Study design.

Details of the study design and protocol have been published previously.[8] The trial was approved by the Institutional Review Board (IRB) of Brigham and Women’s Hospital and the privacy board of Horizon Blue Cross Blue Shield of New Jersey (“Horizon”) and is registered with clinicaltrials.gov (NCT02910089). A waiver of informed consent and authorization to use the study data was granted by the IRB and Horizon privacy board because of the minimal risk nature of the study and because the research could not have feasibly been done without the waiver or access to potentially-identifiable information. The trial protocol was designed, written, and executed by the investigators (S1 Protocol). We designed the trial to be pragmatic using PRECIS-2 trial guidance and reported using CONSORT trial guidelines (S1 Checklist).[19] The authors analyzed the trial data using an independent copy of the study database and vouch for analytic accuracy and fidelity to the study protocol. Study enrollment began in October 2016 and was completed in December 2016. Follow-up of all trial participants ended in December 2017. Final data were received by June 2018.

Study setting and population.

The trial was conducted at Horizon Blue Cross Blue Shield of New Jersey (Horizon), the largest health insurer in New Jersey. We included patients ≥18 years of age who filled 1 or more oral hypoglycemic agents within the 12 months prior to randomization and who had evidence of poor diabetes control (defined as HbA1c ≥8%).[2, 4] Horizon regularly receives this type of laboratory information from over 200 patient-centered medical homes and population health programs as part of routine quality improvement monitoring. Patients were excluded if they were insured by Medicaid or Medicare, had <3 months of continuous enrollment, had recently filled insulin, or had no telephone contact information (which would prevent delivery of the intervention). An HbA1c minimum threshold of 8% was chosen based being the minimum control threshold for major quality measures for health plans.[3, 20]

Randomization and study arms.

Subjects were randomized by Horizon in a 1:1 ratio to the intervention or usual care group using a random number generator. Randomization occurred at the patient level.

The primary component of the multi-faceted intervention was an individually-tailored telephone consultation conducted by a clinical pharmacist. Patients randomized to the intervention group were sent an invitation letter informing them about the study, including a simple pillbox and a shared decision-making tool to prime the patients for the telephone consultations. This tool was postcard-sized and was developed using principles of decision aid design, querying patients about how they manage their diabetes.[13] Within two weeks of the mailing, patients were contacted by telephone to participate in the consultation or schedule it for a later date. Patients were asked to provide verbal consent for the clinical pharmacist consultation.

The consultation was conducted using a semi-structured call guide developed by the study team with direct input from patients. The interventions were delivered by trained pharmacists from Magellan Rx Management, a pharmacy benefit management company that specializes in pharmacist-delivered telephonic disease management services. Prior to study launch, the pharmacists underwent a training program that included script development and role-playing exercises.

The overall structure of the behaviorally-tailored telephone intervention was based on the principles of brief negotiated interviewing.[8, 9, 21, 22] The telephone consultation began with a review of study medications and then proceeded into a conversation eliciting the reasons for poor disease control, using open-ended questions to identify patient preferences. Based on this, patients identified one of 3 strategies to improve their diabetes control: 1) treatment intensification, 2) adherence improvement, or 3) lifestyle improvement. Patients could also identify being ‘not ready’ to select a strategy at present as a fourth option.

The subsequent interventions were tailored based on the strategy that each patient chose. For patients who chose to focus on improving medication adherence or lifestyle modifications, a second, separate brief negotiated interviewing process was used to help patients identify their adherence barriers and to propose solutions tailored to their specific barriers. These solutions included pillboxes, simple alarm reminders, mail order use, dietitian support, counseling, or other exercise recommendations. For patients who chose treatment intensification, shared decision-making was used to identify treatment options they would prefer.[13, 14] For patients who were not ready to select a strategy, the pharmacists repeated the consultation at a follow-up call. For all patients who selected a strategy, the proposed plan was communicated by the pharmacist to the patient’s provider via fax. Follow-up “booster” phone calls by the pharmacists were used up to 3 more times, which continued to engage the patient in discussions about their barriers to disease control. The pharmacists were also provided patients’ updated HbA1c laboratory data from Horizon throughout the 12-month follow-up period in order to better tailor the discussions.

