CMStatistics 2022: Start Registration
View Submission - CMStatistics
Title: An influence function based instrumental variable estimator of censored medical costs Authors:  Nicholas Illenberger - NYU Langone Health (United States) [presenting]
Nandita Mitra - University of Pennsylvania (United States)
Abstract: Studies aimed at estimating medical costs accrued under different treatments are critical to making informed healthcare policy decisions. Because cost analyses often use data from observational sources, their results may be biased due to unmeasured confounding or informative cost censoring. We introduce a partitioned, instrumental variable estimator of the complier average treatment effect on costs. Given a valid instrument, our estimator provides unbiased estimates of the target treatment effect in the presence of unmeasured confounding. Additionally, the use of a partitioned cost estimator allows us to address informative cost censoring and improve efficiency by utilizing data from patients with partially observed medical costs. Our proposed estimator is based on influence functions, allowing for multiple robust, efficient, and flexible semiparametric estimation. We present results from simulation studies to assess the performance of our proposed estimator under varying degrees of censoring and strength of IV. We apply our approach to a study assessing the costs of surgical and non-surgical interventions for gallstones and hemorrhaging using observational data.