 (60%), validation data set = 187.

Data for all time points (1, 3 and 12 months) were pooled to increase the sample size and the statistical precision of the estimates with correction for the clustering of several responses per patient by using the option ‘cluster’ in the regress command in Stata version 11.112

  • Prediction models using the main effects with and without squared terms and interaction term (using models estimated from 3-month data and used to predict EQ-5D scores at 12 months):

    for (1), EQ-5D and pain VAS, validation data set = 171

    for (2), EQ-5D and SF-36 PCS and MCS, validation data set = 163.

Assessment of models' performance Model fit (explanatory performance) and predictive/mapping performance were based on assessments of the adjusted R2, mean error (ME), mean absolute error (MAE) and root-mean-square error (RMSE), as well as the percentage of individual predictions within 0.10, 0.05 and 0.01 of their actual EQ-5D utility values (arbitrary cut-off values). Descriptive statistics and plots of actual versus predicted EQ-5D scores were also generated to examine the pattern of errors across EQ-5D scale, for example to assess whether they were evenly distributed across poor/better health states.

Results from the mapping exercise

Pattern of EQ-5D, pain visual analogue scale and SF-36 PCS and MCS scores

EQ-5D and pain visual analogue scale The mean EQ-5D scores (ordered according to health state severity) in the SAPPHIRE population vary over a wide range (e.g. from negative utility values to ‘perfect health’ as indicated by utility values of 1.0; Table 61 and Figure 25). In contrast, the SAPPHIRE trial mean pain VAS scores vary over a narrower range (e.g. highest mean VAS score = 60, when a maximum pain VAS of 100 is possible). An increasing EQ-5D score (i.e. better health status) appears to be generally accompanied by decreasing pain VAS scores (i.e. lower scores = less pain).

TABLE 61. Pattern of EQ-5D scores and pain VAS scores within the SAPPHIRE data set (with complete responses at 3 months).

TABLE 61

Pattern of EQ-5D scores and pain VAS scores within the SAPPHIRE data set (with complete responses at 3 months).

FIGURE 25. Pattern of EQ-5D scores and pain VAS scores within the SAPPHIRE data set (for complete responses) at 3 months' follow-up (base case).

FIGURE 25

Pattern of EQ-5D scores and pain VAS scores within the SAPPHIRE data set (for complete responses) at 3 months' follow-up (base case).

EQ-5D and SF-36 PCS and MCS Figure 26 and Table 62 show that mean EQ-5D scores (ordered according to health state severity) in the SAPPHIRE population vary over a wide range. In contrast, the SAPPHIRE trial mean SF-36 PCS and MCS scores vary over a relatively narrow range. An increasing EQ-5D score (i.e. better health status) appears to be generally accompanied by increasing SF-36 PCS and MCS scores (i.e. higher scores = a better quality of life).

FIGURE 26. Pattern of EQ-5D scores and SF-36 PCS and MCS scores within the SAPPHIRE data set (for complete responses) at 3 months' follow-up (base case).

FIGURE 26

Pattern of EQ-5D scores and SF-36 PCS and MCS scores within the SAPPHIRE data set (for complete responses) at 3 months' follow-up (base case).

TABLE 62. Pattern of EQ-5D scores and SF-36 PCS and MCS scores within the SAPPHIRE data set (with complete responses at 3 months).

TABLE 62

Pattern of EQ-5D scores and SF-36 PCS and MCS scores within the SAPPHIRE data set (with complete responses at 3 months).

Regression analyses

EQ-5D and pain visual analogue scale Table 63 shows the results of the regression of the pain VAS scores onto the EQ-5D using the 3-month data set (base case). The explanatory power and fit of both OLS models was poor, with an adjusted R2 ranging from −0.0014 to 0.0081. The TOBIT and CLAD models including main effects and squared terms were still relatively poor although they produced the smallest MAEs (0.2158 and 0.1877 respectively) compared with the two OLS models. However, all models predicted only between 16% and 23% of scores to within 0.05 of actual EQ-5D utility values.

TABLE 63. Regression models using pain VAS at night main effects with and without squared terms (using individual-level data at 3 months).

TABLE 63

Regression models using pain VAS at night main effects with and without squared terms (using individual-level data at 3 months).

