Rachel Willard-Grace, MPH

 Health coaches can help patients gain knowledge, skills, and confidence to manage their chronic conditions. Coaches may be particularly valuable in resource-poor settings, but they are not typically reimbursed by insurance, raising questions about their budgetary impact.

The Health Coaching in Primary Care (HCPC) study was a randomized controlled trial that showed health coaches were effective at helping low-income patients improve control of their type 2 diabetes, hypertension, and/or hyperlipidemia at 12 months compared with usual care.

We estimated the cost of employing 3 health coaches and mapped these costs to participants. We tested whether the added costs of the coaches were offset by any savings in healthcare utilization within 1 year. Healthcare utilization data were obtained from 5 sources. Multivariate models assessed differences in costs at 1 year controlling for baseline characteristics.

Coaches worked an average of 9 hours with each participant over the length of the study. On average, the health coach intervention cost $483 per participant per year. The average healthcare costs for the coaching group was $3207 compared with $3276 for the control group (

 = .90). There was no evidence that the coaching intervention saved money at 1 year.

Health coaches have been shown to improve clinical outcomes related to chronic disease management. We found that employing health coaches adds an additional cost of $483 per patient per year. The data do not suggest that health coaches pay for themselves by reducing healthcare utilization in the first year.

Studies have shown that health coaches yield modest clinical benefits. Many clinics, especially those with limited resources, have questions about the budgetary impact of hiring additional staff. Linked with a coaching intervention, we found that:

		Health coaches cost $483 per participant per year.

		There was no evidence that coaches saved money at 1 year (through a hypothesized reduction in care).

		In 2015, CMS approved a monthly payment of $42.60 for chronic care management per qualified patient (defined as having 2 or more significant chronic conditions). Such payments would enable many clinics to train coaches to provide essential services for patients with multiple chronic illnesses.

Primary care clinicians, especially in resource-poor settings, often face competing demands and resource constraints that impede their ability to provide high-value care that improves patient outcomes while minimizing the use of resources. These clinicians are tightly scheduled to manage patients, many of whom have multiple chronic conditions, limited means, and low health literacy. Frequently, a clinician cannot address a patient’s needs in a single visit.

 New evidence-based models of care are needed to provide self-management support in primary care that is culturally and linguistically appropriate, as well as financially sustainable in resource-poor settings.

Health coaches represent a unique resource for self-management support in primary care. They help patients gain the knowledge, skills, and confidence to manage their chronic conditions.

 Coaches are trained in collaborative communication to improve understanding of, and adherence to, mutually agreed upon care plans. They may provide patients with health-related information, navigational support though the healthcare system, connections to community resources, and emotional support. Although health coaches may come from a variety of training backgrounds, medical assistants are emerging as a common and relatively economical workforce that may meet the demand for self-management support. Health coaching has been proposed as an inexpensive and effective means to improve control of chronic conditions, including risk factors for cardiovascular disease, such as diabetes, hypertension, and hyperlipidemia.

 Coaches may be particularly valuable in resource-poor settings, where minority and low-income communities bear a disproportionate burden of chronic disease and its complications and are less likely to engage in effective self-management of their conditions.

 In these settings, clinics can employ coaches that culturally and linguistically match the patients’ characteristics.

Previous studies have found health coaching to be effective in improving outcomes for chronic conditions, including diabetes, hypertension, asthma, and hyperlipidemia.

 However, for health coaching to be adopted on a wider basis, more must be understood about the budgetary impact for clinics interested in this model. For this reason, we analyzed cost data from a randomized controlled trial of health coaches versus usual care. Our goal was to determine the added costs associated with implementing a health coaching program in 2 primary care clinics and then to examine whether the added costs of the program were offset by any changes in short term healthcare utilization (within 1 year).

The Health Coaching in Primary Care (HCPC) study was a randomized controlled trial testing the efficacy of health coaching versus usual care to help low-income patients with uncontrolled type 2 diabetes, hypertension, and/or hyperlipidemia to better manage their condition(s) at 12 months. The study was conducted at 2 San Francisco safety net clinics from April 2011 to June 2012. Patients were eligible if they were between the ages of 18 and 75, spoke Spanish or English, could be reached by phone, and had poorly controlled diabetes (glycated hemoglobin [A1C] ≥8%), hypertension (systolic blood pressure [SBP] ≥140 mm Hg), and/or hyperlipidemia (low-density lipoprotein cholesterol [LDL-C] ≥160 mg/dL for patients without diabetes or ≥100 mg/dL for patients with). A total of 441 (66.4%) patients provided informed consent, completed baseline measures, and were randomized to usual care plus health coaching (n = 224) or usual care alone (n = 217). More details on the HCPC study design and methods have previously been published.

