Content area
Background
Sodium-glucose cotransporter-2 inhibitor (SGLT2i) and glucagon-like peptide-1 receptor agonist (GLP-1RA) medications reduce the risk of cardiovascular and renal complications among patients with type 2 diabetes but are underutilized. There are numerous barriers to prescribing including insurance coverage, medication availability, comfort with prescribing, and diffusion of responsibility of prescribing across specialists. Methods are needed to support prescribing in primary care.
MethodsThis was a pragmatic, randomized controlled trial testing interventions to increase appropriate SGLT2i and GLP-1RA prescribing. Primary care providers (PCPs) were randomized to 1 of 3 arms: (1) peer champion support (2) peer champion support and information on insurance coverage, or (3) usual care (no intervention). PCPs in both intervention arms received a welcome email and electronic health record (EHR) messages before visits with patients who had sub-optimally controlled diabetes and an indication for 1 of these medications. In the peer champion support only arm the EHR messages included prescribing tips. In the arm that provided peer champion support and information on insurance coverage, EHR messages contained information on medications in each class that would be most affordable for the patient based on their insurance coverage and offered support for prior authorizations if needed. The primary outcome was prescriptions for an SGLT2i or GLP-1RA medication, beginning 3 days before the targeted visit and continuing through 28 days, in each intervention arm compared to control.
Results191 primary care providers were included in the study. 1,389 patients had at least 1 visit scheduled with their PCP during the 6-month intervention period; of these 1,079 patients attended at least 1 of these visits and will be included in the primary outcome analysis. 66 providers (484 patients) received the peer champion intervention alone, 63 providers (446 patients) received the peer champion intervention and information on insurance coverage, and 62 providers (459 patients) received usual care. On average, patients were 66 years old, 46% were female, 61% were white, and 16% were Hispanic. There were small differences between groups with regards to patient sex, race, ethnicity, partner status, and percent with Medicare insurance.
ConclusionsThese medication classes have the potential to reduce cardiovascular and kidney disease among patients with type 2 diabetes. This study tests interventions to support prescribing of these medications in primary care.
Clinical Trial
Type 2 diabetes is highly prevalent in the United States and contributes to morbidity and mortality, particularly though atherosclerotic cardiovascular disease (ASCVD) and chronic kidney disease (CKD). 1 , 2 Two newer categories of medications, sodium-glucose cotransporter-2 inhibitors (SGLT2is) and glucagon-like peptide-1 receptor agonists (GLP-1RAs) improve glycemic control, offer benefits for common comorbid conditions, and reduce mortality. 3-9 Specifically, SGLT2is reduce heart failure admissions, CKD progression, cardiovascular events, and death. 3-11 GLP-1RAs reduce cardiovascular events, major kidney disease events, and mortality, and induce weight loss. 3-9 , 12-15 The American Diabetes Association, American College of Cardiology and the American Heart Association recommend the use of these medication classes for patients with type 2 diabetes and ASCVD, CKD, heart failure, or cardiovascular risk factors. 5 , 16 , 17
Despite these strong recommendations, adoption has been slow. 18-21 While over 50% of patients with diabetes would benefit from one of these medications, 10-50% of eligible patients receive one. 19 , 20 , 22-29 Though prescribing has increased over time, there remain significant gaps and variation in prescribing, particularly in the primary care setting. 25-28
There are many factors that contribute to this suboptimal adoption. 30-36 Safety concerns, lack of understanding of risks and benefits, and competing priorities in visits contribute to under-prescribing. 30-32 Additionally, while diabetes is commonly managed by primary care providers, SGLT2is and GLP-1RAs are particularly indicated for individuals with diabetes and comorbid conditions, and so multiple specialists are frequently involved in their care. 31 , 32 This can result in “diffusion of responsibility,” a behavioral concept that refers to the tendency of people (in this context, prescribers) to feel less responsible for individual actions when there are multiple actors and responsibility is unclear. 31 , 32 , 37-41 This behavioral barrier is increasingly recognized in the medical context 42-47 and may in part explain suboptimal prescribing in primary care.
