Six‐Month Prognostic Tool for Community‐Dwelling People With Dementia

People with dementia (PWD) with less than a 6-month prognosis stand to benefit from discussions about goals of care, medications, and deprescribing, and hospice [1]. Hospice eligibility requires a 6-month prognosis, but existing 6-month prognostic models are not accurate [2] nor easily replicable [3] for community-dwelling PWD. The “Deardorff 1–10 year Mortality in Dementia Index” has been shown to be effective at assessing mortality at 1, 2, 5, and 10 years among community-dwelling PWD [4]. As indicators of its use in clinical practice, the original manuscript has been cited 36 times between 2022 and 2025; in addition, monthly use of the Index in ePrognosis (https://eprognosis.ucsf.edu/dementia.php) ranges from 8000 and 11,000 people in 2025. We tested its performance for a 6-month prognosis. As previously reported, we used the Health and Retirement Study (HRS) 1998–2016 (internal validation) and the National Health and Aging Trends Study (NHATS) 2011–2019 (external validation). We also replicated the analyses with newer data: HRS 2018–2022 and NHATS 2020–2023 (temporal validation). Participants were age 65+, community-dwelling, with a 50%+ probability of dementia per the Wu-Glymour algorithm [5]. Predictors were assessed at the time of dementia identification: age, sex, body mass index, smoking status, count of activities of daily living dependencies (1–5), count of instrumental activities of daily living difficulties (1–5), difficulty walking several blocks, participation in vigorous physical activity, and presence of cancer, heart disease, diabetes, or lung disease. The primary outcome was death within 6 months. Missing variables were multiply imputed (m = 20). We assessed model performance for 6-month prognosis by applying the previously developed Deardorff index to obtain predicted risks and evaluated discrimination (time-specific area under the receiver operating characteristic curve [AUC]), accounting for survey weights, calibration (plots of predicted and observed mortality), and calculation of sensitivity, specificity, positive and negative predictive value for various 6-month mortality risk thresholds. We re-estimated coefficients and tested the model in a subpopulation with greater disability, but performance was not superior. Our reporting was guided by the Transparent Reporting of a multivariable prediction model for Individual Prognosis or Diagnosis (TRIPOD) guidelines [6]. The study was approved by the UCSF Human Research Protection Program. Of 4267 PWD in HRS (internal validation), the mean (SD) age was 82.2 (7.6) years, 2930 (69.4%) were female, and 785 (survey-weighted 12.1%) identified as Black. Median (IQR) follow-up time was 3.9 (2.0–6.8) years. Of these, 239 died within 6 months (survey-weighted 5.3%). The time-specific AUC at 6 months was 0.76 (95% confidence interval [CI], 0.74–0.77). Of 2404 PWD in NHATS (external validation), 176 individuals died within 6 months (survey-weighted 7.1%); 25% died within 43 days and 75% died within 138 days. The time-specific AUC at 6 months was 0.71 (95% CI, 0.67–0.75). Calibration plots demonstrated strong alignment between observed and predicted risk across deciles of predicted risk (Figure 1). Above a 5% risk threshold, specificity and negative predictive value were high, while sensitivity and positive predictive value were low (Table 1). Temporal validation in newer data identified an additional n = 498 newly diagnosed PWD in HRS, 31 of whom died within 6 months; and n = 412 newly diagnosed PWD in NHATS, 17 of whom died within 6 months. Time-specific AUCs were similar (in HRS 0.70 (95% CI, 0.61–0.79); in NHATS 0.77 (95% CI, 0.64–0.89)). The Deardorff 1–10 year Mortality in Dementia Index performed well for 6-month outcomes via traditional measures of model performance and similarly to the 1-year Deardorff index (AUC statistics 0.73). Yet, 6-month mortality was rare (~5%) in this sample of community-dwelling PWD, and accordingly, the index had high specificity, low sensitivity, and low positive predictive value. Limitations include the potential misclassification of dementia despite the use of a validated algorithm, a lack of information on dementia etiology or severity, and a low mortality rate at 6 months (which affects estimates of positive and negative predictive values but not sensitivity or specificity). In clinical practice or health systems, the index could be used as a screening tool to identify community-dwelling adults with dementia for palliative interventions with a high likelihood of benefit and a low risk of harm, including conversations about goals, opportunities for deprescribing, and planning for future care. Sun Y. Jeon had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis. Study concept and design: Krista L. Harrison, W. James Deardorff, Kenneth E. Covinsky, Alexander K. Smith, Sei J. Lee. Acquisition of data: Alexander K. Smith, Sei J. Lee, Kenneth E. Covinsky. Analysis and interpretation of data: All. Drafting of the manuscript: Krista L. Harrison. Critical revision of the manuscript for important intellectual content: All. Statistical analysis: Sun Y. Jeon and Irena Cenzer. Obtained funding: Alexander K. Smith, Sei J. Lee, Kenneth E. Covinsky, Krista L. Harrison, and Melissa D. Aldridge. Administrative, technical, or material support: Alexander K. Smith, Sei J. Lee, W. James Deardorff, and Irena Cenzer. Study supervision: Alexander K. Smith and Sei J. Lee. The funders had no role in the design and conduct of the study; in the collection, analysis, and interpretation of the data; or in the preparation, review, or approval of the manuscript. The authors declare no conflicts of interest. Our study was performed using participants enrolled in the Health and Retirement Study (HRS) and National Health and Aging Trends Study (NHATS). HRS data products are available publicly at https://hrs.isr.umich.edu/. NHATS public use files are available at https://www.nhats.org/researcher/data-access.