(143) The prevalent new-user study design: a novel application to estimating the average treatment effect of statin discontinuation on mortality among traditional Medicare beneficiaries who discontinued statins

Background: Randomized trials of statin discontinuation are difficult to conduct and may not be generalizable to real-world populations. Approaches like marginal structural models estimate the average treatment effect of discontinuation in everyone but estimating the average treatment effect of discontinuing (ATD) among discontinuers may be more clinically relevant. In prior simulations, we combined the prevalent new-user (PNU) design with odds weights and was able to obtain an unbiased estimate of the ATD among discontinuers.

Objectives: To use a PNU approach to estimate the ATD on mortality among older adults who discontinued statins.

Methods: From a 20% random sample of traditional Medicare beneficiaries, we assembled a cohort of new statin users (i.e., no use in the prior year) who go on to have ≥5 years of continuous statin use (i.e., use gaps ≤90 days). We used the PNU design with odds weights to estimate the ATD on mortality among statin discontinuers, treating discontinuation like switching to a new drug and continuation as the comparator. For each month after 5 years of continuous statin use, we created exposure sets of (a) discontinuers and (b) continuers who had a prescription dispensed in that month, with equal duration of prior statin use. We estimated a monthly-stratified propensity score for discontinuation. We estimated mortality rates per 100 person-years (MR), rate differences (MRD), and rate ratios (MRR) for discontinuers and continuers with and without odds weighting for the continuers. We estimated 95% CIs with the 2.5th and 97.5th percentiles from 500 bootstrap iterations. We also estimated the predicted probability of frailty (PPF) using the Faurot Frailty Index for each patient and repeated the analyses among frail and non-frail patients, PPF ≥0.29 and ≤0.04, respectively.

Results: We identified 113,109 new statin users with ≥5 years of continuous statin use. Crude MRs were 12.48 among discontinuers and 6.77 among continuers. The weighted MRD was 3.31 (95% CI: 2.98, 3.62) and MRR was 1.36 (1.32, 1.40). Crude MRs were 36.13 and 21.55 among frail discontinuers and continuers, respectively. The weighted MRD was 11.37 (9.98, 13.26) and MRR was 1.46 (1.40, 1.55). Crude MRs were 2.99 and 2.43 among non-frail discontinuers and continuers, respectively. The weighted MRD was 0.65 (0.35, 3.12) and MRR was 1.27 (1.14, 1.40).

Conclusions: We combined the PNU study design with odds weights to estimate the ATD on mortality among statin discontinuers age ≥65 years and found that the MRs were higher among discontinuers compared to continuers. This association was less pronounced after controlling for measured confounders and persisted when stratified by frailty status, but much less pronounced in non-frail patients, compatible with residual confounding by frailty.