Combining complete multivariate outcomes with incomplete covariate information: a latent class approach

Area of Science:

• Biostatistics
• Epidemiology
• Health Services Research

Background:

• Combining outcome data from reference populations with risk factor data from screened subpopulations presents analytical challenges.
• Existing methods may introduce ambiguity and bias when analyzing associations between risk factors and multiple binary outcomes.

Purpose of the Study:

• To propose a novel two-stage latent class procedure for integrating disparate population data.
• To analyze the association between risk factors and multiple binary outcomes efficiently.

Main Methods:

• A two-stage latent class procedure is developed to first summarize outcome commonalities in a reference population.
• A pseudo-maximum likelihood approach is utilized for estimating model parameters.
• The method's performance is assessed via simulation studies and real-world data analysis.

Main Results:

• The proposed method effectively combines outcome and risk factor information, improving accuracy and precision.
• It successfully summarizes commonalities among multiple categorical outcomes.
• Ambiguity and bias in risk factor inferences are significantly diminished.

Conclusions:

• The developed two-stage latent class procedure offers a robust approach for analyzing complex health data.
• This method enhances the ability to make reliable inferences about risk factors and their associations with multiple outcomes.
• It provides a valuable tool for epidemiological and health services research.