Modeling Multivariate Distributions of Lipid Panel Biomarkers for Reference Interval Estimation and Comorbidity...

Area of Science:

• Biostatistics
• Population Health
• Clinical Laboratory Science

Background:

• Laboratory tests are crucial for medical diagnosis, relying on reference intervals (RIs) from healthy populations.
• Traditional RI derivation is expensive, time-consuming, and often ignores demographic influences like age, sex, and ethnicity.
• This study presents a data-driven method to derive RIs from existing laboratory data.

Purpose of the Study:

• To establish a data-driven approach for deriving population-specific, sex- and age-stratified reference intervals (RIs) from routine laboratory results.
• To demonstrate the utility of Gaussian Mixture Models (GMM) and network analysis in separating healthy from pathological subpopulations and accounting for comorbidities.
• To explain counterintuitive age trends in lipid biomarkers by examining selective mortality patterns.

Main Methods:

• Utilized large-scale, real-world laboratory data from the Puerto Rican population.
• Employed Gaussian Mixture Models (GMM) to estimate multidimensional joint distributions of lipid biomarkers.
• Applied statistical analyses, including selective mortality examination and comorbidity network construction, to refine RI derivation without diagnostic codes.

Main Results:

• Generated sex- and age-stratified RIs for lipid panel biomarkers (total cholesterol, LDL, HDL, triglycerides).
• Explained apparent biomarker improvements post-midlife through selective survival.
• Quantified the impact of comorbidities on RI ranges and captured interdependencies using network analysis.
• Enabled continuous risk assessment by mapping individual results to percentiles within full biomarker distributions.

Conclusions:

• Population-specific, sex- and age-segmented RIs can be derived from real-world laboratory data, eliminating the need for dedicated healthy cohorts.
• Incorporating selective mortality and comorbidity network analysis provides deeper insights into population health dynamics.
• This data-driven approach enhances the precision and applicability of laboratory test interpretation for personalized risk assessment.