A new study demonstrates the feasibility of utilizing unsupervised machine learning to differentiate between normal and pathological results in unlabeled screening laboratory tests, specifically for hepatic biomarkers. Researchers have developed a methodology that estimates reference intervals—the range of values considered normal for healthy individuals—by modeling result distributions as transformed Gaussian mixtures. This approach leverages routinely generated clinical laboratory data, which provides a more granular and precise alternative to traditional methods that are often costly, time-consuming, and prone to using generic, one-size-fits-all intervals that fail to account for significant demographic factors such as ethnicity, age, and sex.
By applying this unsupervised machine learning model to the Puerto Rican population, the research team successfully estimated reference intervals for six specific hepatic biomarkers. The methodology extends beyond conventional one-dimensional reference intervals by creating multivariate reference surfaces. These surfaces represent the 95% boundary of the joint distribution of multiple biomarkers, thereby capturing complex correlations between them that rectangular combinations of univariate intervals simply cannot account for. These reference surfaces offer a more holistic assessment compared to traditional one-dimensional reference intervals.
The study, published in BMC Medical Informatics and Decision Making, emphasizes that the reference surfaces correspond to level sets of the inverse survival function defined via Mahalanobis distance. This conceptual shift allows the traditional binary abnormal flag—often used in clinical settings—to be generalized into a survival probability, which provides clinicians with a more quantitative and continuous measure of abnormality.
Instead, this method taps into the vast amount of clinical data produced during everyday medical practice. While this real-world data is often unlabeled, the Bayesian Gaussian mixture modeling allows the study to extract parameters for the healthy distribution directly from the collective data, effectively separating pathological signals from healthy baseline variations with high efficiency.
Despite these advancements, the authors acknowledge important limitations to their current findings. While univariate reference intervals have long been established and remain highly useful for comparability with current clinical practices, the diagnostic utility of these newly constructed multivariate reference surfaces has yet to be rigorously established against clinical outcomes.
The study demonstrates that real-world clinical data, when processed through advanced probabilistic models, can yield reference surfaces that account for biomarker interdependence in the Puerto Rican population.
Subject of Research: Machine learning methodology for multivariate reference intervals in clinical diagnostics.
Article Title: Machine learning method to estimate multivariate reference surfaces from real-world data applied to liver disease diagnostics in the Puerto Rican population
Article References: Velev, J., LeBien, J., Hernandez-Suarez, D., & Roche-Lima, A. (2026). Machine learning method to estimate multivariate reference surfaces from real-world data applied to liver disease diagnostics in the Puerto Rican population. BMC Medical Informatics and Decision Making. https://doi.org/10.1186/s12911-026-03955-7
Image Credits: AI Generated
DOI: 10.1186/s12911-026-03955-7
Keywords: Machine Learning, Liver Disease, Puerto Rico, Biomarkers, Clinical Diagnostics, Reference Intervals, Gaussian Mixture Modeling, Medical Informatics, Unsupervised Learning, Multivariate Analysis, Hepatic Function, Real-World Data, Bayesian Modeling, Reference Surfaces, Survival Probability
Cite Scienmag News
Teresa Odom. (November 18, 2026). Machine Learning Estimates Multivariate Hepatic Reference Surfaces in Puerto Rican Population. Scienmag. https://scienmag.com/machine-learning-estimates-multivariate-hepatic-reference-surfaces-in-puerto-rican-population/
Teresa Odom. "Machine Learning Estimates Multivariate Hepatic Reference Surfaces in Puerto Rican Population." Scienmag, 18 November 2026, https://scienmag.com/machine-learning-estimates-multivariate-hepatic-reference-surfaces-in-puerto-rican-population/. Accessed 18 November 2026.
Teresa Odom. "Machine Learning Estimates Multivariate Hepatic Reference Surfaces in Puerto Rican Population." Scienmag. November 18, 2026. https://scienmag.com/machine-learning-estimates-multivariate-hepatic-reference-surfaces-in-puerto-rican-population/








