Causal inference of latent classes in complex survey data with the estimating equation framework

Area of Science:

• Statistics
• Epidemiology
• Health Survey Methodology

Background:

• Latent Class Analysis (LCA) is effective for clustering survey items.
• Causal inference with LCA-identified exposures presents challenges due to unobserved variables and confounding.
• Complex survey designs and weights add further complexity.

Purpose of the Study:

• To introduce a statistical procedure for causal inference with LCA-identified exposures in complex survey data.
• To address challenges of unobserved exposure uncertainty and confounding bias.
• To demonstrate the method using the National Health and Nutrition Examination Survey (NHANES) data.

Main Methods:

• Utilizing the expected estimating function approach to assess point estimates.
• Modifying survey design weights with LCA-based propensity scores.
• Applying the estimating equation approach for effect assessment.

Main Results:

• The proposed statistical procedure provides a framework for causal inference with LCA-identified exposures.
• The method accounts for unobserved exposure uncertainty and confounding bias.
• The approach is demonstrated to be applicable to complex survey data like NHANES.

Conclusions:

• This statistical procedure enhances causal inference capabilities when using LCA-identified exposures in complex survey data.
• The methodology offers a robust approach to handle unobserved variables and confounding in health research.
• The findings have implications for improving the accuracy of health outcome assessments from survey data.