Systolic Blood Pressure

Five of the 6 trials examined the effect of health coaching on systolic blood pressure.^{16,41,51,53,59} One trial, rated as low ROB,¹⁶ targeted patients with hypertension and compared the effects of small-group, in-person health coaching focused on dietary behavior change with a physician-focused quality improvement intervention. This study found that patients who received health coaching had a 2.6mmHg drop in systolic blood pressure (95% CI -4.4 to -0.7; p=0.01) at 6 months, although this change did not persist at 18 months. Another trial rated as low ROB⁵¹ compared a mixed intervention of phone and in-person individual counseling with usual care. This study found that at 6 months, patients who completed the coaching intervention had a 6.4-mmHg reduction in systolic blood pressure compared with the control group. Three additional trials were of mixed quality (high or unclear ROB) and concluded that health coaching was not associated with a significant reduction in blood pressure.^41,53,59 Two of these focused on patients with heart disease^41,53 while the third focused on a mixed population of participants with uncontrolled diabetes, hypertension, or hyperlipidemia.⁵⁹

Cholesterol

Four of the 6 trials examined the effects of health coaching on cholesterol (total cholesterol or LDL).^41,52,53,59 Of these, 1 was rated as low ROB,⁵³ 2 as unclear ROB,^52,59 and 1 as high ROB.⁴¹ These studies produced mixed findings, with most reporting no statistical or clinically significant effects of health coaching on cholesterol. The trial rated as low ROB⁵³ recruited patients immediately following revascularization procedures and compared personalized phone-based health coaching to usual care for changes in cholesterol. This study found that at the 6-month follow-up, patients who received health coaching had a 14mg/dL (0.36mmol/L) greater drop in mean total cholesterol level compared with those who received usual care (0.328mmol/L to 0.163mmol/L; p <0.02).

One trial rated as unclear ROB⁵⁹ examined the proportion of patients meeting cholesterol-reduction goals in a health coaching intervention for a mixed population. No significant difference was found between the health coaching (43%) and the usual care groups (37%) (95% CI -4 to 25, p=0.15). The second trial rated as unclear ROB⁵² found a positive effect of health coaching on mean cholesterol level in the health coaching group compared with control (5.0mmol/L vs 5.54mmol/L; p<0.0001) as well as LDL cholesterol (3.11 vs 3.57; p=0.0004); no positive effect was found on HDL cholesterol. The trial rated as high ROB⁴¹ compared phone-based health coaching with usual care in patients with coronary artery disease or congestive heart failure. This study found that cholesterol-reduction goals were achieved more often in the health coaching arm compared with the control arm, yet there were no significant reductions in cholesterol endpoints between the 2 groups.

Physical Activity Change

The 10 trials evaluated in the physical activity change meta-analysis comprised 6 inactive comparators^{31,32,35,47,55,58} and 4 active comparators.^29,30,43,48 The 6 trials in the inactive group contained 1215 participants and were judged either unclear (n=3) or high (n=3) ROB. The 4 trials in the active group contained 940 participants and were all judged as high ROB. Figure 3 shows the forest plot examining the effect of health coaching on physical activity change stratified by inactive and active comparator subgroups. When compared with inactive controls, the pooled estimate demonstrated a small positive effect of health coaching interventions on physical activity change that was statistically significant (n=6; SMD 0.29; 95% CI 0.15 to 0.43). This summary estimate exhibited no heterogeneity (I²'d=0.0%). This effect disappeared when health coaching was compared with active controls (n=4; SMD 0.17; 95% CI -0.32 to 0.67). This summary estimate showed moderate heterogeneity (I²'d=53.2%).

Figure 3

Effect of Health Coaching on Physical Activity Change. Abbreviations: CI=confidence interval; SD=standard deviation; SMD=standardized mean difference

Three trials that examined the effect of health coaching on physical activity change had more arms than were included in the meta-analysis.^29,30,48 These were all conducted in the United States on populations with obesity^29,48 or diabetes.³⁰ All trials were judged as high ROB and had active control groups. The first study on obesity (n=481)²⁹ examined a second mode of coaching (via the phone) versus in-person group coaching or the VA's weight control program, MOVE!, over a year. The second study on obesity (n=63)⁴⁸ examined a second coaching duration length (10 weeks instead of 20 weeks) versus a self-directed control group, all using the same study materials with follow-up at 6 months. The study on diabetes (n=320)³⁰ examined a second mode of coaching (from a peer rather than a professional) versus self-directed access to all materials via the internet. However, none of these additional comparisons showed any statistically significant differences between groups, in keeping with the result of the meta-analysis for active comparators (Figure 3 above).

Physical Activity Threshold

The 5 trials evaluated in the physical activity threshold meta-analysis all used inactive comparators.^{20,21,34,45,54,61} These trials contained 711 participants and were judged either unclear (n=3) or high (n=2) ROB. Figure 4 shows the forest plot examining the effect of health coaching on physical activity threshold. The pooled estimate demonstrated a small positive effect of health coaching interventions on physical activity threshold when compared with inactive controls, but it was not statistically significant (n=5; SMD 0.33; 95% CI -0.54 to 1.19). This summary estimate exhibited high heterogeneity (I²'d=87.9%).

