Integration of an objective cognitive assessment into a prognostic index for 5-year mortality prediction

Abstract

Background/Objectives:

Prognostic indices rarely include cognition. We determined if a comprehensive cognitive screen or brief individual items were associated with improved mortality predictions of a widely-used prognostic index.

Design, Setting, Participants:

The National Social life Health and Aging Project Wave 2 (NSHAP-W2), a nationally-representative, cross-sectional, in-home survey conducted in 2010–11 on 3,199 community-dwelling adults age 60–99 years.

Measurements:

Cognition was measured using a Survey-Adapted Montreal Cognitive Assessment (MoCA-SA) grouped into three screened categories: Screen Normal (24+ pts), Screen Positive for Mild Cognitive Impairment (MCI) (18–23 pts), and Screen Positive for Dementia (<18 pts). Single-item cognitive measures included clock-draw and 5-word delayed-recall. We constructed a modified Lee Prognostic Index (range 0–18 points) based on age, behavior, function, and comorbidities shown to predict long-term mortality. We used logistic regression and the fraction of new information provided to determine if each cognitive measure improved the Lee index’s five-year mortality prediction.

Results:

The sample was 54% female and had a mean age of 72 years, MoCA-SA score of 22 (SD=4.5), and Lee index of 7 (SD=3). Regression analysis indicated the MoCA-SA modestly improved the Lee Index’s mortality prediction (p<0.001, fraction of new information provided=0.06); for low Lee Index scores (<4 points) the absolute mortality rate difference was 7% by cognitive status, and for higher Lee Index scores (4–7 points or 8–12 points) the absolute mortality rate difference was 15% by cognitive status. The clock-draw and delayed-recall items added similar value to mortality predictions as the longer MoCA-SA. Cognition had the third highest fraction of new information of all 13 Lee index items.

Conclusion:

Incorporating a brief measure of cognition into a modified Lee Index, even with single items, resulted in more accurate five-year mortality risk predictions. Cognition should be included in prognostic calculators in older adults given its independent association with mortality risk.

INTRODUCTION

Dementia is common at older ages with prevalence estimates of 10% in adults 75–84 years old and 30% in those ≥85 years.¹ There is robust epidemiologic evidence independently linking cognition to mortality risk,^2–4 even at early stages of cognitive impairment, such as Mild Cognitive Impairment (MCI).^5–11 Despite growing evidence of this link, prognostic models for predicting life expectancy among older adults have not incorporated objective measures of cognition. In a recent systematic review of prognostic indices, only 1 of the 6 indices developed for community-dwelling older adults included self-reported dementia, and no indices included measures of cognition capable of assessing MCI.¹² Risk stratification tools examining frailty similarly have focused on physical function and have not consistently assessed cognition.¹³ Cognitive measures have likely not been incorporated into prognostic models due to limitations in data from large population-based surveys, including lack of clinically-relevant screening tools and relying on self-reported diagnoses. Better integration of cognitive measures into prognostic models for older adults might improve the models’ ability to predict mortality and assist in decision-making.

The Montreal Cognitive Assessment (MoCA), a widely-used, multidomain clinical tool developed to screen for MCI and dementia, is a candidate measure to be included in prognostic models.¹⁴ However, to date, there are no studies examining the association of the MoCA with mortality in a representative sample of older adults. Moreover, it is unclear if it is necessary to use a comprehensive cognitive screen like the MoCA which can take on average 10–15 minutes to administer,¹⁴ or whether a shorter assessment such as the clock-draw, subtract-7’s or delayed recall would be sufficient. This is particularly relevant given the new recommended training and certification costs for the MoCA, which have prompted consideration of alternative cognitive assessments.¹⁵ Single items such as delayed-recall, clock-draw, subtract-7’s, and orientation items are common to several cognitive assessments (e.g. MoCA, Mini-Cog, Saint Louis University Mental Status (SLUMS) exam, Telephone Interview of Cognitive Status, Mini-Mental Status Exam, etc.).^16–19