The study employed 3 medical assistant health coaches. Each coach attended 40 hours of training over 6 weeks based on a curriculum of: active listening and nonjudgmental communication; self-management skills for diabetes, hypertension, and hyperlipidemia; social and emotional support; lifestyle change; medication understanding and adherence; clinic navigation; and community resources. The study team developed the curriculum, which is available online.

 Each health coach managed a panel of 40 to 60 patients; no limitations were placed on the time that a health coach could spend with a patient. The coaches attended medical visits with patients, met with them before and after the visit, and called or met with them between visits. Health coaches helped patients review their medications, ensured that they understood their lab results and their goals, assisted in creating action plans for personalized behavior change, and assisted with navigation of clinic and community resources. Patients faced no co-payments or other financial barriers for meetings with their health coach. The coaches were paid a salary, and the study did not employ any financial incentives (eg, bonuses) for coaches to meet performance targets.

After 12 months, patients randomized to health coaching were more likely to have met the primary outcome of control for 1 or more of the conditions for which they were enrolled than patients in usual care (46.4% vs 34.3%; 

 This result was driven primarily by the benefit of health coaching on control of diabetes (change in A1C of —1.2% vs –0.5%) and, to a lesser extent, by control of hyperlipidemia (change in LDL-C of –28 mg/dL vs –18 mg/dL).

Our first goal was to assess the cost of implementing a health coach program. We estimated the cost of the health coaches based on labor, training, supplies, and space. We tracked the hours of work spent by the health coaches as related to the intervention, excluding research-specific activities. The total time spent coaching (estimated as 5376 hours, or 60% of the total full-time equivalent [FTE]) was calculated based on time studies conducted at 3 intervals during the study; it excluded time that was spent assisting with other study activities, such as chart review for eligibility screening or assisting with trainings in other locations.

The amount of health coaching time spent per patient was derived from interaction forms that health coaches completed after each encounter. The health coaches were paid $19 per hour, plus benefits (30% of the FTE). The coaching program included 40 hours of training time for each health coach and the course trainer. In addition, there was ongoing mentoring and observation, estimated at 1 hour per month per coach and the trainer. Because wages are higher in San Francisco compared with the national median, cost analyses were also run using median salary data from the Bureau of Labor Statistics, which was $13.28 per hour, plus an additional 30% for benefits. The coaches shared an office of 222 square feet, which was the equivalent of three 8×9 cubicles with additional filing space; this space is consistent with government benchmarks.

 Office space was valued at a cost of $25 per square foot. Accounting databases from San Francisco General Hospital were queried to estimate the costs associated with telephones ($24/FTE/month), computer hardware, network and support ($90/FTE/month), and supplies ($50/FTE/month).

Our second goal was to determine the budgetary impact of the health coaching program. We tested whether the added costs of the coaches were offset by any savings in healthcare utilization within 1 year. Utilization data were obtained from clinics, health systems, and insurers. Of the 2 clinics involved in this trial, one (Site A) was an independent federally qualified health center; we obtained utilization data for primary care services from its practice management and pharmacy system. At the second clinic (Site B), utilization data—including primary care, specialty, inpatient, and emergency department (ED) care&mdash;were drawn from the health system.

The inpatient and ED care for Site A that occurred within the health system were also available through this data set; however, pharmacy data were not available for Site B. Visits to external hospitals and EDs were identified through patient report, and diagnosis and visit information was abstracted from discharge reports. For the utilization data, we estimated Medicare payments based on Current Procedural Terminology codes, when available. California Medicaid (MediCal) estimated payments were used for services not covered by Medicare, namely maternal services.

Payer information was also collected for 239 participants who were covered under either MediCal or Healthy San Francisco, a local health insurance system funded by the county. This data set included inpatient, ED, outpatient, and pharmacy claims across all reporting hospitals and clinics. Of these 239 patients, 192 were covered by one of these programs for at least half of the year (at least 182 days). These data included payments or charges. To estimate costs, we used payments when available; otherwise we cost-adjusted the charges using Medicare’s hospital-specific cost-to-charge ratio.