Figure 4

Effect of Health Coaching on Physical Activity Threshold. Abbreviations: CI=confidence interval; SMD=standardized mean difference

Qualitative Findings for Physical Activity

Two of the 17 RCTs examining the effect of health coaching on physical activity could not be pooled with the other studies.^25,26 One was rated as low ROB²⁵ and the other as high ROB.²⁶ These trials were conducted in populations with rheumatoid arthritis (n=228)²⁶ or systemic lupus erythematosus (n=32)²⁵ and were majority or entirely female. Both trials examined categorical variables: attainment of a “healthy” goal (moderate to high intensity physical activity 4 times/week) or a specific frequency category for high or moderate-to-high intensity physical activity versus usual care. Both interventions lasted 1 year. One intervention²⁶ was monthly coaching by phone. The other²⁵ consisted of coaching every 6 weeks for 3 months decreasing over time. In addition, participants received supervised exercise, a heart rate monitor and a physical activity diary. Despite differences in study size, quality, and intervention intensity, neither study found any significant differences in physical activity between intervention and control groups. One possible reason might be the high exercise intensity level set for reaching the goal or moving between categories, which would be difficult to attain for these populations.

Qualitative Findings for BMI

Five trials that examined the effect of health coaching on BMI change had more coaching-related arms than were included in the meta-analysis (n=3^23,29,56) or compared health coaching with more robust, active comparators (n=1⁵⁷), and could not be included in the summary estimate with the inactive comparators. One 2-arm trial of unclear ROB⁵⁷ compared a health coaching intervention with a more intensive coaching intervention that involved real-time, on-demand access to the coach through a phone application for patients with diabetes; this study showed no significant difference between the 2 coaching interventions.

The other 3 remaining trials with more than 2 coaching-relevant arms were all conducted among populations with obesity. Results were mixed. The first 3-arm study²³ compared in-person coaching with phone-delivered coaching or usual care. In both coaching arms, the frequency of the interventions was the same (12 weekly coaching sessions followed by monthly coaching session for the duration of the 24-month intervention). Both coaching arms produced the same statistically significant reduction in BMI compared with usual care (1.7 vs 0.4 decrease in BMI; p-values not reported). This trial was rated as unclear ROB. Another study rated as unclear ROB⁵⁶ compared usual care with brief monthly lifestyle coaching with or without meal replacement or weight loss medications. (Figure 5 above displays the contrast between usual care and the brief lifestyle coaching without meal replacement condition.) There were no significant differences between arms on change in BMI. Last, one 3-arm study at high ROB²⁹ examined a second mode of coaching (via the phone) versus in-person group coaching or the VA's MOVE! weight control program over a year among Veterans with obesity. Participants in all three groups achieved statistically significant reductions in BMI at 12 months. Group coaching outperformed both MOVE! and phone coaching, but the contrast with MOVE! was not statistically significant (Figure 5).

Qualitative Findings for Weight Management in Pounds or Kilograms Only

There were 4 studies conducted in North America, Scandinavia, or Australia that presented data on change in pounds/kilograms but not BMI.^19,24,33,48 These findings are synthesized qualitatively. All 4 health coaching interventions were delivered via telephone. Two studies were conducted in patients with obesity, one was conducted in patients with type 2 diabetes, and one study looked at the effect of the same intervention on 2 populations, one with type 2 diabetes and one with cardiovascular disease (CVD).

The 2 obesity studies had conflicting results: one study with high ROB⁴⁸ showed a positive effect of health coaching, while the other study with high ROB¹⁹ showed a positive effect for the active control group. The latter study¹⁹ assessed the effect of weekly health coaching sessions that emphasized motivational interviewing versus structured lifestyle change instruction over 12 weeks. The noncoaching lifestyle change arm decreased weight more (-3.5 kg vs -1.1 kg, p=0.01) at post-intervention than the coaching arm. However, there was a 73% dropout rate among these participants. The other study⁴⁸ assessed 3 groups, 2 of which received weekly phone calls from a coach, either 10 or 20 sessions, over 6 months. The control group received all of the materials—an instructional manual, a pedometer, and a log book for self-monitoring—but did not receive any personal contact. On average, the 20-call group lost twice as much weight as the self-directed group (-4.9 kg vs -2.3 kg), while the average weight loss of the intermediary 10-call group was -3.2 kg (p values not given).

Neither diabetes study, one with low ROB²⁴ and one with unclear ROB,³³ found positive effects on weight loss in kilograms for health coaching versus an inactive control. The latter study,³³ which also looked at CVD, did not find a positive effect of health coaching. The diabetes studies^24,33 had inactive, usual care control groups, used change in weight (kg) as a secondary outcome, and had completion rates over 90%. One study²⁴ was a cluster RCT (n=473 from 30 primary care practices, mean age 47 to 48, 30% women) that assessed the effect of 8 nurse-led, structured health coaching sessions via phone over 18 months. They found no significant difference in weight between groups (p=0.89); however, the median number of sessions received was only 3 (interquartile range, 1 to 5). The other study³³ used the same intervention on 2 populations, diabetes (n= 250, mean age 66 years, 49% women) and CVD (n=267, mean age 69 years, 44% women). The intervention consisted of a 30-minute phone call from a trained health coach every 4 to 6 weeks over 12 months. They found no significant difference in weight between groups in either population.