Here, we use data from a nationally-representative cohort of community-dwelling older adults which includes a validated survey-adaptation of the MoCA (MoCA-SA).²⁰ In addition, we examine individual items within the MoCA-SA to determine if single items can be integrated into a prognostic index as a substitute for the larger cognitive measure. We focus on the Lee Prognostic Index, a widely-used mortality prognostic index for the general population of older adults previously developed from the Health and Retirement Study (HRS).^{21, 22} The Lee index is similar to two other available indices to predict 5- or 10-year mortality and is readily available online.^{12, 23, 24} During the original selection of items for the Lee Index, self-reported memory-related diseases (such as dementia) were found to not substantially improve mortality predictions after accounting for age, comorbidities and other factors.²² Therefore, an improvement in mortality predictions by adding cognition would suggest utility in conducting an objective cognitive test or a test capable of detecting mild impairment as opposed to relying on self-report. In summary, our objective was to determine if the MoCA-SA or individual items within the MoCA-SA were associated with five-year mortality and added predictive value to the Lee Prognostic Index for older adults.

METHODS

Study Sample.

We used data from the National Social life Health and Aging Project, Wave 2 (NSHAP-W2), an in-home, nationally-representative probability sample of older adults residing in households across the United States. Data was collected between August 2010 and May 2011.²⁵ The weighted overall response rate was 76.9%. Individuals with missing values for key covariates were excluded (n=178), yielding a final sample of 3,199 individuals. All respondents provided written informed consent and the protocol was approved by the institutional review boards at the University of Chicago and National Opinion Research Center (NORC).

Cognitive measures.

The MoCA is designed to screen for cognitive changes consistent with MCI and early dementia.¹⁴ Following extensive testing, an 18-item Survey-Adapted Montreal Cognitive Assessment (MoCA-SA) was included in NSHAP-W2.²⁶ The MoCA-SA was designed for administration by non-medical individuals and to reduce respondent burden within the context of a large, time-limited national survey, while preserving the MoCA’s overall sensitivity to a range of cognitive abilities. MoCA-SA scores are highly correlated with the MoCA (r=0.98) and scores can be accurately converted into full MoCA scores.²⁰ We used the scale as both a continuous measure and divided into three screening categories based on prior literature: 24+ points (screened as normal), 18–23 points (screen positive for MCI), <18 points (screen positive for dementia).^{27, 28} Sensitivity analyses of MoCA-SA as a categorical variable with higher cutoffs yielded similar results.

In addition, we examined individual items within the MoCA-SA with particular attention to the 5-word delayed-recall and clock-draw, given their use in multiple cognitive screening tests. The 5-word delayed-recall was administered using standard instructions on the MoCA with the words Face, Velvet, Church, Daisy, and Red. Participants were given two trials of immediate recall, then asked to recall as many of the words at the end of the administration of the cognitive battery. Participants could recall the words in any order, were not given cues, and could score a range of 0–5 points. For the clock draw item, participants were provided a large, blank white paper and asked to draw a clock, put in all the numbers, and set the time to “10 past 11.” Clocks were manually scored after the interview for the correct contour, number placement, and location of hands (Range: 0–3 points) by two independent reviewers using MoCA scoring instructions. In cases of disagreement, clocks were re-examined and reviewers would come to a consensus on the final score or seek input from a third reviewer.

Prognostic Model

The Lee Prognostic Index was originally developed in the Health and Retirement Study (HRS) and includes 12 items that together predict 5- and 10-year life expectancy.^{22, 29} We constructed a slightly modified Lee Prognostic Index utilizing available items from NSHAP. The following 10 items were included directly from NSHAP: 1) Age (ages 60–64: 1 point, 65–69: 2 points, 70–74: 3 points, 75–79: 4 points, 80–84: 5 points, >84: 7 points); 2) Male sex (2 points); 3) Current tobacco use (2 points); 4) Body mass index <25 (1 point); 5) Diabetes (1 point); 6) Non-skin cancers (2 points); 7) Chronic lung disease (2 points); 8) Heart failure (2 points); 9) Difficulty bathing (2 points); and 10) Difficulty managing finances (2 points). The following two items in the original Lee Index were substituted with available items from NSHAP: 1) “difficulty walking several blocks” was substituted with the NSHAP item “difficulty walking one block” (2 points); 2) “difficulty pushing/pulling large objects” was substituted with the NSHAP item “difficulty performing housework” (1 point). We examined the Lee index as a continuous measure as well as in categories of <4 points (<5% 5-year mortality), 4–7 points (5–20% 5-year mortality), 8–12 points (20–50% 5-year mortality), and >12 points (>50% 5-year mortality).²² A combined ≥8 point category reflects a >50% 10-year mortality risk based on a prior published study.²⁹

Mortality.

