Development and External Validation of a Mortality Prediction Model for Community-Dwelling Older Adults With Dementia

W James Deardorff; Deborah E Barnes; Sun Y Jeon; W John Boscardin; Kenneth M Langa; Kenneth E Covinsky; Susan L Mitchell; Elizabeth L Whitlock; Alexander K Smith; Sei J Lee

doi:10.1001/jamainternmed.2022.4326

. 2022 Sep 26;182(11):1161–1170. doi: 10.1001/jamainternmed.2022.4326

Development and External Validation of a Mortality Prediction Model for Community-Dwelling Older Adults With Dementia

W James Deardorff ^1,^2,^✉, Deborah E Barnes ^3,⁴, Sun Y Jeon ^1,², W John Boscardin ^1,^2,⁴, Kenneth M Langa ^5,^6,^7,⁸, Kenneth E Covinsky ^1,¹², Susan L Mitchell ^9,¹⁰, Elizabeth L Whitlock ¹¹, Alexander K Smith ^1,², Sei J Lee ^1,²

¹Division of Geriatrics, University of California, San Francisco

²Geriatrics, Palliative and Extended Care Service Line, San Francisco Veterans Affairs Health Care System, San Francisco

³Department of Psychiatry and Behavioral Sciences, University of California, San Francisco

⁴Department of Epidemiology and Biostatistics, University of California, San Francisco

⁵Department of Internal Medicine, School of Medicine, University of Michigan, Ann Arbor

⁶Institute for Healthcare Policy and Innovation, University of Michigan, Ann Arbor

⁷Veterans Affairs Ann Arbor Center for Clinical Management Research, Ann Arbor, Michigan

⁸Institute for Social Research, University of Michigan, Ann Arbor

⁹Hinda and Arthur Marcus Institute for Aging Research, Hebrew SeniorLife, Boston, Massachusetts

¹⁰Department of Medicine, Beth Israel Deaconess Medical Center, Boston, Massachusetts

¹¹Department of Anesthesia and Perioperative Care, University of California, San Francisco

¹²Associate Editor, JAMA Internal Medicine

Accepted for Publication: August 6, 2022.

Published Online: September 26, 2022. doi:10.1001/jamainternmed.2022.4326

^✉

Corresponding Author: William James Deardorff, MD, 490 Illinois St, Floor 08, San Francisco, CA 94158 (william.deardorff@ucsf.edu).

Author Contributions: Drs Deardorff, Lee, and Smith had full access to all of the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis. Drs Smith and Lee contributed equally as co-senior authors.

Concept and design: Deardorff, Barnes, Smith, Lee.

Acquisition, analysis, or interpretation of data: All authors.

Drafting of the manuscript: Deardorff.

Critical revision of the manuscript for important intellectual content: All authors.

Statistical analysis: Deardorff, Boscardin.

Obtained funding: Smith, Lee.

Administrative, technical, or material support: Covinsky, Lee.

Supervision: Boscardin, Mitchell, Smith, Lee.

Conflict of Interest Disclosures: Dr Deardorff reported grants from National Institute on Aging T32-AG000212 during the conduct of the study. Dr Langa reported personal fees from the University of California, San Francisco (UCSF) as a consultant for this project during the conduct of the study; grants from the National Institutes of Health (NIH)/National Institute on Aging (NIA) related to epidemiology of dementia; and grants from Alzheimer's Association outside the submitted work; and served as an expert witness for a legal case related to cognitive decline and decisional capacity. Dr Covinsky reported grants from NIA during the conduct of the study. Dr Whitlock reported grants from National Center for Advancing Translational Sciences (KL2TR001870) during the conduct of the study; grants from NIA (R03AG059822), grants from Foundation for Anesthesia Education and Research, grants from UCSF Department of Anesthesia & Perioperative Care, and nonfinancial support from NIA (P30AG044281) outside the submitted work. Dr Lee reported grants from NIA (R01AG057751) during the conduct of the study and grants from VA HSR&D IIR 15-434 outside the submitted work. No other disclosures were reported.

Funding/Support: This work was supported by NIA (T32-AG000212 to Dr Deardorff, R01AG057751 to Drs Smith and Lee, K24AG068312 to Dr Smith, K24AG066998 to Dr Lee, P01AG066605 to Drs Covinsky, Smith, Lee, Jeon, and Boscardin, and P30AG044281 to Drs Covinsky, Smith, Lee, Jeon, and Boscardin) and the National Center for Advancing Translational Sciences (KL2TR001870 to Dr Whitlock). The Health and Retirement Study is funded by NIA grant (U01AG009740) and performed at the Institute for Social Research, University of Michigan. The National Health and Aging Trends Study was produced and distributed by https://www.nhats.org/ with funding from the NIA (grant number U01AG32947).

Role of the Funder/Sponsor: The funding agencies had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.

Disclaimer: Dr Covinsky is Associate Editor of JAMA Internal Medicine, but he was not involved in any of the decisions regarding review of the manuscript or its acceptance.

^✉

Corresponding author.

PMCID: PMC9513707 NIHMSID: NIHMS1840633 PMID: 36156062

Key Points

Question

What is the accuracy of a novel model for predicting time to death in community-dwelling older adults with dementia?

Findings

We used 2 nationally representative cohorts including a total of 6671 community-dwelling older adults with dementia from 1998 to 2016 (n = 4267) and 2011 to 2019 (n = 2404) to develop and externally validate a mortality prediction model. The final model included readily available clinical predictors (demographics, health factors, functional measures, and chronic conditions) to predict mortality from 1 to 10 years with good discrimination and calibration.

Meaning

Mortality predictions using this prognostic model may help inform conversations between clinicians, patients, and families related to advance care planning and clinical decisions such as cancer screening.

This cohort study examines the accuracy of a novel model for predicting time to death in community-dwelling older adults with dementia.

Abstract

Importance

Estimating mortality risk in older adults with dementia is important for guiding decisions such as cancer screening, treatment of new and chronic medical conditions, and advance care planning.

Objective

To develop and externally validate a mortality prediction model in community-dwelling older adults with dementia.

Design, Setting, and Participants

This cohort study included community-dwelling participants (aged ≥65 years) in the Health and Retirement Study (HRS) from 1998 to 2016 (derivation cohort) and National Health and Aging Trends Study (NHATS) from 2011 to 2019 (validation cohort).

Exposures

Candidate predictors included demographics, behavioral/health factors, functional measures (eg, activities of daily living [ADL] and instrumental activities of daily living [IADL]), and chronic conditions.

Main Outcomes and Measures

The primary outcome was time to all-cause death. We used Cox proportional hazards regression with backward selection and multiple imputation for model development. Model performance was assessed by discrimination (integrated area under the receiver operating characteristic curve [iAUC]) and calibration (plots of predicted and observed mortality).

