NCBI Bookshelf. A service of the National Library of Medicine, National Institutes of Health.

Lewis R, Williams N, Matar HE, et al. The Clinical Effectiveness and Cost-Effectiveness of Management Strategies for Sciatica: Systematic Review and Economic Model. Southampton (UK): NIHR Journals Library; 2011 Nov. (Health Technology Assessment, No. 15.39.)

Cover of The Clinical Effectiveness and Cost-Effectiveness of Management Strategies for Sciatica: Systematic Review and Economic Model

The Clinical Effectiveness and Cost-Effectiveness of Management Strategies for Sciatica: Systematic Review and Economic Model.

Show details

8Review of existing economic evaluations: results


It was anticipated that the existing evidence relating to the cost-effectiveness of treatments would have a number of limitations that would make it insufficient to inform decision-making regarding the most appropriate management strategy for patients with sciatica. The findings from this review, alongside the review of clinical effectiveness, are intended to assist in informing the basis for the economic model.

Summary of results

Twelve studies were reviewed, data extracted and appraised.62,100,173,275283 A brief summary of these studies is presented in Table 164. A full summary is presented in Appendix 10. Studies evaluated the cost-effectiveness of single interventions for the treatment of sciatica (i.e. pair-wise comparisons) rather than mixed treatment effects. There was significant variation in the quality of studies presented as economic evaluations.

The majority of studies (9/12) were conducted primarily from a health-care or payer perspective. Several studies considered employment-related losses related to work days lost owing to sciatica; with three studies conducted from a societal perspective. The studies covered a diverse range of population settings, with some variation in age range and gender within the studies. Most studies considered a relatively short time horizon. One of the limitations of all studies was the lack of data relating to the longer-term outcome of sciatica. There was little distinction made in most studies between acute and chronic sciatica.

With the exception of one earlier study which employed a decision tree to represent potential pathways, all studies were based on individual patient data derived from RCTs and observational studies. As the majority of identified studies focused on intermediate or surgical interventions, resource utilisation and costs were commonly evaluated with respect to secondary care contacts and associated resource usage. Only one study focused specifically on primary care. Outcomes varied across studies, but the majority considered a global outcome and condition-specific or health-related quality of life (HRQoL). The measures used varied considerably from instruments designed specifically for the study to the use of established generic measures.

Of considerable importance to the review was the quality and robustness of the cost-effectiveness analysis (CEA). Only five studies were considered as full economic evaluations, i.e. reported incremental cost-effectiveness ratios (ICERs), when reviewed against established guidelines.29,31 The other seven studies reported costs per adjusted outcome,62 unsuccessful outcome,281 cost per response276 or costs per extra success,275 with no ICERs presented. One study, published 16 years previously,277 reported a decision-analytic model to compare chemonucleolysis with surgical disctectomy. Again, this study did not present ICERs.

Economic evaluation conducted alongside trials, modelling studies and analyses of administrative databases were included if they compared two or more treatments, and considered both costs and consequences (including cost-effectiveness, cost–utility, cost–benefit and cost-consequences analysis). Some comparative studies included in the effectiveness section of the review also reported cost data, but the data on costs and consequences were not combined. Although not conforming to a full economic evaluation under our definition, two studies warrant specific attention as providing useful information on the cost–utility of interventions for sciatica.

Hansson and Hansson100 undertook a cost–utility analysis (CUA) of 92 individuals who underwent surgery for lumbar disc herniation in a cohort of 1822 individuals aged between 18 and 59 years and selected consecutively in five regions of Sweden between 1994 and 1995. All participants had been off work for at least 28 days as a result of either low back pain or neck problems. The intervention was surgery with conservative treatment as the comparator. Outcome measures were HRQoL using European Quality of Life-5 Dimensions (EQ-5D); functional restrictions because of back problems using the Hannover Activities of Daily Living questionnaire; and pain experienced during the previous 6 months using the Von Korff pain scale. Medical costs for back pain were estimated (appointments, admission, examination and treatment) over a 2-year study period. Cost of work absenteeism was also estimated. A 5% discount rate and an assumed annual increase in productivity of 1.5% were used to convert future years' production loss to present values. Costs of illness, HRQoL and cost–utility (presented as difference in utility between 28 days and 2 years) were used as the gain in QALY.

The findings showed that the total cost of surgical treatment of lumbar disc herniation during a 2-year period was lower than the cost of non-surgical treatment. The direct cost of surgery was much higher than the direct cost of non-surgical treatment, whereas the indirect cost was lower. Lower indirect costs were the effect of lower rates of recurrence of work absence episodes and permanent disability benefits. Surgery reduced pain and improved back function and HRQoL to a greater extent than non-surgical treatments. The effects on HRQoL in combination with lower costs for surgery resulted in a better cost–utility for surgical treatment. The authors concluded that surgery for lumbar disc herniation is quite cost-effective.

Patients were drawn from a cohort study100 with explicit selection criteria in place, although the well-reported difficulties of selecting appropriate controls was acknowledged. The EQ-5D was used with utility values derived from a time trade-off (TTO) method, although a UK (rather than Swedish) population was used. Resource costs appear limited and methods to collect cost information were not fully described. Discounting was applied, but not at comparable NHS rates. Costs of illness were reported based on mean costs over 2 years (no CIs were presented). Cost per QALY were then calculated by calculating the difference between 28 days and 2 years. It is not clear why baseline values were not used. In addition, no ICERs were presented to explore QALY gain/loss over a longer time period. No sensitivity analysis was presented, with the authors stating that the Swedish cohort had a lower frequency of disc surgery within the starting 3 months than other national cohorts.