Five-year mortality was determined after speaking with the respondent in the following data collection, NSHAP Wave 3 (2015–2016), through a proxy interview with a neighbor or family member, or through public records or news sources. Additional investigation occurred to determine whether respondents were alive, but unavailable for re-interview. Of the 3,377 Wave 2 respondents, 604 were deceased five years later (17.9%), 2,370 were alive and re-interviewed (70.2%), 160 were too sick to interview (4.7%), and 185 (5.5%) were determined to be alive but not accessible for re-interview. There were 57 cases (1.7%) in which it was unknown if the respondent was alive; these were excluded from the analysis.

Statistical Analysis.

We assessed the bivariate association of the MoCA-SA, individual items within the MoCA-SA, and the Lee Index with five-year mortality. We then conducted a logistic regression to test for an interaction term between the MoCA-SA and the Lee index in predicting mortality and calculated the new model’s Area Under the Curve (AUC). Using 10-fold cross-validation we found the risk of overfitting was low.³⁰ We determined the value of including the MoCA-SA in the prognostic model by calculating the fraction of new information provided, described by Harrell (2019), which is the proportion of predictive information added by the MoCA-SA to the Lee Index.^{31, 32} In addition, we conducted likelihood ratio tests between models and a test of equality of the AUCs.³³ We used a scatter-plot to illustrate the difference in the predicted probability of five-year mortality from the two models. Analyses were replicated to determine if the delayed-recall or clock-draw items added similar predictive power to the Lee Index as the full MoCA-SA. Finally, we determined appropriate point modifications to a Lee index incorporating cognition, and calculated the Net Reclassification Index (NRI) for our sample in clinically-relevant categories of the Lee index using the modified scoring system.^{34, 35} All statistical analyses were conducted using Stata 15.1.³⁶

RESULTS

The sample was 54% female and had a mean age of 72 years, MoCA-SA score of 22 points (SD=4.5), and mean Lee index of 7 points (SD=3) (Table 1). The MoCA-SA was highly associated with five-year mortality, even after adjustment for each individual Lee Index component (24+ points: 10%, 18–23 points: 23%, <18 points: 35%, p<0.001) (Supplementary Figure S1). Each MoCA-SA cognitive domain was associated with five-year mortality (Supplementary Figure S2), with clock-draw (range 11%−36%) and delayed-recall (range 8–35%) having the highest association, and attention items (subtract-7’s and digit-span) having the lowest association (range 15–27%).

Table 1.

Sample characteristics in the NSHAP Wave 2 cohort (n=3,199)

Characteristics		No. (%)
Age	Median (IQR)	72 (66–79)
	<60	89 (2.8)
	60–69	1235 (38.6)
	70–79	1177 (36.8)
	80–84	451 (14.1)
	85+	247 (7.7)
Gender	Female	1742 (54.4)
Race/Ethnicity	White	2284 (71.4)
	Black	492 (15.4)
	Hispanic	349 (10.9)
Modified Lee Index Score	<4 points	658 (20.6)
	4–7 points	1,560 (48.8)
	8–12 points	840 (26.2)
	>12 points	141 (4.4)
MoCA-SA	Median (IQR)	23 (19–26)
Score	<18 points	560 (18)
	18–23 points	1171 (37)
	24–30 points	1468 (46)
5-word Delayed Recall	0–1 points	540 (16.8)
	1–2 points	775 (24.1)
	3–5 points	1884 (59.0)
Clock Draw	0–1 points	653 (20.3)
	2 points	1211 (37.7)
	3 points	1335 (42.0)
BMI	<25	2463 (77)
Smoking	Current	425 (13.3)
Medical	Diabetes	757 (23.7)
Conditions	Non-skin Cancer	437 (13.7)
	Chronic Lung Disease	493 (15.4)
	Heart Failure	157 (4.9)
Functional	Bathing	294 (9.2)
Impairment	Finances	294 (9.2)
(Difficulty)	Walking 1 Block	832 (26)
	Housework	504 (15.8)
5-year mortality	Deceased	582 (18.2)

Abbreviations: NSHAP – National Social life Health and Aging Project, MoCA-SA – Survey-Adapted Montreal Cognitive Assessment, IQR – Interquartile Range, BMI – Body Mass Index.

