STRENGTH OF EVIDENCE

Table 16 presents an overview of findings and strength of evidence (SOE) by major outcomes prioritized by key stakeholder partners. We found moderate SOE for small increases in HbA1c (MD -0.30; 95% CI -0.50 to -0.10) and small decreases in BMI (MD -0.52; 95% CI -0.91 to -0.14) when health coaching interventions were compared with inactive controls. We found insufficient SOE for the impact of health coaching on HbA1c and BMI when compared with active control conditions.

Table 16

Summary of Intervention Effects and Strength of Evidence Ratings.

SUMMARY OF EVIDENCE BY KEY QUESTION

KQ 1 assessed the impact of self-identified health coaching interventions on key clinical, health behavior, and self-efficacy outcomes. Compared to inactive comparators, health coaching had a statistically significant effect on HbA1c (MD -0.30; 95% CI -0.50 to -0.10); physical activity change as measured in metrics such as step counts or minutes of activity (SMD 0.29; 0.15 to 0.43); BMI reduction (MD -0.52; -0.91 to -0.14); dietary fat reduction (SMD -0.21; -0.31 to -0.10); and self-efficacy (SMD 0.41; 0.21 to 0.62). For the outcome of achieving or exceeding physical activity thresholds, health coaching showed a positive trend when compared with inactive controls, but the contrast was not statistically significant (n=5; SMD 0.33; 95% CI -0.54 to 1.19). Similarly, the effect of health coaching on adherence to a prespecified dietary plan was also not significant when compared with an inactive comparator (SMD 0.05; 95% CI -0.08 to 0.19). Only change in physical activity had sufficient studies to compare effects against trials with active comparators. When compared to active controls, physical activity change was not significant (SMD 0.17; -0.32 to 0.67). Many pooled estimates exhibited moderate to high statistical heterogeneity (I² ≥50%). In qualitative syntheses, results were mixed or inconclusive for health coaching effects on functional status, smoking, and medication adherence. However, qualitative evidence suggests that coaching has a positive effect on systolic blood pressure, cholesterol, and total calorie reduction. These trends are based on a limited number of studies, and findings are inconsistent for systolic blood pressure and cholesterol.

For KQ 2, we looked at 5 potential moderators of health coaching: study population, intervention dose operationalized as number of planned contacts, primary mode of intervention delivery, type of individual conducting the coaching intervention, and intervention concordance score. None of these factors were robust predictors of treatment effects; however, some qualitative patterns of effects emerged. While results on dose of intervention are inconclusive, there is some evidence that doses that were in the middle of the range in number of planned sessions may yield more benefit than those with smaller or larger numbers of planned sessions. Also, health coaching delivered by either telephone or in-person yielded similar small to moderate positive effects across several outcomes. However, not all estimates were statistically significant. For the type of individual conducting the coaching intervention, the majority of analyses identified no clear pattern of effect. We did find some limited evidence from studies that reported HbA1c and physical activity outcomes that use of behavioral healthcare providers may positively influence the effect of health coaching. Likely training of personnel is a key factor in treatment effects; however, training was highly variable across studies. We were unable to explore the type or level of training as a moderator of treatment effects. Also we were not able to assess the impact of using a certified health coach because only one study reported using such personnel. Moreover, because of the nascent state of health coaching, there is no single certification standard for certifying coaches, so even this training and personnel distinction is fraught with problems.

CLINICAL IMPLICATIONS

While there has been one recent review of health coaching,⁶⁸ ours is the first to attempt to quantitatively synthesize the evidence on health coaching for adults with chronic medical conditions. The results of our review provide important quantifiable, new information on the impact of self-identified health coaching across clinical outcomes, patient health behaviors, and self-efficacy. Overall, we found some small effects of health coaching that are both statistically significant and within acceptable ranges for clinically significant changes. For HbA1c, there is consensus that improvements of 0.3%—the summary effect found in this report—are clinically relevant changes and, as such, health coaching appears to be a clinically relevant intervention for diabetes management. However, other systematic reviews of nonpharmacologic interventions (eg, shared medical appointments, chronic disease self-management) have shown somewhat greater effects.^69,70

Studies that assessed key cardiovascular outcomes were not amenable to quantitative syntheses. Qualitative synthesis suggests, however, that health coaching also produces small but clinically relevant changes in systolic blood pressure and cholesterol similar in magnitude to those seen for HbA1c in those studies that showed an impact (effect size range: 0.36 to 0.46 mmol/dl of cholesterol, 2.6 to 6.4 mmHg for systolic blood pressure). Yet results were inconsistent across the included trials.