Results

Of 4267 participants with probable dementia in HRS, the mean (SD) age was 82.2 (7.6) years, 2930 (survey-weighted 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, and 3466 (81.2%) participants died by end of follow-up. The final model included age, sex, body mass index, smoking status, ADL dependency count, IADL difficulty count, difficulty walking several blocks, participation in vigorous physical activity, and chronic conditions (cancer, heart disease, diabetes, lung disease). The optimism-corrected iAUC after bootstrap internal validation was 0.76 (95% CI, 0.75-0.76) with time-specific AUC of 0.73 (95% CI, 0.70-0.75) at 1 year, 0.75 (95% CI, 0.73-0.77) at 5 years, and 0.84 (95% CI, 0.82-0.85) at 10 years. On external validation in NHATS (n = 2404), AUC was 0.73 (95% CI, 0.70-0.76) at 1 year and 0.74 (95% CI, 0.71-0.76) at 5 years. Calibration plots suggested good calibration across the range of predicted risk from 1 to 10 years.

Conclusions and Relevance

We developed and externally validated a mortality prediction model in community-dwelling older adults with dementia that showed good discrimination and calibration. The mortality risk estimates may help guide discussions regarding treatment decisions and advance care planning.

Introduction

Dementia is a leading cause of morbidity and mortality worldwide and is associated with increased mortality risk.^1,2,3 An estimated 6.5 million Americans aged 65 years or older are living with Alzheimer disease and related dementias in 2022.^4,5 The clinical course of individuals with dementia is highly variable, with median survival time from age at diagnosis ranging from 3.3 years to 11.7 years.⁶ Heterogeneity in prognosis has important implications for decisions related to financial planning, advance care planning, long-term care admission, and clinical care. For interventions that have immediate risks or burdens and delayed benefits, such as cancer screening and tight glycemic control in patients with diabetes, life expectancy can help differentiate between patients most likely to benefit vs most likely to be harmed by interventions.⁷ Therefore, accurate estimates of life expectancy in people with dementia are important for providing the foundational data for patients, family, and clinicians to help define patient-centered care goals in their remaining lifetime.^7,8

Mortality prediction models are one tool that can help estimate an individual’s prognosis. Several non–disease-specific mortality indices have been developed among community-dwelling older adults, and they are readily available on websites such as ePrognosis.^9,10,11,12 However, their utility may be limited if the populations in which they were developed were not entirely inclusive of individuals with conditions that have a dramatic effect on mortality (eg, dementia or advanced cancer).⁹

Several mortality prediction models have been developed specifically in individuals with dementia.¹³ The most well known is the Advanced Dementia Prognostic Tool (ADEPT), which estimates 6-month mortality in nursing home patients with advanced dementia.^14,15 However, for the approximately 70% of older adults living with dementia in the community, there are currently no well-established mortality models.⁴ This population is more heterogenous in terms of cognition, comorbidities, and daily functioning compared with a nursing home population. Previously published models of community-dwelling older adults with dementia have several limitations, including small sample sizes, outdated data, lack of validation, and poor analytic strategies.¹³

Therefore, the aim of this study was to develop and externally validate a mortality prediction model in 2 nationally representative cohorts of community-dwelling older adults with dementia in the US. Information from this model can be used to facilitate discussions regarding clinical care and advance care planning.

Methods

This study’s reporting was guided by the Transparent Reporting of a Multivariable Prediction Model for Individual Prognosis or Diagnosis (TRIPOD) reporting guidelines (eTable 1 in the Supplement).¹⁶

Study Design and Population

We used data from the Health and Retirement Study (HRS) between 1998 and 2016, a nationally representative survey of US adults aged 51 years or older in which participants were interviewed every 2 years.¹⁷ For individuals who could not participate in the interview (eg, physical and/or cognitive reasons), HRS performed proxy interviews typically with a spouse or other family member to help reduce attrition bias.¹⁸ We included information from both sample respondents and proxy interviews.

We included community-dwelling older adults aged 65 years or older with probable dementia ascertained by a previously validated algorithm (eMethods in the Supplement).¹⁹ This algorithm has high accuracy compared with gold-standard dementia diagnoses from the Aging, Demographics, and Memory Study.^19,20,21 Participants were included from the date of interview when predicted probability of dementia exceeded 0.5.

For external validation, we used the National Health and Aging Trends Study (NHATS), a nationally representative longitudinal study of Medicare beneficiaries aged 65 years or older from 2011 to 2019.²² We included individuals classified as having probable dementia based on defined cutoffs (eMethods in the Supplement).²³

Outcomes

The primary outcome was all-cause mortality. Information on mortality in HRS was collected until May 31, 2019, in multiple ways to ensure accurate and complete ascertainment. We primarily used information from the HRS Tracker File, which ascertains mortality through exit interviews with next of kin, and the National Death Index (NDI). This method has been shown to be highly accurate and effectively complete.²⁴ To further increase accuracy, we searched Medicare data files (Denominator File, Master Beneficiary Summary File, and Inpatient files) and the NDI.

Candidate Predictors

Candidate predictor variables were identified through previously published systematic reviews of risk factors for mortality in individuals with dementia and performing an independent review of mortality models for community-dwelling individuals with dementia (eMethods in the Supplement).^{6,13,25,26,27,28,29} All candidate predictors were collected at the time of the HRS interview wave when dementia probability was greater than 0.5. Detailed definitions of all candidate variables are provided in eTable 2 in the Supplement.

Briefly, predictors included demographic factors, such as age category (65-69, 70-74, 75-79, 80-84, 85-89, ≥90 years), sex, and marital status. Health and/or behavioral factors included body mass index (BMI, calculated as weight in kilograms divided by height in meters squared) category (<18.5, 18.5 to <25, 25 to <30, and ≥30), smoking status (never smoker, former smoker, current smoker), and participation in vigorous physical activity (hardly ever/never vs any vigorous physical activity). Functional status was assessed by count of difficulties or dependencies with activities of daily living (ADL, including dressing, bathing, eating, getting in/out of bed, and toileting with range 0-5), and instrumental activities of daily living (IADLs, including preparing meals, managing medications, managing money, grocery shopping, and using telephone with range 0-5). In general, difficulty and dependency with individual ADLs/IADLs were defined by self or proxy report of difficulty performing the task or needing help with the task, respectively. Self-reported comorbidities included cancer (excluding minor skin cancer), diabetes, heart disease, lung disease, hypertension, and stroke. Definitions of these predictors largely overlapped in NHATS (eTable 3 in the Supplement).

Model Development

We used Cox proportional hazards regression with all candidate predictors to develop the model. To create a parsimonious model that can more efficiently be used in clinical practice, we performed variable selection using backward selection. Because this process may lead to a final set of predictors that is unstable, we combined the backward selection process with bootstrapping and multiple imputation.^30,31,32,33 Because some variables had incomplete baseline data, we first performed multiple imputation (MI) assuming the data were missing at random to create 20 imputed data sets. We created 30 bootstrap samples for each of the 20 imputed data sets. We performed a backward selection process with a P value of <.05 for elimination in each of the imputed data sets within each bootstrap sample. Variables that were selected in more than 60% of the 20 imputed data sets (≥12 times) and more than 50% of the 30 bootstrap samples (≥15 times) were included. We explored different percentage thresholds based on previous research without any significant changes in predictors included.^30,34 To account for the complex survey design of the HRS, we incorporated survey weights and strata at each stage. The final model coefficients were obtained by pooling results across imputations. To evaluate the effect of proxy responses, we included a proxy variable as a candidate predictor and separately evaluated models in proxy and nonproxy respondents. Because the results were not substantially different with either method, we reported the pooled model without an indicator for proxy status. We also performed a sensitivity analysis only including incident cases of probable dementia.