Manca et al.280 reported HRQoL, resource consumption and costs of spinal cord stimulation compared with conventional medical management in 100 patients aged ≥ 18 years participating in the PROCESS (prospective, randomised controlled multicentre study of patients with failed back surgery syndrome) trial. Conservative medical management included oral medications, nerve blocks, epidural corticosteroids, physical and psychological rehabilitative therapy, or chiropractic care. HRQoL using the Short Form questionnaire-36 items (SF-36) and EQ-5D was measured at baseline and 3 months and 6 months after initiation of treatment. Unit costs were calculated using UK and Canadian figures. Health resource-data were prospective and collected over a comprehensive range of resources. Because of the time line, discounting was not performed.

The 6-month mean total costs were significantly higher (£15,081) in the spinal cord stimulation group than in the conservative management group (£3573), with a statistically significant adjusted differential mean cost of £11,373. However, the gain in HRQoL with spinal cord stimulation over the same period was considerably greater in this group, with a mean EQ-5D score difference of 0.25 (p < 0.001) and 0.21 (p < 0.001), respectively, at 3 and 6 months after adjustment for baseline characteristics. The authors concluded that the addition of spinal cord stimulation to conservative medical management in patients resulted in higher costs to health-care systems, but generated important improvements in patients EQ-5D over the same period.

Resource data were collected in detail and unit costs were undertaken using Canadian and UK figures, although patient resource data were derived from eight countries participating in the study. However, analysis of ‘country effect’ suggested that the differences in the total cost for UK and Canada did not appear to be statistically significantly different from the trial overall mean. The study did not take into account the patients' perspective in the economic evaluation. EQ-5D data were collected and utilities were derived from a UK sample. The analysis of cost and HRQoL were presented separately. The limited follow-up period was the main limitation of this study and the authors acknowledged that a full CEA would need to consider how costs and HRQoL difference developed beyond the 6-month period.

With significant heterogeneity across these studies, it was difficult for any reliable conclusions from the results to be drawn from the existing economic evaluation evidence base.

A summary of the main issues identified include:

  • studies were undertaken across different countries
  • variability in the population settings across studies
  • lack of information on the clinical management pathways with many studies not indicating the previous treatment strategies or the timing of the intervention since diagnosis (e.g. patients who received conservative management for longer periods may be less likely to receive surgery which could lead to differences in costs and QALYs)
  • different perspectives were adopted (a significant limitation; of particular relevance for this review was the lack of a NHS and personal social services perspective in the studies)
  • unclear distinctions between acute and chronic sciatica
  • different comparators were used across studies
  • usual care was often poorly defined and variable across studies
  • short time horizons for studies with little consideration for the longer-term outcomes of sciatica
  • lack of discounting
  • the difficulty in blinding patients in the RCTs reported (patients' preferences for treatment may have influenced the reported utilities and costs)
  • different approaches to measuring resource utilisation and unit costs
  • different outcome measures used across studies
  • limited data (particularly in earlier studies) of preference-based valuations
  • lack of information on the overall duration of symptoms and how these varied across different patient groups and treatments in order to adjust for these durations in any estimation of QALYs
  • the potential for crossover between interventions and additional co-interventions (e.g. owing to recurring or worsening symptoms/relapse/complications over time) has been overlooked in the majority of economic evaluations
  • variability in the CEA presented, with nearly 60% of studies not presenting an ICER
  • lack of sensitivity analysis in these evaluations (where sensitivity analysis is performed, there was considerable variability in the parameters used for changing the base-case analysis).

A recognised limitation in reporting this review is the relevance of these studies and data to current decision-making in the UK NHS. However, even with the significant heterogeneity precluding any formal comparison or conclusions from the results, the ICER estimates reported in Table 165 suggest marked differences between treatments. The approaches, assumptions and results of these five studies are reviewed in detail to identify possible key differences and issues in order to assist in the development of the new model. Five studies were reviewed. One study compared PT with GP care,278 one study compared an intermediate intervention (ESI with placebo)173 and three studies compared surgery with conservative treatment,283 usual care282 or chemonucleolysis.279

TABLE 165. Summary of cost-effectiveness studies.


Summary of cost-effectiveness studies.

Review of full economic evaluations

Primary care

Luijsterburg et al.

Luijsterburg et al.278 undertook an economic evaluation as part of an RCT with 112 GPs in Rotterdam. One hundred and thirty-five patients aged between 18 and 65 years with duration of symptoms of < 6 weeks were randomised to PT and GP care compared with GP care alone. PT consisted of exercise therapy with information and advice provided by physical therapists. Passive therapies were not allowed. GP care was defined as care according to GP clinical guidelines and included information, advice and, if necessary, prescribed analgesia. A societal perspective was taken to the economic evaluation.

Source of effectiveness data

The primary outcome measure was global perceived effect (GPE) measured on a seven-point scale, dichotomised to improved and much improved versus not improved. GPE was rated as the percentage of patients who reported improvement. The EQ-5D was a secondary outcome measure that measured health utilities in order to calculate QALYs. Outcome measures and costs were assessed at baseline and at 3, 6, 12 and 52 weeks. Longer time horizons were not examined and discounting was not applied.

