Practice Innovations, Vol 10(1), Mar 2025, 1-14; doi:10.1037/pri0000276
Nursing home/assisted living placement (NHP/ALP) for dementia patients is costly and may be unfeasible for many. Various patient characteristics have been identified as contributors of NHP/ALP in dementia. Longitudinal prognostic models, which estimate the time from diagnosis to NHP/ALP based on characteristics and changes over time in these factors, are scarce yet may be valuable for clinicians in the planning of community-based interventions to delay future admissions. A multivariate Bayesian joint longitudinal and survival modeling approach was applied to develop an algorithm and web-based application and estimate individualized patient survival probability and time to NHP/ALP. Data were analyzed from the National Alzheimer’s Coordinating Center’s data set. Four thousand four hundred twenty-one participants with a diagnosis of dementia were included in algorithm development. The model was validated on a separate hold-out sample (n = 780). Fourteen predictors examining patients’ sociodemographic factors, caregiver age and relationship, level of global cognitive impairment, behavioral and psychiatric disturbances, mobility and cardiovascular factors, and instrumental and basic activities of daily living were entered into the model. Older age (hazards ratio [HR] = 1.05), living alone (HR = 2.75), motor (HR = 1.16), cognitive (HR = 0.92), and neuropsychiatric impairments (HR = 1.14) increased the probability of NHP/ALP. Racial (HR = 0.27), ethnic minority identification (HR = 0.40), and greater cardiovascular risk (HR = 0.94) reduced this likelihood. The final model demonstrated excellent predictive accuracy at 3-year postbaseline diagnosis (0.74–0.85). Our algorithm and web-based application may help in conceptualizing long-term patient needs by approximating time to care-facility admissions and viewing dynamic predictive probability plots at distinct time points. Further research with competing risk approaches, diverse samples, neuropsychological, and disease-related characteristics would enhance this model’s prediction sensitivity. (PsycInfo Database Record (c) 2025 APA, all rights reserved)