Developing the Total Health Profile, a Generalizable Unified Set of Multimorbidity Risk Scores Derived From Machine...

Area of Science:

• Computational biology and bioinformatics
• Health informatics and data science
• Clinical epidemiology

Background:

• Existing multimorbidity scores often rely on single data types (diagnoses, labs) and may not capture comprehensive patient health.
• Current scores can be limited by demographic specificity (e.g., age) and lack nuanced risk stratification.
• There is a need for more integrated and generalizable risk assessment tools for clinical decision-making.

Purpose of the Study:

• To develop a physiologically diverse and generalizable set of multimorbidity risk scores using comprehensive electronic health record (EHR) data.
• To create integrated risk scores reflecting multiple organ systems and overall health.
• To improve the granularity and predictive accuracy of multimorbidity risk assessment.

Main Methods:

• Utilized a nationwide cohort of 794,294 patients from EHRs, including diagnosis, lab, prescription, procedure, and demographic data.
• Developed six machine learning models to predict inpatient hospitalizations within a 2-year follow-up period.
• Created a Total Health Score (THS) and organ-specific risk scores (heart, lung, neuro, kidney, digestive) evaluated on a separate cohort of 198,574 patients.

Main Results:

• The developed risk scores demonstrated strong performance, with the Total Health Score (THS) achieving an AUROC of 0.83.
• The THS outperformed traditional scores like the Charlson Comorbidity Index (CCI) and Elixhauser Comorbidity Index (ECI) across all demographic subgroups.
• Performance improvements were particularly notable in middle-aged and lower-income patient populations.

Conclusions:

• Machine learning applied to large-scale EHR data enables the creation of practical, highly predictive, and generalizable multimorbidity risk scores.
• The developed scores offer enhanced personalization and intuitive explanations for clinical use.
• These advanced risk scores can significantly aid clinicians in patient assessment and care management decisions.