Model Evaluation

Model performance was assessed through discrimination, calibration, and the Brier score. For discrimination, we calculated the integrated area under the receiver operating characteristic curve (iAUC), which averages all available AUC statistics over time.^35,36,37 We also reported time-specific values at clinically relevant time points of 1, 2, 5, and 10 years. Calibration refers to the agreement between observed and predicted probability of mortality. This was assessed graphically at fixed time points.³⁸ As an overall measure of model performance, we calculated the integrated Brier score, a scoring rule affected by both discrimination and calibration, with lower scores indicating better model accuracy.³⁹

We assessed internal validity via bootstrapping to quantify any optimism in model performance. We repeated the entire modeling process (including variable selection) in 30 bootstrapped samples and averaged the individual iAUC values to obtain an optimism-corrected iAUC. We externally validated the final prediction model by examining model performance in NHATS. Because the length of follow-up time in NHATS did not allow us to provide estimates at 10 years, we present time-specific AUC values and calibration plots at 1, 2, and 5 years. To present the final model, we have provided β coefficients and baseline survival at specific time points to facilitate calculation of individual predictions. We also created a web application to input individual patient characteristics to obtain mortality risk estimates. Statistical analyses were performed using SAS (version 9.4; SAS Institute, Inc), Stata (version 17.0; StataCorp, LLC), and R statistical software (version 4.05, R Project for Statistical Computing).

Results

Cohort Characteristics

Of the 43 398 participants aged 65 years or older in the HRS between 1998 and 2016, we included 4267 participants with probable dementia (eFigure 1 in the Supplement). The mean (SD) age when participants were classified as having dementia was 82.2 (7.6) years (Table 1). In the HRS cohort, 2930 (survey-weighted 69.4%) were female, and 785 (survey-weighted 12.1%) identified as Black. The amount of missing information was generally less than 1%. The median (IQR) follow-up time was 3.9 (2.0-6.8) years in the overall cohort, 3.5 (1.6-6.2) years among those who died, and 5.2 (3.0-8.6) years among those who did not die by end of follow-up. A total of 3466 participants (81.2%) died by the end of follow-up.

Table 1. Baseline Characteristics of Participants With Probable Dementia in the Health and Retirement Study 1998 to 2016 (Derivation Cohort) and the National Health and Aging Trends Study 2011 to 2019 (Validation Cohort).

Characteristic	No. (weighted %)^a
Characteristic	Health and Retirement Study (n = 4267)	National Health and Aging Trends Study (n = 2404)
Age, mean (SD), y	82.2 (7.6)	NR^b
Age category, y
65-69	286 (6.5)	108 (4.4)
70-74	436 (10.0)	235 (10.5)
75-79	771 (17.2)	430 (18.6)
80-84	1052 (24.8)	593 (25.4)
85-89	993 (23.5)	553 (21.6)
≥90	729 (18.0)	485 (19.5)
Female sex	2930 (69.4)	1432 (59.3)
Race/ethnicity
Non-Hispanic White	2846 (75.1)	1274 (45.4)
Non-Hispanic Black	785 (12.1)	715 (32.8)
Hispanic	527 (10.1)	261 (13.9)
Other	109 (2.6)	93 (5.4)
Missing	0	61 (2.5)
BMI
<18.5	308 (7.3)	135 (5.3)
18.5 to <25	1987 (47.4)	943 (39.0)
25 to <30	1262 (29.3)	706 (29.3)
≥30	592 (13.4)	447 (18.8)
Missing	118 (2.6)	173 (7.5)
Smoking status
Never	2158 (51.5)	1335 (57.2)
Former	1762 (40.6)	920 (36.4)
Current	310 (7.1)	144 (6.2)
Missing	37 (0.8)	5 (0.2)
Marital status
Married or partnered	1655 (38.8)	889 (37.3)
Single or widowed	2607 (61.7)	1513 (62.6)
Missing	5 (0.1)	2 (0.1)
Lives alone	1475 (37.7)	615 (25.2)
≥1 ADL difficulty at baseline	1976 (45.2)	1479 (62.7)
Missing	20 (0.4)	2 (0.1)
≥1 ADL dependency at baseline	1445 (32.7)	840 (35.8)
Missing	25 (0.5)	3 (0.1)
≥1 IADL difficulty at baseline	2575 (60.3)	1868 (78.3)
Missing	24 (0.6)	2 (0.1)
≥1 IADL dependency at baseline	2438 (57.2)	1655 (69.7)
Missing	30 (0.7)	2 (0.1)
Difficulty reaching above shoulder	1380 (31.7)	986 (43.3)
Difficulty with walking several blocks	2863 (66.5)	1689 (70.9)
Vigorous physical activity	611 (14.7)	354 (14.3)
Comorbidities
Cancer	772 (18.9)	369 (15.3)
Diabetes	1041 (23.6)	758 (32.5)
Heart disease	1679 (39.8)	783 (31.9)
Hypertension	2775 (64.7)	1755 (74.4)
Lung disease	528 (12.9)	498 (20.6)
Stroke	990 (23.5)	405 (17.6)

Open in a new tab

Abbreviations: ADL, activities of daily living; BMI, body mass index (calculated as weight in kilograms divided by height in meters squared); IADL, instrumental activities of daily living; NR, not reported.

^{^a}

The numbers in each column represent the raw unweighted number of individuals within each cohort with a particular characteristic. The weighted percentages in each column represent the column percent based on the weighted sample size after using national survey weights from the cohort studies.

^{^b}

Mean age for individuals in the National Health and Aging Trends Study is not reported because continuous age is not available.

The external validation cohort from NHATS included 2404 individuals with probable dementia (eFigure 2 in the Supplement). Demographics were similar to HRS, although a greater proportion of individuals had functional impairments and were Black (33% in NHATS vs. 12% in HRS) (Table 1). Kaplan-Meier survival curves are shown in eFigures 3 and 4 in the Supplement. The histogram of the prognostic index showed decreased case-mix heterogeneity in NHATS (eFigure 5 in the Supplement). By the end of follow-up, 1426 participants (59.3%) had died.

Model Development

eTable 4 in the Supplement shows the unadjusted hazard ratios for all candidate predictors. The final variables following variable selection were age category, sex, BMI category, smoking status, number of ADL dependencies, number of IADL difficulties, difficulty walking several blocks, participation in vigorous physical activity, and chronic conditions including diabetes, heart disease, cancer, and lung disease. Table 2 displays the multivariable-adjusted hazard ratios and β coefficients for variables in the model.

Table 2. Multivariable-Adjusted Hazard Ratios for Variables Included in the Final Mortality Prediction Model for Community-Dwelling Older Adults With Dementia^a.

