NEW YORK — Using electronic health record data for millions of individuals over a nearly two-decade period, scientists from the Weizmann Institute have developed a computational system for the personalized interpretation of standard lab tests.
The work, which was reported in Nature Medicine on Monday, represents a new tool that can help clinicians evaluate lab results that fall within a normal range on a per-patient basis and determine disease risk in still-healthy individuals, according to the researchers.
Clinical lab tests, a mainstay of patient diagnosis and management, are interpreted in two steps: First, values are classified as normal or abnormal based on predefined reference ranges, then the results are interpreted by a physician in the context of an individual patient's medical history, current health status, and other diagnostic tests.
"While the goals of setting reference ranges [are] to provide a context to the test result using population-based distributions, analysis of intra-individual trends and variation within these ranges is challenging and typically approached qualitatively," the Weizmann team wrote. "Physician decisions on management of such within-normal changes are typically not supported by quantitative and precise decision algorithms."
As such, a holistic system for interpreting normal lab test trends and how they affect disease potential could enhance the quality and precision of medical decision making, they stated.
To build such a resource, the researchers — along with collaborators from Tel Aviv Sourasky Medical Center and the Clalit Research Institute — performed a retrospective analysis of an integrative electronic health record, or EHR, resource from the Clalit Healthcare system that included 2.1 billion measurements from 92 lab tests performed on 2.8 million adults between 2002 and 2019.
To compensate for the bias of EHR data toward patients with active chronic diseases, a research engine was developed for the systematic and unbiased classification of patients' lab test trajectories, filtering segments representing 131 major pathological conditions or the effects of specific drugs.
With the filtered dataset, the investigators created computational tools for multivariate longitudinal analysis that could predict patients' within-normal lab trajectories with very high accuracy. Personalized lab models built using these tools, the team wrote, can be applied to patients with still-normal lab readouts to evaluate risks for future lab test abnormalities, the development of multiple types of chronic diseases, and overall mortality.
"Our analysis shows unequivocally that variation in lab tests values between individuals is mixing heritable and environmental effects that intensify with age to define highly individualized patient trajectories," they wrote. "Even when lab values are well within the normal ranges, their quantitative interpretation is instructive and predictive for overall survival, future emergence of pathological lab results, and progression toward chronic diseases."
Applying this approach for personalized and quantitative lab readout interpretation is within reach for healthcare systems using modern EHR management software, which already accumulates the necessary data, the researchers stated. "The described methodology can be deployed flexibly in systems with different testing policy and population characteristics and the benefit of the approach is not limited to systems with very deep longitudinal coverage."
The ultimate clinical impact of personalized lab modeling, however, will rely on improved guidelines for using quantitative metrics in preventive medicine and diagnostic workup, as well as the development of new disease predictors and specific calculators, the researchers noted.
Additionally, "a unified quantitative framework for understanding patient trajectories in lab space will become increasingly important. This will be synergistic with new molecular markers and continuous sensors integration into the medical data acquisition and decision algorithms," they wrote.