Abstract
<title>Abstract</title> <p>Background Frailty is highly prevalent in nursing home residents and is associated with excess mortality, yet scalable measurement from routine records remains uncommon in many long-term care settings. We developed an automatable 31-item Frailty Index for Nursing Homes in Spain (FI31-NH) using routinely collected clinical data and assessed its initial prognostic validity for all-cause mortality. Methods This retrospective cohort study used anonymised clinical and administrative data from a large network of nursing homes in Spain. FI31-NH was constructed following the deficit accumulation approach using multidomain routine clinical information recorded during the first 90 days after admission. Baseline was defined individually as the earliest date within this window at which sufficient information was available to compute FI31-NH. Cox models adjusted for age and sex estimated hazard ratios (HRs), presented per 0.10 increase in FI31-NH for clinical interpretability. Discrimination was assessed using the concordance index (C-index). Kaplan-Meier estimates compared cumulative mortality using the FI31-NH ≥ 0.25 threshold. Results FI31-NH was computable in 11,145 residents. Mean age was 84.6 years, 69.7% were women, and mean FI31-NH was 0.257 (SD 0.094), with a right-skewed distribution (p99 0.484; observed range 0.032–0.661). Mortality follow-up was available for 10,863 residents, among whom 6,754 deaths occurred; median follow-up was 749 days. In age- and sex-adjusted Cox models including 10,728 residents and 6,688 deaths, each 0.10 increase in FI31-NH was associated with higher mortality (HR 1.23, 95% CI 1.21–1.27), with moderate discrimination (C-index 0.643, SE 0.004). Using FI31-NH ≥ 0.25, 1-year cumulative mortality was 17.5% versus 11.5%, and 5-year mortality was 69.6% versus 54.6%. Conclusions FI31-NH can be derived from routine nursing home records and shows expected distributional properties and moderate prognostic validity for mortality. These findings support its potential use for scalable, automatable frailty stratification among long-stay nursing home residents after sufficient routine clinical assessment has been completed. However, external validation and further evaluation of calibration, predictive performance and clinical utility are needed before implementation as a decision-support tool.</p>