Comparisons Between Hypothesis- and Data-Driven Approaches for Multimorbidity Frailty Index: A Machine Learning...

Area of Science:

• Gerontology and Public Health
• Health Services Research
• Biostatistics and Machine Learning

Background:

• The multimorbidity frailty index (mFI) is widely used but faces challenges in selecting critical determinants and stratifying risk.
• Developing a robust mFI requires identifying key factors and establishing clear dose-response relationships for clinical practice.

Purpose of the Study:

• To develop a machine learning multimorbidity frailty index (ML-mFI) using optimal variable selection.
• To establish four distinct risk categories with a demonstrable dose-response relationship for adverse outcomes.

Main Methods:

• Utilized Taiwan's National Health Insurance Research Database for a cohort aged 65-100 years.
• Employed the random forest method for data-driven selection of 38 influential diseases/deficits for the ML-mFI.
• Developed distance and coverage indices to stratify individuals into fit, mild, moderate, and severe frailty categories for survival analysis.

Main Results:

• The ML-mFI, comprising 38 diseases/deficits, demonstrated similar age and sex distribution to conventional mFI.
• Survival analysis confirmed the ML-mFI's significant predictive ability for all-cause mortality, unplanned hospitalizations, and ICU admissions over 8 years.
• A clear dose-response relationship was observed between the four ML-mFI risk groups and adverse health outcomes.

Conclusions:

• The developed ML-mFI effectively stratifies older individuals into risk groups for mortality and hospitalizations.
• This data-driven approach supports the implementation of precise, patient-centered medical care in an aging population.