Two study phases were conducted. Phase I focused on decision-making within the MedSims patient simulation, and phase II focused on treatment decisions in the real world (Figure 1).

A paired, pre-post study was conducted from July 21, 2021, to December 31, 2021. PCPs and endocrinologists who made at least one decision in the simulation and were shown feedback were included.

A retrospective, matched case-control study was conducted from April 21, 2021, to January 31, 2022. The participation period was July 21, 2021, to October 31, 2021. The participation date upon which the 3 months prior to and after are calculated is herein referred to as the index date. Physicians were included if they had International Statistical Classification of Diseases and Related Health Problems, Tenth Revision (ICD-10) claims for T2D and prescription data available. Patients were required to have had at least one visit with either a patient simulation CME intervention (intervention group) clinician or a matched control group (comparison group) clinician. In addition, during those visits T2D was coded, and in the 3 months prior to or after the index date an oral T2D treatment was prescribed. Clinician- and patient-level data were obtained through licensed claims data that contains data from over 320 million US patients. A data specialist provided aggregated deidentified clinician-level data to the study lead author (KSL), who analyzed the data.

The intervention was a web-based MedSims CME simulation activity: Beyond Glycemic Control...Comprehensive Management of T2D [33].

This MedSims patient simulation, which allows learners to order lab tests, make diagnoses, and prescribe treatments in a manner matching the scope and depth of actual practice, contains 2 cases—one focused on primary prevention of CV events in a patient with T2D and one focused on secondary prevention of CV events in a patient with T2D. MedSims simulation provides a customized learning experience with point-of-decision (formative) feedback. Decision-making is open-ended, and opportunities to review decisions allow learners to make real-time changes based on formative feedback. Learners also go through a full review at the end of each case to make adjustments when considering the patient visit as a whole. The intervention addressed the following learning objectives: improve performance associated with ordering appropriate tests to assess glycemic control and CV risk in patients with T2D and improve performance associated with selecting appropriate treatments for primary and secondary prevention of CV events in patients with T2D and CVD.

For the purposes of this study, we limited our investigation to the case that focused on secondary prevention, for which the learners were expected to choose the GLP-1 RA as the most appropriate treatment because GLP-1 RAs are indicated for T2D and secondary CV risk reduction.

According to the US Department of Health and Human Services, this study was exempt from institutional review board approval because it was compliant with the Code of Federal Regulations (CFR) and Medscape privacy policy and leveraged study of existing data that were deidentified to the investigator under 45 CFR 46.104(d)(4); this study qualified for educational research exemption under 45 CFR 46.104(d)(1) [36].

Across physicians, there was a 10 percentage point increase in correct responses from pre- to postfeedback (313/435, 71% to 352/435, 80.9%; P<.001) for assessing glycemic control and CV risk. There was a 29 percentage point increase in correct responses from pre- to postfeedback (178/435, 40.9% to 304/435, 69.9%; P<.001) for selecting treatment that addresses both glycemic control and CV event protection.

Descriptive results showed that after participation, 69 of 157 (44%) intervention-group clinicians used GLP-1 RAs with their patients with T2D compared with 51 of 157 (32.5%) comparison-group clinicians (Multimedia Appendix 2). Overall, 39 of 157 (24.8%) intervention-group clinicians increased their use of GLP-1 RAs with their patients with T2D compared with 20 of 157 (12.7%) comparison-group clinicians.

Logistic regression results showed that being in the intervention group predicted GLP-1 RA use (odds ratio [OR] 4.49, 95% CI 1.45-13.97; P=.001). The intervention group was 4.2 times more likely to use GLP-1 RAs for patients with T2D than the comparison group, controlling for the number of total patients with T2D, being an endocrinologist, and prior use of GLP-1 RAs (Table 1).

Results from decisions made within the CME simulation show an improvement in assessment of CV risk and glycemic control and selection of GLP-1 RAs within the simulation, and the matched case-control study shows that participation in the CME simulation was associated with significant increases in use of GLP-1 RAs.

Previous research has shown that short, case-based, web-based CME activities improved knowledge, competence, and self-reported performance in T2D management among health care professionals (HCPs) [37]. The results of this study go beyond the limitation of HCPs’ self-reported performance to suggest that patient simulation CME is reflective of real-world practice behavior, as there was concordance (within the same phase I time frame) in decision-making between the percentage of clinicians who selected GLP-1 RAs in the CME simulation prior to feedback (178/435, 41%) and the entire population of clinicians using GLP-1 RAs for patients with T2D in the real world (394,133/947,437, 40.6%) prior to the intervention. However, there was not concordance between the percentage of clinicians who selected GLP-1 RAs in the CME simulation after feedback (304/435, 69.9%) and the percentage of clinicians using GLP-1 RAs in patients with T2D in the real world who also participated in the CME simulation (69/157, 44%). This difference is likely due to barriers faced in the real world that may limit prescribing (such as insurance coverage and patient readiness for injectables), as well as the limitation inherent in presenting only 1 patient type in the simulation, whereas many different patient types and possible care scenarios exist in the real world. For example, real-world patients may refuse certain treatments, but this possibility was not a factor in the CME simulation; thus, clinicians who participated in the simulation may need additional education that presents them with several patient types.