Adherence to a Prespecified Diet Plan

Figure 6 shows the forest plot of the meta-analysis examining the effect of health coaching on diet adherence stratified by inactive and active comparator subgroups. The pooled estimate for 5 studies indicated a nonsignificant effect for health coaching interventions on diet adherence when compared with an inactive comparator (SMD 0.05; 95% CI -0.08 to 0.19). This summary estimate did not exhibit heterogeneity (I²=0%). Of the 5 studies using inactive comparators, 1 was rated as low ROB¹⁶ 2 as unclear ROB,^21,45 and 2 as high ROB.^20,58

Figure 6

Effect of Health Coaching on Adherence to a Prespecified Diet Plan. Abbreviations: CI=confidence interval; SD=standard deviation

In the one active comparison study in adults with type 2 diabetes,³⁰ participants were randomized to one of 3 groups: web-based information alone, or web-based information with either peer support or tailored self-management from a coach. There were no significant differences between those who received peer support compared with those who did not. There was a small positive effect (SMD 0.26; 95% CI -0.08 to 0.60) that was not statistically significant for health coaching in the form of tailored self-management when compared with an active comparator that did not include health coaching in the form of tailored self-management.

Change in Dietary Fat Consumption

Figure 7 shows the forest plot of the meta-analysis examining the effect of health coaching on change in dietary fat intake stratified by inactive and active comparator subgroups. The pooled estimate indicated a small statistically significant pooled effect for health coaching to decrease dietary fat intake when compared to an inactive comparator (SMD -0.21; 95% CI -0.31 to -0.10). This summary estimate had low heterogeneity (I²=4%). There was also a small pooled effect (SMD -0.22; 95% CI -0.41 to -0.03) that was statistically significant for health coaching to reduce dietary fat intake when compared with an active comparator. This summary estimate also had low heterogeneity (I²=0%).

Figure 7

Effect of Health Coaching on Fat Consumption. Abbreviations: CI=confidence interval; SD=standard deviation; SMD=standardized mean difference

Two of the studies using active comparators had more than 2 arms.^29,30 As stated for diet adherence, one study³⁰ also examined the effect of peer support but did not find any effect on grams of daily fat intake. In the other,²⁹ coaching was delivered by phone as well as in a group setting compared with the intensive weight control program currently offered by the VA (MOVE!). The figure shows that group coaching decreased fat intake more than MOVE!. Phone coaching also decreased fat intake more than MOVE! at 12 months (MD -1.6gm vs -0.8gm), but this difference was not statistically significant between groups. The difference between the 2 types of coaching, phone (-1.6gm fat) and group (-2.3gm fat), also was not statistically significant. Both of these studies had relatively large sample sizes (n>300) and were rated as high ROB.

Change in Total Calories

Only 2 studies measured total energy or kilocalorie intake; we summarize them qualitatively.^16,19 The study rated as high ROB¹⁹ had 45 participants with obesity and compared phone coaching with a certified health coach to a scripted phone education lesson based on the LEARN manual, both provided once/week for 12 weeks. Diet was measured via a 24-hour recall that was reviewed with a staff member at 5 time points: baseline, 6 weeks, 12 weeks (end of treatment) and 3 and 6 months post-treatment. There was a significant difference (p<.05, r=0.32) between groups at 12 weeks (coaching mean: -626.8kcal, SE=167.4 vs LEARN mean: -105.5kcal, SE=180.3). Both groups maintained reduced caloric intake at 3 and 6 months post-treatment, but differences between groups were no longer significant. The study rated as low ROB¹⁶ had a 2×2 factorial design study with 574 hypertensive participants and assessed the separate and combined influences of a physician training intervention and a patient intervention with lifestyle coaching over 6 months. Both groups that contained coaching elements decreased daily caloric intake more than the groups that did not contain coaching at both 6 (250-287kcal vs 72-171kcal, p<0.05) and 18 months (159-261kcal vs 73-119kcal), but was no longer significant at 18 months. Both of these studies indicate that coaching has a positive effect on total calorie reduction.

Change in Fruit and Vegetable Consumption

Five studies examined the outcome of fruit and vegetable consumption: 3 used an inactive comparator^16,20,31 and 2 used an active comparator.^19,29 However, measurement varied considerably between studies and meta-analysis could not be performed. We summarize these qualitatively.

The 3 studies with inactive comparators were all moderately large (n=270 to 574) but varied on population, how fruit and vegetable consumption was measured, and quality. A study rated as unclear ROB³¹ examined 410 colorectal cancer patients using a cancer-specific food frequency questionnaire that asked about fruit and vegetable intake separately. There were no significant differences at 6 or 12 months in fruit intake between the phone coaching intervention and usual care. However, at 6 months patients in the coaching group ate 0.4 more servings of vegetables per day (p=0.001) than usual care. This difference was not maintained at 12 months.

A study rated as high ROB²⁰ examined 270 patients with diabetes who were either Hispanic or African American. The study compared culturally tailored self-care coaching delivered by medical assistants over 6 months to enhanced treatment as usual. They used the 5-item diet subscale of the Summary of Diabetes Self-care Activities, which does not isolate fruit and/or vegetable consumption. There were no significant differences between groups at either time point. The last study, rated as low ROB,¹⁶ examined 574 patients with hypertension in a nested 2×2 design over 18 months: physician intervention (MD-I) or control (MD-C) and patient intervention (PT-I) (which included lifestyle coaching) or control (PT-C). Intake was measured by the Block Food Frequency Questionnaire. At 6 months, both MD-I and PT-I showed significant increased fruit and fruit juice consumption over control groups (p<0.05 and p<0.001, respectively). A significant difference was maintained at 18 months in the PT-I group but not the MD-I group.