Similarly, health coaching produced small, statistically significant effects on some of the prioritized health behaviors when compared with inactive controls. The 6 trials of health coaching in the pooled analysis that evaluated physical activity change as measured in metrics such as step counts or minutes of activity demonstrated improvements of 0.29 SD compared with usual care. To contextualize this, a meta-analysis of observational studies found that the pooled SD of number of steps was 2295⁷¹; thus, our results would suggest that health coaching showed an improvement equivalent to about 665 steps/day. The minimum clinically important difference in steps/day, in one study,⁷² was about 600. This suggests that health coaching is weakly potent on physical activity. Similarly, we found that health coaching produced 0.52 decrease in BMI. While promising, this decrease in BMI likely falls short of the reductions in body weight deemed clinically significant. Reduction in calories is the most noncontroversial outcome of dietary interventions. The 2 studies that evaluated the effect of health coaching on total calories both showed benefit, at the level of ∼100 kcal/day in one study, and ∼500 kcal/day in the other. Reduction of caloric intake by 500 kcal/day would clearly be clinically meaningful. For self-efficacy, health coaching had a moderate impact, with a pooled SMD of 0.41. However, the association of self-efficacy with disease control has proved challenging to assess, and so the clinical relevance of this moderate change in intermediate outcome is uncertain.

Only one study actively recruited Veterans, yet all studies were conducted among populations recruited for at least one underlying chronic medical condition including obesity, diabetes, and cardiovascular disease, so our results likely apply to a broader group of Veterans. It is likely that the results of these studies are highly applicable to the VA, because these conditions are common among VA users. However, having so few studies with large sample sizes leaves unanswered questions of feasibility around integrating health coaching into a healthcare system with large, heterogeneous patient populations and multiple types of providers and number of providers.

Overall, it is likely premature to either dismiss or adopt health coaching as a strategy for producing clinically significant improvements in key clinical and health behavior outcomes. Beyond HbA1c and weight management outcomes (ie, BMI, kilograms), many comparisons were based on a small number of studies and study quality was poor or unclear across most of the included studies. Further, many pooled estimates exhibited moderate to high statistical heterogeneity (I² ≥50%), limiting conclusions that can be drawn from these pooled estimates. The changes seen beyond usual care were similar to those seen in the literature for a number of other self-management education interventions.⁷³ Thus, our results suggest that health coaching may be an effective self-management approach. It is important to note that many of the interventions used multiple noncoaching components (eg, meal replacements, pedometers, supervised exercise sessions) as part of the overall intervention package, which makes it difficult to isolate the impact of health coaching alone. Further work is needed on how health coaching distinguishes itself from other behavioral, patient-focused approaches and when it may be the optimal behavioral approach.

LIMITATIONS

Our review has a number of strengths, including a protocol-driven design, a comprehensive search, and careful quality assessment. Also, we conducted both quantitative and qualitative synthesis when possible. Our review, and the literature, have limitations. Our review was limited to English-language publications, but the likelihood of identifying relevant data unavailable from English-language sources is low. We also limited our study to RCTs only, which excluded some evidence from nonrandomized designs. The number of identified studies for many outcomes was small, and most trials had design limitations that affected study quality (51% judged to be unclear risk of bias; 34% judged as high risk of bias). It should be noted that many of the studies evaluated as unclear or high risk of bias did not provide adequate information needed to fully judge risk of bias related to key intervention design elements, including randomization, blinding, and reporting.

Many pooled estimates exhibited moderate to high statistical heterogeneity (I² ≥50%), limiting conclusions that can be drawn from these pooled estimates. We explored if the effects of health coaching varied by intervention characteristics, including, patient chronic disease status, intervention dose (ie, the number of coaching sessions), mode of coaching delivery, individuals conducting health coaching (eg, healthcare providers, peers), and concordance of health coaching intervention with an a priori list of key elements. However, none of these individual factors was a robust predictor of heterogeneity. Thus, the observed heterogeneity is likely attributable to a combination of factors that relate to underlying differences in trial populations, comparators, interventions, inconsistency in how outcomes were measured or operationalized, and study design and quality issues. Further, many of the outcomes included in these analyses were secondary outcomes of the included trial. As such, it is important to note studies included variability in baseline levels of secondary outcomes that ranged from normal to out-of-acceptable ranges, which likely contribute to the variability seen in treatment effects.

