Covid-19 risk factors: Statistical learning from German healthcare claims data

TL;DR

This study employs a detailed statistical learning approach on German healthcare claims data to identify risk factors for severe COVID-19 outcomes, avoiding prior variable grouping and enhancing predictive accuracy.

Contribution

It introduces a novel methodology that uses hierarchical medical classification data with over 33,000 covariates, improving risk prediction without prior subject-matter assumptions.

Findings

Better predictive ability than morbidity groups

Utilizes over 33,000 covariates from medical classifications

Provides coefficients for risk factor prioritization

Abstract

We analyse prior risk factors for severe, critical or fatal courses of Covid-19 based on a retrospective cohort using claims data of the AOK Bayern. As our main methodological contribution, we avoid prior grouping and pre-selection of candidate risk factors. Instead, fine-grained hierarchical information from medical classification systems for diagnoses, pharmaceuticals and procedures are used, resulting in more than 33,000 covariates. Our approach has better predictive ability than well-specified morbidity groups but does not need prior subject-matter knowledge. The methodology and estimated coefficients are made available to decision makers to prioritize protective measures towards vulnerable subpopulations and to researchers who like to adjust for a large set of confounders in studies of individual risk factors.