Variable	Adjusted hazard ratio (95% CI)	β coefficient (95% CI)	P value
Age category, y
65-69	1 [Reference]	0 [Reference]
70-74	1.37 (1.09 to 1.72)	0.32 (0.09 to 0.54)	.007
75-79	1.94 (1.58 to 2.38)	0.66 (0.46 to 0.87)	<.001
80-84	2.65 (2.14 to 3.28)	0.97 (0.76 to 1.19)	<.001
85-89	3.32 (2.71 to 4.07)	1.20 (1.00 to 1.40)	<.001
≥90	4.34 (3.46 to 5.45)	1.47 (1.24 to 1.70)	<.001
Female sex	0.71 (0.64 to 0.79)	−0.34 (−0.44 to −0.24)	<.001
BMI category
<18.5	1.56 (1.31 to 1.85)	0.44 (0.27 to 0.62)	<.001
18.5 to <25	1 [Reference]	0 [Reference]
25 to <30	0.85 (0.78 to 0.92)	−0.16 (−0.24 to −0.08)	<.001
≥30	0.75 (0.67 to 0.85)	−0.28 (−0.41 to −0.16)	<.001
Smoking status
Never	1 [Reference]	0 [Reference]
Former	1.18 (1.08 to 1.29)	0.17 (0.08 to 0.26)	.001
Current	1.37 (1.15 to 1.64)	0.32 (0.14 to 0.50)	.001
ADL dependency count (range 0-5)^b	1.15 (1.10 to 1.20)	0.14 (0.10 to 0.18)	<.001
IADL difficulty count (range 0-5)^b	1.11 (1.08 to 1.14)	0.10 (0.07 to 0.13)	<.001
Difficulty walking several blocks	1.09 (0.99 to 1.20)	0.09 (−0.01 to 0.18)	.083
Vigorous physical activity^c	0.87 (0.80 to 0.96)	−0.13 (−0.23 to −0.04)	.005
Diabetes	1.24 (1.11 to 1.39)	0.22 (0.11 to 0.33)	<.001
Heart disease	1.24 (1.15 to 1.33)	0.22 (0.14 to 0.29)	<.001
Non-skin cancer	1.24 (1.13 to 1.36)	0.21 (0.12 to 0.30)	<.001
Lung disease	1.35 (1.21 to 1.51)	0.30 (0.19 to 0.41)	<.001

Open in a new tab

^{^a}

The baseline survival probability S_o(t) equals: 0.9716 at time = 1 year, 0.9338 at time = 2 years, 0.8001 at time = 5 years, and 0.5369 at time = 10 years.

^{^b}

The corresponding hazard ratio indicates the increase in hazard of death for each additional ADL dependency and IADL difficulty (range 0-5). ADL dependencies include needing help with dressing, bathing, eating, getting in or out of bed, and toileting. IADL difficulties include difficulty using a telephone, preparing a meal, managing medications, managing money such as paying bills, and shopping for groceries.

^{^c}

This variable involves taking part in sports or activities that are vigorous, such as running or jogging, swimming, cycling, aerobics or gym workout, tennis, or digging with a spade or shovel. Individuals were characterized as either participating in vigorous physical activity vs hardly ever or never performing vigorous physical activity.

Model Evaluation

The iAUC prior to internal validation was 0.76 (95% CI, 0.75-0.76) (Table 3). After bootstrap internal validation, the optimism-corrected iAUC was 0.76 (95% CI, 0.75-0.76), suggesting good discrimination. Overall, AUC values tended to increase over time (eFigure 6 in the Supplement). For example, the optimism-corrected, time-specific AUC at 1 year was 0.73 (95% CI, 0.70-0.75), 0.75 (95% CI, 0.73-0.77) at 5 years, and 0.84 (95% CI, 0.82-0.85) at 10 years. Figure 1 displays the predicted vs observed risk of death at the 1-, 2-, 5-, and 10-year time points across deciles of predicted risk. This showed that the model was well calibrated with good agreement between observed and predicted risk. The integrated Brier score averaged from 20 imputations was 0.10. Results from the sensitivity analysis only including incident probable dementia cases showed similar model coefficients and performance (eResults 1, eTables 5 and 6, and eFigure 7 in the Supplement).

Table 3. Performance of the Mortality Prediction Model in the Health and Retirement Study (Derivation Cohort) and the National Health and Aging Trends Study (External Validation Cohort)^a.

Timeframe	No. remaining at risk in HRS (n = 4267)^b	Apparent AUC in HRS (95% CI)	AUC in HRS after internal validation (95% CI)	No. remaining at risk in NHATS (n = 2404)^b	AUC in NHATS (95% CI)
iAUC	NR	0.76 (0.75-0.76)	0.76 (0.75-0.76)	NR	NR^c
1 y	3753	0.73 (0.70-0.75)	0.73 (0.70-0.75)	2006	0.73 (0.70-0.76)
2 y	3192	0.73 (0.72-0.75)	0.73 (0.71-0.75)	1589	0.72 (0.70-0.75)
5 y	1651	0.75 (0.74-0.77)	0.75 (0.73-0.77)	643	0.74 (0.71-0.76)
10 y	464	0.84 (0.82-0.86)	0.84 (0.82-0.85)	NR	NR^d

Open in a new tab

Abbreviations: AUC, area under the receiver operating characteristic curve; HRS, Health and Retirement Study; iAUC, integrated AUC; NHATS, National Health and Aging Trends Study; NR, not reported.

^{^a}

The time-specific AUC values represent a measure of discrimination up until time point t (ie, the ability of the linear predictor from the mortality model to separate the cumulative cases through time t from the controls at time t). Therefore, the value for the time-specific AUC at 10 years includes all the cumulative information available from baseline up until 10 years.

^{^b}

The number remaining at risk indicates the number of individuals remaining in the study cohorts at specific time points as individuals either die (outcome event) or are censored. See eFigures 3 and 4 in the Supplement for Kaplan-Meier curves along with the number remaining at risk.

^{^c}

We report time-specific AUC for the NHATS external validation rather than overall iAUC because several participants only had information on month of death rather than exact date of death in NHATS. Additionally, only month of interview was known in NHATS. This led to inaccurate estimates of model discrimination for very short-term mortality risk (due to short survival times in multiples of 30 or 31 days compared with HRS, where exact date of death was known), which was not the primary focus of our mortality model.

^{^d}

The AUC at 10 years is not reported for the NHATS external validation because the time period of observation was not long enough to obtain estimates at 10 years.

Figure 1. — The calibration plots at the (A) 1-, (B) 2-, (C) 5-, and (D) 10-year time points indicating the agreement the agreement between the predicted mortality risk using the prediction model and observed mortality at various time points. Individual dots represent deciles of predicted risk. Perfect predictions would be at the 45-degree line. Each plot also contains a density plot displaying the distribution of predicted risk in the individual cohort. Data from the HRS are presented as blue dots and blue shaded density plots, and data from the NHATS are presented as orange dots and orange shaded density plots. Calibration in NHATS was not assessed at 10 years because the duration of follow-up was not long enough.