Widely variable cost, insurance coverage, and shortages also make prescribing these medication classes more difficult. 34-36 For example, at the time of prescribing, providers often do not know if a medication will be covered by a patient's insurance, the out-of-pocket cost to the patient, or if the medication is available at the patient's pharmacy. Need for prior authorizations, unexpectedly high copays, and shortages (particularly for GLP-1RAs) frequently result in multiple rounds of communication between the patient, prescriber, pharmacy, and insurance company to troubleshoot these barriers. This series of denials and changes can prevent initiation and delay the start of the medication for those who ultimately manage to acquire it. 34-36 , 48 The complexity and work required for this process is burdensome for providers and can deter future prescribing. 49
Addressing these barriers could promote evidence-based prescribing. Interventions to address diffusion of responsibility have been evaluated in a few medical and several nonmedical contexts. Modeling the desired behavior, assigning responsibility to individuals or smaller groups, highlighting individuals’ competence to act, and highlighting the perceived harm of the situation to be addressed are all effective approaches. 43 , 50-52 Interventions from peers, such as opinion leaders and peer champions, can incorporate each of these factors and have been shown to positively influence a broad range of provider behaviors. 53-56 While interventions at the policy and health system level are critical for improving the availability and affordability of these medications in the long term, pharmacists or other healthcare workers can provide more accurate insurance information and assist with prior authorization processes now to help physicians select the best, least expensive option for each patient. 57 , 58
Thus, the goal of this 3-arm randomized controlled trial is to test the effectiveness of a peer champion support intervention with and without additional information on insurance coverage, compared to usual care, on SGLT2i and GLP-1RA prescribing in primary care.
Methods Overall study designThis was a pragmatic, randomized controlled trial testing 2 interventions to increase appropriate SGLT2i and GLP-1RA prescribing compared to usual care. We included tirzepatide, a dual GLP-1 and GIP receptor agonist, in the group of GLP-1RAs. 191 primary care providers (PCPs) caring for patients with sub-optimally controlled diabetes and an indication for one of these medications were randomized in a 1:1:1 ratio to 1 of 3 arms: (1) peer champion support, (2) peer champion support and information on insurance coverage, or (3) usual care (no intervention). Intervention delivery began in May 2023 and continued through November 2023. Collection of follow-up data continued through August 2024.
This research is supported by the National Institute on Aging of the National Institutes of Health under Award Number P30AG064199 to BWH (Choudhry PI). Dr. Lauffenburger (K01HL141538) and Dr. Haff (K23HL161480) are supported in part by career development grants from the NIH. This study was approved by the institutional review board at Mass General Brigham and is registered on clinicaltrials.gov (NCT05463705).
Study setting and participantsThis study was conducted within primary care practices at Massachusetts General Hospital (MGH). MGH has more than 200 PCPs providing care for over 200,000 patients in the Boston area. MGH has several routine system-wide population health and quality initiatives that all PCPs receive, and which continued through the course of the trial. These include quarterly dashboards sent to PCPs by email containing their performance metrics for several chronic disease and preventive health measures, including diabetes care, but they do not specifically focus on SGLT2i or GLP-1RA prescribing. Physicians also have access to a real-time benefits check function in the electronic health record that can estimate the out-of-pocket cost of medications for some patients at the time of prescribing, but this information is not available for all patients in the system and there is variability in the use of this tool. These activities are coordinated centrally across all practices and were delivered equally across study arms, including the usual care arm.
Provider eligibilityMGH PCPs caring for adult patients were enrolled in the study. Providers who were involved in the design or conduct of the study (eg, the peer champions and study authors) were excluded. Two of the 16 MGH adult primary care clinics chose not to participate in the study and their providers were also excluded.
Patient eligibilityWe used data from the electronic health record (EHR) to identify patients aged 18 years and older with type 2 diabetes, whose most recent hemoglobin A1c (HbA1c) was >7.0%, who had an indication for an SGLT2i or GLP-1RA and who were not currently prescribed one of these therapies. Patients with end-stage renal disease, dementia, or EHR indicators hospice care were excluded. Although an A1c above 7.0% may represent acceptable diabetes control for some patients, we chose this threshold because SGLT2is and GLP-1RAs can be used to treat or prevent comorbid conditions even when diabetes is well controlled. 3-10 , 12 , 13 , 59-62 Clinical indications for these agents were determined from ICD-10 diagnosis codes in the preceding 12 months and included cardiovascular disease, heart failure, kidney disease, or obesity.
