Evaluating methods to quantify the effect of deprescribing a medication on patient clinical outcomes using a simulation study

Observational studies using electronic health records are being increasingly used to answer research questions of interest in deprescribing, where a medication is intentionally discontinued or the dose reduced. Target trial emulation (TTE) is particularly attractive as it is guided by sound theoretical principles based on causal inference. In the field of deprescribing, the evaluation of TTE within a simulation setting can provide evidence on optimal approaches for implementation and compare different estimation approaches which theory cannot easily inform. Deprescribing studies are susceptible to time-dependent confounding, competing risks due to early adverse events, time-dependent effects and other data complexities which may render estimates biased if such aspects are not properly addressed. In this work, a Monte-Carlo simulation study was conducted to estimate the performance (bias and coverage) of seven methods within several synthetic data generating mechanisms involving the above characteristics. The methods implemented were Cox regression, inverse probability of treatment weighted (IPTW) Cox regression, naive sequential Cox regression, IPTW weighted sequential Cox regression, extended Cox regression and the parametric g-formula. The IPTW Cox regression and IPTW sequential Cox regression were the only methods achieving nominal coverage and unbiasedness under the scenarios evaluated.