On external validation in NHATS, the calibration slope was 0.92 (95% CI, 0.84-1.00). The time-specific AUC values were 0.73 (95% CI, 0.70-0.76) at 1 year, 0.72 (95% CI, 0.70-0.75) at 2 years, and 0.74 (95% CI, 0.71-0.76) at 5 years (Table 3). Calibration was similar to HRS (Figure 1).

Users can input individual patient characteristics on ePrognosis (https://eprognosis.ucsf.edu/dementia.php) to obtain mortality risk estimates (eFigures 8 and 9 in the Supplement). eResults 2 in the Supplement shows how to manually compute mortality risk. For illustration, we selected 10 participants at random in each decile of predicted risk and display their baseline characteristics and median predicted time to death (Figure 2). For example, patient G was a woman in her 80s with a BMI of 19.5, never smoker, 3 ADL dependencies, 4 IADL difficulties, and walking difficulty. Her median (IQR) time to death was 3.8 (1.9-6.3) years with predicted mortality risk of 12% at 1 year, 64% at 5 years, and 94% at 10 years.

Figure 2. — A, Baseline characteristics are displayed for 10 randomly selected individuals from the Health and Retirement Study in each decile of predicted risk. The color and shading indicate the direction and magnitude of the specific characteristic’s effect on mortality risk, respectively. For example, going from light to dark pink indicates increasing hazard of mortality, whereas gray indicates lower hazard of mortality. B, The median predicted time to death in years based on the individual’s baseline characteristics is displayed. The ends of the whiskers indicate the 25th and 75th percentile of the median predicted time to death. Although age has a strong prognostic effect, individuals may have high predicted mortality risk due to multiple other factors including functional impairments and comorbidities. ADL Indicates activities of daily living; BMI, body mass index (calculated as weight in kilograms divided by height in meters squared); IADL, instrumental activities of daily living.

Discussion

We developed and externally validated a mortality prediction model in community-dwelling older adults with dementia that can be used to estimate mortality risk across a time frame of 1 to 10 years. We included variables that are readily obtainable in clinical practice, including demographics, functional status, health and/or behavioral factors, and comorbidities. Measures of model performance, including discrimination and calibration, were good and generally exceeded the discrimination in previous dementia mortality prediction models.

Our model provides accurate prognostic estimates for a heterogenous population of community-dwelling older adults with dementia that may be used to enhance conversations with patients and families and inform decision-making. Several clinical guidelines related to cancer screening and management of chronic diseases now encourage the use of life expectancy when making treatment decisions.^40,41,42,43 Given the broad range of mortality risk in this cohort, providing these estimates in a clinical context can help frame screening and treatment decisions by targeting interventions to those most likely to benefit.⁴⁴ Similarly, they can help families prepare for the future in terms of advance care planning and financial planning. Nonetheless, we recognize that the mortality estimates from our model are only one of many factors that go into decision-making for older adults with dementia and should be incorporated within the broader context of patient and family goals and preferences.

The final predictors in our mortality model represent variables commonly used in prediction models involving community-dwelling older adults and those previously identified as predictive of mortality among individuals with dementia.^6,9,25,26 In line with previous studies, older age had a strong effect on mortality, with female sex being a protective factor.^13,25 A BMI lower than 18.5 compared with 18.5 to 25.0 was associated with increased mortality risk, which is consistent with previous studies.^10,45,46,47 We included several functional measures (ADL dependencies, difficulties with cognitively demanding IADLs, difficulty walking several blocks, and participation in vigorous physical activity), which have been consistently shown to influence prognostic estimates in mortality indices, likely because they represent the end effect of chronic diseases.^48,49

Given that this was a community-dwelling cohort covering a wide range of dementia severity, our focus was to estimate prognosis over years rather than months. Therefore, we did not include predictors such as pressure ulcers or time spent in bed that are often included in models assessing hospice eligibility in those with severe dementia. The final predictors in our model suggest that prognosis in a community-dwelling cohort is often driven by comorbid disease.

A 2021 systematic review and our review of mortality models in persons with dementia identified many models at high risk of bias (ROB) owing to methodologic limitations, including small sample sizes, lack of internal and/or external validation, and not reporting model calibration (eMethods and eTables 7-9 in the Supplement).¹³ A few mortality models are worth noting in comparison to our model. The model by Haaksma et al⁵⁰ developed in Sweden presents 3-year survival probabilities from a simplified model that includes age, sex, Mini-Mental State Exam score, Charlson Comorbidity Index (CCI), and dementia type. The model by van de Vorst et al⁵¹ developed in the Netherlands presented 1-year mortality risk using age, sex, setting of care, and modified CCI. The concordance statistics ranged from 0.69 to 0.72, which is somewhat lower than the findings in our study. Advantages of our model include longer follow-up duration, greater applicability for individuals in the US, and inclusion of functional measures that might explain improved model performance.

Notable models in US populations include Newcomer et al (2003), Armstrong et al (2022), and several longitudinal Grade of Membership (L-GoM) models.^{52,53,54,55,56,57,58} Newcomer et al⁵² included participants enrolled in the Medicare Alzheimer Disease Demonstration Evaluation. The primary limitations are that the data came from 1989 to 1994, and the article did not easily allow calculations of individual mortality risk. Armstrong et al⁵⁸ used data from the National Alzheimer Coordinator Center (NACC) and reported 5-year survival probabilities for individuals with different types of dementia. A possible limitation is the generalizability of the NACC population. The L-GoM models offer a flexible approach to modeling future trajectories across multiple end points, and the most recent update demonstrated accurate mortality predictions based on a single comprehensive visit.⁵³ Disadvantages include unclear generalizability (developed using approximately 250 participants in the predictors 1 [1989-2001] and predictors 2 [1997-2007] cohorts at 3 academic research centers) and a large number of variables (79 clinical signs/symptoms).⁵⁹

Strengths and Limitations

This study has several strengths. We included a large cohort of community-dwelling individuals with dementia to predict mortality with good discrimination and calibration up to 10 years. The final variables in the model are readily obtainable and can be input into an online calculator to obtain prognostic estimates. We performed an external validation in NHATS and found similar discrimination and calibration.

This study has a few limitations. First, participants were classified as having dementia using a validated algorithm that may be subject to misclassification. Although this algorithm has shown high accuracy in validation studies (eg, AUC of 0.93 in training data and 0.84 in validation data), the overall accuracy is reduced in certain subgroups, such as racial and ethnic minorities and less-educated individuals.^19,20,60 Second, information on dementia etiology (eg, Alzheimer disease or vascular disease) or dementia severity was not available. We also chose not to include cognitive tests as candidate predictors, instead relying on cognitively demanding IADLs as a proxy of cognitive status. We did this because cognitive tests were only available for self- (nonproxy) responses, and it is often challenging to guide patients with dementia through cognitive testing in routine clinical settings. The fact that our model performance was an improvement on previous models that included measures assessing cause and severity of dementia suggests that our model can still provide predictions that inform decision-making without this data. Third, the follow-up duration in NHATS precluded external validation at 10 years. Given that model performance after internal and external validation was relatively similar at 5 years, it is likely that our internal validation results from HRS at 10 years will remain accurate. However, future validation studies in other cohorts with longer follow-up times should be performed.