Source of cost data

Direct health-care costs included the costs of PT, GP care, medication, additional visits to other health-care providers and hospitalisations. Prices were obtained from Dutch guidelines284 or from the Professional Association.285 The currency was euros (€), but the year was not reported. Indirect costs outside the health-care system included the costs of production losses caused by absence from work. Costs for paid work were calculated by using the friction cost approach (period 154 days) based on the overall mean income of the Dutch population.

Summary of cost-effectiveness analysis

Analysis was undertaken using the ITT principle. Difference in resource utilisation between the two groups was assessed using non-parametric methods because of the skewed nature of the cost data. For the CEA, GPE and EQ-5D were used to calculate benefits. Utilities derived from the EQ-5D allowed a CUA to be performed, although this was not reported. ICERs were constructed and CIs were calculated using Fieller's methods using bootstrapping methods with the construction of cost-effectiveness acceptability curves. No sensitivity analysis was undertaken because it was claimed that most variations in cost or health effects were included in the bootstrap estimates of the ICER.

Summary of the findings

Total costs (direct and indirect) at 3, 6, 12 and 52 weeks consisted mainly of production losses with significant differences between groups for PT visits in favour of the control groups. Total direct costs were also significantly different at the four follow-up time points in favour of the control group. At baseline and 6 and 12 weeks, the mean utility score was higher in the control group (0.41, 0.70 and 0.73 compared with 0.39, 0.34 and 0.65), but the difference was statistically significant only at 6 weeks. At 52 weeks, the utility in the intervention group was higher (0.76 compared with 0.73).

The ICERs were: for direct costs €837 (95% CI −€732 to €3186) per improved patient gained and for total costs €6224 (95% CI €10,419 to €27,551) per patient improvement gained. The ICERs and CIs estimated by bootstrap and Fieller's methods were similar. The cost-effectiveness acceptability curve constructed for direct costs showed, for a threshold of €600 per patient improved, an ICER acceptable with 35% certainty and, for a threshold of €1200 per patient improved, an ICER acceptable with 69% certainty. For total costs, the curve showed, for a threshold of €4000 per patient improved, an ICER acceptable at 37%, and for a threshold of €12,000 per patient improved, an ICER acceptable at 68%.

The authors concluded that treatment of patients with lumbar radicular syndrome (LRS) with PT and GP care was not more cost-effective than GP care alone.

Critique of Luijsterburg et al.

The study research question was justified because there was a lack of knowledge concerning the cost-effectiveness of PT in sciatica. The economic evaluation has been conducted alongside a RCT which appeared to have good internal validity.

However, some clear issues were identified. The data collection methods used to collect resource utilisation and cost data were not well explained and the reliability of this information could be questioned. For example, the authors recognised that some aspects that may have affected absence from work and productivity costs (e.g. waiting times) were ignore. The authors conceded that future studies should pay more attention to analysing the effect of these factors on absence from work and costs. Costs were cumulative so recall bias from patients may have occurred, but the authors did state that differences between the groups would be minimised by the randomisation process. The authors did not clarify why only a 1-year time horizon was considered, apart from the implicit reason of length of follow-up for the RCT. The collection of outcome measures was also highlighted as a possible limitation, with the EQ-5D criticised as not being sensitive enough to capture the health effects of the additional PT, but no information was given about how benefits were valued. A CUA was not undertaken as there was no effect on QoL between the two groups with higher costs for the intervention group, and in the case of no effect the authors suggested that interventions with the lowest cost were the preferred option. However, despite no significant differences reported, the authors could have estimated an ICER based on best information available, and this highlights the continued criticism that few studies are adequately powered to detect a difference in QoL outcomes.

The issue of uncertainty around the ICER was assessed using the bootstrap method. However, although this allowed CIs to be estimated, and reliability confirmed by comparison with the results of the parametric Fieller's method, it did not allow changes in the base-case assumptions to be explicitly examined (e.g. to take into account increased waiting time).


Malter and Weinstein

Malter et al.286 undertook a review of published studies and estimated the cost-effectiveness of lumbar discectomy for herniated intervertral disc. The study was of 126 patients randomly assigned to medical or surgical treatment for radicular pain unresponsive to conservative therapy and was supplemented by data from a second trial to account for early surgery. Estimates of effectiveness were derived from a survey of 42 surgeons. This US-based study took the perspective of the health payer.

Source of effectiveness data

Effectiveness was defined as the number of QALYs gained with surgical treatment versus medical treatment. The comparator was chemonucleolysis. To determine effectiveness, results from the two trials were adjusted by QoL values obtained in a separate study of 83 subjects reporting an episode of severe back pain. A TTO utility measure was administered to estimate QoL. Mean TTO values were calculated and self-assessed outcomes reported in the trials were weighted by corresponding QoL values. For discectomy, a 2-week postoperative period was included in the base-case model. Benefits were discounted by an annual 5% rate.

