Latent class distributional regression for the estimation of non-linear reference limits from contaminated data...

Area of Science:

• Biostatistics
• Medical Informatics
• Quantitative Health Sciences

Background:

• Reliable reference intervals are crucial for medical decision-making based on quantitative test results.
• Current methods for estimating reference limits face challenges with contaminated datasets and age-dependent analyte dynamics.
• Existing approaches often require extensive data or are computationally intensive.

Purpose of the Study:

• To develop a novel method for estimating indirect reference limits from contaminated datasets.
• To address non-linear dependencies on covariates in reference interval estimation.
• To combine mixture models and distributional regression for improved accuracy.

Main Methods:

• Proposed a new method combining mixture models and distributional regression.
• Utilized latent class distributional regression for indirect reference limit estimation.
• Applied the method to estimate pediatric reference limits for hemoglobin concentration.

Main Results:

• Simulation results demonstrated accurate quantile approximation, improving with sample size and component separation.
• The method successfully estimated clinically useful pediatric reference limits for hemoglobin.
• The approach is computationally less expensive than alternative methods requiring more samples.

Conclusions:

• Latent class distributional regression models offer a novel way to estimate indirect non-linear reference limits in a single model fit.
• The developed framework is flexible and applicable to various scenarios with latent heterogeneity.
• This method provides a more efficient and accurate approach to establishing reference intervals.