Enrollment and randomizationThe study was presented to primary care clinic leaders within MGH, who could opt their clinic out of participation before randomization. An organization-wide announcement was then circulated to inform all primary care providers of study launch. Individual providers could also opt out of the study, or selectively opt out of interventions but continue to have their data analyzed, by replying to study emails or messages beginning from announcement of the study through the completion of intervention delivery. Formal consent was not obtained.
Physicians were randomized in April 2023. We used stratified permuted block randomization to help achieve balanced numbers of providers and patients in each arm. 63 Each clinic served as a separate stratum. Within each clinic, a random number generator was used to create a randomly ordered allocation within blocks of 3. Within each block, physicians were randomized in a 1:1:1 ratio to 1 of the study arms. Randomization was performed by the study analyst for all providers simultaneously. The remainder of the study staff were blinded to arm allocation until knowledge of assignment was required to deliver study interventions. PCP participants were notified of allocated arm in the first intervention email. In September 2023, new providers who began employment after the initial randomization were additionally randomized following the same process. These providers received a condensed 2.5-month intervention, and their outcomes will be handled as outlined in the analysis sections.
Peer champion selection and trainingTo identify peer champions for the study, consistent with studies of effective opinion leaders, 64-69 we identified PCPs in the system with evidence of strong connections to their peers, extensive clinical experience, and a strong presence in the community using a combination of methods. We first generated lists of PCPs who were the most frequent GLP-1RA and SGLT2i prescribers prior to study start. Primary care leadership reviewed this list and suggested PCPs that they felt might best serve as champions. Practice leaders were purposefully not selected, but PCPs with longer tenures in our system or whose names were otherwise more widely recognized due to involvement in research, teaching, or service were preferred. We also ensured that each peer champion practiced in a different clinic. PCPs were individually invited to serve as a peer champion until 4 were confirmed. Each primary care practice was then assigned to a peer champion. Peer champions were assigned their own clinics, and other practices were then allocated to each champion, prioritizing geographic proximity when possible, so that each champion served approximately 30 PCPs, half in the peer champion support alone arm and half in the peer champion support and information on insurance coverage arm. Allocation was performed in this way so the same peer champion administered both intervention arms within a clinic and thus any variation in effectiveness of the peer champions would not bias outcome comparisons between arms. The peer champions met with study staff and the study endocrinologist before study launch to review and train on study procedures and provide feedback on study intervention messages, and then regularly during intervention delivery to report on study progress and troubleshoot issues with intervention delivery.
Intervention and usual care armsAfter randomization, PCPs received the interventions or usual care for 6 months. In the usual care arm, there was no additional contact with the PCPs.
Peer champion supportIn this arm, PCPs received support for SGLT2i and GLP-1RA prescribing from the peer champions including elements to address diffusion of responsibility across providers, reminders, and interactive education. Peer champion interventions have been shown to impact provider behaviors, though the content of the interventions vary widely. 53-55 , 70
PCPs in this arm first received a welcome email from their peer champion offering encouragement and support in prescribing GLP-1RAs and SGLT2is ( Table 1 ). This email adapted elements from interventions that mitigate diffusion of responsibility in other contexts, including: (1) assigning responsibility to individuals or smaller groups, (2) increased perceived harm of the situation to be addressed (3) highlighting competence to act, and (4) modeling the desired behavior. 43 , 50-52 EHR messages with prescribing tips ( Table 2 ), which provided a reminder to prescribe as well as education, were signed by the peer champion and sent by research assistants 3 business days prior to upcoming visits with eligible patients. Prescribing tips ( Appendix A), developed with input from the peer champions and the study's consulting diabetologist, were designed to address common questions and pitfalls in SGLT2i and GLP1-RA prescribing. Providers received 1 prescribing tip per patient about either SGLT2i or GLP1-RA prescribing based on the patient's qualifying indication (heart failure or CKD for SGLT2is, obesity or ASCVD for GLP-1RAs). If a patient had an indication for both medication classes, SGLT2i tips were selected preferentially. There were 11 tips for SGLT2is and 10 tips for GLP1-RAs. Providers received the tips sequentially; if providers saw enough eligible patients to receive all the tips, then repeat tips were sent. Additional education was provided throughout as needed by contact with the peer champion. PCPs in the study could reply to the email or the EHR messages, or reach out to the peer champion by email, phone, or EHR message any time. In the 3rd month after the start of the intervention, PCPs also received a printed chart summarizing SGLT2i and GLP-1RA prescribing with a hand-signed note from the peer champion again offering support as needed ( Appendix B).
