Reflection on modern methods: when worlds collide-prediction, machine learning and causal inference.

Area of Science:

• Statistics
• Epidemiology
• Machine Learning

Background:

• Causal inference traditionally relies on theory and prior knowledge, not prediction modeling.
• Contemporary causal inference methods, based on potential outcomes, often incorporate prediction steps.
• Machine learning offers advanced 'best prediction' approaches.

Purpose of the Study:

• To overview the emergence of prediction in causal inference.
• To explore the role of machine learning in causal inference.
• To examine specific causal inference methods.

Main Methods:

• Overview of contemporary causal inference methods.
• Exploration of machine learning applications within these methods.
• Focus on propensity scores, IPTWs, G computation, and TMLE.

Main Results:

• Prediction modeling is an emerging and useful perspective in causal inference.
• Machine learning is already utilized in propensity score and TMLE methods.
• There is potential for greater machine learning application in G computation and IPTW estimation.

Conclusions:

• Prediction modeling, especially machine learning, is becoming integral to causal inference.
• The integration of 'best prediction' enhances causal analysis.
• Further adoption of machine learning in causal inference methods is anticipated.