In total, the analysis pooled utilization data from 5 data sets. We extracted data for the year prior to the intervention and for the year following randomization. We standardized costs to 2013 using the US general consumer price index. Given the possibility of duplicate records across these data sets, we examined claims for an individual based on dates, and we examined records based on the providing clinician and date. All data cleaning was blind to treatment assignment.

We compared participants randomized to usual care or to health coaches.  We examined differences based on gender, age, whether they were born in the United States, race/ethnicity, marital status, formal educational attainment, yearly income, English as a first language, and employment status. We used bivariate statistics to compare treatment groups’ health services costs at 1 year; we examined the cost across the 5 data sources to ensure consistency of results. We also estimated counts of services from the claims data, but were less confident that each claim represented a unique visit or stay—we treated this as a check on the cost data.

We used multivariate models to control for possible observed differences and to run sensitivity analyses, and ordinary least squares for the primary multivariate analysis, controlling for clinic and the covariates listed above. Sensitivity analyses used semi-log and generalized linear models with a log link and a gamma distribution. We ran 2 subsample analyses: first, we examined whether there were differential effects by clinic; second, we examined the impact of the intervention on patients who had healthcare costs in the top quartile in the year prior to the intervention to determine whether the intervention had a differential effect for high-cost patients.

Participants had an average age of 52.7 years, with a broad range of 22 to 75 years (

). Approximately half of the participants were women. The population exemplified those seeking care at resource-poor settings, with over half reporting a household income of $10,000 or less and only 12% reported an income of more than $20,000 per year. There were no significant differences in characteristics of the sample between the intervention and the control groups.

Health coaches worked, on average, 9 hours with each participant over the length of the study. There was considerable variation in how much time coaches spent with participants: at the upper end, a coach spent 27.4 hours with a participant. Intervention costs were allocated to the patients based on the hours the coaches spent with each patient. On average, the health coach intervention cost $483 per participant per year period (

). Wages are higher in San Francisco than the national median, so based on the national median wage rate, we expect the coaching intervention would average $356 per participant per year.

The majority (70%) of the intervention was attributable to direct labor costs: the time and effort coaches spent working with patients. Of the remainder, 6% was related to training coaches and 24% was related to staff benefits, staff office space, and supplies. These costs and percentages were based on national wages. If we consider costs in San Francisco, where the labor costs are higher than the national average, the percentage of total costs related to labor increases from 70% to 74.5% and there is a corresponding small increase in staff benefits, whereas the other relative percentages drop slightly.

Despite the substantial investment in health coaching time, utilization and healthcare costs were relatively similar between the 2 treatment groups. The average 1-year healthcare costs for patients in the coaching group was $3207 compared with $3276 for the control group (

 = .90) (Table 2)—this total does not include the intervention. As shown in Table 2, there was a higher maximum in the control group, but the median costs were slightly higher in the coaching group (median = $823) than the control group (median = $735). We tested for cost differences across the different administrative data sets and found similar results, confirming that the results were not obscuring differences across clinics or data sets. The results were highly robust to the choice of statistical model. Using semi-log regression or general linear models had no substantive change in the results compared with ordinary least squares. In the subsample analysis, we found no significant differences by clinic. We also found no evidence to suggest that the intervention saved money for the high-cost patients, defined as patients in the top 25% of healthcare costs in the prior year.

Health coaches have been shown to improve clinical outcomes related to chronic disease management. Although many of the gains in clinical outcomes are modest, health coaches are relatively inexpensive compared with nurses or physicians. Adding health coaches costs approximately $483 per patient per year ($356 per patient per year using nationalized costs). These estimates are similar to the cost for health coaching by community health workers in previous studies. One evaluation of a community-based support program for Latinos with diabetes (Project Dulce) estimated an intervention cost of $507 per patient per year (2002 dollars).

 Another study of a community health worker program to improve control of diabetes found a mean cost of $1176 per patient over 18 months ($784/patient/year).

 A recent report of the costs of diabetes self-management programs in 4 community primary care settings found first-year costs ranging from $832 to $2340 per patient.

 Given a cost of $483 per patient per year, one question is how to pay for these coaches.

 One possible way is through downstream savings, perhaps by engaging the patients in timely services that avert inappropriate care. In the current study, we did not find significant savings in health costs for the coached group over the 1-year study period. Although many might hope that health coaches would save money in the short term, the rationale is often predicated on the expectation that patients will reduce their utilization of inappropriate care. The corollary to this is that patients often increase their use of appropriate care, including medications and scheduled visits to their doctor. Of course, it is challenging, using administrative records, to distinguish between inappropriate and appropriate care.

In Project Dulce, healthcare costs were actually higher in the intervention group by a mean of $842 per year, primarily due to the high costs of prescription medications

—this higher cost could be appropriate if patients in the intervention group were filling prescriptions for needed medications they would otherwise have lacked.

 ED and inpatient utilization are often used as proxies for inappropriate utilization; however, utilization of these services is typically rare, as was the case in this study, leaving little room to save money for most patients. When we limited the analysis to patients in the highest cost quartile, the results were not significantly different; however, the number was too small to make a definitive statement about the impact of health coaching on costs for “high-utilizers.” A study focused on high-utilizers would be needed to answer this question.

In the current study, we did not examine costs beyond the 1-year of health coaching, and 1 year is probably too short a time to expect to see differences in complication rates related to glycemic control. In addition, we did not model cost-effectiveness.

 Previous investigators analyzing data from the UK Prospective Diabetes Study found that a 1% reduction in mean A1C was associated with reductions of 21% for diabetes-related deaths, 14% for myocardial infarctions, and 37% of microvascular complications of diabetes (eg, renal failure) over an approximately 10-year period.

A lifetime cost-effectiveness analysis using data from Project DULCE for the subgroup of patients most similar to those in our study (receiving county medical services and a mean improvement in A1C of 0.8%) found the cost to be $24,584 per quality-adjusted life-year [QALYs]), assuming the intervention was continued and the improvement in A1C persisted.

 The investigators who reported the results of 4 community-based diabetes self-management programs also ran the CDC/Research Triangle Institute model to estimate lifetime cost savings assuming an on-going reduction of 0.5% in A1C, a 10% reduction in LDL-C, and no change in SBP. The estimated lifetime savings in healthcare costs was $3385 per patient, while the lifetime intervention cost was estimated at $15,031 per patient, resulting in a net cost of $29,563 per QALY.

 In our study, we found a similar effect (mean reduction of 0.7% in A1C, 7% in LDL-C, and no significant change in SBP).

Another way to pay for coaching is through reimbursement. To date, few insurers provide reimbursement for health coaching, but recent changes by CMS may open the door for further coverage. For 2015, CMS approved a monthly payment rate of $42.60 for chronic care management per qualified patient (defined as having 2 or more significant chronic conditions).

 CMS continues to clarify which clinical staff qualifies for reimbursement and the required level of supervision. However, if coaches are capable of being reimbursed, then annualized Medicare reimbursement would offset the cost of the coach—of course, this would only cover services provided by the coaches to adults covered by Medicare, but other insurers might also follow suit.

A key limitation in this study is the reliance on administrative records, drawn from multiple sources that were not mutually exclusive. Although efforts were taken to eliminate duplication using provider names and dates, there remains the possibility that utilization was double-counted or there was missing information. In addition, ED and hospital visits at hospitals other than the county-administered hospital were identified through patient self-reporting; thus, it is possible that this method could lead to errors, even though self-report is often accurate for these highly salient, uncommon events. In addition, it seems unlikely that any errors would systematically differ for either the experimental or control arm. Finally, this cost analysis covered a time period of only 1 year, and it may have failed to capture downstream savings from improved glycemic control or cholesterol after 1 year.

The incremental costs from health coaching in this study were relatively small compared with total health costs, but no cost savings were found during the year of coaching. Previous studies and models suggest benefits of continual control would accrue over time; however, there are few models that guide how to project the future costs of coaching. If coaches stopped working with patients after 1 year, the benefits might erode over time. Additional research is needed to understand the long-term benefits and costs of health coaching programs.

 Department of Health Research and Policy, Stanford University (THW), Stanford, CA; Health Economist, VA Palo Alto Health Care System (THW), Palo Alto, CA; Department of Family and Community Medicine, University of California, San Francisco School of Medicine (RW-G, EC, TB, DHT), San Francisco, CA; Silver Avenue Health Center (EC), San Francisco, CA.

 This study was funded by the Gordon and Betty Moore Foundation (GBMF2492). 

 Ms Willard-Grace’s salary is paid in part by research grants related to health coaching, including a grant from PCORI, which was received for 2014-2017 for COPD-related coaching; she has also attended meetings by ADA, Academy Health, and was a presenter at NAPCRG. Ms Willard-Grace’s employer is a Center for Excellence in Primary Care and has a training arm that is paid to conduct trainings and receives licensing fees. The remaining authors report no relationship or financial interest with any entity that would pose a conflict of interest with the subject matter of this article.

 Concept and design (TB, EC, DT, TW); acquisition of data (DT, RW-G, EC); analysis and interpretation of data (TW, DT, EC); drafting of the manuscript (TW); critical revision of the manuscript for important intellectual content (DT, RW-G); statistical analysis (TW, DT); provision of patients or study materials (DT, RW-G); obtaining funding (DT, TB, EC); administrative, technical, or logistic support (DT, RW-G, EC); and supervision (TW, DT, EC).

Articles

On average, the health coach intervention cost $483 per participant per year. There was no evidence that the coaching intervention saved money at 1 year.

Costs for a Health Coaching Intervention for Chronic Care Management

We sought to test the hypothesis that training medical assistants to provide health coaching would improve patients’ experience of care received and overall satisfaction with their clinic.

Low-income English- or Spanish-speaking patients aged 18 to 75 years with poorly controlled type 2 diabetes, hypertension, and/or hyperlipidemia were randomized to receive either a health coach or usual care for 12 months. Patient care experience was measured using the Patient Assessment of Chronic Illness Care (PACIC) scale at baseline and at 12 months. Patient overall satisfaction with the clinic was assessed with a single item asking if they would recommend the clinic to a friend or family member. PACIC and satisfaction scores were compared between study arms using generalized estimating equations to account for clustering at the clinician level. 

PACIC scores were available from baseline and at 12 months on 366 (76%) of the 441 patients randomized. At baseline, patients receiving health coaching were similar to those in the usual care group with respect to demographic and other characteristics, including mean PACIC scores (3.00 vs 3.06) and the percent who would “definitely recommend” their clinic (73% and 73%, respectively). At 12 months, coached patients had a significantly higher mean PACIC score (3.82 vs 3.13; 

 <.001) and were more likely to report they would definitely recommend their clinic (85% vs 73%; 

 Using medical assistants trained in health coaching significantly improved the quality of care that low-income patients with poorly controlled chronic disease reported receiving from their healthcare team.

	The Chronic Care model (CCM) has been widely applied to the transformation of primary care. Health coaching can incorporate functions of the CCM, including patient education, navigation, collaborative goal setting, and personal support. We found that medical assistants trained as health coaches substantially improved patient-reported quality of care under the CCM, as measured by the Patient Assessment of Chronic Illness Care (PACIC) in a randomized controlled trial. 

		Mean PACIC scores (range 1 to 5) increased over 12 months from 3.00 to 3.83 in the coached group versus from 3.06 to 3.13 in the usual care group, an effect size of 0.77 which was highly significant (

		The proportion of patients reporting they would definitely recommend their clinic to family and friends increased from 73% to 85% in the coached group, but was unchanged in the usual care group (

		We conclude that medical assistants trained as health coaches can substantially improve patient-reported quality of care received from their healthcare team and satisfaction with their clinic.

The provision of high-quality, patient-centered primary care for patients with chronic conditions has been identified as a priority for US healthcare.

 A widely used model for improving care is the Chronic Care model (CCM),

 which is based on 6 elements identified from an extensive literature review of interventions: self-management support, decision support, delivery system design, clinical information systems, healthcare organization, and community resources. The CCM has been broadly applied as a model for improving chronic care and has been linked to better outcomes of care.

 Many studies have examined interventions to improve the quality of care under the CCM, including practice redesign,

 The key elements of the CCM are part of the primary care medical home movement,

 which has demonstrated success in improving quality of care and patient satisfaction while holding down costs.

One approach to improving the quality of chronic care is the use of health coaches as part of the healthcare team.

 Health coaching by health workers or peers trained as coaches has emerged as an effective model to improve the management of several chronic conditions, including asthma in children,

 or a combination of cardiovascular risk conditions

 in adults.  Health coaches incorporate functions of patient education, navigation, collaborative goal setting, and personal support,

In contrast to health educators, health coaches help patients choose health goals and create action plans to achieve those goals, in addition to supporting the patient in achieving those goals. Health coaches frequently have contact with patients outside of medical visits, and while they do provide substantial emotional support, they are not therapists. In contrast to case managers who are usually health professionals (eg, registered nurses, pharmacists), health coaches can be unlicensed members of the support staff, such as medical assistants or health workers, who are more likely to be culturally and linguistically concordant with patients. Medical assistants—one of the fastest growing allied health professions

—are likely to be available in primary care practices,

 so we sought to assess the extent to which medical assistants trained as health coaches could improve the quality of care received by the patient based on the CCM.

The Health Coaching in Primary Care (HCPC) study was a randomized controlled trial of 12 months of health coaching versus usual care for low-income patients with uncontrolled type 2 diabetes, hypertension, and/or hyperlipidemia. Patients who received health coaching were more likely to achieve the primary outcome of control for 1 or more of the conditions for which they were enrolled.

 In this paper, we reported on the effect of health coaching on patient-reported quality of care received from their primary care clinic, as well as satisfaction with their clinic.

Setting, Participants, Enrollment, and Randomization

A detailed description of the HCPC study design and methods has previously been published.

 Briefly, the study was conducted at 2 safety net clinics in San Francisco from March 2011 to May of 2013. Patients were considered eligible if they were between the ages of 18 and 75 years, spoke Spanish or English, could be reached by phone, and had poorly controlled diabetes (glycated hemoglobin >8%), hypertension (systolic blood pressure ≥140 mm Hg), or hyperlipidemia (low-density lipoprotein ≥160 mg/dL for patients without diabetes or ≥100 mg/dL for patients with diabetes). A total of 664 eligible patients were identified at the 2 clinic sites, of which 441 (66.4%) consented and were enrolled (see 

). After enrollment and completion of baseline measures, participants were randomized to the health coaching arm (n = 224) or the usual care arm (n = 217) by opening the next randomly ordered, sealed envelope.

The health coaches included 3 certified medical assistants who attended 40 hours of health coach training over 6 weeks using a curriculum developed by the study team that included instructions on using active listening and nonjudgmental communication; helping with self-management skills for diabetes, hypertension, and hyperlipidemia; providing social and emotional support; assisting with lifestyle change; facilitating medication understanding and adherence; navigating the clinic; and accessing community resources. Training incorporated aspects of the CCM,

Health coaches interacted with patients at medical visits, individual visits, and by phone calls. The minimum required frequency of contact was once every 3 months for in-person visits (often as part of a medical visit) and monthly for additional contacts such as phone calls. The health coach met with the patient before the visit for medication reconciliation, agenda-setting, and the review of lab numbers; usually stayed in the exam room during the medical visit and met with the patient afterward to review the care plan and check for patient understanding; and assisted the patient in making action plans

 to increase physical activity, improve healthy eating, reduce stress, or improve medication adherence. In addition, the health coach facilitated navigation of other resources, such as diagnostic imaging or referrals to specialists, by making follow-up appointments or facilitating introductions to behaviorists or other clinic resources.

After each contact with a patient, the health coach reported the type of contact (phone, medical visit, or in-person visit with the health coach only), approximate length of the visit, and topics covered from a list of possible activities (eg, review medication adherence, create or follow-up on an action plan).

Patients randomized to usual care continued to have visits with their clinician over the course of the 12-month period and had access to any additional resources that were part of usual care at the clinic, including diabetes educators, nutritionists, chronic care nurses, or educational classes.

Patient demographic characteristics were assessed by survey at the time of enrollment and at 12 months. Quality of care received was assessed by patient report using the previously validated Patient Assessment of Chronic Illness Care (PACIC) scale.

 The PACIC scale measures patient-reported receipt of services included in the CCM over the past 6 months using a 5-point Likert-type response scale ranging from 1 (“almost never”) to 5 (“almost always”). Three examples of items are: “Asked to talk about your goals in caring for your condition,” “Helped to make a treatment plan that you could carry out in your daily life,” and “Contacted after a visit to see how things were going.” The PACIC scale has been associated with increased physical activity, receiving appropriate laboratory assessments,

 greater engagement in all self-management behaviors, and higher quality of life.

 Scores are reported as the mean Likert score (range = 1-5) for all 20 items, as well as scores for each of the 5 domains (patient activation, delivery system design/decision support, goal-setting items, problem-solving/contextual counseling, and follow-up/coordination). Patient satisfaction with their primary care clinic was assessed by a single item: “How likely would you recommend your clinic to your friend or relative?” with a response scale from 1 (“definitely not recommend”) to 5 (“definitely recommend”) and analyzed as a dichotomous variable (“definitely recommend” vs “not definitely recommend”).

Analyses were in accordance with the Consolidated Standards of Reporting Trials (CONSORT) guidelines for reporting results from clinical trials.

 Group comparisons were conducted using χ

 test for categorical data and analysis of variance for normally distributed continuous variables. PACIC data were treated as missing if less than two-thirds of the items were answered. Effect size (Cohen’s d) was calculated as the difference in change in PACIC score divided by the pooled standard deviations from both study arms at baseline and 12 months.

 Unadjusted change in PACIC scores between study arms was compared using 

 tests, while differences in the proportion of patients reporting they would definitely recommend their clinic at 12 months were compared using χ

 testing. Generalized Estimating Equation models were used to account for clustering by clinician and to additionally control for baseline levels of the outcome. All 

 values are 2-sided. The primary analysis was done without replacement of missing data (data assumed to be missing at random). Analyses were repeated using multiple imputation to test the assumption of missing at random using NORM version 2 software,

 which imputes data through the expectation-maximization algorithm. All other statistical analyses were performed using SPSS version 19.0 (SPSS Inc, Chicago, Illinois).

The PACIC score was available at 12 months for 200 (89%) of the 224 patients enrolled in the health coaching arm and 166 (76%) of the 217 patients enrolled in the usual care arm. Patients with missing data were more likely to be in the usual care arm and more likely to speak Spanish, but otherwise did not differ significantly from patients without missing data.

On average, health coaches had a mean of 12.4 interactions with patients over the 12-month period. Nearly half of (45%) the interactions occurred during meetings between the health coach and patient while slightly over a quarter occurred during a medical visit (29%) or as a phone call (26%). The average length of interaction was approximately 15 minutes for phone calls, 30 minutes for individual meetings, and 45 minutes for medical visits (including time spent in the room during the patient visit with their clinician). The most common topics discussed during interactions were medications (75%), reviewing clinical values such as blood pressure and laboratory test results (60%), and discussing lifestyle changes such as health eating and physical activity (59%).

Study participants in each study arm were similar with respect to demographic characteristics (

). Over two-thirds (69%) were Spanish-speaking, 57% had less than a high school education, and 88% reported a household income of less than $20,000 in the past year.

The mean overall PACIC scores were nearly identical in the 2 study arms at baseline but increased by 0.83 for patients assigned to receive health coaching compared with 0.07 in the usual care group (effect size = 0.79) (

). This pattern was seen in all 5 domains of the PACIC, with the greatest difference in the patient activation domain. Similarly, the proportion of patients who reported they would highly recommend their primary care provider was similar at baseline but increased significantly more in the health coach group. The differences remained significant after adjustment for baseline levels of the PACIC score and satisfaction, and for clustering by clinicians.

When the association between health coaching and PACIC scores was examined separately for English- and Spanish-speaking participants, a significant difference (interaction) was found, with the association being substantially stronger in the Spanish-speaking group (

 <.01) even after controlling for differences in baseline PACIC scores, which were lower in Spanish-speaking patients compared with English-speaking patients (2.94 ± 1.04 vs 3.21 ± 0.96).  The effect size for the PACIC score was 1.00 (

 <.001) for Spanish-speaking participants compared with 0.40 for English-speaking participants (

 value not significant). This pattern was seen across all 5 PACIC domains and for the measure of patient satisfaction with their clinic. Repeating the analyses using imputed data did not materially affect the results.

Health coaching has been shown to improve chronic disease management and the use of health coaches in primary care has become more popular.

 To our knowledge, this is the first study to show that adding health coaching by a trained member of the patient care team also improves patients’ reported overall quality of care and satisfaction with their clinic. Other studies have looked at interventions to improve PACIC scores, but they have not utilized unlicensed personnel, have not emphasized health coaching techniques, and have generally found smaller improvements in PACIC scores.

Previous studies have examined other interventions aimed at improving quality of chronic care as measured by the PACIC. One study, using practice nurses to implement components of the CCM for patients with diabetes found no improvement in PACIC scores.

 Another study of a guided care intervention by practice nurses, based on the CCM, for patients aged over 65 years with complex chronic conditions, found an adjusted difference in PACIC score of just 0.2 after 18 months compared with usual care.

 A disease-management program based on the CCM for patients with chronic obstructive pulmonary disease found a small improvement in PACIC score (0.08) compared with the control group (—0.05)

—a result similar to Cramm et al’s study of other chronic disease management programs.

 Another study, by Schillinger et al, compared automated telephone self-management support or group medical visits to usual care for patients with diabetes and reported effect sizes of 0.51 and 0.53, respectively, for total PACIC score.

 In our current study, we found that, compared with patients receiving usual care, patients receiving health coaching had a statistically significant improvement in their mean total PACIC score with a standard effect size of 0.77—close to the conventional threshold for a large effect size of 0.8.

Health coaching had a significantly greater impact on patient-reported quality of care and satisfaction in Spanish-speaking patients. This is not surprising, as language often presents a barrier to receiving high-quality care, as evidenced by the lower baseline PACIC scores of Spanish-speaking patients; with our health coaches being bilingual, they were likely able to reduce this barrier. Although there was also a positive effect of health coaching for English-speaking patients, the magnitude was smaller and not significant.

One strength of the current study is the randomized controlled design, which avoided the potential biases due to the patient self-selecting to receive health coaching or usual care. The study included English- and Spanish-speaking low-income patients who received care through safety net clinics—a group for whom improvement in the quality of care is especially important. However it should be noted that the impact of health coaching might vary in a different population. A limitation was that coached patients were more likely have outcome information available compared with usual care patients (89% vs 76%). It is possible that a differential response by study arm could have introduced bias, though analyzing the data using imputed values for missing data did not change the results.

Our results support the view that patients perceive services received from medical assistant health coaches as part of their primary health. The magnitude of the improvement in chronic care for patients who received health coaching is greater than what has been reported from most other practice improvement studies, suggesting that this model of health coaching should play an important role in improving the quality of care for patients with chronic conditions, particularly for non—English-speaking patients.

The authors gratefully acknowledge the work of health coaches Christina Araujo, Adriana Najmabadi, and Dalia Canizalez; study research assistant Marissa Pimentel; as well as the support of medical directors Dr Elsa Tsutaoka and Dr Ricardo Alvarez and the staff at the participating clinics. The study was conducted under the auspices of the UCSF Center for Excellence in Primary Care.

 Department of Family and Community Medicine, University of California at San Francisco School of Medicine (DHT, DH, RW-G, DD, CP, TB, EHC), San Francisco, CA; Silver Avenue Family Health Center (EHC), San Francisco, CA.

 This study was supported by a generous grant from the Gordon and Betty Moore Foundation (Grant #2492).  

 The authors report no relationship or financial interest with any entity that would pose a conflict of interest with the subject matter of this article.

 Concept and design (DHT, TB, EHC); acquisition of data (DHT , RW-G, CP,  DD, EHC); analysis and interpretation of data (DHT, DH, RW-G, EHC); drafting of the manuscript (DHT); critical revision of the manuscript for important intellectual content (DHT, RW-G, CP, DD, TB); statistical analysis (DH); provision of patients or study materials (CP, DD); obtaining funding (DHT, TB, EHC); administrative, technical, or logistic support (RW-G, CP, DD, EHC); and supervision (DHT, RW-G, EHC).

 David Thom, MD, PhD, MPH, Department of Family and Community Medicine, University of California, San Francisco School of Medicine, 1001 Potrero Ave, Bldg 80/83, San Francisco, CA 94110. E-mail: dthom@fcm.ucsf.edu.

Medical assistants trained as health coaches substantially improved patient-reported primary care under the Chronic Care model measured by the Patient Assessment of Chronic Illness Care.

Rachel Willard-Grace, MPH

Articles

Costs for a Health Coaching Intervention for Chronic Care Management

Health Coaching by Medical Assistants Improves Patients' Chronic Care Experience