Heterogeneous treatment effects in observational studies (MA)
Background:
While generating high-quality evidence used to be reserved to RCTs, real world evidence nowadays bears large potentials in that regard, too. The dispersion of machine learning methods and expanding data is improving the ability to identify patients at risk of healthcare events so that providers can intervene to prevent adverse health outcomes. Researchers utilizing machine learning approaches are increasingly exploring heterogeneous treatment effects (HTE) as opposed to overall average treatment effects. Heterogeneous treatment effects are those that are systematically different within different groups of study subjects. Prominent examples for detecting HTE include random forests, regression trees, and neural networks. Traditional methods to identify subpopulations through regression modeling or interaction terms will likely produce multiple testing bias. Machine learning on the other hand uses algorithmic approaches instead of hypothesis testing. That means that the evaluation of as many variables as desired is possible without increasing the statistical error.
Objective:
Apply a machine learning based method to real-world intensive care data (e.g. MIMIC – IV, AmsterdamUMCdb, eICU) with the intention to examine heterogeneous treatment effects. Examples could be characteristics that explain re-admission, survival analysis, effect of a particular intervention, etc.
Literature:
Athey, S. & Imbens, G. W. (2019). “Machine Learning Methods That Economists Should Know About”, Annu. Rev. Econ., vol. 11, pp. 685-725. doi: 10.1146/annurev-economics-080217-053433
Mullainathan, S. & Spiess, J. (2017). “Machine Learning: An applied Econometric Approach”, Journal of Econ. Persp., vol. 31, no. 2, pp. 87-106. doi: 10.1257/jep.31.2.87
Salditt, M., Eckes, T. & Nestler, S. (2023). „A Tutorial Introduction to Heterogeneous Treatment Effect Estimation with Meta-learners”. Adm Policy Ment Health. doi: 10.1007/s10488-023-01303-9
Note: Writing in English language desired.