¹Lee Index was modified from index provided in Lee, SJ et al. 2006 JAMA [22]. Categories correspond to 5-year mortality rates (<4 pts: <5%, 4–7 pts: 5–20%, 8–12 pts: 20–50%, >12 pts: >50%). In addition, ≥8 points corresponds to a >50% 10-year mortality rate [29].

²MoCA-SA categories reflect <18 (screen positive for dementia), 18–23 (screen positive for MCI), and 24–30 (screen normal).

The Lee Index was highly predictive of five-year mortality (<4 points: 2%, 4–7 points: 10.7%, 8–12 points: 34.5%, >12 points: 57.6%, AUC: 0.795) (Figure 1). The MoCA-SA modestly improved the Lee Index’s overall five-year mortality prediction (fraction of new information provided = 0.06, AUC 0.807, p<0.001). For context, the MoCA-SA had the third highest fraction of new information of all Lee index items, following age (0.27) and sex (0.08), and was higher than cancer (0.04) (Supplementary Table S1). Logistic regression analysis indicated a significant interaction term between the Lee Index and the MoCA-SA in predicting five-year mortality (p<0.001) (Figure 1). Among those with lowest Lee Index scores (<4 points) there was a 7% absolute difference in mortality rates by screened cognitive status, and for scores of 4–7 points or 8–12 points there was a 15% absolute difference in mortality rates (Figure 1). There was less absolute difference between mortality rates between cognitive groups in the highest Lee index group (>12 points), although comparisons were limited by sample size. A scatter-plot of mortality predictions for Lee Index with and without the MoCA-SA (Supplementary Figure S3) indicated that worse MoCA-SA scores augment Lee index predictions most at lower Lee index scores (with worse MoCA scores increasing the predicted probability of mortality). From a logistic regression model, we derived on average the point modification of the Lee index for different MoCA-SA categories in comparison to existing Lee index items. By adding 2 points for MoCA-SA of <18 points to the Lee score and subtracting 2 points for MoCA-SA of >23 points to the Lee score, we were able to capture the majority of discrimination from the full scale (AUC=0.80).

Figure 1.

Modified Lee Prognostic Index and five-year mortality stratified by screened cognitive status.

The graph shows marginal probabilities of five-year mortality drawn from a logistic regression model.

¹The Lee index is modified from Lee, SJ et al. JAMA 2006 [22]. Categories correspond to 5-year mortality rates (<4 pts: <5%, 4–7 pts: 5–20%, 8–12 pts: 20–50%, >12 pts: >50%). In addition, ≥8 points corresponds to a >50% 10-year mortality rate [29].

²Cognitive categories represent screening groups based off the Survey-Adapted Montreal Cognitive Assessment (screen as Normal: 24+ points, screen positive for MCI: 18–23 points, screen positive for Dementia: <18 points).

Abbreviation: pts – points, CI – confidence interval, MCI – Mild Cognitive Impairment

We conducted additional analyses to determine if the most predictive single items within the MoCA-SA, the clock-draw or delayed-recall, could significantly improve Lee Index predictions (Table 2). Both items added similar predictive value as the full MoCA-SA. The clock-draw increased the predictive power of the Lee Index (fraction of new information provided = 0.06) (Supplementary Figure S4), as did the delayed-recall item (fraction of new information provided = 0.06) (Supplementary Figure S5). Modifying the Lee Index to add two points for a normal clock and subtract two points for a score of 0–1 points on the clock improved the discrimination of the Lee Index similar to the MoCA-SA (AUC = 0.80). Similarly, adding 2 points for delayed-recall of 0 and subtracting 2 points for delayed-recall of 3–5 words significantly improved discrimination of the Lee index (AUC = 0.80). We show the difference in classification by Lee risk categories using a proposed scoring system (Table 3) in our sample along with the NRI (Table 4). Using the modified scoring system, more individuals are classified in the very low risk (<5%) or highest risk (>50%) for 5-year mortality, and the NRI reflects an overall improvement in classification with values ranging 0.14–0.16.

