Eligible participants were characterized by the following: SFHP membership; ≥1 primary care clinic visit in the preceding 24 months at one of our designated clinics; age 18 or above; a diagnosis of diabetes (type 1 or 2); English-, Cantonese-, or Spanish-speaking; access to a touch-tone phone; and plans to remain in the region during the evaluation period (12 months). Patients who were pregnant or unable to provide verbal consent were excluded. The diagnosis of diabetes was assessed in two ways. First, we searched for SFHP members in the CHNSF electronic diabetes registry. Second, we searched for members with SFHP claims data consistent with diabetes (ICD-9 codes for utilization and pharmacy claims for oral or injectable glycemic medications or glucose self-monitoring supplies). If members were identified through SFHP claims data but were not found in the registry, we validated the diabetes diagnosis by reviewing electronic health records for fasting glucose ≥ 126 mg/dl and/or clinician-documented diagnosis of diabetes based on HbA1c ≥ 7% [43]. During the course of the evaluation trial, the American Diabetes Association revised diagnostic criteria to include HbA1c ≥6.5% [44], but we did not modify our eligibility criteria to avoid introducing selection bias and because evidence suggests a variable delay in the adoption of guideline criteria by practicing providers [45,46].

SFHP mailed to potentially eligible members' homes and clinicians' offices SMARTSteps post cards describing this new benefit in English, Spanish, and Cantonese and offering an SFHP number to call for information. SFHP enrollment workers fluent in these three languages also actively recruited potentially eligible health plan members through scripted telephone calls. The enrollment workers confirmed eligibility for SMARTSteps by phone and offered $25 gift cards as incentives for participation. All eligible members who reported interest in SMARTSteps were then randomized by SFHP to one of four intervention statuses (described below) regardless of interest in the UCSF evaluation.

SFHP workers also assessed participant willingness to complete evaluation interviews administered by UCSF staff. Those members willing to participate in the interviews were then telephoned by UCSF research assistants to obtain verbal consent for participation in the evaluation interviews and were offered $50 gift cards as incentives for each interview completed (up to $150).

The Committee on Human Research at the University of California, San Francisco (H9894-32044-01) approved collection of interview data by UCSF from consenting participants after obtaining verbal consent. Because all recruitment and implementation was conducted by SFHP by phone, it was felt that requiring a separate visit to obtain written consent at UCSF or the primary care clinic sites reduce participation, leading to selection bias, and disrupt clinic workflow. Because the interviews offered minimal risk, the Committee approved verbal telephone consent for these interviews. In addition, ethics approval included waiver of informed consent to abstract and analyze CHNSF clinical data from the CHNSF electronic health record, CHNSF diabetes registry, SFHP claims data, and SFHP health coaching data for all SFHP members eligible for SMARTSteps.

Those randomized to the wait-list continued to receive usual care through their clinics, as well as all existing SFHP benefits (reminders and incentives for receipt of recommended health services, including laboratory testing, eye and foot examination, and influenza vaccination) [43,47]. At the end of the 6-month wait-list period, each participant "crossed-over" to begin SMARTSteps-ONLY or SMARTSteps-PLUS, depending on initial randomization arm.

Participants randomized to SMARTSteps-ONLY received the ATSM intervention within 2 weeks. Developed with extensive input from patients to be sensitive to literacy, language, and culture in the target populations [32], this ATSM system provided 27 weeks of 8-12 minute weekly calls in English, Cantonese, or Spanish. Patients specified the weekday and time convenient for their schedules or called toll-free into the system if they missed their scheduled call. The content consisted of rotating sets of queries about self-care (such as diet, exercise, and medication adherence), psychosocial issues (such as depressive symptoms), and access to preventive services (such as eye care). Patients responded via touch-tone commands, and based on their answers, patients heard automated health education messages in the form of narratives.

Patients answering "out-of-range" on an item or selecting a "call back" option received a telephone call within 3 days from a language-concordant SFHP health coach. This system was designed to promote health coach efficiency and effectiveness by focusing outreach calls to patients needing further support based on ATSM responses. Health coaches provided education and engaged in collaborative goal-setting to form patient-centered action plans, a core process for self-management support by which patients set short-term goals to improve their self-management [20,35].

SFHP used a health coach for responding to ATSM triggers, rather than a nurse practitioner care manager as was done in the IDEALL trial [33]. Health coaches - SFHP employees without a medical background or postgraduate training - conducted phone calls under the supervision of a registered nurse care manager. SFHP registered nurse supervisors and UCSF study physicians taught health coaches about diabetes and trained them in behavior change counseling and overcoming barriers to health communication. UCSF staff also guided the development of written protocols and scripts to respond to potential ATSM call triggers, such as assessing hypoglycemia symptoms or causes.