Source of resource utilisation and costs

Rates of service utilisations were obtained, from a commercially available database, using data from 2175 patients diagnosed with a herniated disc. Demographic details of these patients were reported as similar to the trial participants. From this database, patients operated within 6 weeks of treatment were defined as surgically treated. Those patients who never underwent surgery and those operated on after 6 weeks were categorised as medically treated. Operation costs for medical patients requiring late surgery were counted as costs of initially choosing medical treatment. Direct costs were not discounted. Direct costs reflected costs for all services related to disc herniation (patient visits, diagnostic tests, procedures and hospitalisations). The quantity–cost boundary adopted was that of the hospital. The estimation of quantities and costs was based on actual data. Costs and rates of service utilisation were derived from MEDSTAT (January 1987–December 1989) and data on 78 patients diagnosed at a health maintenance organisation (HMO). Costs were adjusted to 1993 prices using the medical component of the Consumer Price Index and presented in US dollars ($). A 10-year time horizon was undertaken.

Summary of cost-effectiveness analysis

A model-based cost-effectiveness analysis was undertaken. Sensitivity analyses were conducted on efficacy (± 25%), QoL (± 50%) and costs. Additional estimates were obtained from a survey of spine surgeons, who were presented with case scenarios and asked to estimate the probabilities of excellent to poor outcome after surgical or medical treatment. However, these estimates were not reported, but were available on request from the authors. Additional cost estimates were undertaken from 78 patients diagnosed at a HMO. The authors stated that these were designed to estimate the true resource cost and may have reflected the actual costs more accurately than those used in the base-case analysis.

Summary of the findings

Patients treated with surgical discectomy or chemonucleolysis experienced faster improvement than patients treated medically. The probability of a good outcome varied between 0.36 and 0.56 after medical treatment and between 0.64 and 0.70 after discectomy. For a poor outcome, the probability varied between 0.06 and 0.20 after medical treatment and between 0.07 and 0.14 after discectomy. QoL values associated with a good outcome were 0.95, with a fair outcome 0.77, with a poor outcome 0.62 and with a bad outcome 0.5.

During the 10 years after surgery the average surgical patient experienced 8.7 QALYs whereas the average medical patient experienced 8.27 QALYs, with the difference of 0.43 representing the non-discounted improvement in QALYs associated with surgery. Total costs for the 18-month period beginning 6 months before diagnosis, were $17,020 for the surgical group compared with $4470 for the medical group. The non-discounted cost-effectiveness ratio of surgical over medical therapy was $29,200 per QALY. The discounted cost-effectiveness was $33,900 per QALY Cost-effectiveness of discectomy remained < $100,000 as long as surgery produced an incremental quality-adjusted benefit of at least 0.125 years. The authors concluded that, for carefully selected patients with herniated discs, surgical discectomy was a cost-effective treatment with favourable cost-effectiveness results obtained from its effect on QoL coupled with moderate costs.

Critique of Malter and Weinstein

There are key limitations of Malter and Weinstein's study which limit its relevance to current practice. It is a US study, involving a comparator not currently available to the UK NHS. In addition, the effectiveness data were from the 1970s and 1980s; improvements in surgical management may be important, so caution would be needed if attempting to generalise these findings to current management.

Although not reported in accordance with accepted current guidelines, the paper reasonably reported the economic evaluation undertaken. One possible issue was the robustness of the review undertaken, with effectiveness estimates derived from a qualitative synthesis. Effectiveness data were collected from different subjects, combined, then the estimation of benefits was modelled. The reporting of this process was limited; however, the TTO method used to derive the measure of benefits appears to be appropriate.

All costs relevant to the perspective adopted appeared to have been included in the analysis. The authors were unable to assess costs incurred more than 1-year after diagnosis from the MEDSTAT database. A sensitivity analysis was conducted on prices, but not on costs. The authors did make appropriate comparisons of their findings with those from other studies at the time of publication.

van den Hout et al.

van den Hout et al.283 examined the cost-effectiveness of early surgery compared with 6 months of prolonged conservative care, for patients aged 18–65 years with sciatica for 6–12 weeks because of lumbar disc herniation. Economic evaluation was conducted alongside a RCT.

Source of effectiveness data

The source of clinical effectiveness data was a RCT undertaken in nine hospitals in the Netherlands.87 Two hundred and eighty-three patients were randomised with 142 patients (mean age 43 ± 10 years; 68% men). Patients were followed up in the trial for 12 months. A CUA was undertaken from the perspectives of the health-care system and society.

Source of resource utilisation and cost data

Costs included the costs of hospital stay, visits to health-care professionals, home care, paid domestic help, informal care, drugs and aids, out-of-pocket expenses as a result of the disc hernia (e.g. swimming) and hours of absenteeism from work. Resource-use data were collected using patient-completed diaries and collected at several time-points over the study period. Nine per cent of patients who did not return resource diaries were equally distributed across the two comparator groups and less likely to have undergone surgery. Correction for selected non-response was made by multiple imputation of data on costs from patients in the same group with same surgical status who returned diaries. This did not substantially change the results compared with excluding these patients. For patients who did return cost diaries, the diaries covered 97%, 91%, 83% and 84% at 3, 6, 9 and 12 months respectively. For periods that were not covered, data were imputed from the closest available diary from the same patient.

Hospital costs were obtained following diagnosis using treatment prices available from 75 different centres, excluding the two highest and two lowest prices. Other health-care costs were based on Dutch standard prices. The costs of absenteeism were valued using the human capital approach. All costs were presented in euros and at 2008 Dutch consumer index prices. As a 1-year time horizon was used, costs were not discounted.