The 2 studies with active comparators were both rated as high ROB.^19,29 The first was a 3-arm study²⁹ that used a food frequency questionnaire to measure diet. It is described under fat intake. The other study¹⁹ used 24-hour recalls to measure diet. It is described under calorie intake (above). Neither study found any difference between groups in fruit and vegetable consumption.

Random Sequence Generation

Almost all included studies (95%) described treatment allocation as random. However, 13 (33%) of these 39 studies did not give the methods used to generate the random sequence and were thus judged as unclear ROB. Two studies were judged as high ROB because they did not address randomization at all.

Allocation Concealment

In 20 of the 41 trials (49%), methods for allocation concealment were described in sufficient detail to determine whether the intervention allocation could have been foreseen in advance of or during enrollment, resulting in a judgment of low ROB. In many trials (18 of 41 [44%]), there was an unclear ROB due to inadequate detail about allocation concealment provided by authors. The remaining 3 trials^26,34,48 (7%) were judged as high ROB because they either stated a procedure that could have caused allocation to become unblinded or they did not state whether the trial was randomized at all.

Variation by Population Characteristics

Of the 20 trials, we had sufficient studies to pool effects for one population subgroup: those with diabetes. This subgroup had 17 studies. The other group comprised studies that recruited populations with a variety of chronic medical conditions and only contained 2 studies. Both subgroups displayed a similar direction of effect but the magnitude of the effect was different (Figure 11). Those recruited for diabetes had a smaller effect size (-0.21; 95% CI -0.40 to -0.02) compared to the rage of effect sizes for the studies comprised of mixed populations (range -0.90 to -0.80).

Figure 11

Effect of Health Coaching on A1c by Population Characteristics.

Variation by Dose

Eighteen studies had enough information to determine the dose of the intervention measured as planned number of intervention contacts. The number of planned contacts ranged from 5 to 156, with a median of 15. Qualitatively, we did not see any evidence of a dose response number of session on the outcome of HbA1c (Figure 12).

Figure 12

Effect of Health Coaching on A1c by Dose (Number of Planned Contacts).

Variation by Mode

Studies generally delivered their coaching intervention by phone (n=11),^{20,21,24,33,34,44-46,55,57,61} in person (n=4),^39,40,58,59 or with some mix of those 2 (n=3).^27,28,49,52 One study used a web-based coaching intervention.³⁰ Subgroups displayed a similar direction of effect, but the magnitude of the effect was slightly different. Studies that used a mix of phone and in-person sessions had a slightly greater impact, but all pooled subgroup effects were not statistically significant (Figure 13).

Figure 13

Effect of Health Coaching on A1c by Mode of Delivery.

Variation by Types of Individuals Conducting Coaching Interventions

Studies used a wide variety of personnel in the health coaching role (Figure 14). Nine used a nurse or other licensed healthcare provider^{20,21,24,27,39,44,55,58,59} as the coach. Four used a licensed behavioral provider, typically a psychologist or social worker.^28,45,57,61 Two used peer coaching,^46,49 and 4 used some other personnel as coaches (nurse or lay health worker at study discretion³⁴; unspecified employees of the healthcare system³³; professional life coaches⁴⁰; masters'-level health science students³⁰). Qualitatively, most coach types appeared roughly equally effective, with behavioral health providers having the largest pooled effects.

Figure 14

Effect of Health Coaching on A1c by Type of Coach.

Variation by Concordance

An equal number of studies received concordance scores of 3 or the maximum, 4 (Figure 15). No clear pattern emerged when we conducted exploratory subgroup analysis by concordance score; however, only the summary estimate for a concordance score of 4 was statistically significant and was similar in magnitude for the pooled estimate for the impact of health coaching compared to an inactive comparator in KQ 1 (-0.36 vs -0.30).

Figure 15

Effect of Health Coaching on A1c by Concordance.

Variation by Population Characteristics

Systolic blood pressure. Five studies reported the impact of health coaching on systolic blood pressure across the following conditions: cardiovascular disease,⁵³ hypertension,^16,51 coronary artery disease or congestive heart failure,⁴¹ or a mixture of conditions.⁵⁹ Only the 2 studies that examined the effects of health coaching on cardiovascular outcomes in patients with hypertension found a positive impact on this outcome.^16,51 The other 3 studies in the following populations found no significant effect of health coaching on systolic blood pressure outcomes: cardiovascular disease,⁵³ coronary artery disease or congestive heart failure,⁴¹ and mixed population (patients with one or more of uncontrolled diabetes, hypertension, or hyperlipidemia).⁵⁹

Cholesterol. Four of the 6 studies examined the impact of health coaching on change in cholesterol across these conditions: cardiovascular disease,^52,53 coronary artery disease or congestive heart failure,⁴¹ and mixed population.⁵⁹ Results were mixed. Both studies conducted in populations with cardiovascular disease demonstrated positive results only on changes in cholesterol,^27,52,53 while the other 2 studies conducted in mixed populations did not yield statistically significant findings.^41,59 Table 11 summarizes the finding for both systolic blood pressure and cholesterol.