Health coaches documented the content of their phone interactions, including patient-generated action plans and whether they were achieved, using an SFHP care management database system. SFHP health coaches also contacted the designated clinic contact and/or the primary care provider (PCP) using standardized templates for patients with pre-specified safety issues (such as a new medical symptom) or access concerns (such as need for refills or appointments). Non-urgent issues were communicated by email or fax, while urgent issues were also conveyed by phone.

Patients who indicated on their ATSM responses that they missed their diabetes, blood pressure, or cholesterol medications for 3-7 days in the previous week received calls from health coaches to troubleshoot barriers to adherence. These calls were only triggered by patient self-disclosure in response to ATSM queries, unlike adherence counseling calls triggered for SMARTSteps-PLUS participants described below.

The goal of the SMARTSteps-PLUS intervention was to detect and intervene for participants whose medication treatment was suboptimal, either because of non-adherence or potential missed opportunities to intensify their regimens. Because SFHP does not have prescribing authorization for its members, the intervention focused on enhancing medication regimens and adherence through collaborative goal-setting with patients and feedback to PCPs. Participants randomized to SMARTSteps-PLUS immediately received the ATSM intervention described above, as well as medication activation and intensification coaching triggered by refill non-adherence or suboptimal achievement of cardiometabolic treatment goals. Based on review of evidence-based guidelines [43], SFHP and UCSF collaborated to develop a protocol to improve medication adherence and promote patient-centered intensification by harnessing electronically available data from two additional sources: monthly SFHP pharmacy claims and clinical laboratory and blood pressure data from the electronic health records of the CHNSF.

The medication activation protocol targeted 3 groups of patients within the SMARTSteps-PLUS arm (see Figure 2):

1. Patients who indicated in response to weekly ATSM queries that they missed 3-7 days of medications in the last week.

2. Patients with a 15-day to 6-month gap in refilling specific cardiometabolic prescriptions: oral hypoglycemic, insulin, anti-hypertensives (including ACE inhibitors and angiotension-receptor blockers), cholesterol-lowering medications, or glucose testing strips (for patients receiving insulin or sulfonylureas), based on pharmacy claims data. We chose a lower limit of 15-days to account for potential delays in SFHP receipt of pharmacy claims and an upper limit of 6 months in case healthcare providers actively discontinued these medications.

3. Patients with suboptimal achievement of cardiometabolic goals [43]:

• Systolic blood pressure ≥ 130 mmHg at last recorded check or no measurement in clinical registry within preceding 6 months

• Diastolic blood pressure ≥ 80 mmHg at last recorded check or no measurement in clinical registry within preceding 6 months

• Hemoglobin A1c (HbA1c) > 7.0% on last measurement or no measurement in clinical registry within preceding 6 months

• Low-density lipoprotein (LDL) > 100 mg/dL on last measurement or no measurement in clinical registry within preceding 12 months

• Urinary albumin/creatinine > 30 mcg/mg for patients who were not prescribed or late to refill an ACE inhibitor or angiotensin-receptor blocker

An individual SMARTSteps-PLUS participant could fall into none, one, or more than one of these groups at any given time. SFHP queried the pharmacy claims and clinical registry data monthly to provide health coaches with a list of SMARTSteps-PLUS participants to be called.

For patients with evidence of non-adherence, SFHP health coaches were trained with specific written protocols and scripts to promote self-disclosure of medication non-adherence and troubleshoot barriers such as confusion about doses or frequency of medications, forgetfulness, concerns about side effects, or health beliefs (see Additional File 1). For patients who had not had cardiometabolic measurements within guideline-concordant timeframes, health coaches counseled patients about the reasons for these measures, encouraged them to talk with their providers, and notified PCPs of patients willing to obtain these tests. For patients with suboptimal achievement of treatment goals, health coaches were trained to counsel patients about their blood pressure or lab values and assess their willingness to discuss with their PCPs the possibility of taking more medication. Health coaches also encouraged patients to undergo potential lifestyle changes. Based on their discussions, health coaches engaged patients in collaborative goal-setting and action planning [20,35]. Health coaches then notified the clinic contacts and/or PCPs about interactions with any patients who needed laboratory or blood pressure assessments, reported barriers to medication adherence, or indicated a willingness to discuss medication intensification.

Table 1 lists study outcomes.

