Methodological Medical Research while Intergrating ML

Causal Research question

Use of ML in residual confounding bias reduction

Example research topics in advanced epidemiological methods include the comparison of supervised deep learning architectures with autoencoders, the development of double crossfitting (DCF) guidelines focusing on the optimal number of folds/splits and repetitions, and strategies for reducing residual confounding bias using high-dimensional propensity score (hdPS) approaches (M. E. Karim 2024) versus machine learning (ML) extensions (M. E. Karim, Pang, and Platt 2018). Additionally, enhancements to hdPS are explored through various methods, including the original hdPS, pure ML approaches, Targeted Maximum Likelihood Estimation (TMLE) without ML, TMLE with a choice of Super Learner, and TMLE combined with DCF and a choice of Super Learner (M. Karim 2023).

References

Karim, ME. 2023. “Rethinking Residual Confounding Bias Reduction: Why Vanilla hdPS Alone Is No Longer Enough.” https://doi.org/10.5281/zenodo.7877767.

Karim, Mohammad Ehsanul. 2024. “High-Dimensional Propensity Score and Its Machine Learning Extensions in Residual Confounding Control.” The American Statistician, no. just-accepted: 1–38.

Karim, Mohammad Ehsanul, Menglan Pang, and Robert W Platt. 2018. “Can We Train Machine Learning Methods to Outperform the High-Dimensional Propensity Score Algorithm?” Epidemiology 29 (2): 191–98.