As there is no consensus on how to define health coaching or the elements that constitute a health coaching intervention, we included studies that self-identified primarily as coaching interventions. Thus, we included and evaluated a diversity of interventions that varied by content, theoretical orientation, approach, and other factors that may impact overall effects. Any method of identifying literature for complex behavioral interventions has strengths and limitations. This is even more pronounced when the complex behavioral intervention has not been well-defined and there is no consensus on what constitutes key elements of the approach. Health coaching is not immune to these complexities. As illustrated in Wolever's seminal 2011 Archives of Internal Medicine commentary,¹⁰ there is currently no agreement on what comprises health coaching. To date, there has also been no research to establish the active ingredients of a health coaching intervention. Thus, in close consultation with our key stakeholders and our technical expert panel, we weighed our options for identifying this literature and jointly decided on use of self-identified interventions. This approach is supported in the literature; it has been used in at least 2 other recent systematic reviews of health coaching.^9,68

We recognize that any approach to identifying this literature would introduce heterogeneity. We sought to unpack this variability and, in consultation with our content experts and stakeholders, we developed an a priori list of potential moderators of intervention effects to explore. Yet, the number of studies precluded any analyses of variability by more than one characteristic at a time. Thus, we sought to further explore variability in treatment effects by applying a health coaching concordance standard across the identified literature. We co-developed this concordance score with stakeholders, technical expert panel members, and local experts in health coaching. As many behavior change approaches share common elements, the key elements identified by our stakeholders were not unique to health coaching. While this exploratory concordance score was not a robust predictor of variation in treatment effects, the inconsistency in the application of these elements across the 41 included trials underscored the overall heterogeneity in the included studies.

RESEARCH GAPS/FUTURE RESEARCH

This comprehensive review of the literature identified several gaps in the current evidence that warrant future investigation. We used the framework recommended by Robinson et al⁷⁴ to identify gaps in evidence and classify why these gaps exist (Table 17). This approach considers the population, intervention, comparator, outcome, timing, and setting (PICOTS) to identify gaps and classifies them as due to (1) low strength of evidence or imprecise information, (2) biased information, (3) inconsistency or unknown consistency, and (4) not the right information. VA and other healthcare systems should consider their clinical and policy needs when deciding whether to invest in research to address gaps in evidence.

Table 17

Evidence Gaps and Future Research.

CONCLUSIONS

Overall results suggest that self-identified health coaching interventions have the potential to produce small positive, statistically significant effects on HbA1c decreases, BMI reductions, physical activity increases, dietary fat reductions, and self-efficacy improvements when compared with inactive controls. This trend did not extend to studies with more robust comparators. We also saw a small positive, qualitative trend toward impact on total calorie reductions; however, we found only 2 studies that assessed this outcome. Some of these findings may result in effects that cross the clinically significant threshold. However, the relatively large number of studies at high or unclear risk of bias and the moderate to high heterogeneity in pooled estimates limit certainty about the interpretation of our findings and the conclusions that may be drawn.

None of the moderators were strong drivers of variability in treatment effects, suggesting that moderate to high heterogeneity in pooled estimates may be driven by a combination of intervention characteristics. We allowed studies to self-identify as health coaching interventions. Thus, variability in what is considered health coaching may contribute to the overall inconsistency and heterogeneity of effects. While health coaching may be a promising intervention modality, additional research is warranted on the impact of health coaching, especially in areas with limited identified literature (eg, medication adherence, smoking, physical function). Compared with usual care, our results suggest that health coaching may be an effective self-management approach; however, variability in the included studies, lack of consistency in what constitutes health coaching, and inclusion of multiple noncoaching components as part of the overall intervention package makes it difficult to draw firm conclusions on the impact of health coaching alone. Further, it is unclear whether health coaching offers additional advantages over other behavioral intervention modalities or when compared with more robust and active comparators. Thus, it may be premature to either dismiss or adopt health coaching in clinical or community-based settings.

Prior to conducting additional studies evaluating the effectiveness of health coaching, some foundational steps should be considered. First, both clinical and research fields would benefit from a consensus definition of health coaching. Next, training and/or credentialing required to become a certified health coach should be codified. Third, more stringent application of publication guidelines requiring full descriptions of study procedures, including randomization, blinding, and analytic methods, would allow for greater transparency and evaluation around risk of bias of complex behavioral interventions. Together, these steps would promote greater consistency in health coaching interventions, allow for more direct comparisons across studies, and promote more accurate evaluation of risk of bias. Finally, future studies should employ innovative and rigorous designs (refer to Table 17) to explore the central elements that distinguish health coaching from other behavior change and health promotion interventions and examine how these unique elements impact clinical and behavioral outcomes. Health coaching is an emerging field with shifting definitions across time. Our approach in this evidence review offers a snapshot of the literature at this time. The heterogeneity of the identified studies we included underscores the importance of better efforts to distinguish this approach from other common behavioral interventions.