Summary of cost-effectiveness analysis

Utilities were obtained from the same patients participating in the RCT, through the administration of the EQ-5D (US and UK), the SF-6D (derived from the SF-36) and the VAS. Utilities were derived at several time points from baselines to 52 weeks after randomisation. Missing data were present in 4%, 5% and 5% of the EQ-5D, SF-36 and VAS, respectively, and inputted using the rounded average within the same randomisation group at the same time. QALYs were derived, using the area under the curve (AUC) method, for each separate quarter of the year after randomisation and during the entire year as the summary benefit measure.

Uncertainty was addressed by calculating CIs around the cost–utility ratios. Cost-effective acceptability curves were presented. Sensitivity analysis was carried out on the different utility measures and on the included cost categories using a health-care or societal perspective.

Summary of the findings

Over 12 months, the differences in QALYs and all four utility measures during all four quarters were consistently more favourable after early surgery. The differences in QALYs reported according to the utility measure used were UK EQ-5D 0.044 (95% CI 0.0005 to 0.083), US EQ-5D 0.032 (95% CI 0.005 to 0.059), SF-6D 0.024 (95% CI 0.003 to 0.046) and VAS 0.032 (95% CI −0.003 to 0.066).

From the perspective of the health-care system, total health-care costs remained significantly higher than the costs of prolonged conservative care, with a difference in costs of €1819 (95% CI €842 to €2790) per patient. Total societal costs were −€12 (95% CI −€4029 to €4006): slightly in favour of early surgery. The probability that early surgery is cost-effective compared with conservative care varies with willingness to pay. From a societal perspective it was 76% at €40,000 per QALY and was 87% at €80,000 per QALY Smaller differences were seen with other utility measures.

From the health-care perspective, according to the UK EQ-5D and US EQ-5D, the incremental cost per QALY gained with early surgery was estimated at €41,000 (95% CI €14,000 to €430,000) and €57,000 (95% CI €19,000 to €436,000), respectively.

The authors concluded that faster recovery from sciatica makes early surgery more cost-effective than prolonged conservative care. The estimated differences in health-care costs were acceptable and were compensated for by the difference in absenteeism from work. For a ‘willingness-to-pay’ ceiling ratio of €40,000 or more per QALY, early surgery need not be withheld for economic reasons.

Critique of van den Hout et al.

The source of economic data, methodology and interpretation of findings from this study were generally of good quality in this well-presented paper.

The economic evaluation was performed alongside a RCT, so selection bias was unlikely with comparable clinical, demographic and economic characteristics at baseline. The comparators were well defined and justified on the basis that prolonged conservative care is often advocated with no evidence available on the optimal timing of disc surgery.

There were clear inclusion criteria, robust power calculation and analysis undertaken using ITT principles. The internal validity of the study underpinning the economic evaluation was good. One of the strengths of the paper was the considered approach taken to the instruments used to derive utilities. In the absence of a condition-specific measure of health utility, three different generic instruments were used to measure patient preferences, which were compared in a sensitivity analysis.

Costs were considered within the two perspectives. Although there are inherent difficulties associated with the collection of resource data using patient diaries, adherence was high and, where necessary, appropriate analysis was undertaken to account for missing data. A detailed breakdown of costs was presented in the paper including sources of data, price year and statistical analysis. A limitation of the paper, which was clearly acknowledged by the authors, was the considerable variation depending on the method used for assigning costs.

Cost and benefits were appropriately analysed using an ICER. These were clearly presented. Uncertainty was addressed by calculating CIs; however, these were extremely wide. The authors did caution about the limitation of this study owing to the particular characteristics of the Dutch health-care system, citing a high rate of surgery, quicker waiting times and legislation which protects employees resulting in higher absenteeism, but not necessarily lower productivity.

Other limitations acknowledged were the 1-year time horizon for the study; a longer time horizon would have reduced statistical power and the clinical evaluation showed no differences after year 1. Another limitation was that patients were inevitably aware of the randomised group they were in; their reported utilities and costs may have been influenced by their preference for treatment. A final limitation identified was that 40% of patients randomised to receive prolonged conservative care underwent disc surgery at some time, although this was similar to other reported studies. The authors stated that this was an expected clinical consequence, as the study compared two different management strategies and that persistent or increasing symptoms that caused some patients to cross over should be part of the economic evaluation.

Tosteson et al.

Tosteson et al.282 reported a cost-effectiveness analysis based on data derived from the pooled analysis of the SPORT randomised and observational cohorts, based in the USA. The interventions compared were standard open laminectomy, laminectomy with removal of herniation and examination of the involved nerve root, and non-operative treatment, defined as usual care chosen individually by patients and physicians. Participants were aged ≥ 18 years, diagnosed with herniated intervertebral disc and confirmed as surgical candidates with a symptom history of at least 6 weeks.

Source of effectiveness data

Cost-effectiveness analysis was based on data from 1191 participants, including 775 who underwent surgery and 416 who were treated non-operatively for the entire follow-up period of 2 years. Clinical effectiveness was evaluated using QALYs at baseline, 6 weeks and 3, 6, 12 and 24 months. Health-utility values were obtained using the EQ-5D with US scoring. Time-weighted sums of EQ-5D values, adjusted to the overall mean baseline health-state value, provided the estimate of QALYs for each treatment group. CEA was based on the perspective of the health insurer and society.

At baseline, differences in patient demographic and clinical status were noted. Surgical patients were significantly younger, more likely to work full-time and to receive or be in receipt of social security compensation. Clinically, surgical patients were more likely to have L5–S1 (lumbar segment 5 to sacral segment 1) herniation, worse bodily pain, physical function, mental health and ODI and EQ-5D scores compared with non-operative patients.