To provide context for physicians’ use of GLP-1 RAs postintervention, it is helpful to note the reasons they offered within the CME simulation for selecting a treatment for secondary prevention of CV events. The top 6 reasons given by endocrinologists were drug efficacy (25/53, 47%), clinical trials supporting drug use (23/53, 43%), patient profile (20/53, 38%), guideline recommendation (17/52, 33%), familiarity with use (17/52, 33%), and indication for primary and secondary prevention (3/55, 5%). The top 6 reasons given by PCPs were guideline recommendation (80/195, 41%), clinical trials supporting drug use (76/195, 39%), patient profile (72/190, 37.8%), indication for primary and secondary prevention (70/195, 35.9%), efficacy (61/190, 32.1%), and familiarity with use (10/195, 5.1%).

Notably, drug efficacy held nearly inverse positions for endocrinologists and PCPs as a reason for treatment selection. This may indicate that endocrinologists take a more granular view of drug-specific factors described in guidelines [38], which show high efficacy for GLP-1 RAs but only intermediate efficacy for sodium-glucose cotransporter-2 (SGLT2) inhibitors. As nonspecialists, PCPs may be more reliant on guideline decision trees [38], which present the selection of GLP-1 RAs and SGLT2 inhibitors as an “either/or” treatment choice for patients with or at high risk for atherosclerotic CVD (ASCVD). Indeed, only 10 of 195 (5.1%) PCPs who participated in the CME simulation chose familiarity with use as a reason for treatment selection.

The most common reasons offered by endocrinologists for not selecting a treatment for secondary prevention of CV events were being unfamiliar with use (9/20, 45%), the drug being unavailable on the formulary (8/22, 36%), the patient not needing secondary prevention (4/23, 18%), and drug cost (4/23, 18%). The most common reasons given by PCPs were cost (36/92, 39%), being unfamiliar with use (36/92, 36%), the patient being uncomfortable with injection (21/90, 23%), and the drug being unavailable on the formulary (18/91, 20%).

Although in the real world physicians were using GLP-1 RAs with patients with T2D, the majority of their patients with CV event risk factors were still not receiving treatment; only 205/4372 (4.68%) of patients with T2D and ASCVD received GLP-1 RAs in the CME group after the intervention. More education is needed to address barriers to use of this class of drugs with patients who would benefit. Our results suggest that education for endocrinologists should emphasize familiarity with the use of GLP-1 RAs and recognition of patients who would benefit; for PCPs, education should aim to improve familiarity with use and comfort with injection.

Possible limitations of our study are the inability to determine clinicians’ rationale for selecting GLP-1 RAs in practice, lack of randomization, small sample size for non-PCP participants, inability to determine if increased real-world use of GLP-1 RAs could be associated with all patient simulation CME interventions (ie, our results are localized to this intervention), and a follow-up limited to a 3-month period. The 3-month follow-up limits our ability to evaluate whether increased and new prescribing were a durable result associated with the intervention. However, durability of results for this type of study typically wanes the further out from the intervention period the results are measured. Unmeasured confounders such as motivation to take the intervention for the intervention group versus not take it for the control group were not considered. Simply the motivation to undertake the CME simulation could contribute to the effect found. Finally, the study design helped minimize the sampling bias, but there still may be unmeasured confounders. These may include consumption of other content, such as other CME activities, published studies, and collegial conversations, as well as level of motivation to adjust treatment selection.

A strength of this study is that claims data indicated that before the intervention, the percentage of physicians in the intervention group who treat patients with T2D with GLP-1 RAs (59/157, 37.6%) and the percentage of the total population of US physicians who treat patients with T2D with GLP-1 RAs were comparable (394,133/947,437, 40.6%). In addition, claims data showed that before the intervention, the percentage of patients treated with GLP-1 RAs by the intervention-group physicians (1056/17,004, 6.21%) and the percentage of patients treated with GLP-1 RAs in the US population were also comparable (848,607/13,731,508, 6.18%). These similarities between groups indicate that our study has direct implications for impact on public health. Ultimately, the results are generalizable for clinicians who treat patients with T2D, are members of clinical news and medical education platforms, and engage in web-based, simulation-based CME on the topic of newer T2D treatments that have potential benefits for glycemic control and CV protection.

A case-based virtual patient simulation CME intervention focused on secondary prevention of CV events in a patient with T2D was associated with increased selection of cardioprotective antihyperglycemic treatments in clinical practice among both endocrinologists and PCPs.