Peer champion support and information on insurance coveragePCPs in this arm received the peer support intervention, but rather than receiving prescribing tips, patient-specific information about affordability and insurance coverage was sent in the previsit EHR messages.
Cost and lack of insurance coverage are 2 common barriers to prescribing. 34-36 , 48 One potential solution is to provide PCPs with information on cost and coverage, tailored to each patient, before the visit. With this information, PCPs could then select the medication most likely to be affordable, potentially avoid surprise costs, and reduce prior authorizations, postvisit follow-up, and loss to follow-up.
To do this, in this arm, charts of eligible patients with upcoming visits with their PCP were reviewed by an insurance specialist in a pre-existing administrative support team with expertise in insurance coverage and prior authorization. This team currently provides support to the endocrine specialty physicians, but not primary care. They maintain updated lists of which medications are covered by common insurance plans, and in cases where coverage is not clear, they contact insurance plans and pharmacies to determine coverage for each patient.
For each patient with an upcoming visit, the administrative support team generated a list of SGLT2is and GLP1-RAs that were covered without a prior authorization, and a list that would require prior authorization, and sent this information to the PCP in an EHR message 3 business days in advance of the visit ( Table 2). The message also included information on the real-time benefits estimator available to help determine out-of-pocket costs for the patient in the case that multiple medications were covered but might have different copays. If the provider prescribed a medication that required prior authorization, they could ask the administrative support team for assistance, or they could proceed through their clinic's usual prior authorization workflow, which was typically completion by practice medical assistants. PCPs could reply to any of these messages with questions. If the question related to insurance coverage, it was routed to the insurance specialist for response; if clinical it was routed to the peer champion.
Intervention period conclusionThe intervention period concluded in November 2023. In January 2024, all PCPs in the study, including in the control arm, received an email that notified them that the intervention period was complete, included an electronic copy of the SGLT2i/GLP-1RA prescribing chart, and invited them to participate in a survey as part of an explanatory sequential study design. 71 For providers in the intervention arms the email also included a list of their patients who we considered eligible for an SGLT2i or GLP-1RA but did not have a visit during the intervention period. Providers received 1 additional reminder email if they had not completed the survey after 4 weeks.
Qualitative and implementation data collection End-of-study surveyUpon completion of the study intervention period, all PCPs in the study were invited to complete a survey about their experiences prescribing SGLT-2i and GLP-1RA medications over the preceding 3 months (Appendix C). PCPs selected from a 5-point Likert scale, from strongly disagree to strongly agree, how much they agreed with statements about their responsibility for prescribing these medications versus deferring to specialist colleagues, access to colleagues for clinical questions, difficulties with and ability to navigate insurance coverage, confidence in prescribing, and limitations from medication shortages. PCPs were also asked open-ended response questions about other barriers to SGLT-2i and GLP-1RA prescribing, helpful existing resources, and how to best support prescribing in the future. PCPs in the intervention arms were asked their opinions of the peer champion and insurance coverage specialist support, as applicable. PCPs were given an electronic $25 gift card for survey completion.
Interactions with peer champions and insurance specialistEach week of the intervention period, the peer champions and insurance coverage specialist completed surveys in which they reported if there were interactions with PCPs in the prior week. If there was an interaction, they reported the approximate duration (<5 minutes, 5-10 minutes, >10 minutes) and a brief description of the content discussed.
Chart reviewCharts of all patients who received a prescription for an SGLT2i or GLP-1RA will be reviewed to determine if these medications were filled using pharmacy fill data integrated into the EHR. This will be used to examine rates of successful filling and time to filling in each arm compared to usual care.
We will conduct additional qualitative exploratory work to better understand the complexities of starting patients on 1 of these medication classes. We will randomly select 20 patients from each of the study arms, 10 who received a prescription for an SGLT2i or GLP-1RA and 10 who did not, and review their charts, including primary care visits, patient portal messages, and prescription fill data, to qualitatively assess how conversations were documented, how prescribing decisions were made, and what happened in prescribing over time.
