A deep-learning-augmented electronic frailty index (eFI) built from 193,629 Finnish health records spanning ages 35–103 demonstrates that severe frailty carries a 7.31-fold elevated mortality risk, a 9.22-fold higher severe infection risk, and a 2.75-fold fracture risk compared to non-frail individuals. The 53-item index uniquely incorporates free-text clinical notes via natural language processing alongside standard diagnosis codes and lab values, outperforming both the Hospital Frailty Risk Score and Charlson Comorbidity Index on discrimination across all outcomes. Crucially, frailty acceleration begins meaningfully at age 65, yet elevated risks appear in younger cohorts too.
Frailty measurement has long been constrained by reliance on structured data and geriatric populations, missing the gradual accumulation of deficit that begins in midlife. This tool's extension into adulthood from age 35 is a meaningful conceptual shift — frailty is reframed as a life-course process, not solely an old-age syndrome. The NLP extraction of unstructured notes captures clinical signals routinely discarded by registry-based indices, potentially closing a significant information gap. For health systems, automated eFI scoring could enable proactive intervention decades earlier than current practice.
Limitations are notable: the cohort is geographically confined to central Finland, raising questions about generalizability across healthcare systems and ethnicities. The design is observational, so causal inference is restricted. As a preprint not yet peer-reviewed, the NLP validation methodology and potential overfitting warrant independent scrutiny. If findings hold post-review, this represents a genuinely paradigm-shifting approach to population-level frailty surveillance.