Conclusions

In this cohort study, we developed and externally validated a mortality model with good discrimination and calibration that can be used to estimate mortality risk up to 10 years in community-dwelling older adults with dementia. Final predictors in the model are readily obtainable in clinical practice, including demographics, behavioral/health factors, functional status, and comorbidities. Estimates can be used by clinicians, patients, and families to help aid in discussions around clinical management of medical comorbidities, preventive screening, and advance care planning.

Supplement.

eMethods. Additional details on the methods

eResults 1. Sensitivity analysis accounting for incident vs. prevalent dementia cases

eResults 2. Example calculation using the final model

eFigure 1. Flow chart of participants from the Health and Retirement Study 1998-2016

eFigure 2. Flow chart of participants from the National Health and Aging Trends Study 2011-2019

eFigure 3. Kaplan-Meier curve displaying the survival probability over time in the Health and Retirement Study cohort (N=4,267)

eFigure 4. Kaplan-Meier curve displaying the survival probability over time in the National Health and Aging Trends Study cohort (N=2,404)

eFigure 5. Histogram of the prognostic index in the Health and Retirement Study (derivation cohort) and National Health and Aging Trends Study (validation cohort)

eFigure 6. Graph of the area under the receiver operating characteristic curve over time with 95% confidence limits in the Health and Retirement Study for the final mortality prediction model following bootstrap internal validation

eFigure 7. Calibration plots and time-specific AUC values at the 1, 2, 5, and 10-year time points in the model restricted to those with incident probable dementia in the Health and Retirement Study from the sensitivity analysis

eFigure 8. Representation of how the dementia mortality risk calculator appears on ePrognosis which allows users to input specific patient characteristics and obtain mortality risk estimates

eFigure 9. Representative output from the dementia mortality risk calculator on ePrognosis

eTable 1. Transparent reporting of a multivariable prediction model for individual prognosis or diagnosis (TRIPOD) checklist for prediction model development and validation

eTable 2. Full description of predictors in the Health and Retirement Study (HRS), including categorization scheme

eTable 3. Full description of predictors in the National Health and Aging Trends Study (NHATS) that were included in external validation

eTable 4. Unadjusted hazard ratios of all the candidate predictors and adjusted hazard ratios of the selected variables in the final model

eTable 5. Number of individuals with dementia by interview wave at which they were first classified as probable dementia and whether it was considered an incident or prevalent case

eTable 6. Multivariable-adjusted hazard ratios for variables included in the Cox regression model for mortality prediction when restricted to those with incident probable dementia (n = 3,665) in the sensitivity analysis

eTable 7. Characteristics of studies reporting the development and/or validation of a mortality prediction model in predominantly community-dwelling individuals with dementia

eTable 8. Characteristics of model performance from studies reporting the development and/or validation of a mortality prediction model in predominantly community-dwelling individuals with dementia

eTable 9. Risk of bias assessment for mortality prediction models among predominantly community-dwelling individuals with dementia using the Prediction Model Risk of Bias Assessment Tool (PROBAST)

eReferences.