Patients randomized to the usual care arm were not contacted in any way.

Study subjects and pharmacists interacting with patients were not blinded to group assignment. Study investigators and data analysts remained blinded until all follow-up data were obtained, and the analyses were finalized.

Study outcomes.

The trial’s primary outcome was the change in HbA1c from baseline (i.e., the closest value before randomization) to 12 months after randomization. In keeping with the pragmatic nature of the trial[19], HbA1c was assessed using laboratory values collected as part of routine care, rather than at study-specific visits.[23] The HbA1c result recorded closest to the end of the 12-month follow-up was used. As expected with the use of routinely-collected data, clinical HbA1c data was missing for some (28.6%) patients, including 28.7% in the usual care arm and 28.5% in the intervention arm. Accordingly, we used multiple imputation to handle the missing data.[24] In specific, we used 20 imputations with Proc MI in SAS version 9.4 (Cary, NC) to fill in any estimated values using fully conditional specification. Analyses were then conducted on each imputed dataset using generalized estimating equations with an identity link and normally-distributed errors. The results from each analyzed dataset were combined using Rubin’s rules.[24] This approach achieved in-range values and a 98% relative efficiency.

The prespecified secondary outcomes included the proportion of patients achieving HbA1c <8%[3] and medication adherence in the 12-month follow-up period. Adherence was measured by linking all observed fills based on dispensing date and days’ supply. Different drugs in the same chemically-related therapeutic class (e.g., sulfonylureas) were considered to be interchangeable.[25] From these supply diaries, we calculated the proportion of days that patients had ≥1 oral glucose-lowering medication available, or the proportion of days covered (PDC), by dividing the number of days with medication available by the number of days during follow-up.[26] This adherence metric is used by Medicare Star ratings and health plans for quality improvement.[27] Patients were censored when they lost continuous eligibility. Medication adherence was measured as both a continuous measure and as the proportion of patients achieving optimal adherence (≥80%).[8, 25, 26] There was no missing data for the adherence outcomes.

Statistical analysis.

We randomized 1,400 patients in a 1:1 ratio to achieve more than 80% power to detect an average HbA1c change of 0.5% between the intervention and usual care arms (alpha = 0.05, HbA1c standard deviation of 1.9, two-sided significance).[8] We powered the study assuming that approximately 30–35% of subjects in the intervention arm would agree to participate. Prior studies of telephonic pharmacist interventions have indicated a 30–55% acceptance rate by patients.[15, 28] Using Proc Power (two-sample means) in SAS version 9.3 (Cary, NC), we estimated that we would need 228 subjects per arm, prior to factoring in an estimated 30–35% acceptance rate in the intervention group (i.e., the midpoint of this range, 33%, would require 691 patients/group). Therefore, approximately 700 study subjects per arm were therefore estimated to be necessary after considering the estimated acceptance rate.

For the analyses, we used intention-to-treat principles for all randomized patients, i.e., regardless of whether patients actually received the intervention. We calculated the means and frequencies of pre-randomization variables separately by arm and compared them using absolute standardized differences.

We evaluated change in mean HbA1c using generalized estimating equations with an identity link and normally-distributed errors. Analyses of medication adherence were compared using an identify link function and normally distributed errors.

We also conducted additional sensitivity and secondary analyses, including adjustments for imbalanced baseline covariates with a standardized difference >0.1. For HbA1c, we conducted a “complete case” analysis. Subgroup analyses were performed according to age, sex, and baseline HbA1c control. For adherence, we measured PDC beginning with the first fill after randomization, censored patients upon initiation of insulin, and measured average adherence across all eligible diabetes medications.