For participants consenting to the UCSF evaluation trial, measures of all primary and secondary patient-reported outcomes were derived from structured interviews conducted in English, Spanish or Cantonese. For Spanish and Cantonese interviews, survey questions were translated and back-translated in their respective languages. Interviews were administered using computer-assisted telephone interview (CATI) software or paper survey, with responses later entered into the CATI-created database. Consenting participants completed baseline interviews within 2 weeks of being randomized, prior to the first ATSM call. SMARTSteps-ONLY and SMARTSteps-PLUS participants completed the 6-month follow-up interviews after the conclusion of their ATSM intervention, and wait-list participants completed the 6-month follow-up interviews before the start of their ATSM calls. Wait-list groups received an additional 6-month follow-up interview (12 months after enrollment) after completion of intervention period. Participants received a $50 gift card for a pharmacy store at the completion of each interview.

For all eligible SFHP members, we collected socio-demographic variables and cardiometabolic outcomes data using monthly downloads of SFHP claims data and CHNSF clinical registry data. At the conclusion of the project, we will obtain SFHP administrative and claims data for utilization outcomes.

Self-reported socio-demographic variables included age, gender, race/ethnicity, language, English proficiency, years since immigration to the U.S., marital status, educational attainment, employment status, annual household income, self-reported health literacy [49,50], and duration of diabetes. We obtained SFHP administrative and CHNSF data regarding age and gender. Preferred language is captured in both SFHP and CHNSF data, and where this preference was discrepant, we used the language reported to SFHP enrollment workers.

The primary outcome variables are physical and mental functional status (SF-12) and the number of days spent in bed due to illness [51]. Secondary outcomes include diabetes self-efficacy and self-management behavior, as well as medication adherence in the preceding 7 days [52-54]. Secondary outcomes also include patient perspectives on the structure of their care, as measured by two instruments which have been translated and validated for our population in our prior work: Patient Assessment of Care for Chronic Conditions (PACIC) and interpersonal processes of care (IPC) [33,55,56].

These secondary outcomes will be analyzed on the larger set of SFHP participants randomized in the trial (N = 362), regardless of whether or not they participated in the evaluation interviews. Cardiometabolic outcomes include measures of HbA1C, LDL, and blood pressure obtained by clinical providers. Data will be derived from CHNSF electronic health record, the clinical registry database, and paper chart review. Values will be used at baseline only if obtained within one year before program enrollment. Follow-up laboratory and blood pressure values will be included only if obtained within 90 days before or after the 6-month and 12-month follow-up dates.

Finally, for all eligible participants, we will use SFHP administrative claims data to examine measures of care delivery processes: proportion receiving blood pressure checks, laboratory testing, retinal examinations, influenza vaccination, and pneumococcal vaccination at guideline-concordant frequencies [43]. We will also examine utilization, including the frequency of emergency department visits and hospitalizations.

Using data from the ATSM system, we will assess patient engagement as measured by exposure intensity (defined as the proportion of the 27 weeks of exposure, participants replied to the weekly calls) and the proportion of weekly calls that are determined to be complete (calls answered and responses to 100% of queries), partially complete (call answered and responses to < 50% of queries), and incomplete (call answered but responses to 0% of queries) [32]. In addition, using data from the SFHP health coach database, we will analyze at the participant level the number of health coach calls that were triggered by ATSM responses; the number of calls attempted and/or completed by health coaches; and the number of action plans created and/or achieved, as reported by patients and documented in the database. We will conduct fidelity assessments to examine potential differential engagement across languages, clinics and study arms. Finally, we will assess satisfaction and perceived usefulness of the intervention as reported during evaluation interviews at 6- and 12-month follow-up.

Sample size calculations were based on comparison of primary outcomes for the combined intervention groups (SMARTSteps-ONLY and SMARTSteps-PLUS) with the combined wait-list control groups. Based on an eligible population of approximately 500 SFHP members, we anticipated recruitment of 260 participants to intervention and control with a 10% drop-out or loss to follow-up rate at 6 months. Using a two-tailed test, we calculated that we would have 80% power to detect a standardized effect size (SE) of 0.35 in the primary outcome of SF-12 scores, comparing the combined intervention groups with the combined wait-list groups. In the prior RCT, we observed standardized effect sizes in this range [33].

We estimated that over the course of the two-year observation period, approximately 20% of enrolled participants would have insufficient data for secondary cardiometabolic outcomes to be included in these analyses. Under this assumption, we would have 80% power to detect a difference in HbA1c of 0.51% (SE 0.28) when comparing wait-list vs. combined ATSM interventions participants. This effect size is intermediate between that estimated to be the effect on glycemic control of diabetes disease management programs that involve team changes (-0.3% HbA1c improvement) and those that involve medication intensification strategies (-0.6%) [57].