OutcomesThe primary outcome is the rate of SGLT2i or GLP-1RA prescribing among eligible patients who attended the visit with their PCP, beginning 3 business days prior to the visit (day of intervention delivery for providers in intervention arms) and continuing through 28 days after the targeted visit. If multiple visits were identified, all could be intervened upon, but only the first attended visit will be included in the primary outcome. Prescriptions will be identified from EHR data during the outcome window. We will conduct several sensitivity analyses including (1) shortening the observation window to 3 business days before the visit through 7 days after the visit, (2) lengthening the observation window from 3 business days before the visit through the end of follow-up (28 days after the last visit date), and (3) excluding the 13 providers who received the condensed intervention.
We will also measure several secondary outcomes including: (1) whether patients who received a prescription for an SGLT2i or GLP-1RA filled their prescription (determined through chart review); (2) if prescribing and filling occurred more rapidly in any group compared to usual care; (3) the change in HbA1c between baseline and 6 months after the intervention (determined from EHR data); and the rate of prescribing of these medication classes among (4) all patients considered study eligible, regardless of whether the patient attended the visit and (5) across all patients with diabetes cared for by each PCP (without specific HbA1c or comorbidity requirements).
Implementation outcomes focus on assessing the acceptability of and fidelity to the intervention and include (1) the frequency, duration, and nature contact between PCPs and peer champions and the insurance specialist during the intervention period; (2) qualitative analysis of comments from PCPs about the interventions on the end-of-study survey; and (3) qualitative analysis of differences in documentation and patient prescribing journeys from chart review.
Statistical considerations Power and sample sizeWe estimated that 178 providers caring for 1,249 eligible patients will provide 80% power to observe an 11 percentage-point difference in prescribing rates between each intervention arm and control, assuming a control arm prescribing rate of 30%, an intraclass correlation coefficient of 0.07 (estimated based on prior studies in this system), 72 , 73 an average of 7 patients per provider, and a type I error rate of 5%. Intervention delivery required advanced notice of upcoming visits by at least 1 week. In a previous trial this time frame resulted in a 29% no-show or late cancel rate, 72 and so we anticipated needing 1611 expected encounters with patients to achieve the target number of unique patients with visits attended. Providers with prolonged periods away from practice or who left the health system during the intervention period could contribute few or no patients to the sample, but this is not expected to differ across arms.
Our assumed effect estimate was based upon prior studies of interventions to encourage appropriate use of GLP-1RAs and SGLT2is which increased prescribing by 5-12 percentage points for lower-touch interventions 74 , 75 and up to 30-50 percentage points for high-touch multifaceted interventions. 57 , 76 , 77 Because the use of GLP-1RA medications can reduce major adverse cardiovascular events by 12%-14% 4 , 78 and the use of SGLT2i medications can reduce heart failure admissions by 20%-25% and CKD progression by 20%-30%, 4 , 79-81 increases in use of these medications even smaller than demonstrated in prior trials are likely to have important clinical impact at the population level. 80 , 81
We assumed a control group prescribing rate of 30% based on preliminary data from our health system prior to study start. This is higher than rates observed in other studies that focused on either class alone, 82-85 but in this outcome we included GLP-1RA and/or SGLT2i prescribing together, and prescribing rates have gradually increased over time. 82 , 84 , 85 If the usual care arm intensification rate is higher than 30%, we will still be sufficiently powered for reasonable effect sizes. For example, if the rate of the primary outcome in the usual care arm is 40%, we will still have 80% power to detect an effect size of 12 percentage points or more.
Analytic planBaseline demographic and clinical characteristics of patients were determined from the EHR using structured fields for demographic, laboratory, and biometric data and ICD 10 codes to assess for comorbid conditions and calculate a Combined Comorbidity Score. 86 Provider characteristics will be collected from provider data files within the EHR. Descriptive statistics were used to report the means and frequencies of baseline variables for eligible patients.
