Application of Targeted Maximum Likelihood Estimation using Varying Lookback Windows in Pharmacoepidemiology

In observational research, confounding control is critical to appropriate inference.In the field of pharmacoepidemiology, determining the appropriate lookback period is essential as it dictates how far back we examine potential confounding factors from the time of exposure.This study aims to compare treatment effect estimates using two different approaches: targeted maximum likelihood estimation (TMLE) and propensity score method under inverse probability weighting approach (PS-IPW), considering varying lookback periods (short term and long term).To conduct the comparison, simulation settings were established, considering eight different lookback periods (1,3,6 months, and 1,2,5,7,9 years), with 10 years as the ideal reference lookback period.We applied two different approaches: PS-IPW and TMLElogistic, both within the logistic regression framework.Additionally, we included TMLE using SuperLearner (TMLE-SL) as part of our simulation.To assess the effect of lookback on propensity score models, propensity score quantile estimates were computed, revealing that longer lookback periods exhibited less bias compared to shorter ones.Subsequently, the average treatment effect (ATE) along with standard error was estimated using PS-IPW, TMLE-logistic, and TMLE-SL.It was observed that TMLE-logistic and TMLE-SL produced lower standard errors than the PS-IPW approach across the varying lookback periods.Finally, the study applied both methods to the CPRD (Clinical Practice Research Datalink) database to evaluate how TMLE and PS-IPW perform in real-life scenarios for each of the specified lookback periods.In conclusion, this research contributes valuable insights into the impact of lookback periods on treatment effect estimates, highlighting the advantages of using TMLE approaches over PS-IPW in certain lookback scenarios.Furthermore, the application of these methods to real-world data from CPRD aids in understanding their performance in practical healthcare research settings.i AbrgEn recherche observationnelle, le contrle des facteurs de confusion est essentiel pour obtenir des infrences appropries.Dans le domaine de la pharmacopidmiologie, la dtermination de la priode de rfrence approprie est essentielle, car elle dtermine quel point nous examinons les facteurs de confusion potentiels partir du moment de l'exposition.Cette tude vise comparer les estimations des effets du traitement en utilisant deux approches diffrentes : l'estimation du maximum de vraisemblance cible (TMLE) et la mthode du score de propension avec approche de pondration par probabilit inverse (PS-IPW), en tenant compte de diffrentes priodes de rfrence (court terme et long terme).Pour mener la comparaison, des paramtres de simulation ont t tablis en considrant huit priodes de Above all else, I wish to convey my profound appreciation to my supervisor Dr. Robert Platt, for being an unwavering source of support throughout my graduate study.

Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.