Background: Non-benzodiazepine hypnotics (i.e., “Z-drugs”) are often prescribed to treat insomnia and may increase the risk of motor vehicle crash (MVC) through prolonged drowsiness and delayed reaction times, though the effect is unknown.
Objective: To estimate the effect of Z-drug initiation on the risk of MVC among older adults in a sequential target trial emulation.
Methods: We linked New Jersey (NJ) driver licensing and police-reported MVC data to Medicare claims. We emulated a new target trial each week from July 1, 2007 through October 7, 2017 (for a total of 537 sequential target trials) in which Medicare fee-for-service beneficiaries aged ≥ 66 years and licensed in NJ were classified as Z-drug initiators or non-initiators at baseline and (for the intention-to-treat [ITT] estimand) followed for the MVC outcome, death, disenrollment from Medicare, or 12 weeks of follow-up, whichever occurred first. For the per-protocol (PP) estimand, we added deviating from treatment assignment (by discontinuation or switching) as censoring criteria. We used inverse probability of treatment and censoring weighted pooled logistic regression models to estimate risk ratios (RR) and risk differences (RD) with 95% bootstrap confidence limits (CLs) adjusting for potential confounding and selection bias by 30 investigator-specified and empirically identified (via the Bross formula) covariates, including age, sex, comorbidities, and prior medication use and healthcare utilization.
Results: Among 181,473 unique persons there were a total of 257,554 person-trials (median age 74 years; 62% female; 87% white), of which 103,371 were Z-drug initiator and 154,183 were non-initiator person-trials giving rise to 976 and 1,249 MVCs respectively. The ITT RR was 1.06 (95%CLs 0.95, 1.16) and the RD was 0.05 MVCs per 100 person-trials (95%CLs -0.04, 0.13). For the PP estimand, there were 800 MVCs among initiator person-trials and 1,241 MVCs among non-initiator person-trials (RR 0.83 [95%CLs 0.74, 0.92]; RD -0.15 MVCs per 100 person-trials [95%CLs -0.24, -0.06]). There was no evidence of meaningful effect heterogeneity on either the relative or the absolute scales across strata of either age or sex.
Conclusions: Sustained use of Z-drugs after initiation resulted in a lower MVC risk, but initiation alone resulted in a slightly higher risk. More work is necessary to understand the divergence of the ITT and PP effect estimates. Based on our findings, Z-drugs should be prescribed to older patients judiciously but not withheld entirely over concerns about MVC risk.