For our primary outcome, we will utilize generalized estimating equations (GEE) to adjust for the clustering of patients within providers and fixed effects of primary care practices to adjust for the randomization design. Each intervention arm will be compared to usual care, using intention-to-treat principles and a 2-tailed test. Additional models will adjust for patient demographic characteristics including sex, race, ethnicity, partnered status, and Medicare insurance to account for small differences between groups that emerged from the provider-level randomized design. If there are other strong predictors of the outcomes not balanced by randomization, we will also adjust for these in additional analyses. We will utilize a similar statistical approach for sensitivity analyses. The study is powered for the primary outcome testing only. Because the peer champion intervention component is common to both intervention arms, we will use a Holm-Bonferroni correction for statistical significance thresholds. This approach addresses concerns for multiple testing while maintaining power. 87-89 Sensitivity and subgroup analyses are considered exploratory.
Subgroup analyses will be conducted by patient and provider demographic characteristics, peer champion and clinic groups, among patients who did not have an SGLT2i or GLP-1RA prescription in the preceding 6 months and 12 months before study start, and among groups based on providers’ baseline SGLT2i and GLP-1RA prescribing rates.
Secondary outcomes will be modeled similarly using GEE with fixed effects of primary care practices to adjust for the clustering of patients within providers and the randomization design. For change in diabetes control over time, based on prior studies in our system we anticipate approximately 20% missingness rate for the follow-up HbA1c within 6 months. 90-93 It is possible that patients who have medications added may be more likely to have a follow-up HbA1c measured and so we will explore several options to handle missing data. We will first evaluate for differential missingness by study arm and by whether a patient was started on an SGLT2i or GLP-1RA, and we will examine clinical and demographic characteristics of patients with and without missing data. If data appear to be missing at random, we will then use multiple imputation using Proc MI in SAS, to impute estimated values using fully conditional specification. 94 , 95 Analyses will then be conducted on each imputed dataset and combined using Rubin's rules. 96 This approach with 20 imputations has produced reliable and efficient results in prior work using the same types of data with similar degrees of missingness. 90-93 If data appear to be missing not at random, we will use additional methods including the last value carried forward and the mean value to complete missing values, as well as a complete case analysis, and evaluate the robustness of findings to the varied imputation strategies.
Responses to survey questions using Likert scales will be calculated and compared between groups. Free-text responses including intervention feedback will be summarized and reported to facilitate interpretation of efficacy results. Barriers and facilitators of prescribing obtained from chart review will be qualitatively summarized. Findings from both these sources will be used to explore reasons behind the trial results as well as to identify additional barriers to GLP-1RA and SGLT2i use.
Baseline resultsA total of 191 primary care providers were included in the study, 66 allocated to the peer champion alone intervention, 63 to the peer champion and information on insurance coverage intervention, and 62 to usual care ( Figure 1 ). 178 primary care providers were randomized at the study start. Between study start and September 2023 13 PCPs had newly joined the system, were randomized, and completed 2.5 months of the study through the completion date of November 2023. Five PCPs in the peer champion alone arm opted out of receiving ongoing intervention messages, but their patients were still followed for outcome ascertainment and the providers and patients were included in the main analyses consistent with an intention to treat approach. Sixteen PCPs did not have a visit with an eligible patient during the intervention period, 8 in the peer champion alone arm, 4 in each of the other arms.
1,389 unique eligible patients had at least 1 visit scheduled with their PCP during the 6-month intervention period. Of these 1079 patients attended at least 1 of these visits and will be included in the primary outcome analysis, 377 in the peer champion arm, 346 in the peer champion and information on insurance coverage arm, and 356 in usual care ( Figure 1). Among the cohort of 1389 patients, there were an average of 7.76 (SD 6.20) patients per provider. On average, patients were 66 years old, 46% were female, 61% were white, and 16% were Hispanic. There were small differences between groups with regards to sex, race, ethnicity, partner status, and percent of patients with Medicare insurance ( Table 3 ).
DiscussionThis study tests interventions to improve the appropriate prescribing of SGLT2i and GLP-1RA medications in primary care. 191 providers and 1,389 unique patients will be analyzed for trial outcomes. If successful, these interventions could be implemented in health systems to assist with prescribing, with the goal of ultimately helping prevent cardiovascular and renal complications of diabetes.
