TOPIC DEVELOPMENT

Improving the management of multiple chronic conditions continues to be a top priority for VA researchers and clinicians alike.¹³ This evidence report was commissioned to examine the effectiveness of health coaching on changes in clinical health outcomes, health behaviors, and other key outcomes of interest to stakeholders. The report is intended to inform clinical practice decisions and develop guidelines on how best to incorporate the use of health coaching within the VA healthcare system. It is also intended to identify key program elements associated with variable intervention effects, including the patient groups most likely to benefit, the optimal dose (ie, the number and frequency of coaching sessions), the mode of coaching delivery, and the most effective types of people/professionals to conduct health coaching (eg, physicians, social workers, nurses, dieticians, peers). This report identifies gaps in evidence that warrant further research, which may help the Office of Patient Centered Care and Cultural Transformation (OPCC&CT) and the National Center for Health Promotion and Disease Prevention (NCP) prioritize future research projects.

The key questions (KQs) for this systematic review were developed after a topic refinement process that included a preliminary review of published peer-reviewed literature, consultation with internal partners and investigators, and consultation with content experts and key stakeholders at OPCC&CT and NCP.

The final KQs were:

• KQ 1: Among adults, what is the effectiveness of health coaching on
  1. Clinical health outcomes (eg, HbA1c, blood pressure)
  2. Patient health behavior (eg, physical activity, weight management, diet, smoking, medication adherence)
  3. Self-efficacy
• KQ 2: Among adults, does the impact of health coaching vary by
  1. Characteristics of the population (eg, type of chronic medical illnesses)
  2. Dose of the intervention (eg, number and frequency of sessions, minutes of contact)
  3. Mode of delivery (eg, individual visits vs group visits, face-to-face vs telephone)
  4. Types of individuals conducting coaching interventions (eg, peers, nurses, health educators, health coaches)
  5. Concordance with key elements of health coaching (ie, patient-centeredness, patient-determined goals, self-discovery process)

We followed a standard protocol for this review, and each step was pilot-tested to train and calibrate study investigators. The PROSPERO registration number is CRD42016036119.

SEARCH STRATEGY

In consultation with an expert librarian, we conducted searches of MEDLINE (via PubMed), Embase, CINAHL, and PsycINFO. Because health coaching is a relativity new intervention approach, we limited the search to the year 2000 forward.¹⁴ We evaluated the bibliographies of included primary studies and any systematic or nonsystematic reviews that were identified. We used a combination of MeSH keywords and selected free-text terms to search titles and abstracts. To ensure completeness, search strategies were also informed by search strategies recommended by the Cochrane Effective Practice and Organization of Care Group. To assess for publication bias, we searched ClinicalTrials.gov to identify completed but unpublished studies meeting our eligibility criteria, an indicator of possible publication bias. All citations were imported into 2 electronic databases (for referencing, EndNote^® Version X7, Thomson Reuters, Philadelphia, PA; for data abstraction, DistillerSR; Evidence Partners Inc., Manotick, ON, Canada). The exact search strategies used are in Appendix A.

STUDY SELECTION

Using prespecified inclusion/exclusion criteria (Table 1), titles and abstracts of RCTs identified through our search were reviewed by 2 reviewers for potential relevance to the KQs. Articles included by either reviewer underwent full-text screening. At the full-text screening stage, 2 independent reviewers were required to agree on a final inclusion/exclusion decision and the rationale for this decision. Disagreements were resolved by discussion or by a third investigator. Articles meeting eligibility criteria were included for data abstraction.

Table 1

Inclusion and Exclusion Criteria.

DATA ABSTRACTION

Data from published reports were abstracted into a customized DistillerSR database by one reviewer and overread by a second reviewer. Disagreements were resolved by discussion or by a third investigator. Data elements included descriptors to assess applicability, quality elements, intervention/exposure details, and outcomes. Each included primary article was abstracted for date of publication, sample size, location of study, and key outcomes measured (Appendix B). Key characteristics abstracted were participants' age, sex, and chronic medical illness status. We collected details about the coaching intervention such as the number and frequency of sessions, mode of session delivery, type of provider conducting coaching intervention (eg, peer, nurse), whether there was collaboration with a primary care team, communication (eg, motivational interviewing) or theoretical orientation (eg, social cognitive theory, self-determination theory), training of the coach, content of coaching calls (eg, goal-setting, problem-solving, health education, self-monitoring). Multiple reports from a single study were treated as a single data point.