Table 11

Impact of Health Coaching on Key Cardiovascular Outcomes by Population.

Variation by Dose

Systolic blood pressure. Of the 5 studies that reported systolic blood pressure outcomes, 3 studies had 10 or more planned contacts (range 11 to 20)^16,41,59 and 2 had fewer than 10 contacts (range 5 to 6).^36,51,53 Results were mixed; only one of the 3 interventions with 10 or more contacts had positive findings.¹⁶ Similarly, only one of the 2 interventions with fewer than 10 contacts had a positive impact on systolic blood pressure.⁵¹

Cholesterol. Of the 4 studies that reported cholesterol outcomes, 2 had 10 or more planned contacts,^41,59 and 2 had fewer than 10 planned contacts.^52,53,55 No clear pattern emerged; both studies that had more contacts did not product a significant impact on cholesterol, while the group with the smaller dose produced one study with statistically significant findings.⁵³ Table 12 summarizes the finding for both systolic blood pressure and cholesterol.

Table 12

Impact of Health Coaching on Key Cardiovascular Outcomes by Intervention Dose.

Variation by Mode

Systolic blood pressure. Of the 5 trials that assessed systolic blood pressure, 2 studies used primarily in-person health coaching,^16,59 and 3 studies used primarily phone-based coaching.^41,51,53 Across both modes of delivery, results were mixed. Only one phone-delivered study⁵¹ and one in-person study¹⁶ produced significant impacts on systolic blood pressure.

Cholesterol. Of the 4 trials that assessed changes in cholesterol, only one study used primarily in-person health coaching,⁵⁹ and 3 studies used primarily phone-based coaching.^41,52,53 Again, results were mixed. Only 2 phone-based studies produced significant effects.^52,53 Table 13 summarizes the finding for both systolic blood pressure and cholesterol.

Table 13

Impact of Health Coaching on Key Cardiovascular Outcomes by Intervention Delivery Mode.

Variation by Type of Coach

Systolic blood pressure. Of the 5 trials that assessed systolic blood pressure, 3 studies used healthcare providers (ie, nurse or medical assistant) to deliver the coaching intervention.^41,53,59 One study used a peer coach,⁵¹ and another used a trained health educator.¹⁶ There was a consistent pattern of effects. The 3 interventions delivered by a healthcare provider did not have significant effects on systolic blood pressure, while the 2 interventions delivered by a non-healthcare provider did report significant effects of health coaching on systolic blood pressure.

Cholesterol. All 4 studies that reported cholesterol outcomes used some form of a healthcare provider, including nurse,^41,53 dietician,⁵² or medical assistant,⁵⁹ to deliver the coaching intervention. No clear pattern emerged from the data. Only one of the 2 nurse-led interventions reported a positive impact on cholesterol.⁵³ The other study with a positive outcome was delivered by a dietician.⁵² Table 14 summarizes the finding for both systolic blood pressure and cholesterol.

Table 14

Impact of Health Coaching on Key Cardiovascular Outcomes by Type of Individual Conducting Coaching Intervention.

Variation by Concordance

Systolic blood pressure. The 5 studies of health coaching that reported impacts on systolic blood press had the following range of concordance scores: one study each for a score of 1,⁴¹ 2,¹⁶ or 3,⁵¹ and 2 studies with a score of 4.^53,59 No clear pattern emerged. Of the trials reporting no statistically significant effects, one had a concordance score of 1,⁴¹ and 2 had the highest possible concordance score of 4.^53,59 The 2 positive studies had concordance scores of 2¹⁶ and 3.⁵¹

Cholesterol. The 4 studies of health coaching that reported impacts on cholesterol had the following concordance scores: one study with a scores of 1⁴¹ and 3 studies with a score of 4.^49,52,53,59 Similar to systolic blood pressure, no clear pattern of effects emerged by concordance score. While both positive impact studies had scores of 4, a no impact study also had a score of 4.⁵⁹ The only consistent finding was that, across both outcomes, the study with the concordance score of 1 did not have a statistically significant impact on either of the prioritized cardiovascular outcomes. Table 15 summarizes the finding for both systolic blood pressure and cholesterol.

Table 15

Impact of Health Coaching on Key Cardiovascular Outcomes by Concordance Score.

Variation by Population Characteristics

Both coaching interventions that reported effects on functional status sought to increase physical activity in individuals with physically disabling conditions or rheumatoid arthritis³⁵ and multiple sclerosis.⁴⁷ Results were mixed. The study of patients with multiple sclerosis found a positive effect of health coaching on functional status.⁴⁷ However, the study of patients with rheumatoid arthritis demonstrated no positive effect of health coaching on functional status, as indicated by self-reported disability scores.³⁵

Variation by Dose of Intervention

One study had fewer than 10 planned contacts³⁵ and one study had 10 or more planned contacts.⁴⁷ The latter study, with 15 planned contacts,⁴⁷ found a positive effect of health coaching on functional status, while the other study did not.³⁵

Variation by Mode of Delivery

One study delivered the health coaching intervention via video chat using Skype and found a positive effect of health coaching on functional status.⁴⁷ The second study delivered the health coaching intervention using a mix of in-person group sessions and individual phone calls but did not find a positive effect of health coaching on functional status.³⁵

Variation by Type of Individual Conducting Coaching Intervention

One study did not report on the type of personnel used as a coach.⁴⁷ The second study used healthcare providers to deliver the health coaching intervention.³⁵ No positive effects on functional status were found.

