Variable selection and prediction using a nested, matched case-control study: Application to hospital acquired...

Area of Science:

• Epidemiology
• Neuroimaging
• Biostatistics

Background:

• Matched case-control designs enhance efficiency in epidemiologic studies.
• Modern neuroimaging and genomic studies present challenges due to high-dimensional covariates.
• Statistical methods that adjust for matching in these complex studies are not widely adopted.

Purpose of the Study:

• To identify specific brain regions of acute infarction associated with hospital-acquired pneumonia (HAP) in acute ischemic stroke patients.
• To investigate penalized logistic regression approaches for variable selection in matched case-control studies with high-dimensional interactions.
• To develop a predictive score for HAP in future stroke patients by integrating imaging and clinical data.

Main Methods:

• A matched case-control study of 430 acute ischemic stroke patients at Massachusetts General Hospital (MGH).
• Analysis of 138 brain regions and nearly 10,000 two-way interactions.
• Application of penalized conditional and unconditional logistic regression for variable selection, accounting for matching.

Main Results:

• Identification of specific brain regions associated with HAP in acute ischemic stroke.
• Demonstration of penalized regression methods for handling high-dimensional interactions in matched studies.
• Development of a framework for a HAP prediction score using combined imaging and clinical data.

Conclusions:

• Penalized logistic regression methods effectively address variable selection in matched case-control neuroimaging studies.
• Integrating neuroimaging and clinical data from nested studies can yield robust HAP prediction scores for stroke patients.
• The study provides a methodological approach for analyzing complex matched cohort data in clinical research.