Source of resource utilisation and cost data

Costs were collected on health-care costs (visits to health-care professionals, diagnostic tests, other health-care services, medications and surgery, including repeat surgery costs). Other costs included lost productivity, measured as missed work, unpaid caregiving time and missed housekeeping. Resource-use data were collected at each follow-up visit for health-care costs. A nurse-administered survey collected detail on medication usage. Recall time for self-reports of resource utilisation and time away from work/usual activities were 6 weeks for the 6-week and 3-month visits. For all other times a 1-month recall was used. Participants were provided with a diary to assist in tracking resource utilisation and missed work/housekeeping days.

Direct medical costs were estimated by multiplying patient-reported medical resource use by unit costs for each cost component. These were presented in the paper. Unit costs for office visits, hospitalisation, diagnostic test and procedures are based on 2004 Medicare national allowable payment amounts and medication prices on 2004 Red Book prices.287 Costs were adjusted for inflation, expressed in 2004 US dollars with a 3% annual discount rate used in the analysis of costs and QALYs. The differences in surgical costs were considered in terms of the procedure performed and the cost of intraoperative complications, which determined their diagnostic-related group (DRG). This was handled in the following manner: (1) a cost approximating the value paid by non-Medicare insurers was estimated to be 70% of the mean amount billed to Medicare in 2004; and (2) the observed 2004 Medicare mean total DRG price was used to reflect hospital-related surgery costs population aged > 65 years. Surgeons' costs were based on 2004 Medicare amounts; anaesthesiology costs were estimated using operative time with a fixed amount added if an intraoperative complications occurred. For non-spine-related hospitalisations, costs were based on the DRG and priced using mean observed Medicare prices in 2004 for each admission.

Loss of productivity costs due to spine-related problems were calculated by recording missed days of work (for those employed) and missed homemaking days. Use of unpaid caregivers (including spousal care given) were obtained and costs were estimated using the standard human capital approach; for work days lost this was estimated by multiplying change in hours worked by the gross of tax wage rate on self-reported wages at study entry. For homemaking and caregiving these were valued using the average wage plus non-health benefits for individuals aged ≤ 35 years.

Summary of cost-effectiveness

Owing to the high rates of non-adherence in the original randomised and observational cohorts, the two cohorts were combined and analysed according to treatment received using regression modelling of longitudinal data via generalised estimating equations. Separate models were fitted for EQ-5D and 30-day cost rates; measured at 6 weeks and 3, 6, 12 and 24 months. Cost rates were based on reported utilisation rates at each time period taking into account the recall period used.

Outcomes were assigned to the surgical group with follow-up times measured from the surgery date. To take into account the windows for scheduled visits and crossover, the actual time of the outcome assessment varied. This was included as adjusting variables in the longitudinal variables. To adjust for potential confounding baselines, variables associated with missing data or treatment received were included as covariates.

Based on the adjusted mean differences in EQ-5D from the longitudinal regression, an AUC/ time-weighted average was undertaken to estimate QALY differences between surgical and non-operative costs, adjusted to a common baseline value. ICER CIs were estimated using bootstrapping methods. Sensitivity analyses were undertaken to consider the impact of limiting costs included in the analysis to direct medical cost or direct medical costs plus costs of work loss for those employed.

Summary of the findings

Mean health scores improved over time for both groups of patients. Total mean discounted QALYs were 1.64 (95% CI 1.62 to 1.67) for surgical patients and 1.44 (95% CI 1.40 to 1.47) for non-operative patients, a difference of 0.21 (95% CI 0.16 to 0.25).

Ninety-six per cent of surgical procedures were back and neck without complications (DRG 500) with a mean cost of $12,754 (95% CI $12,740 to $12,760). Three per cent had complications (DRG 499) with mean costs estimated at $19,063 (95% CI $18,960 to $19,160). Repeat surgery occurred in 6.8% of surgical patients with a mean cost of $28,019 (95% CI $19,950 to $26,730).* Total mean costs were $27,273 (95% CI $26,009 to $28,644) for surgical patients and $13,135 (95% CI $11,244 to $14,902) for non-operative patients. Total direct costs were $20,237 (95% CI $19,314 to $21,160) for surgery and $5804 (95% CI $4639 to $6969) for non-operative patients. Total loss of productivity costs were $7089 (95% CI $6155 to $8022) for surgical patients and $7399 (95% CI $6221 to $8577) for non-operative costs. Over the 2-year period, indirect costs contented for 26% of costs for surgical patients and 57% of non-operative patients. The distribution of non-surgical direct costs was similar across both groups. Both types of cost were highest following the first 6 weeks among those undergoing surgery. Mean indirect costs for non-operative patients were higher over time than for surgically treated patients.

When all costs were considered, the cost per QALY gained for surgical treatment relative to non-operative care in the general population was $69,403 (95% CI $4923 to $94,999). For those aged ≥ 65 years, the cost per QALY gained decreased to $34,355 (95% CI $20,419 to $25,512).* Limiting costs to direct costs alone for general population ($72,181, 95% CI $56,473 to $92,394) and Medicare ($37,285, 95% CI $28,364 to $48,993) or direct costs with lost work days (general population $77,300, 95% CI $60,009 to $99,544) or Medicare ($42,111, 95% CI $30,976 to $56,284) had little change. This also had little impact on the ICER, which was estimated at $33,176 (95% CI $18,348 to $54,157) under Medicare pricing.

