Development and validation of prediction model (VALOR-m) for all-cause mortality following very late-onset schizophrenia-like psychosis: A 23-year population-based study.

OBJECTIVES

Individuals with very late-onset schizophrenia-like psychosis (VLOSLP) experience a persistent mortality gap compared to the general population, yet no validated prognostic tools tailored to this growing population exist. We aimed to develop and validate a 3-year all-cause mortality prediction model (VALOR-m) for individuals with VLOSLP.

DESIGN

We developed VALOR-m with LASSO-penalised Cox regression. To evaluate temporal drift and geographic generalisability, we employed a two-stage approach: 1)temporal validation stratified by two time periods (period 1: 2002-2013; period 2: 2014-2024), and 2)temporal recalibration with internal-external cross-validation (IECV) stratified by seven catchment areas in period 2. Discrimination was assessed by time-dependent AUC and Harrell's c-statistic; calibration by calibration slope and observed-to-expected ratio (O/E ratio).

PARTICIPANTS

Using a population-based electronic health record cohort from Hong Kong (2002-2024), we identified 33,594 older adults (aged≥60 years) with VLOSLP.

RESULTS

Of 33,594 individuals included (16,262[48·4%] male; mean age of onset 76·4 years), 11,272 (33·6%) died within 3 years. VALOR-m retained 11 predictors and demonstrated good discrimination (pooled time-dependent AUC:0·772 95%CI[0·765-0·779]; Harrell's c-statistic 0·734 95%CI[0·728-0·741]) and satisfactory calibration (calibration slope 0·885 95%CI[0·859-0·911]; O/E ratio 0·972 95%CI[0·908-1·041]) in IECV.

CONCLUSIONS

VALOR-m provides the first validated individualised mortality risk estimates for individuals with VLOSLP, applicable to real-world clinical practice using routinely collected data. VALOR-m achieved comparable discrimination to conventional comorbidity indices with fewer predictors, suggesting that it better captures the unique mortality profile of VLOSLP, and supporting its use to guide antipsychotic treatment decisions, personalised healthcare management in this understudied yet growing population.