Appendix 14 shows the results of the regression of the pain VAS scores onto the the EQ-5D using individual-level data at 1, 3 and 12 months. Compared with using data from 3 months only, the explanatory power and fit improves but only slightly. The two OLS models generated an adjusted R2 of 0.1009. Errors were generally similar across all models with RMSEs ranging from 0.26403 to 0.26689. Models 1–4 were able to predict between 20% and 26% of the scores to within 0.05 of actual EQ-5D utility values. The predictive performances of all models were generally similar.

Appendix 14 also shows the results of the regression of the pain VAS scores onto the EQ-5D using models estimated from 3-month data and used to predict EQ-5D scores at 12 months. Predicting EQ-5D scores at 12 months based on the mapping functions derived from the 3-month data set did not perform as well as the analyses using data from all time points. Models 1–4 were able to predict between 11% and 21% of the scores to within 0.05 of actual EQ-5D utility values.

EQ-5D and SF-36 PCS and MCS Table 64 shows the results of the regression of the SF-36 PCS and MCS scores onto the EQ-5D using the 3-month data set (base case). The explanatory power and fit of the three OLS models was poor, with an adjusted R2 ranging from 0.0147 to 0.0830. The simplest OLS model including only main effects produced the smallest MAEs (0.21135) compared with the other models. However, models 1–5 were only able to predict between 13% and 25% of the scores to within 0.05 of actual EQ-5D utility values.

TABLE 64. Regression models using SF-36 PCS and MCS main effects with and without squared terms and interaction term (using individual-level data at 3 months).

TABLE 64

Regression models using SF-36 PCS and MCS main effects with and without squared terms and interaction term (using individual-level data at 3 months).

Appendix 14 shows the results of the regression of the SF-36 PCS and MCS scores onto the EQ-5D using individual-level data at 1, 3 and 12 months. Compared with using data only from 3 months, the explanatory power and fit of all models improves. The three OLS models generated an adjusted R2 ranging from 0.3840 to 0.4284. Errors are smaller, with the smallest MAE and RMSE for model 5 (CLAD) of 0.1814 and model 1 (OLS) of 0.1889 respectively. Models 1–5 were able to predict between 70% and 78% of the scores to within 0.05 of actual EQ-5D utility values but only 41–55% to within 0.01 of actual EQ-5D utility values. The predictive performances of all models are generally similar.

Appendix 14 also shows the results of the regression of the SF-36 PCS and MCS scores onto the EQ-5D using models estimated from 3-month data and used to predict EQ-5D scores at 12 months. Predicting EQ-5D scores at 12 months based on the mapping functions derived from the 3-month data set was somewhat less accurate than the analyses using data from all time points. Models 1–5 were able to predict between 31% and 58% of the scores to within 0.05 of actual EQ-5D utility values.

Finally, the data suggest that all models (whether they are based on pain VAS or SF-36 PCS and MCS scores) predict better for less severe health states, but overpredict the value of more severe EQ-5D states.

Potential implications for application to frozen shoulder populations

The population used in mapping studies should cover the range of clinical and demographic characteristics of the population on which the mapping function is ultimately to be applied. The two populations should be sufficiently homogeneous that relationships between outcomes in one population can be used to infer relationships about outcomes in the other population. The current exploratory analyses were undertaken on a UK data set of patients with rotator cuff disease and not on the specific population of interest in the current review, that is, frozen shoulder patients. However, we considered the patients in the SAPPHIRE trial data set to exhibit similar enough characteristics to patients with frozen shoulder (see Table 60) to be potentially useful for drawing inferences for this population from the mapping analyses. Our findings indicated that, of the models estimated, the simple OLS model including only the main effects of (1) pain VAS mapped onto the EQ-5D and (2) SF-36 PCS and MCS mapped onto the EQ-5D performed just as well as some of the more complex models/approaches. If we apply this mapping function to the two studies in the review that did report data for (1) mean pain VAS16,51 and (2) mean SF-36 PCS and MCS scores35,51 then hypothetically it becomes possible to derive utilities based on the EQ-5D, which in principal can then be used in performing assessments of the cost-effectiveness (cost–utility) of two or more treatments in patients with frozen shoulder.