The authors concluded that surgery for intervertabral disc herniation was moderately cost-effective over 2 years, but expressed caution about the different values for surgery according to the method used for assigning surgical costs.

*There was obviously an error in the published paper for the figures, but no erratum could be found; therefore, we do not know whether it is the mean estimate or the CI that is correct.

Critique of Tosteson et al.

The approach and interpretation of the data and findings in the paper appeared to be of good quality. Efforts were made by the authors to capture the different resource costs associated with different surgery, and also indirect costs. The justifications for taking into account the high non-adherence rates and the variations encountered during follow-up (e.g. missed visits, delaying surgery, timing of assessment and confounding variables) were well explained.

The rationale for the study is based upon critiquing the findings from Malter and Weinstein's study.286 In this paper, the comparators could be better described. The type of surgical technique was not controlled for. There is also little description of what constituted non-operative care beyond ‘usual care chosen individually by patients and physicians’.

The data were derived from two cohorts of patients: randomised and observational. The demographics of the cohorts showed significant differences. Although these were considered in the analysis, there was little interpretation beyond a descriptive analysis of these differences. Possible reasons for the decision to have surgery (e.g. surgical patients were younger, less likely to be working full-time or to be receiving or have applied for compensation, and generally had worse clinical signs and symptoms) may have resulted in worse outcomes, which in turn influenced QALYs.

The authors considered resource usage. However, the limitations of using patients' self-reporting of resource use are referred to. The paper mentions the data collection approaches to obtain patient-reported data, but provides little information on how reliable or valid the data were. Recall bias is a potential concern, and the authors attempted to minimise this by limiting the recall window to 6 weeks after early visits and 1 month after annual visits. The authors expressed reasonable confidence that chronic problems were captured as they incurred ongoing costs, and that large costs including hospitalisation and repeat surgery were not limited by the recall period. However, some acute costs could have been missed and the small but important biases when reporting indirect costs may be a factor to take into account. However, it would seem likely this bias was applicable to both groups. The authors considered better ways of capturing resource costs, e.g. linking with electronic billing records, but this would have been likely to have biased cost ascertainment with near-complete capture of surgery compared with non-operative care.

Epidural steroids

Price et al.

Price et al.173 undertook a multicentre, double-blinded RCT of ESIs versus placebo in 228 patients with clinically diagnosed unilateral sciatica aged between 18 and 70 years who had duration of symptoms between 4 weeks and 18 months. The justification for the study was that, although 45,938 ESIs were performed in the NHS in 2002–3, there was a lack of evidence of their benefit, with safety and cost-effectiveness not previously evaluated.

Source of effectiveness data

The intervention was up to three ESIs compared with normal saline. The primary outcome was the ODI with measures of pain, physical and psychological function collected alongside objective measures of sciatic root irritation, neurological deficit and procedural side effects. QoL was determined using the SF-36.

Source of resource utilisation and cost data

A pilot was undertaken to inform the data collection method. Resource-use data were collected using an instrument completed by all clinical staff which recorded their time spent on patient consultation, aiding the patient before or after the consultation, the time associated with patient administration for all patients presenting with sciatica not included in the trial, pathology tests and imaging. Data were collected across all three centres during July–October 2000. Costs of initial radiology and pathology, if not already performed by the referring centre, were included. Analgesic costs were examined and assumed not to differ between the two groups, so were not considered in the economic analysis.

Cost data were used to calculate a cost per patient for treating sciatica with epidural injections from the perspective of health provider and purchaser. An average cost per patient was based on two management practices. Under each management practice it was assumed that patients had an initial consultation and follow-up. Owing to the short time horizon when costs and benefits were incurred, discounting was not performed.

Summary of cost-effectiveness analysis

Cost-effectiveness was undertaken from the perspective of the health provider and purchaser (NHS).

QALYS were derived from SF-6D health-utility scores using SF-36 raw data by the Brazier et al.288 technique. CUA was undertaken using the standard gamble (SG) method to derive incremental cost per QALY ratios for managing a patient with an ESI. Sensitivity analysis was undertaken to explore how cost estimates changed, given the assumptions that underlay resources, resource-base costs were relaxed. Sensitivity analysis was not undertaken for purchaser costs.

Summary of the findings

The study found ESIs conferred a short-term benefit only. The resource savings could be substantial even with a modest change to treatment. For example (from the purchasers' perspective), the saving from moving from an assumed model of current pragmatic practice (maximum of three ESIs) to a patient management strategy suggested by the trial (one ESI) would represent a saving of £16,505,700 in the sector.

The estimated average cost per patient treated from the provider's perspective was £265.30 per patient for the trial protocol and £152.80 per patient assuming a management strategy based on trial costs. Using NHS recharge cost from the purchaser's perspective, the estimated average cost was £2102 per patient to deliver treatment based on the trial protocol and £992 per patient for one epidural injection, based on the trial results.

The incremental analysis is shown in Table 166.

TABLE 166. Incremental analysis from Price et al.


Incremental analysis from Price et al.

To obtain an improvement at 3 weeks in one patient based on the trial protocol is £16,816–23,963 [depending on number needed to treat (NNT) assumed (8–11.4)], or one epidural to improvement in one patient at 3 weeks is £936–11,306.

