Causality on longitudinal data: Stable specification search in constrained structural equation modeling.

Area of Science:

• Causal inference
• Machine learning
• Biostatistics

Background:

• Causal modeling often suffers from structural instability with finite data.
• Existing methods struggle with reliable causal structure learning from limited samples.

Purpose of the Study:

• To develop a novel causal modeling algorithm for longitudinal data that is robust to finite sample instability.
• To improve the accuracy and reliability of causal discovery in complex datasets.

Main Methods:

• Utilized stability selection with subsampling and selection algorithms for robust causal discovery.
• Employed structural equation models and multi-objective evolutionary algorithms to identify Pareto optimal models.
• Incorporated prior knowledge and stability selection to refine causal substructures for visualization via causal graphs.

Main Results:

• The novel algorithm demonstrated comparable or significantly improved performance over state-of-the-art methods on simulated data.
• Applied to real-world longitudinal datasets (chronic fatigue syndrome, Alzheimer's, chronic kidney disease), findings align with existing research.
• Identified potential novel causal relationships warranting further investigation in clinical datasets.

Conclusions:

• The proposed causal modeling algorithm offers a robust and effective approach for analyzing longitudinal data.
• This method enhances causal discovery's reliability, particularly in scenarios with limited or noisy data.
• The findings suggest new avenues for research into complex diseases using advanced causal inference techniques.