Variation by Concordance

Both studies demonstrated low concordance with key health coaching elements, with scores of 1⁴⁷ and 2.³⁵ The study with lower concordance was the only study to find a positive effect of health coaching on functional status.⁴⁷

Variation by Population Characteristics

Physical activity change. Change was reported in the following medical conditions: diabetes^30,55,58 (n=3), obesity^29,61 (n=2), multiple sclerosis⁴⁷ (n=1), breast cancer⁴³ (n=1), colorectal cancer³¹ (n=1), arthritis³⁵ (n=1), and mixed conditions³² (n=1). Figure 16 shows the forest plot organized by medical condition. Studies showed that same trend of a positive impact of health coaching on physical activity change; however, across the 7 populations, there were major differences in effect sizes (SMD range 0.02 to 0.63). No clear pattern of effects by population emerged; subgroups with more than one study produced a mix of significant and not significant results.

Figure 16

Effect of Health Coaching on Physical Activity Change by Population Characteristics.

Physical activity threshold. Threshold was reported in 5 studies^{20,21,45,54,61}; all 5 were conducted among populations with diabetes. Thus we were not able to explore the differential impact of health coaching by population on physical activity threshold (Figure 17).

Figure 17

Effect of Health Coaching on Physical Activity Threshold by Population Characteristics.

Two additional trials assessed physical activity threshold through categorical variables and therefore could not be combined with the studies above. Neither study found statistically significant impacts of health coaching on threshold. In brief, one study²⁵ examined the effects of health coaching on physical activity in adult females with systemic lupus erythematosus and found no statistically significant differences between the intervention and inactive control groups. A second study²⁶ assessed the effects of health coaching on physical activity in patients with rheumatoid arthritis. Although the intervention group increased the number of patients who attained the physical activity “health goal,” the increase was not significantly different from the increase in the control group.

Variation by Dose

Physical activity change. Change was reported in 10 studies^{29-32,35,43,47,48,55,58} whose “intervention dose” ranged from 5 to 80 planned sessions of health coaching. Figure 18 shows the forest plot organized by number of sessions. The SMD range was -0.05 to 0.63. With one exception (SMD -0.03; 95% -0.50 to 0.43),⁵⁵ all studies with 20 or fewer planned sessions showed a small positive effect of intervention dose on health coaching (all SMD ≥0.20). Three of these results,^32,43,47 with intervention doses of 6, 12, or 15 sessions, were significant. Conversely, 2 studies with the highest numbers of planned sessions, 28 and 80, found negligible effect sizes (SMD-0.05 and 0.02) that were not significant.

Figure 18

Effect of Health Coaching on Physical Activity Change by Intervention Dose.

Physical activity threshold. Threshold was reported in 5 studies^{20,21,45,54,61} whose intervention dose ranged from 3 to 18 planned sessions. Figure 19 shows the forest plot organized by number of sessions. Two studies^21,54 had fewer than 6 planned sessions and exhibited negative effects (SMDs -0.16 and -0.21) that were not significant. The other three studies^20,45,61 had 14 to 18 planned sessions and exhibited positive effects (SMDs 0.12 to 1.50), 2 of which were significant.^45,61

Figure 19

Effect of Health Coaching on Physical Activity Threshold by Intervention Dose.

The 2 trials that assessed physical activity threshold through categorical variables have been described previously. Both had an average of 12 planned sessions. In brief, one study²⁵ provided individual coaching every 6 weeks for 3 months decreasing over a year. The other study²⁶ provided monthly coaching sessions. Neither found a statistically significant difference between the intervention and control groups on physical activity.

Variation by Mode

Physical activity change. Change was reported in 10 studies^{29-32,35,43,47,48,55,58} wherein mode of delivery was sorted into 5 categories: telephone^31,43,48,55 (n=4), in-person^29,32,58 (n=3), web³⁰ (n=1), video⁴⁷ (n=1), and mixed³⁵ (n=1). Figure 20 shows the forest plot organized by type of delivery mode. Across the delivery modes, there were differences in effect sizes (SMD range -0.05 to 0.54). Meta-analysis was possible for 2 types of delivery mode, telephone and in-person; both subgroups displayed a similar magnitude of effects, but neither pooled estimate was significant. The other 3 modes examined had only one eligible trial each. One of these studies, one showed a moderate positive effect of health coaching via video that was significant.⁴⁷ The other 2 studies, which used the web³⁰ and “mixed” mode of delivery,³⁵ found negligible to small effect sizes that were not significant.

Figure 20

Effect of Health Coaching on Physical Activity Change by Mode of Delivery.

Physical activity threshold. Threshold was reported in 5 studies,^{20,21,45,54,61} all with inactive comparators, 4 of which used the telephone as the mode of delivery^20,21,45,61 while one used in-person as the mode of delivery.⁵⁴ The 2 types of delivery modes produced effects that were different in magnitude and direction. The pooled estimate for the 4 telephone-delivered studies produced a small, positive effect that was not statistically significant. The one study that used in-person health coaching as the mode of delivery found a small negative effect for in-person coaching that was not significant (Figure 21).

