Related Experiment Videos
Thirteen Questions About Using Machine Learning in Causal Research (You Won't Believe the Answer to Number 10!).
American Journal of Epidemiology
|March 22, 2021
Summary
Machine learning offers new ways to reduce bias in health science research beyond prediction. This guide helps epidemiologists use these causal inference techniques, even with complex models, and provides code examples.
Area of Science:
- Health Sciences
- Epidemiology
- Biostatistics
Background:
- Machine learning (ML) is increasingly used in health sciences, primarily for predictive modeling.
- Concerns exist regarding ML's "black box" nature and potential for bias in causal inference.
- Traditional methods may suffer from model misspecification, introducing bias.
Purpose of the Study:
- To introduce epidemiologists to machine learning for causal inference.
- To address concerns about bias and rigor when using ML in health research.
- To provide practical guidance and software examples for applying ML in causal analyses.
Main Methods:
- The study employs a question-and-answer format for clarity.
- It focuses on embedding machine learning within causal inference frameworks.
- Sample software code is provided to facilitate practical application.
Main Results:
- Machine learning can be utilized to reduce bias stemming from model misspecification in causal analyses.
- The "black box" concern can be mitigated by understanding ML's role in causal inference.
- The provided guidance and code aim to lower the barrier for epidemiologists.
Conclusions:
- Machine learning presents a valuable tool for enhancing causal inference in epidemiology.
- Careful application of ML can improve the rigor and reduce bias in health research.
- Accessible code examples empower researchers to adopt these advanced techniques.