Limits to Causal Inference with State-Space Reconstruction for Infectious Disease

Area of Science:

• Epidemiology
• Complex Systems Science
• Dynamical Systems Theory

Background:

• Infectious diseases exhibit complex dynamics, hindering hypothesis testing with traditional models.
• State-space reconstruction methods, including convergent cross-mapping (CCM), offer model-free approaches to infer causal interactions in nonlinear systems.

Purpose of the Study:

• To assess the practical limitations and sensitivity of convergent cross-mapping (CCM) for causal inference in ecological and epidemiological dynamics.
• To identify conditions under which CCM may produce spurious causal inferences.

Main Methods:

• Simulated dynamics of two interacting pathogen strains with varying interaction strengths.
• Evaluated the original CCM method and alternative criteria for inferring causality.
• Analyzed time series data of childhood infections in New York City and Chicago (pre-vaccine era).

Main Results:

• The original CCM method is highly sensitive to periodic fluctuations, falsely inferring interactions between independent oscillating strains.
• Alternative causality criteria mitigate sensitivity to periodic signals but CCM remains vulnerable to high process noise and attractor changes.
• Challenges in estimating noise and attractor quality in natural systems limit CCM's reliability.

Conclusions:

• Convergent cross-mapping (CCM) has practical limitations for causal inference in real-world infectious disease dynamics due to sensitivity to noise and system changes.
• While CCM offers a promising model-free approach, statistical and conceptual challenges must be addressed for robust application.
• Further research is needed to refine state-space reconstruction methods for reliable causal inference in complex biological systems.