QUALITY ASSESSMENT

Quality assessment was done by the researcher abstracting or evaluating the included article; this initial assessment was overread by a second, highly experienced reviewer. Disagreements were resolved between the 2 reviewers or, when needed, by arbitration from a third reviewer.

We used the key quality criteria described in the Cochrane Collaboration Risk of Bias Tool.¹⁵ This tool was designed to evaluate the risk of bias in RCTs. The tool evaluates 6 domains: (1) adequacy of random sequence generation, (2) allocation concealment, (3) blinding of participants and study personnel, (4) incomplete outcome data, (5) reporting bias due to selective outcome reporting, and (6) other forms of bias such as differences in relation to baseline measures, reliable primary outcomes or protection against contamination. The Cochrane Collaboration provides guidelines to score each item.¹⁵ Each domain is evaluated as low risk of bias, high risk of bias, or unclear risk of bias (Table 2). To draw conclusions about the overall risk of bias within trials, we summarized assessments across items in the tool for each outcome within each trial and used the approach outlined below to formulate overall risk of bias for key outcomes separately. Appendix C contains a table of quality assessment responses for the included studies.

Table 2

Approach to Formulating Summary Risk of Bias for Each Outcome Across Domains.

DATA SYNTHESIS

We summarized the primary literature by abstracting relevant data. We developed a summary table describing intervention and control conditions and key outcomes. We then determined the feasibility of completing a quantitative synthesis (ie, meta-analysis) to estimate summary effects. Feasibility depends on the volume of relevant literature, conceptual homogeneity of the studies, and completeness of results reporting. We aggregated outcomes when there were at least 3 studies with the same outcome, based on the rationale that one or 2 studies do not provide adequate evidence for summary effects. If meta-analyses were feasible, we explored the possibility of conducting subgroup analyses to explore the consistency of effects across populations and key intervention components. Because subgroup analyses that involve indirect comparisons (across studies) are subject to confounding, we interpreted results of these moderator analyses cautiously.

Six trials had more than 2 arms, and one trial used a nested 2x2 design.¹⁶ As comparisons with usual care were the most common across other trials, we prioritized these comparisons for quantitative synthesis. If more than one active arm was a coaching intervention, we prioritized the arm with the most intensive dose (eg, 20 vs 10 sessions) or delivery mode (in-person vs telephone) coaching-only arm for quantitative analysis. We qualitatively synthesized other relevant health coaching comparisons with more active comparators. The study with a 2x2¹⁶ design assessed the separate and combined impact of a physician-focused intervention and a patient-focused coaching intervention. For this study, we selected the comparisons between exposure to the patient-focused intervention and usual care control, as this was the comparison that most directly assessed the isolated impact of health coaching.

When quantitative synthesis was possible, we combined continuous outcomes using differences in follow-up means for HbA1c and body mass index (BMI) outcomes and standardized follow-up mean differences (SMD) for physical activity, diet, and self-efficacy outcomes in a random-effects model with the Knapp-Hartung correction for summary standard errors. For KQ 1, we stratified analysis by comparator type of active (eg, counseling, another form of coaching, attention control) or inactive (eg, waitlist, usual care). For KQ 2, we explored potential sources of heterogeneity by key design factors including characteristics of the population (eg, chronic medical illness status), dose of the intervention (ie, number of planned sessions), primary mode of intervention delivery (eg, telephone, in-person sessions), type of coach (eg, certified health coach, peer, healthcare professional), and concordance of health coaching interventions with an a priori list of key elements (see below for more details). We evaluated for statistical heterogeneity using visual inspection and Cochran's Q and I² statistics. Publication bias was assessed using findings from the ClinicalTrials.gov search and using funnel plots (if >10 studies in an analysis).