Figure 21

Effect of Health Coaching on Physical Activity Threshold by Mode of Delivery.

The 2 trials that assessed physical activity threshold through categorical variables have been described previously. One study²⁶ provided in-person coaching sessions, while the other study²⁵ provided coaching via phone. Neither study found a statistically significant difference between the intervention and inactive control groups on physical activity, which is congruent with the findings reported above.

Variation by Type of Individual Conducting Coaching Intervention

Physical activity change. Change was reported in 10 studies that used the following types of individuals as coaches: healthcare provider^32,35,55,58 (n=4), “other”^29,30,48 (n=3), behavioral health provider³¹ (n=1), peer coach⁴³ (n=1), and “not reported”⁴⁷ (n=1). Figure 22 shows the forest plot organized by type of coach. Across the coach types, effect sizes were consistently positive although varying in magnitude and statistical significance (SMD range: 0.03 to 0.63). There were 2 categories of coach type, healthcare provider and “other,” for which there were enough studies to perform a meta-analysis. The meta-analysis for healthcare provider (n=4) found a significant positive effect for health coaching on physical activity (SMD 0.30; 95% CI 0.09 to 0.52) with negligible heterogeneity (I²=0.0%). The meta-analysis for “other” provider type found a negligible effect of health coaching on physical activity (SMD 0.03; 95% CI -0.31 to 0.36) that was not significant.

Figure 22

Effect of Health Coaching on Physical Activity Change by Type of Coach.

There was one study each for peer coaches,⁴³ behavioral health providers,³¹ and unidentified type of coach.⁴⁷ All found a small to a moderate positive effect of health coaching (SMD range 0.20 to 0.63), but only the peer coach study and the unidentified type of coach study produced moderate effect sizes that were statistically significant. The third study found a small, positive effect that was not significant for behavioral health providers.

Physical activity threshold. Threshold was reported in 5 studies^{20,21,45,54,61}; 2^20,21 used healthcare providers, 2^45,61 used behavioral health providers, and one⁵⁴ used peer coaches. Results are converse to those found for physical activity change above. The 2 studies that used healthcare providers^20,21 both found negligible effects (SMD range -0.16 to 0.12) that were not significant. The 2 studies that used behavioral health providers^45,61 both found sizeable positive effects that were significant (SMD range 0.62 to 1.50). The study using peer coaches found a negative effect that was not significant (SMD -0.21) (Figure 23).

Figure 23

Effect of Health Coaching on Physical Activity Threshold by Type of Coach.

Two trials assessed physical activity threshold through categorical variables and therefore could not be combined with the other studies.^25,26 Both used healthcare providers as coaches and both found no significant differences between the intervention and control groups, which is consistent with the 2 continuous variable physical activity threshold studies.

Variation by Concordance

Physical activity change. Change was reported with the following concordance scores: 0-1 (n=2),^47,48 2 (n=4),^32,35,43,55 3 (n=2),^30,58 and 4 (n=2).^29,31 Figure 24 shows the forest plot organized by concordance score. Across the 4 scores, there were major differences in effect sizes (SMD range -0.05 to 0.54), but they did not form any type of consistent pattern by level of concordance score.

Figure 24

Effect of Health Coaching on Physical Activity Change by Concordance.

Physical activity threshold. Threshold was reported in 5 studies,^{20,21,45,54,61} all of which had a concordance score of either 3 (n=3)^20,45,54 or 4 (n=2).^21,61 Again, no clear pattern emerged by level of concordance (Figure 25).

Figure 25

Effect of Health Coaching on Physical Activity Threshold by Concordance.

The 2 trials^25,26 that assessed physical activity threshold through categorical variables both had concordance scores of 0-1, and both found no significant differences between the intervention and control groups. This is different from the results for physical activity change, but still adds evidence to no specific pattern of effect for concordance score on physical activity change.

Variation by Population Characteristics

Change in BMI. Change was reported in the following medical conditions: diabetes^{21,27,40,46,49,55,57,58} (n=8), obesity^23,29,37,56 (n=4), cardiovascular disease⁵³ (n=1), metabolic syndrome³⁶ (n=1), colorectal cancer³¹ (n=1), and “mixed conditions”³² (n=1). Figure 26 shows the forest plot organized by medical condition. No clear pattern of effects emerged. All subgroups demonstrated the same direction of effects. However, across the 6 populations, there were major differences in magnitude of effect sizes (MD range -1.40 to -0.10). Both pooled estimates of the diabetes and obesity subgroups displayed moderate to high heterogeneity as exhibited by an I² >50%.

Figure 26

Effect of Health Coaching on Change in BMI by Population Characteristics.

Change in weight in kilograms. There were 4 additional studies that presented data on weight in kilograms but not on BMI.^19,24,33,48 These findings are synthesized qualitatively. Two studies were conducted in patients with obesity,^19,48 one was conducted in patients with type 2 diabetes,²⁴ and one study looked at the effect of the same intervention on 2 populations, one with type 2 diabetes and one with cardiovascular disease (CVD).³³ Congruent with the BMI studies, no clear pattern emerged. The 2 obesity studies had conflicting results; one⁴⁸ showed a positive effect of health coaching, while the other¹⁹ displayed a positive effect for the active control group. Consistent with the findings above, neither diabetes study^24,33 found positive effects on weight loss in kilograms for health coaching. In addition, a study that also looked at CVD³³ did not find a positive effect of health coaching in this population.