Click here for additional data file.^{(985.3KB, pdf)}

References

1.Liang CS, Li DJ, Yang FC, et al. Mortality rates in Alzheimer’s disease and non-Alzheimer’s dementias: a systematic review and meta-analysis. Lancet Healthy Longev. 2021;2(8):e479-e488. doi: 10.1016/S2666-7568(21)00140-9 [DOI] [PubMed] [Google Scholar]
2.Ganguli M, Dodge HH, Shen C, Pandav RS, DeKosky ST. Alzheimer disease and mortality: a 15-year epidemiological study. Arch Neurol. 2005;62(5):779-784. doi: 10.1001/archneur.62.5.779 [DOI] [PubMed] [Google Scholar]
3.Larson EB, Shadlen MF, Wang L, et al. Survival after initial diagnosis of Alzheimer disease. Ann Intern Med. 2004;140(7):501-509. doi: 10.7326/0003-4819-140-7-200404060-00008 [DOI] [PubMed] [Google Scholar]
4.2022 Alzheimer’s disease facts and figures. Alzheimers Dement. 2022;18(4):700-789. doi: 10.1002/alz.12638 [DOI] [PubMed] [Google Scholar]
5.Hebert LE, Weuve J, Scherr PA, Evans DA. Alzheimer disease in the United States (2010-2050) estimated using the 2010 census. Neurology. 2013;80(19):1778-1783. doi: 10.1212/WNL.0b013e31828726f5 [DOI] [PMC free article] [PubMed] [Google Scholar]
6.Todd S, Barr S, Roberts M, Passmore AP. Survival in dementia and predictors of mortality: a review. Int J Geriatr Psychiatry. 2013;28(11):1109-1124. doi: 10.1002/gps.3946 [DOI] [PubMed] [Google Scholar]
7.Lee SJ, Leipzig RM, Walter LC. Incorporating lag time to benefit into prevention decisions for older adults. JAMA. 2013;310(24):2609-2610. doi: 10.1001/jama.2013.282612 [DOI] [PMC free article] [PubMed] [Google Scholar]
8.Lee SJ, Kim CM. Individualizing prevention for older adults. J Am Geriatr Soc. 2018;66(2):229-234. doi: 10.1111/jgs.15216 [DOI] [PMC free article] [PubMed] [Google Scholar]
9.Yourman LC, Lee SJ, Schonberg MA, Widera EW, Smith AK. Prognostic indices for older adults: a systematic review. JAMA. 2012;307(2):182-192. doi: 10.1001/jama.2011.1966 [DOI] [PMC free article] [PubMed] [Google Scholar]
10.Lee SJ, Lindquist K, Segal MR, Covinsky KE. Development and validation of a prognostic index for 4-year mortality in older adults. JAMA. 2006;295(7):801-808. doi: 10.1001/jama.295.7.801 [DOI] [PubMed] [Google Scholar]
11.Schonberg MA, Davis RB, McCarthy EP, Marcantonio ER. Index to predict 5-year mortality of community-dwelling adults aged 65 and older using data from the National Health Interview Survey. J Gen Intern Med. 2009;24(10):1115-1122. doi: 10.1007/s11606-009-1073-y [DOI] [PMC free article] [PubMed] [Google Scholar]
12.ePrognosis. Accessed April 5, 2022. https://eprognosis.ucsf.edu/
13.Smith EE, Ismail Z. Mortality risk models for persons with dementia: a systematic review. J Alzheimers Dis. 2021;80(1):103-111. doi: 10.3233/JAD-201364 [DOI] [PMC free article] [PubMed] [Google Scholar]
14.Mitchell SL, Miller SC, Teno JM, Kiely DK, Davis RB, Shaffer ML. Prediction of 6-month survival of nursing home residents with advanced dementia using ADEPT vs hospice eligibility guidelines. JAMA. 2010;304(17):1929-1935. doi: 10.1001/jama.2010.1572 [DOI] [PMC free article] [PubMed] [Google Scholar]
15.Mitchell SL, Miller SC, Teno JM, Davis RB, Shaffer ML. The advanced dementia prognostic tool: a risk score to estimate survival in nursing home residents with advanced dementia. J Pain Symptom Manage. 2010;40(5):639-651. doi: 10.1016/j.jpainsymman.2010.02.014 [DOI] [PMC free article] [PubMed] [Google Scholar]
16.Moons KGM, Altman DG, Reitsma JB, et al. Transparent Reporting of a multivariable prediction model for Individual Prognosis or Diagnosis (TRIPOD): explanation and elaboration. Ann Intern Med. 2015;162(1):W1-73. doi: 10.7326/M14-0698 [DOI] [PubMed] [Google Scholar]
17.Fisher GG, Ryan LH. Overview of the Health and Retirement Study and introduction to the special issue. Work Aging Retire. 2018;4(1):1-9. doi: 10.1093/workar/wax032 [DOI] [PMC free article] [PubMed] [Google Scholar]
18.Weir D, Faul J, Langa K. Proxy interviews and bias in the distribution of cognitive abilities due to non-response in longitudinal studies: a comparison of HRS and ELSA. Longit Life Course Stud. 2011;2(2):170-184. doi: 10.14301/llcs.v2i2.116 [DOI] [PMC free article] [PubMed] [Google Scholar]
19.Wu Q, Tchetgen Tchetgen EJ, Osypuk TL, White K, Mujahid M, Maria Glymour M. Combining direct and proxy assessments to reduce attrition bias in a longitudinal study. Alzheimer Dis Assoc Disord. 2013;27(3):207-212. doi: 10.1097/WAD.0b013e31826cfe90 [DOI] [PMC free article] [PubMed] [Google Scholar]
20.Gianattasio KZ, Wu Q, Glymour MM, Power MC. Comparison of methods for algorithmic classification of dementia status in the Health and Retirement Study. Epidemiology. 2019;30(2):291-302. doi: 10.1097/EDE.0000000000000945 [DOI] [PMC free article] [PubMed] [Google Scholar]
21.Langa KM, Plassman BL, Wallace RB, et al. The Aging, Demographics, and Memory Study: study design and methods. Neuroepidemiology. 2005;25(4):181-191. doi: 10.1159/000087448 [DOI] [PubMed] [Google Scholar]
22.Freedman VA, Kasper JD. Cohort Profile: The National Health and Aging Trends Study (NHATS). Int J Epidemiol. 2019;48(4):1044-1045g. doi: 10.1093/ije/dyz109 [DOI] [PMC free article] [PubMed] [Google Scholar]
23.Kasper J, Freedman V, Spillman B. Classification of Persons by Dementia Status in the National Health and Aging Trends Study. Technical Paper #5. Published online 2013. Accessed February 8, 2022. www.nhats.org/sites/default/files/inline-files/DementiaTechnicalPaperJuly_2_4_2013_10_23_15.pdf
24.Weir D. Validating Mortality Ascertainment in the Health and Retirement Study. Published 2016. Accessed February 8, 2022. https://hrs.isr.umich.edu/publications/biblio/9022
25.Kaur P, Tan WS, Gunapal PPG, et al. Deaths in dementia: a scoping review of prognostic variables. BMJ Support Palliat Care. 2021;11(3):242-252. Published online June, 2020. doi: 10.1136/bmjspcare-2020-002217 [DOI] [PubMed] [Google Scholar]
26.Brodaty H, Seeher K, Gibson L. Dementia time to death: a systematic literature review on survival time and years of life lost in people with dementia. Int Psychogeriatr. 2012;24(7):1034-1045. doi: 10.1017/S1041610211002924 [DOI] [PubMed] [Google Scholar]
27.Brown MA, Sampson EL, Jones L, Barron AM. Prognostic indicators of 6-month mortality in elderly people with advanced dementia: a systematic review. Palliat Med. 2013;27(5):389-400. doi: 10.1177/0269216312465649 [DOI] [PMC free article] [PubMed] [Google Scholar]
28.Lee M, Chodosh J. Dementia and life expectancy: what do we know? J Am Med Dir Assoc. 2009;10(7):466-471. doi: 10.1016/j.jamda.2009.03.014 [DOI] [PubMed] [Google Scholar]
29.Dewey ME, Saz P. Dementia, cognitive impairment and mortality in persons aged 65 and over living in the community: a systematic review of the literature. Int J Geriatr Psychiatry. 2001;16(8):751-761. doi: 10.1002/gps.397 [DOI] [PubMed] [Google Scholar]
30.Heymans MW, van Buuren S, Knol DL, van Mechelen W, de Vet HC. Variable selection under multiple imputation using the bootstrap in a prognostic study. BMC Med Res Methodol. 2007;7(1):33. doi: 10.1186/1471-2288-7-33 [DOI] [PMC free article] [PubMed] [Google Scholar]
31.Vergouw D, Heymans MW, Peat GM, et al. The search for stable prognostic models in multiple imputed data sets. BMC Med Res Methodol. 2010;10(1):81. doi: 10.1186/1471-2288-10-81 [DOI] [PMC free article] [PubMed] [Google Scholar]
