Conclusions
This is one of the first real-world, observational studies using EHR combined with claims data to assess multiple options for treatment intensification in type 2 diabetes. By presenting comparative effectiveness evidence on the full range of choices available at this therapeutic decision point, our findings provide valuable guidance to clinicians. Our PS-matched analysis indicates that patients receiving two OADs have a significantly higher chance of reaching HbA1c targets and achieving weight loss at 6 months following intensification with GLP-1 RA rather than additional OAD(s), provided that they had an equal chance of receiving each intensification option. Similar results were observed for patients who were equally likely to intensify treatment with either a GLP-1 RA or insulin. Overall, 33%–38% of patients intensifying with a GLP-1 RA achieved HbA1c <7% (53 mmol/mol). This is similar to rates reported in previous observational studies: 21%–33% at 12 months in patients intensifying OAD treatment with GLP-1 RA and basal insulin,13 and 30% in patients adding lixisenatide to basal insulin (±OADs).14
Patients in our study receiving a GLP-1 RA rather than OAD(s) or insulin were more likely to achieve target HbA1c without weight gain while discontinuing one or more baseline OADs. For baseline OAD discontinuation as a single endpoint, there was no difference between the GLP-1 RA and insulin cohorts, but patients who intensified treatment with a GLP-1 RA were nearly twice as likely to discontinue a baseline OAD as those in the OAD(s) cohort. Although the reasons for this cannot be inferred retrospectively, it is likely that baseline OAD discontinuation in the GLP-1 RA cohorts was partly driven by patients discontinuing DPP-4is, in line with the ADA guidance.4
Patients who were adherent to intensification therapy tended to achieve greater HbA1c reduction and weight loss from baseline. Medication-taking behavior was affected by both mode and frequency of administration: GLP-1 RAs were associated with higher adherence and persistence than insulin, but significantly lower adherence than OADs. This difference was partly driven by low adherence and persistence rates with OD rather than OW GLP-1 RAs; another possible explanation is the gastrointestinal side effects that are often linked to early discontinuation in this treatment class.4 In future, analyzing adherence and persistence separately during days 0–90 and days 91–180 after intensification would allow examination of medication-taking behavior in those who persisted with treatment beyond 90 days.
Our use of linked EHR and claims data was a major strength of this study because it provided confirmation that all included patients had received their medication. Consequently, our study was able to mimic specific points in the treatment pathway, providing evidence to complement the results of previous RCTs that have examined treatment intensification for patients receiving two OADs.15 Although our approach in assessing GLP-1 RAs as a class rather than individually is limited in its specificity and does not take into account the differences between GLP-1 RAs, it accurately reflects clinical decision-making, whereby treating physicians first identify the most appropriate class of antidiabetic medication before selecting a specific therapy.
Lack of randomization and risk of bias are potential limitations of retrospective observational studies. To mitigate against these, we used PS matching to obtain balanced cohorts and importantly performed exact matching by baseline HbA1c and BMI categories, resulting in cohorts with identical baseline values and allowing us to detect clinically significant changes at follow-up. Subsequent adjustment of the models used to calculate differences between treatment cohorts was intended to address residual confounding, which was particularly valuable for analyses in subpopulations of the main cohorts. We conducted sensitivity analyses using unadjusted models and obtained results similar to the main analyses; however, it is possible that residual confounding remained. It should be noted that PS matching effectively created two subpopulations: those for whom treatment intensification with either an OAD or a GLP-1 RA is indicated and those whose disease has progressed to the requirement for either a GLP-1 RA or insulin. This is reflected in the higher mean HbA1c in the insulin cohorts before matching, indicating that these patients had more advanced disease than those in the other treatment cohorts. Therefore, our results should be interpreted in the context of these distinct treatment decisions. We also acknowledge that, perhaps due to use of the Explorys database, the mean BMI of patients in our study was slightly higher than might be expected in a general population with type 2 diabetes. Although the separation of HbA1c and weight/composite outcomes cohorts could be considered to limit the generalizability of the results, we show that baseline demographic and disease characteristics were similar across all cohorts, providing relatively high confidence that HbA1c and weight benefits are realized in similar patient populations.
Another limitation of claims data is the need to infer some aspects of medication-taking behavior. We selected eligible patients by the presence of relevant treatment claims, with no requirement for continuous treatment periods during the baseline period and no restriction by number of days of index medication prescribed. Although this is an established method of patient identification in claims databases and avoids a possible source of selection bias, it meant that some instances of baseline or index treatment discontinuation may have gone undetected. In future, when more eligible patients are available in these databases, sensitivity analyses requiring overlapping treatment periods could be conducted to assess whether this had an impact on our results.
In our PS-matched cohorts from a population receiving two OADs, treatment intensification with a GLP-1 RA provided significant benefits in terms of glycemic control and weight management compared with additional OADs or insulin, respectively. Future analyses including patients receiving newer GLP-1 RAs would be expected to show even greater benefits associated with this treatment class due to their comparatively greater efficacy. Furthermore, the PATHWAY study design and matching methodology will be valuable to compare different treatment intensification options, to examine other points in the treatment pathway, to assess costs and resource use, and to examine how the availability of oral semaglutide might affect future prescribing practices and patient acceptance of GLP-1 RAs.