Variation by Dose

Change in BMI. The number of planned contacts ranged from 5 to 51. Figure 27 shows the forest plot organized by number of planned contacts. Dose was different for all studies except 2 sets of studies that had either 5 or 6 planned contacts. The median dose was 17 planned contacts. The MD range was -1.70 (a study with 24 planned contacts³⁷) to 0.26 (a study with 17 planned contacts⁴⁶). No clear pattern emerged from the data to demonstrate that number of planned contacts explained variation in effects of BMI across studies.

Figure 27

Effect of Health Coaching on BMI by Intervention Dose.

Change in weight in kilograms. There were 4 additional studies that presented data on change in kilograms.^19,24,33,48 The number of planned contacts ranged from 9 to 20 session, with a median dose was 12 planned contacts. Only the study with the greatest number of sessions (n=20) resulted in a statistically significant impact on weight change.⁴⁸ The other 3 studies with doses of 9 or 12 contacts did not produce significant impacts on weight change.

Variation by Mode

There were 3 different major modes of intervention delivery for the studies that reported changes in BMI. Seven studies used primarily telephone-based delivery,^{21,31,36,46,53,55,57} and an additional 7 used primarily in-person coaching.^{23,29,32,37,40,56,58} The other 2 studies used a mix of intervention delivery modes.^27,49 Figure 28 shows the forest plot organized by delivery mode. Both in-person and telephone delivery displayed a similar direction and magnitude of effects; however, only the telephone delivery estimate was statistically significant. Both estimates also had moderate to high heterogeneity. In contrast, the 2 studies that used a mix of intervention delivery modes displayed point estimates that were null.

Figure 28

Effect of Health Coaching on BMI by Mode of Delivery.

Change in weight in kilograms. The 4 additional studies that presented data on weight change in kilograms^19,24,33,48 were all delivered via telephone. Thus, we were unable to assess the impact of intervention mode on these studies.

Variation by Type of Individual Conducting Coaching Intervention

Studies employed a variety of personnel as health coaches. Nine used a nurse or other licensed healthcare provider as the coach.^{21,23,27,32,36,53,55,56,58} Two used a licensed behavioral health provider,^31,57 and another 2 employed peer coaches.^46,49 The final 3 studies used a variety of other personnel as coaches (eg, study-trained lifestyle coach).^29,37,40 The direction and magnitude of effects were similar across all subgroups, except one (Figure 29). Nearly all subgroups displayed a small, positive impact on reductions in BMI. In contrast, both peer-led coaching interventions did not report reductions in BMI.

Figure 29

Effect of Health Coaching on BMI by Type of Coach.

Change in weight in kilograms. The 4 additional studies that presented data on weight change in kilograms^19,24,33,48 were all delivered by the following: certified health coach,¹⁹ study-trained coach,³³ nurse,²⁴ and a coach with an unspecified training or discipline.⁴⁸ We were unable to assess the impact of intervention mode on these studies. Only the study with the coach of unclear training produced a statistically significant impact on reductions in weight.⁴⁸

Variation by Concordance

Figure 30 shows the forest plot organized by concordance score. Qualitatively, no consistent pattern of effects by level of concordance score were found. All 3 pooled estimates for concordance scores of 4, 3, or 2 displayed a similar magnitude and direction of effects and one of the 2 studies with a concordance score of 0 produced one of the largest point estimates.³⁶

Figure 30

Effect of Health Coaching on BMI by Concordance.

Change in weight in kilograms. The 4 additional studies that presented data on weight change in kilograms^19,24,33,48 had the following concordance scores: 2 had scores of 0,^33,48 one had a score of 1,²⁴ and one had a score of 4.¹⁹ Similar to the findings for BMI, no consistent pattern of effects by concordance score emerged. The only study with a statistically significant impact on reductions in weight⁴⁸ had a concordance score of 0 while studies with scores of 4 did not produce significant impacts on weight loss.

Variation by Population Characteristics

All 3 studies examined the effects of health coaching on medication adherence in patients with type 2 diabetes.^21,59,61 Thus we were unable to assess variation by population.

Variation by Dose

One study had fewer than 10 planned contacts²¹ and 2 studies had 10 or more planned contacts.^59,61 The study with the highest number of planned contacts (16 contacts) was the only study to find a positive effect of health coaching on medication adherence.⁵⁹

Variation by Mode

Of the 3 studies that focused on medication adherence, one delivered the health coaching intervention in-person and found a positive effect on the outcome of interest.⁵⁹ The remaining 2 studies delivered the intervention via telephone and did not find a positive effect of health coaching on medication adherence.^21,61

Variation by Type of Individual Conducting Coaching Intervention

All 3 studies used behavioral or healthcare providers to deliver the health coaching intervention. One study used trained medical assistants and found a positive effect of health coaching on medication adherence.⁵⁹ The remaining 2 studies used either trained nurse educators²¹ or behavioral health providers (social workers or master's-level psychologists).⁶¹ Neither study found a positive effect of health coaching on medication adherence.

Variation by Concordance

All 3 studies had a concordance score of 4; thus we were unable to assess variation by this moderator.^21,59,61