Table 65 illustrates the results of this process using data from the two frozen shoulder trials in the clinical effectiveness review that reported mean pain VAS scores16,51 (see Appendix 7 for full details and data extraction tables for these studies). Using the OLS1 model function/equations we mapped the reported mean pain VAS at baseline and after 3 months of follow-up onto the EQ-5D. The corresponding EQ-5D scores can then be predicted at baseline and after 3 months of follow-up.

TABLE 65. Cost-effectiveness results from mapping onto EQ-5D (pain VAS scale 0–100 mm).

TABLE 65

Cost-effectiveness results from mapping onto EQ-5D (pain VAS scale 0–100 mm).

Table 66 illustrates the same using data from the two frozen shoulder trials in the clinical effectiveness review that reported the SF-36 PCS and MCS35,51 (see Appendix 7 for full details and data extraction tables for these studies).

TABLE 66. Cost-effectiveness results from mapping onto EQ-5D (SF-36 PCS and MCS).

TABLE 66

Cost-effectiveness results from mapping onto EQ-5D (SF-36 PCS and MCS).

Based on these EQ-5D estimates QALYs have been approximated by applying the area under the curve (AUC) method, which is implemented by summing the areas of the geometrical shapes obtained by linearly interpolating between utility scores over the study period. Using the UK estimated costs obtained from our resource-use assessment we have presented tentative costeffectiveness results, which may allow some inference about the cost-effectiveness of some of these interventions.

These calculations are based on the assumption that no further downstream costs will be incurred by the patient. Also, because of the methods we are using we have only been able to compare interventions within trials, not across trials. Any future evaluation should consider all relevant interventions within the same analysis.

Tables 65 and 66 show the baseline and 3-month EQ-5D scores that have been obtained using the estimates provided by the mapping and the QALYs obtained based on these estimates. The costs presented in these tables represent the cheapest UK scenarios presented in Resource-use estimation. In some instances, because of interventions requiring the same resource and therefore costing the same, no ICERs have been presented. For the study evaluating interventions that incur different costs,35 the incremental results suggest that physiotherapy (with placebo injection) and steroid injection with physiotherapy were both dominated (less effective and more costly) by steroid injection alone. Steroid injection has an ICER of £288.63, which is well below the widely used threshold of £30,000.

These results are tentative. Their purpose is to try to help inform further research, not to directly inform decision-making. Actual QALY gains, and incremental cost per QALY estimates, can differ based on how the utilities were derived: directly from preference-based measures or predicted on the basis of mapping.113 This is neatly demonstrated by the results of the Dogru study, which, depending on whether mapping from the VAS or SF-36, suggest QALY gains in favour of alternative interventions.51 Furthermore, the choice of direct measure used in cost–utility analyses matters (e.g. EQ-5D, SF-6D, HUI, 15D) and may, in an economic evaluation, lead to different treatment options being considered more or less cost-effective relative to one another.114 These are all issues outside the scope of this research.

Summary

There were no studies identified in the clinical review that provided utilities data needed for generating cost per QALY estimates in assessments of the cost-effectiveness of alternative treatment options for frozen shoulder.

  • The exploratory analysis presents an alternative approach to deriving utility estimates by mapping from outcomes measured in a trial onto a preference-based measure of QoL.
  • For mapping to be of any value, a degree of overlap is required between the descriptive systems of the measures being used. The results of the exploratory analysis suggest that a statistical association between the EQ-5D and pain VAS and the EQ-5D and the SF-36 PCS and MCS measures potentially exists, specifically that a decreasing pain VAS score (less pain) is accompanied by an increasing (better) EQ-5D score and similarly that increasing (better) SF-36 PCS and MCS scores are accompanied by an increasing (better) EQ-5D score.
  • There is insufficient evidence available to make conclusions about the effectiveness of treatments for frozen shoulder in terms of QoL outcomes based on patient preference measures.
© 2012, Crown Copyright.

Included under terms of UK Non-commercial Government License.

Cover of Management of Frozen Shoulder: A Systematic Review and Cost-Effectiveness Analysis
Management of Frozen Shoulder: A Systematic Review and Cost-Effectiveness Analysis.
Health Technology Assessment, No. 16.11.
Maund E, Craig D, Suekarran S, et al.
Southampton (UK): NIHR Journals Library; 2012 Mar.

PubMed Health Blog...

read all...

Recent Activity

Your browsing activity is empty.

Activity recording is turned off.

Turn recording back on

See more...