Prior studies of interventions to increase appropriate prescribing are promising. One trial showed a 48 percentage point increase in SGLT2i and/or GLP-1RA prescribing (compared to 16 percentage point increase in usual care) with the use of an intensive and multifaceted care coordination intervention to optimize cardiovascular preventive medications among patients with type 2 diabetes across cardiology and primary care practices. 57 This intervention included assessing local barriers, developing new care pathways, educating providers, providing feedback reports, and providing prescribing tools, and thus it was highly resource intensive. If successful, the interventions tested in this trial could provide a more scalable approach to improving prescribing.
This study has several important limitations. First, prescribing, our primary outcome, does not indicate that patients successfully started the medications, especially in the context of medication shortages and unexpected out-of-pocket costs. We will determine rates of filling of medications based on chart review, but this information may not be available for all patients. Second, there were marked shortages of GLP-1RA medications during the time this study was conducted, which may have discouraged prescribing overall and made filling more difficult, potentially creating a ceiling effect for prescribing this medication class. Third, this study was conducted within a single health system that had a relatively high rate of SGLT2i and GLP-1RA prescribing at baseline. Intervention effects may differ in other systems with lower or higher baseline prescribing rates, and in this study we will examine intervention effects among subgroups with higher and lower baseline prescribing rates. Fourth, effectiveness of the peer champion intervention may depend on the local culture of whether these medications are generally prescribed by PCPs, by specialists, or by both, and thus the effects of the interventions may also vary in other systems with different prescribing patterns. Fifth, it is possible that we will observe differential missingness of HbA1c by study arm which may limit our ability to interpret this outcome. Lastly, because we randomized at the provider level, there is potential for contamination of the peer champion intervention component between intervention and control groups. However, peer champions were aware of this concern and worked to interact only with intervention PCPs in their study roles, and contamination, if present, would bias the study results towards the null.
Overall, given the potential impact of SGLT2is and GLP-1RAs on cardiovascular and renal outcomes among patients with type 2 diabetes, it is critical to test interventions to support prescribing of these medications in primary care.
CRediT authorship contribution statementNancy Haff: Writing – review & editing, Writing – original draft, Investigation, Funding acquisition, Formal analysis, Conceptualization. Daniel M Horn: Resources, Investigation, Funding acquisition, Conceptualization. Gauri Bhatkhande: Investigation, Formal analysis, Data curation. Meekang Sung: Writing – review & editing, Investigation, Formal analysis, Data curation. Caitlin Colling: Writing – review & editing, Methodology, Investigation, Conceptualization. Wendy Wood: Methodology, Conceptualization. Ted Robertson: Writing – review & editing, Methodology, Investigation, Funding acquisition, Conceptualization. Daniel Gaposchkin: Investigation, Conceptualization. Leigh Simmons: Investigation, Conceptualization. Judy Yang: Investigation, Conceptualization. James Yeh: Investigation, Conceptualization. Katherine L. Crum: Project administration, Investigation, Formal analysis, Data curation. Kaitlin E. Hanken: Project administration, Investigation, Formal analysis, Data curation. Julie C. Lauffenburger: Writing – review & editing, Investigation, Funding acquisition, Formal analysis, Conceptualization. Niteesh K. Choudhry: Writing – review & editing, Supervision, Resources, Investigation, Funding acquisition, Formal analysis, Conceptualization.
Conflict of interestDr. Choudhry serves as a consultant to Veracity Healthcare Analytics and holds equity in RxAnte and DecipherHealth; unrelated the current work, Dr. Choudhry has also received unrestricted grant funding payable to Brigham and Women’s Hospital from Humana. The remainder of the authors report no conflicts of interest.
AcknowledgmentsThe team would like to thank Xiaoyue Wong for her work with delivery of the intervention and Molly Blair and Theresa Oduol for their administrative and background research work on this project.
Supplementary materialsSupplementary material associated with this article can be found, in the online version, at doi:10.1016/j.ahj.2025.02.007.