We also conducted an “as-treated” analysis, following recommended approaches for pragmatic trials, in which we used propensity score matching to identify patients in the usual care arm who had similar sociodemographic and baseline characteristics as those reached in the intervention arm.[29–31] In this approach, we first calculated a propensity score by estimating the probability of receiving the intervention using logistic regression including all baseline covariates, which were selected based on literature and expert opinion. After estimating the propensity score, few patients had propensity scores that did not overlap with patients in the other group. We excluded these patients prior to matching to trim the distribution. We then used a greedy matching algorithm (5:1 digit matching) to 1:1 match patients to provide estimates for the average effect of the intervention in treated individuals.[30, 32] Absolute standardized differences and post-matching C-statistics were compared for the characteristics.

In exploratory analyses, among patients receiving the intervention, we also evaluated differences in effectiveness of the intervention by which of the four decisions patients made to control their diabetes during the consultation, using ANOVA. We hypothesized that the effectiveness could differ among these groups.

Design and sampling.

We reviewed full-text copies of the detailed notes recorded by clinical pharmacists during their initial consultation with patients. We received detailed notes for 96% of all patients who received the initial consultation. Each of the notes, in paragraph form, described which diabetes medications patients were taking, how they are managing their diabetes, their self-reported barriers to optimal glycemic control, their rationale for the strategy they chose (e.g., adherence improvement), their readiness to change, and the shared plan developed with the clinical pharmacist. Many notes included direct quotations from patients. All patient identifiers were removed prior to sharing with the study investigators.

Analysis.

We used a grounded theory approach to evaluate the clinical pharmacist notes. The objective was to elicit factors that could be explain patient’s reasons for their chosen strategy and identify themes associated with the effectiveness of the intervention. In this approach, the notes were coded and analyzed using constant comparative analysis to generate themes.[33] Verbatim quotations from patients and pharmacists were identified that supported or further elaborated on an identified theme. ATLAS.ti qualitative data analysis software was used to help identify the themes (Berlin, Germany). Because these types of telephonic quality improvement interventions are used regularly by healthcare practices and insurers, we combined the qualitative and quantitative findings to provide hypotheses for how to improve the intervention.

Theme 1: Patients attributed their poor disease control primarily to their inability to consistently manage their diet.

Most patients cited dietary choices as the hardest part about managing their diabetes. Patients reported prior meetings with dietary educators and physicians but expressed difficulty consistently adhering to a guideline-concordant diet.

Theme 2: Patients attributed their poor disease control to difficulty exercising.

Patients reported difficulty getting enough exercise given their work and life schedules. Moreover, the weather (i.e., winter) was very commonly referred to in conversations with the clinical pharmacists; many patients noted that they were deferring until January or the spring to get active again so that they could walk outside.

Theme 3: Some patients had recently changed their treatments or lifestyle and were waiting for their next appointment or lab test.

Numerous patients were not ready to choose a strategy with a clinical pharmacist, citing that they had a near-term appointment with their doctor or were waiting until their next laboratory measure. Some patients had made significant changes since their last visit.

Theme 4: Pharmacists documented when they had reached patients who were at work to explain the nature of the conversation.

Several patients were taking a break from work while the intervention was being delivered. On some of these occasions, pharmacists documented when they felt like the conversation was rushed.

Theme 5: Patients sometimes chose treatment intensification even though they really needed to focus on lifestyle adherence.

On a few occasions, pharmacists described that patients reported difficulty with lifestyle management but that the patients were interested in pursuing treatment intensification with their providers. Some of these patients reported regular adherence to medications but trouble with their diet or exercise.

Theme 6: Patients were hesitant to add on a new therapy, especially insulin and were sometimes more receptive to escalations in dose.

In several encounters, the pharmacists commented on how patients wanted to avoid insulin but were more receptive to combination oral medications or increasing the dose of medications to avoid adding any new medications.