32.Austin PC, Lee DS, Ko DT, White IR. Effect of variable selection strategy on the performance of prognostic models when using multiple imputation. Circ Cardiovasc Qual Outcomes. 2019;12(11):e005927. doi: 10.1161/CIRCOUTCOMES.119.005927 [DOI] [PMC free article] [PubMed] [Google Scholar]
33.Schomaker M, Heumann C. Bootstrap inference when using multiple imputation. Stat Med. 2018;37(14):2252-2266. doi: 10.1002/sim.7654 [DOI] [PMC free article] [PubMed] [Google Scholar]
34.Austin PC, Tu JV. Bootstrap methods for developing predictive models. Am Stat. 2004;58(2):131-137. doi: 10.1198/0003130043277 [DOI] [Google Scholar]
35.Guo C, So Y, Jang W. Evaluating predictive accuracy of survival models with PROC PHREG. Published 2017. Accessed July 28, 2021. https://support.sas.com/resources/papers/proceedings17/SAS0462-2017.pdf
36.Heagerty PJ, Lumley T, Pepe MS. Time-dependent ROC curves for censored survival data and a diagnostic marker. Biometrics. 2000;56(2):337-344. doi: 10.1111/j.0006-341X.2000.00337.x [DOI] [PubMed] [Google Scholar]
37.Bansal A, Heagerty PJ. A tutorial on evaluating the time-varying discrimination accuracy of survival models used in dynamic decision making. Med Decis Making. 2018;38(8):904-916. doi: 10.1177/0272989X18801312 [DOI] [PMC free article] [PubMed] [Google Scholar]
38.Van Calster B, McLernon DJ, van Smeden M, Wynants L, Steyerberg EW; Topic Group ‘Evaluating diagnostic tests and prediction models’ of the STRATOS initiative . Calibration: the Achilles heel of predictive analytics. BMC Med. 2019;17(1):230. doi: 10.1186/s12916-019-1466-7 [DOI] [PMC free article] [PubMed] [Google Scholar]
39.Graf E, Schmoor C, Sauerbrei W, Schumacher M. Assessment and comparison of prognostic classification schemes for survival data. Stat Med. 1999;18(17-18):2529-2545. doi: [DOI] [PubMed] [Google Scholar]
40.Eckstrom E, Feeny DH, Walter LC, Perdue LA, Whitlock EP. Individualizing cancer screening in older adults: a narrative review and framework for future research. J Gen Intern Med. 2013;28(2):292-298. doi: 10.1007/s11606-012-2227-x [DOI] [PMC free article] [PubMed] [Google Scholar]
41.Qaseem A, Wilt TJ, Kansagara D, et al. ; Clinical Guidelines Committee of the American College of Physicians . Hemoglobin A1c targets for glycemic control with pharmacologic therapy for nonpregnant adults with type 2 diabetes mellitus: a guidance statement update from the American College of Physicians. Ann Intern Med. 2018;168(8):569-576. doi: 10.7326/M17-0939 [DOI] [PubMed] [Google Scholar]
42.Krist AH, Davidson KW, Mangione CM, et al. ; US Preventive Services Task Force . Screening for lung cancer: US Preventive Services Task Force Recommendation Statement. JAMA. 2021;325(10):962-970. doi: 10.1001/jama.2021.1117 [DOI] [PubMed] [Google Scholar]
43.Davidson KW, Barry MJ, Mangione CM, et al. ; US Preventive Services Task Force . Screening for colorectal cancer: US Preventive Services Task Force Recommendation Statement. JAMA. 2021;325(19):1965-1977. doi: 10.1001/jama.2021.6238 [DOI] [PubMed] [Google Scholar]
44.Gill TM. The central role of prognosis in clinical decision making. JAMA. 2012;307(2):199-200. doi: 10.1001/jama.2011.1992 [DOI] [PMC free article] [PubMed] [Google Scholar]
45.Besser LM, Gill DP, Monsell SE, et al. Body mass index, weight change, and clinical progression in mild cognitive impairment and Alzheimer disease. Alzheimer Dis Assoc Disord. 2014;28(1):36-43. doi: 10.1097/WAD.0000000000000005 [DOI] [PMC free article] [PubMed] [Google Scholar]
46.García-Ptacek S, Kåreholt I, Farahmand B, Cuadrado ML, Religa D, Eriksdotter M. Body-mass index and mortality in incident dementia: a cohort study on 11,398 patients from SveDem, the Swedish Dementia Registry. J Am Med Dir Assoc. 2014;15(6):447.e1-447.e7. doi: 10.1016/j.jamda.2014.03.001 [DOI] [PubMed] [Google Scholar]
47.Jang H, Kim JH, Choi SH, et al. Body mass index and mortality rate in Korean patients with Alzheimer’s disease. J Alzheimers Dis. 2015;46(2):399-406. doi: 10.3233/JAD-142790 [DOI] [PubMed] [Google Scholar]
48.Carey EC, Walter LC, Lindquist K, Covinsky KE. Development and validation of a functional morbidity index to predict mortality in community-dwelling elders. J Gen Intern Med. 2004;19(10):1027-1033. doi: 10.1111/j.1525-1497.2004.40016.x [DOI] [PMC free article] [PubMed] [Google Scholar]
49.Covinsky KE, Justice AC, Rosenthal GE, Palmer RM, Landefeld CS. Measuring prognosis and case mix in hospitalized elders. the importance of functional status. J Gen Intern Med. 1997;12(4):203-208. doi: 10.1046/j.1525-1497.1997.012004203.x [DOI] [PMC free article] [PubMed] [Google Scholar]
50.Haaksma ML, Eriksdotter M, Rizzuto D, et al. Survival time tool to guide care planning in people with dementia. Neurology. 2020;94(5):e538-e548. doi: 10.1212/WNL.0000000000008745 [DOI] [PMC free article] [PubMed] [Google Scholar]
51.van de Vorst IE, Golüke NMS, Vaartjes I, Bots ML, Koek HL. A prediction model for one- and three-year mortality in dementia: results from a nationwide hospital-based cohort of 50,993 patients in the Netherlands. Age Ageing. 2020;49(3):361-367. doi: 10.1093/ageing/afaa007 [DOI] [PMC free article] [PubMed] [Google Scholar]
52.Newcomer R, Covinsky KE, Clay T, Yaffe K. Predicting 12-month mortality for persons with dementia. J Gerontol B Psychol Sci Soc Sci. 2003;58(3):S187-S198. doi: 10.1093/geronb/58.3.S187 [DOI] [PubMed] [Google Scholar]
53.Stern Y, Stallard E, Kinosian B, et al. Validation and demonstration of a new comprehensive model of Alzheimer’s disease progression. Alzheimers Dement. 2021;17(10):1698-1708. doi: 10.1002/alz.12336 [DOI] [PMC free article] [PubMed] [Google Scholar]
54.Stallard E, Kinosian B, Stern Y. Personalized predictive modeling for patients with Alzheimer’s disease using an extension of Sullivan’s life table model. Alzheimers Res Ther. 2017;9(1):75. doi: 10.1186/s13195-017-0302-6 [DOI] [PMC free article] [PubMed] [Google Scholar]
55.Razlighi QR, Stallard E, Brandt J, et al. A new algorithm for predicting time to disease endpoints in Alzheimer’s disease patients. J Alzheimers Dis. 2014;38(3):661-668. doi: 10.3233/JAD-131142 [DOI] [PMC free article] [PubMed] [Google Scholar]
56.Stallard E, Kinosian B, Zbrozek AS, Yashin AI, Glick HA, Stern Y. Estimation and validation of a multiattribute model of Alzheimer disease progression. Med Decis Making. 2010;30(6):625-638. doi: 10.1177/0272989X10363479 [DOI] [PMC free article] [PubMed] [Google Scholar]
57.Kinosian BP, Stallard E, Lee JH, Woodbury MA, Zbrozek AS, Glick HA. Predicting 10-year care requirements for older people with suspected Alzheimer’s disease. J Am Geriatr Soc. 2000;48(6):631-638. doi: 10.1111/j.1532-5415.2000.tb04720.x [DOI] [PubMed] [Google Scholar]
58.Armstrong MJ, Song S, Kurasz AM, Li Z. Predictors of mortality in individuals with dementia in the National Alzheimer’s Coordinating Center. J Alzheimers Dis. 2022;86(4):1935-1946. doi: 10.3233/JAD-215587 [DOI] [PMC free article] [PubMed] [Google Scholar]
59.Stern Y, Folstein M, Albert M, et al. Multicenter study of predictors of disease course in Alzheimer disease (the “predictors study”). I. Study design, cohort description, and intersite comparisons. Alzheimer Dis Assoc Disord. 1993;7(1):3-21. doi: 10.1097/00002093-199307010-00002 [DOI] [PubMed] [Google Scholar]
60.Gianattasio KZ, Prather C, Glymour MM, Ciarleglio A, Power MC. Racial disparities and temporal trends in dementia misdiagnosis risk in the United States. Alzheimers Dement (N Y). 2019;5:891-898. doi: 10.1016/j.trci.2019.11.008 [DOI] [PMC free article] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials