Protocol for the CHORD project (community health outreach to reduce diabetes): a cluster-randomized community health worker trial to prevent diabetes

The HAP will be created collaboratively by the patient and CHW and reviewed by the research team at a weekly case conference. The CHW returns to the patient with the revised plan to make any final changes. The HAP specifies the patient’s goals and how he/she will engage with the intervention activities. The CHW guides the patient in setting well-defined and achievable goals. CHWs will use motivational interviewing techniques when revisiting these goals with patients during the intervention. The HAP, along with 3-month and 6-month progress updates during the intensive phase, will be shared with the patient’s PCP, who may suggest modifications based on their knowledge and relationship with the patients.

CHWs will conduct one-on-one health education sessions on topics including health self-management, exercise, and healthy eating. CHWs will conduct at least one session a month during the 6-month intensive period, but activated patients may request supplementary modules and have more frequent contacts, as often as weekly. Sessions will occur by phone or in person according to patient preference. CHWs will attempt to meet each patient in person at least once during the intervention, at either the patients’ home, neighborhood, or clinic. Patients also will receive a packet of educational materials about DM, and exercise and nutrition to prevent DM, written at less than an 8th grade reading level in English or Spanish, containing material aligned with each monthly health education session.

CHWs will call patients before and/or after PCP visits, in order to encourage discussion about diabetes prevention with the doctor or to review the doctor’s suggestions. CHWs are available to meet patients at clinical sites for in-person conversations.

The intervention toolkit allows for personalization of this core intervention in two ways. First, it contains additional activities CHWs may offer to patients, including home visits and an ongoing schedule of group activities such as neighborhood walks or group grocery shopping. Second, a patient may intensify any activity by increasing their engagement in core intervention components. For instance, patients may request more frequent calls with the CHW, additional health education modules, or enroll in multiple hospital- or community-based programs. Hospital regulations permitting, patients may also text-message with their CHW, reinforcing the social support dimension of the intervention. The toolkit balances delivering a uniform, evidence-based DM prevention intervention with patient-level tailoring expected to increase sustained behavioral change [29, 30]. Standardized protocols will be used to ensure that CHWs dedicate sufficient time to engaging patients with lower activation levels.

Following the 6-month intensive phase, the intervention transitions into a 6-month maintenance phase, during which CHWs will conduct monthly check-in phone calls with their patients; follow-up in-person visits or health education will not be offered during this phase. Patients may also contact their CHW for support. Beginning during the first wave’s maintenance phase, we will recruit patients that have completed the intensive intervention phase as “buddies” that can provide additional social support for new intervention patients in subsequent waves.

The primary outcome of the study is type 2 DM incidence. Any patient with at least two encounters with a DM diagnosis (ICD-9 codes 250, 357.2, 362.0, or 366.41) will be defined as having DM. Each incident case will be verified by looking for evidence of hyperglycemia (at least two fasting glucose readings > 126 mg/dL or an HbA1c ≥6.5%), or the prescription of DM medications, to avoid under- or overestimating DM incidence. A similar approach was used by Miller and colleagues in the VA DM Epidemiology Cohort [28]. Both change in weight in pounds (and body mass index, BMI) and the proportion of overweight (BMI 25.0–29.9) and obese (BMI ≥ 30) patients are measured as secondary outcomes over the course of the study. A range of additional clinical outcomes will be measured – glycemic control (HbA1C), blood pressure, smoking status, and Framingham Risk Score (10-year) – all derived from the respective  CDWs at each hospital practice throughout the study. Health service utilization – number of PC visits, PC telephone encounters, emergency and urgent care visits, visits to hospital-based wellness programs, and visits to the nutritionist – will also be assessed at baseline and conclusion and will be drawn from the medical record. Self-reported health attitudes and behaviors, including health self-perception, diet, exercise, and activation are further patient-centered outcomes, measured in the patient intake survey (see Table 1).

Implementation fidelity and PCPs’ perspectives about the CHW intervention and CHW integration into the clinic for DM prevention will be assessed (see Table 2) considering factors theorized to influence implementation outcomes, including characteristics of the intervention; of PCPs, primary care teams, and CHWs; and of the community and policy context that shapes health care access and resources in the city [31].

Implementation fidelity will be assessed by the content of CHW activities and the coverage (number of patients reached), duration (how long each patient is followed up), and frequency (how often patients are contacted) of CHW-patient contact. The perceived acceptability (perception that the intervention is agreeable or satisfactory); adoption (intention to use or deploy the intervention); appropriateness (perception about the fit or relevance of the intervention for a practice setting); and feasibility (perception about the extent to which the intervention can be successfully used or carried out within a given setting) of the CHW intervention and CHW integration into the clinic will be assessed using surveys of PCPs and interviews with PCPs, clinical staff, CHWs, and practice administrators at each site [32]. Exposure to the CHW intervention may affect how providers individually and PC teams collectively function in preventing DM and working with a CHW, thus potentially altering their usual practices in managing prediabetic patients.

A survey given to all PCPs at baseline and intervention-end and semi-structured interviews every 6 months with a subsample of PCPs and clinical staff in the intervention group are used to assess their attitudes and approaches regarding DM prevention and incorporating peer coaching into clinical care. Semi-structured interviews will be conducted annually with practice administrators at BH and the VA selected as key informants for their expertise in DM prevention and their clinic’s administrative structure.

Through weekly activity logs, CHWs track their patient contacts and case notes, and the researchers will conduct interviews with the CHWs every 6 months to explore challenges, successful and unsuccessful strategies, and their perspective on their role as part of the clinical team.

As a pragmatic, population-based trial, this study aims to understand the impact of our interventions on panels of patients in a real clinic environment and therefore will not collect any clinical data directly from patients. Instead, the study uses measurements taken during regular clinic visits and recorded in the EMR. The study team works directly with data analysts at the VA and BH to extract the study variables from the EMR and to maintain the study database.

Prior to the initial patient contact, CHWs will receive basic demographic and clinical data, drawn from the EMR, for their patient panel. During the initial contacts with patients, CHWs will conduct a comprehensive intake process. The intake consists of demographic background; self-reported health status, physical activity, and diet; DM risk perception; prediabetes awareness; previous use of prevention resources; social determinants of health; and the 6-item Patient Activation Measure (PAM-6) [33]. Intervention group patients complete this survey at baseline, as part of the intake assessment, and at the end of their 1-year intervention period. Control patients will receive, by mail, a self-administered version of the survey only at the close of the 1-year intervention period. As a response incentive, patients will receive $10 gift cards upon completing the survey.

Data to assess implementation fidelity will be collected across each study phase through a variety of measures. CHWs complete standardized reports on their patient interactions in their weekly activity logs, including the number of patients they target, reach, and refer, and the number, type, and length of patient interactions. The rate of patient contact failure will also be tracked by recording returned mail and out-of-service or incorrect phone numbers. Meeting minutes will also be recorded from weekly CHW-supervisor meetings and will document activities, challenges, and lessons learned by CHWs. Lastly, every month, the CHW Coordinator will monitor one of each of the CHWs telephone counseling sessions to ensure counseling skills, offer feedback, and ensure treatment fidelity. The calls will be evaluated using a standardized checklist.

Analyses will be conducted at the patient level, aggregated to their PCP, controlling for the cluster-randomized design. The annual hazard rate of incident cases of DM over time is the primary clinical outcome for power calculations. The current rate of HbA1c prediabetes screening among eligible patients without prevalent DM since 2012 is 73–74% and is likely to approach 80% due to clinical guidelines and increased healthcare systems transformation efforts. Of those with an HbA1c test since 2012, 41–44% of patients fall within the prediabetic range. While the above inclusion and exclusion criteria return 9300 eligible patients, 8200 at BH and 1100 at the VA, CHW caseload limitations will prevent us from reaching all eligible patients during the 3-year intervention. Assuming a caseload of 72 patients per fulltime CHW per wave (with 3.5 fulltime equivalent CHWs across both sites), an estimated 20% unreachable rate, and a 60% consent rate among intervention patients, our total intervention sample is anticipated to be 1008 patients, split evenly at each site (see Fig. 3). Control patients will be randomly selected for each wave at each site. In the prediabetes population, the 3-year DM incidence is estimated to be 15% (based on preliminary analysis of a similar NY VA cohort from 2010 to 2012, followed through 2015). A conservative estimate of the intracluster correlation coefficient (ICC) is 0.01, based on the median value found in a systematic review of cluster randomized trials [34] and the highest value calculated in our previous VA study [35]. This planned sample size will provide at least 80% power to detect an expected difference in annual hazard rates of time to DM incidence between two arms (3.5% per year in intervention vs. 5.4% per year in control), at a 0.05 significance level. By assuming exponential distribution, this difference in annual hazard rates can be approximated by the difference in incidence rates over 3 years (10% in intervention vs. 15% in control).

We will conduct both an intention-to-treat analysis (all patients randomized to either study group, including those who decline to participate or we cannot contact) and a per protocol analysis (only patients who engage in the intervention according to the protocol). Descriptive analysis techniques (measures of central tendency and variability, frequencies, and proportions) will be conducted to describe baseline characteristics of participants (patients and their PCMH staff) in the control and intervention groups. Chi-square tests and t-tests will be used to check if these baseline characteristics are balanced between the two study groups. For continuous outcomes, normal assumptions will be checked, and non-parametric alternatives and transformations will be considered as needed.

To assess the overall effectiveness of the CHW intervention, we will compare cumulative DM incidence rates and time to DM incidence (in addition to other outcomes: weight loss, obesity rate, and health care utilization) between the two study groups, using intention-to-treat and per protocol analyses. DM cumulative incidence rates at 12-month follow-up of the intervention and control arms, aggregated across the 4 study waves, will be compared using a chi-square test. Similar analyses are conducted for other categorical outcomes (e.g. obesity rate) and t-tests will be used to compare continuous outcomes (e.g. weight change). We will further conduct generalized linear mixed-effects modeling [36] for these outcomes, which (1) takes into account the correlation among patients within teams, (2) analyzes repeated measures at several time-points (baseline, midpoint, and post-intervention) simultaneously, (3) includes team level or patient level baseline covariates that are found to be unbalanced between two arms, and (4) deals with attrition and other missing data.

We will use multilevel survival analysis to determine the effect of the intervention on time to DM, as patients are enrolled in waves and the time to DM incidence may be censored by the end of study. Specifically, we will fit Cox proportional hazard regression models with mixed effects [37] to examine the hazard ratio between two groups, adjusting for important covariates and taking into account the correlation among patients within teams.

Similar generalized linear mixed-effects models will be used for the secondary outcomes, to explore the relationship of the intervention’s effectiveness to specific patient and provider or team characteristics. The study investigates interactions between the intervention and the patients’ level of activation, co-morbidities (hypertension, depression, and smoking), health care utilization, and perceptions of the quality of care.

Secondary analyses to better understand the impact of the individual components of this complex set of interventions will be conducted. Since the intervention is designed to test the effectiveness of an overall strategy, the study seeks to determine which of the intervention elements and dose of intervention (e.g., number of CHW contacts, number of mailings, or successful referral to hospital- or community-based programs) are most strongly associated with any improved outcomes. Subgroup analyses will also be conducted, to see which subgroups may benefit more from the intervention than the others. Such subgroup analyses should be considered as exploratory analyses and any resulting findings should be confirmed by independent studies.

To limit bias due to contamination and missing values, the baseline characteristics of patients who switch teams or are removed from the panel or have incomplete repeated outcomes will be assessed and compared with those of the other participants. The (generalized) linear mixed-effect models discussed earlier can deal with attribution problems and other missing data, under the missing-at-random assumption [38]. To better understand the missing mechanism, outcome variables will be summarized by discontinuation status in each group. If the trajectories of the primary outcome variables from those with incomplete data are similar to those with complete information, the data may be missing at random; otherwise, they may be missing not at random. A sensitivity analysis will then be conducted to assess the impact of missing not-at-random data on our analysis. The sensitivity analysis will test whether our primary statistical analysis remains credible if it includes cases where those patients with incomplete data have outcomes that are unfavorable to the intervention group of the study. The sensitivity analysis will be conducted via multiple imputation using sequential modeling [39].

Survey data will be used to assess PCPs’ readiness to work with CHWs for DM prevention. Baseline and post-intervention scores will be compared using Wilcoxon signed-rank tests. To assess implementation fidelity, baseline and 6-monthly scores on the coverage, duration, and frequency of CHW-patient contact for BH and VA patients will be compared using ANCOVA. Content of delivery will be assessed qualitatively based on a sample of CHW-patient contacts. Implementation fidelity will be compared across BH and the VA to examine if differences in organizational setting shape the implementation process.

All interview transcripts and field notes will be analyzed for common themes by two independent researchers using both deductive and inductive approaches. First, initial codes will be identified based on our study aims and on relevant prior implementation research. Two researchers will independently read the transcripts, identifying new codes based on any new themes or subthemes that emerge from the data. Second, the researchers will meet with each other and the study research team to compare and discuss codes, resolve discrepancies, and iteratively develop a codebook with code names and meanings. Third, researchers will apply the codebook to the transcripts. Researchers will identify patterns, common themes, and connections among themes throughout the transcripts. Baseline interviews will be used to refine the intervention toolkit and to explore anticipated barriers and facilitators to incorporating CHWs into the PC teams’ workflow. Mid- and post-intervention interview and case review data will help us to determine key barriers, facilitators and lessons learned from staff and CHWs regarding the implementation of CHW support and the integration of CHWs in PC practice for DM prevention in an underserved population, accounting for organizational setting.