Appendix Supplementary materialsImage, application 1
| Peer champion alone | Peer champion and information on insurance coverage |
| | |
| | |
| Peer champion alone | Peer champion and information on insurance coverage |
| | |
| | |
| Peer champion alone (
| Peer champion + information on insurance coverage (
| Usual care (
| Standardized mean difference | |
| Age, mean (SD) | 66.4 (11.8) | 65.0 (12.6) | 66.4 (11.9) | 0.06 |
| Female, N (%) | 235 (48.6%) | 218 (48.9%) | 191 (41.6%) | 0.15 |
| Race, N (%) | 0.16 | |||
| | 41 (8.5%) | 48 (10.8%) | 38 (8.3%) | |
| | 63 (13.0%) | 55 (12.3%) | 67 (14.6%) | |
| | 299 (61.8%) | 277 (62.1%) | 270 (58.8%) | |
| | 81 (16.7%) | 62 (13.9%) | 80 (17.4%) | |
| Hispanic, N (%) | 0.11 | |||
| | 78 (16.1%) | 61 (13.7%) | 79 (17.2%) | |
| | 24 (5.0%) | 33 (7.4%) | 26 (5.7%) | |
| Partnered, N (%) | 254 (52.5%) | 232 (52.0%) | 276 (60.1%) | 0.19 |
| Primary insurance, N (%) | ||||
| | 184 (38.0%) | 185 (41.5%) | 209 (45.5%) | 0.12 |
| | 290 (59.9%) | 291 (65.3%) | 301 (65.6%) | 0.07 |
| Baseline HbA1c (%), mean (SD) | 8.4 (1.5) | 8.4 (1.5) | 8.4 (1.4) | |
| Systolic blood pressure (mmHg), mean (SD) | 132.9 (16.6) | 134.0 (17.3) | 133.0 (16.0) | −0.03 |
| Diastolic blood pressure (mmHg), mean (SD) | 75.2 (9.8) | 75.9 (9.7) | 74.7 (9.5) | −0.09 |
| Body Mass Index (Kg/m
2), mean (SD),
| 31.8 (6.0) | 32.1 (7.4) | 32.2 (5.8) | 0.04 |
| EGFR (mL/min/1.73m2), mean (SD) | 79.7 (22.0) | 78.8 (23.0) | 79.8 (21.8) | 0.02 |
| Qualifying Condition, N (%) | ||||
| | 139 (28.7%) | 154 (34.5%) | 136 (29.6%) | −0.05 |
| | 466 (96.3%) | 416 (93.3%) | 430 (93.7%) | −0.05 |
| | 121 (25.0%) | 124 (27.8%) | 107 (23.3%) | −0.07 |
| Comorbid conditions, N (%) | ||||
| | 50 (10.3%) | 69 (15.5%) | 61 (13.3%) | 0.08 |
| | 31 (6.4%) | 34 (7.6%) | 25 (5.5%) | 0.07 |
| | 109 (22.5%) | 82 (18.4%) | 102 (22.2%) | 0.06 |
| | 108 (22.3%) | 104 (23.3%) | 119 (25.9%) | 0.07 |
| | 41 (8.5%) | 37 (8.3%) | 34 (7.4%) | 0.04 |
| | 361 (74.6%) | 338 (75.8%) | 354 (77.1%) | 0.05 |
| | 331 (68.4%) | 269 (60.3%) | 307 (66.9%) | 0.09 |
| | 112 (23.1%) | 108 (24.2%) | 109 (23.8%) | 0.01 |
| | 29 (6.0%) | 28 (6.3%) | 14 (3.1%) | 0.15 |
| | 66 (13.6%) | 36 (8.1%) | 47 (10.2%) | 0.09 |
| | 28 (5.8%) | 21 (4.7%) | 20 (4.4%) | 0.05 |
| Combined comorbidity score, mean (SD) | 1.9 (2.1) | 2.2 (2.4) | 1.9 (2.1) | −0.06 |
| Frailty index, mean (SD) | 0.15 (0.04) | 0.15 (0.04) | 0.15 (0.04) | −0.02 |
| Unique medications, mean (SD) | 19.7 (19.0) | 19.8 (19.6) | 19.3 (18.7) | −0.03 |
| Unique diabetes medications, N (%) | 1.4 (0.8) | 1.4 (0.8) | 1.4 (0.8) | −0.02 |
| Number of office visits, mean (SD) | 7.4 (8.1) | 7.4 (7.5) | 7.7 (7.8) | 0.04 |
| Number of hospitalizations, mean (SD) | 0.24 (1.08) | 0.20 (0.75) | 0.16 (0.60) | −0.07 |
| Number of ER visit, mean (SD) | 0.53 (1.28) | 0.49 (1.14) | 0.48 (1.31) | −0.02 |
©2025. Elsevier Inc.