Advantages of Splitting Switchers and New Users in the Prevalent New User Design

Background: The prevalent new user design (PNUD) extends the traditional active comparator new user design to additionally include patients who switch to a drug of interest from a comparator drug (a population often excluded in traditional new-user designs). Initial work combined two distinct types of users: (1) patients switching to the drug of interest after prior use of the comparator drug (counterfactual: continuing the comparator), and (2) patients initiating the drug of interest without prior use of the comparator (counterfactual: initiating the comparator). As proposed, the PNUD conflates two separate causal questions.

Objectives: We examined the impact of excluding new-users of a drug of interest who have no prior use of the comparator from the cohort of prevalent new-users.

Methods: Using MarketScan claims data (2005-2014), we compared the risk of heart failure between patients initiating GLP-1 analogs and patients initiating or continuing sulfonylureas (SU). Baseline comorbidities and medication use were assessed using an all-available lookback with a 365-day minimum. Continuous use and switching were defined using a 30-day grace period beyond the prescription's day supply. We considered two populations of interest: (1) "GLP-1 New+Switch" combined new users of GLP-1 (no prior SU, all-available lookback) with patients switching from SU to GLP-1, and (2) "GLP-1 Switch" consisted only of patients switching from SU to GLP-1. For both populations, we used two separate methods to identify comparator cohorts and control for baseline confounding. First, using time-conditional propensity scores (TCPS) and prescription-based exposure sets we matched (1:1, with replacement) GLP-1 users to SU users. Second, using time-stratified standardized morbidity ratio weighting (SMRW). Cox-proportional hazard models estimated hazard ratios (HR) and 95% CI of hospitalized heart failure.

Results: We identified 76,179 GLP-1 patients (GLP-1 New+Switch) meeting inclusion criteria. Among these patients, 12.1% (9,220) switched directly from SU to GLP-1 (GLP-1 Switch). We identified 483,490 SU users, for a total of 3,015,387 potential index dates. In the GLP-1 New+Switch cohort, GLP-1 patients were at lower risk of heart failure in the TCPS (adjHRTCPS = 0.86 [95% CI: 0.81,0.92]) and SMR analyses (adjHRSMR=0.77[0.72,0.82]). In the GLP-1 Switch cohort, estimates moved towards the null (adjHRTCPS = 0.98 [95% CI: 0.82,1.16], adjHRSMR=0.95 [0.80,1.13]) .

Conclusions: In our assessment of SU and GLP-1 users, we found that switchers were a small proportion of all GLP-1 users. When analyses were conducted only among switchers, the effect estimates substantially shifted towards the null. The importance of separating new users and switchers is likely to vary based on the clinical context.