Back

Data Imputation for Clinical Trial Emulation: A Case Study on Impact of Intracranial Pressure Monitoring for Traumatic Brain Injury

Zhao, Z.; Liu, R.; Groner, J. I.; Xiang, H.; Zhang, P.

2023-01-30 health informatics
10.1101/2023.01.29.23285172 medRxiv
Show abstract

Randomized clinical trial emulation using real-world data is significant for treatment effect evaluation. Missing values are common in the observational data. Handling missing data improperly would cause biased estimations and invalid conclusions. However, discussions on how to address this issue in causal analysis using observational data are still limited. Multiple imputation by chained equations (MICE) is a popular approach to fill in missing data. In this study, we combined multiple imputation with propensity score weighted model to estimate the average treatment effect (ATE). We compared various multiple imputation (MI) strategies and a complete data analysis on two benchmark datasets. The experiments showed that data imputations had better performances than completely ignoring the missing data, and using different imputation models for different covariates gave a high precision of estimation. Furthermore, we applied the optimal strategy on a medical records data to evaluate the impact of ICP monitoring on inpatient mortality of traumatic brain injury (TBI). The experiment details and code are available at https://github.com/Zhizhen-Zhao/IPTW-TBI.

Matching journals

The top 4 journals account for 50% of the predicted probability mass.

50% of probability mass above

"Similar papers" are the closest papers from that journal in the model's embedding space. They show what the match is built on, but the ranking comes mostly from a classifier over the whole training set, not from these examples alone.