When quantitative synthesis was not feasible, we analyzed the data qualitatively. We gave more weight to the evidence from higher quality studies with more precise estimates of effect. A qualitative synthesis focuses on documenting and identifying patterns of the intervention across outcome categories. We analyzed potential reasons for inconsistency in treatment effects across studies by evaluating differences in the study population, intervention, comparator, and outcome definitions.

PRIORITIZATION OF HEALTH COACHING KEY ELEMENTS

The potential key elements for health coaching, such as accountability, consistent coaching relationship, content education, patient-centeredness, patient-determined goals, and use of a self-discovery process, have been described in a systematic review by Wolever et al.¹⁴ However, the relative importance of these elements to each other has not been defined. In order to determine which elements may be key drivers of effects, we used a forced-rank methodology¹⁷ whereby we presented the set of 6 crucial elements from Wolever et al¹⁴ to stakeholders, the members of our technical expert panel, and content expert research team members for ranking. The top 3 elements from this initial ranking were retained, followed by discussion and reranking to designate a proposed main driver of the effect of health coaching.

Our stakeholders included representatives from the National Center for Health Promotion and Disease Prevention and the Office of Patient-centered Care and Cultural Transformation. The technical expert panel included faculty members from the Duke School of Nursing and Vanderbilt University Medical Center; a core investigator from the VA Portland Healthcare System; and representatives from the Pacific Institute for Research and Evaluation and from Healthwise^®.

After the first round of voting, patient-centeredness, patient-determined goals, and use of a self-discovery process emerged clearly as the “key” elements of health coaching, with 8 to 9 votes each, while the other 3 elements received 2 to 3 votes each. After the second round of voting, patient-centeredness was proposed to be the main driver of health coaching, receiving 12 votes, while both patient-determined goals and use of self-discovery received only 8 votes. Thus, our final key elements were (1) patient-centeredness as the proposed main driver, (2) patient-determined goals, and (3) use of a self-discovery process.

To create a concordance score, 2 investigators independently assessed if the 3 prioritized elements were present or not. A study was given 1 point for demonstrated use of patient-determined goal or use of self-discovery process and 2 points for patient-centeredness, as this was rated as the main driver of coaching effects by our stakeholders and content experts. Thus, a study could receive a concordance score ranging from 0 to 4. Informed by the work of Wolever et al,¹⁴ we operationalized the key elements of health coaching as follows:

• Patient-centeredness: Was the coaching patient-centered, whereby coaching strategies and processes were tailored to the individual's specific needs, concerns, circumstances, priorities, or readiness to change—or was the coaching applied uniformly without regard to individual differences?
• Patient-determined goals: Did patients choose their own change goals and action steps as a target of the coaching—or were their goals preset or created by a professional?
• Use of self-discovery: Did the coaching include a process of discovery or active learning (eg, motivational interviewing) to increase patient awareness through examining strengths, values, and assumptions—or was the coaching instructional?

RATING THE BODY OF EVIDENCE

The strength of evidence (SOE) for each key question was assessed using the approach described in AHRQ's Methods Guide for Effectiveness and Comparative Effectiveness Reviews,¹⁸ and we focused on the key outcomes identified by our partners. In brief, this approach required assessment of 4 domains: risk of bias, consistency, directness, and precision (Table 3).

Table 3

Strength of Evidence Required Domains.

Additional domains were used when appropriate: coherence, dose-response association, impact of plausible residual confounders, strength of association (magnitude of effect), and publication bias. These domains were considered qualitatively, and a summary rating was assigned after discussion by 2 reviewers as high, moderate, or low strength of evidence. In some cases, high, moderate, or low ratings were impossible or imprudent to make. In these situations, a grade of insufficient was assigned. This 4-level rating scale consists of the following definitions:

• High—High confidence that the evidence reflects the true effect. Further research is very unlikely to change our confidence in the estimate of effect.
• Moderate—Moderate confidence that the evidence reflects the true effect. Further research may change our confidence in the estimate of effect and may change the estimate.
• Low—Low confidence that the evidence reflects the true effect. Further research is likely to change the confidence in the estimate of effect and is likely to change the estimate.
• Insufficient—Evidence either is unavailable or does not permit estimation of an effect.

PEER REVIEW

This report was reviewed by technical experts and clinical leadership. A transcript of their comments and our responses is provided in Appendix D.