In the sensitivity analysis, relaxation of the base-case assumptions of labour time, using the maximum recorded time for nurses and clinicians, more than doubled the average patient cost under each management strategy. Changing from day case to overnight stay also increased average patient costs. Assuming that QALYs remain unchanged, the effect would be to increase the cost-utility ratio further. The authors concluded that although ESIs are relatively safe, they confer only transient benefits in symptoms and self-reported function in a small group of patients with sciatica at substantial costs. ESIs failed the QALY threshold recommended by NICE and do not represent good value for money if NICE recommendations are followed.

Critique of Price et al.

Reporting of the economic evaluation conforms to accepted guidelines and is presented in detail. The authors recognised the limitations of the pragmatic study design and attempted to overcome this through their recruitment strategy. The intention was to compare epidural corticosteroid injections with placebo. The duration of symptoms varied from 4 weeks to 18 months, with patients who had previous back surgery excluded. There was a clear acknowledgement that the intention was to consider only patients who presented with sciatica at the point of referral to secondary care, and for the economic analysis a standard package of care was assumed. Costs associated with this package were not considered, as it was assumed that these would be incurred regardless of whether or not the patient received an epidural. Costs of health-service utilisation after week 52 were not included as no significant difference was found. There was a variability in resource usage across the three centres, reflecting the persistent limitation of a lack of clinical consensus in the management of sciatica.

The perspective taken in the economic evaluation was clearly defined and resource data appeared to have been systematically collected across the three centres. Direct costs were appropriately collected based on the perspective chosen. Indirect costs were not obtained, as it was argued that inclusion of indirect costs could overstate potential costs savings and that such savings were not relevant to resource allocation decisions. The authors clearly stated that resource data did not reflect resources expended in the trial per se, but represented the costs to normal practice.

Where differences occurred, these have been highlighted in the study. One of the most notable differences was the difference in clinicians' and nurses' time across the three centres, which probably reflected differences in practice and culture rather than marked differences in the quality of patient care. Although the justification of staff costs were made explicit, several resource costs appeared to have been generalised across several categories.

Cost–utility analysis was clearly presented. SF-6D scores were derived from the SF-36 using an established technique with SG scores calculated, assuming the trial protocol of three injections. The authors note the variability in the number in each sample, so average SG score were derived for patients with observations for all visits up to week 12 to correct for possible sample bias. One of the possible issues was the lack of sensitivity of this generic measure to detect small but important changes that may have affected the findings of limited changes in QoL. QALYs were derived and benefits were appropriately analysed using an incremental analysis.

Cost per QALY gained to the provider using a patient management strategy administering only one epidural injection. These results assumed that gain in QALY calculated would approximate that under a patient management strategy based on the trial results (one ESI). This was not considered an unreasonable assumption by the authors as change in SG score after week 3 was lower in the active group than the placebo group. However, only 21 patients received one injection to confirm this from the clinical data. Costs derived using NNT recognised the fact that ESI was compared with placebo and may therefore increase NNT and subsequent costs.

Sensitivity analysis was appropriately carried out to take into account how costs would change if base-case assumptions were relaxed. These examined changes in variation of clinical labour practices and resource use. The base-case assumption had implied that patients would be treated as day cases, so this assumption was changed. However, in practice this was felt to be too extreme, as in reality there was more likely to be a mix of day-case care and inpatient stay. In both cases, the cost increases. Assuming that QALYs remained unchanged, the effect of this would be to increase the cost–utility ratio further.

As noted by the authors, indirect costs and return to work were not considered. This was justified in terms of the recognised difficulties in using such an outcome measure owing to its definition and collection in a population of mixed age, gender and socioeconomic groups, and that there are many risk factors associated with chronic work disability apart from the level of pain. The study clearly acknowledges that the UK NHS charges differ from the actual resource used. In addition, some strategies for sciatica can be purchased from the private sector. Although these are not true resource costs (in terms of a UK NHS perspective), these may still have an opportunity cost attached. Such costs are substantial for a short period of pain relief. The lack of an individual perspective might limit the interpretation of findings, as a small chance of short-term pain relief (1 in 8 to 1 in 11) based on NNT might be welcomed by some patients. As would be expected, these findings cannot be translated into private clinical practice.


Although some economic evaluations identified in the systematic review were of reasonable to good quality, they were not able to fully address our research question. Although individual studies raised a number of important issues, it was difficult to draw meaningful conclusions across these studies because of their heterogeneity. Although there was some indication of favourable benefit such as with disc surgery, robust findings could not be reliably drawn. Although an evidence base is emerging, there remains a lack of well-designed economic evaluations. The majority of evaluations were undertaken in conjunction with clinical trials, with a lack of published decisions models. There was considerable variation with each of the studies to the management of patients with sciatica, thus limiting the lessons that can be drawn from current evidence in order to understand the relative cost-effectiveness of current management strategies that reflect current practice. Of particular note is the relevance of these studies to the UK NHS setting.

© 2011, Crown Copyright.

Included under terms of UK Non-commercial Government License.

Bookshelf ID: NBK99321
PubReader format: click here to try


  • PubReader
  • Print View
  • Cite this Page
  • PDF version of this title (19M)

Other titles in this collection

Recent Activity

Your browsing activity is empty.

Activity recording is turned off.

Turn recording back on

See more...