Table 2.

Association of Lee Index with five-year mortality stratified by clock-draw and delayed recall scores

Modified Lee Index1	Overall (95% CI) n=3,199	Clock Draw2			Delayed Recall Score3
		3 points (95% CI) n=1,347	2 points (95% CI) n=1211	0–1 Points (95% CI) 0–2 n=653	3–5 points (95% CI) n=1,315	1–2 points (95% CI) n=1364	0 points (95% CI) n=532
<4 pts (n=652)	2% (0.8–3)	1% (0–2)	3% (0–5)	3% (0–6)	0.4% (0–0.9)	5% (1–9)	14% (1–26)
4–7 pts (n=1,541)	11% (9–12)	8% (6–10)	12% (9–14)	16% (12–20)	7% (5–9)	15% (11–18)	19% (14–24)
8–12 pts (n=825)	34% (31–38)	28% (23–34)	33% (28–38)	44% (38–50)	28% (24–32)	36% (30–43)	45% (38–51)
>12 pts (n=139)	58% (49–66)	39% (23–55)	60% (47–74)	67% (55–80)	58% (43–72)	59% (44–74)	56% (43–69)

¹The Lee index is modified from Lee, SJ et al. JAMA 2006 [22]. Categories correspond to 5-year mortality rates (<4 pts: <5%, 4–7 pts: 5–20%, 8–12 pts: 20–50%, >12 pts: >50%). In addition, ≥8 points corresponds to a >50% 10-year mortality rate [29].

²Clocks scored for the correct contour, number placement, and location of hands (Range: 0–3 points)

³Delayed recall was administered using standard instructions on the MoCA with the words Face, Velvet, Church, Daisy, and Red.

Table 3.

Lee Prognostic Index for Older Adults including cognition

Original Items		Points
1. Age	60–64 years	1 point
	65–69 years	2 points
	70–74 years	3 points
	75–79 years	4 points
	80–84 years	5 points
	≥85 years	7 points
2. Sex (Male/Female)	Male	2 points
3. Body Mass Index	< 25	1 point
4. Diabetes		1 point
5. Cancer (excluding minor skin cancer)		2 points
6. Chronic Lung Disease		2 points
7. Congestive Heart Failure		2 points
8. Current smoker		2 points
9. Difficulty Bathing/Showering		2 points
10. Difficulty managing finances		2 points
11. Difficulty walking several blocks		2 points
12. Difficulty pushing/pulling large objects like living room chair		1 point
Proposed addition:
13. Cognition	MoCA ≥24 points	−2 points
	MoCA ≤18 points	+2 points
	Alternatives:
	Delayed Recall 3–5 words	−2 points
	Delayed Recall 0 words	+2 points
	Clock draw 3 points	−2 points
	Clock draw 0–1 points	+2 points

Abbreviations: MoCA – Montreal Cognitive Assessment

Table 4.

Comparison of categorizations for 5-year mortality risk of Lee Index with modified scoring of the Lee Index to include Cognitive Measures (N=3,199)

Score	Lee Index	Lee + MoCA-SA	Lee + Delayed Recall	Lee + Clock
	N	N	N	N
Very Low Risk (<5%)	657	1,048	1,136	972
<4 points
Low Risk (5–20%)	1,559	1,210	1,163	1,258
4 to 7 points
Medium (20–50%)	838	741	710	762
8 to 12 points
High (>50%)	103	158	148	165
>12 points
Overall reclassification (%)	-	26%	31%	25%
Net Reclassification Index	-	0.16	0.15	0.14

Abbreviations: MoCA-SA - Survey adapted Montreal Cognitive Assessment

Net Reclassification Index is calculated using the formula: NRI = P(up | event) − P(down | event) + P(down | nonevent) − P(up | nonevent) [35]

Lee Index was modified from index provided in Lee, SJ et al. 2006 JAMA [22]. Categories correspond to 5-year mortality risk (<4 pts: <5%, 4–7 pts: 5–20%, 8–12 pts: 20–50%, >12 pts: >50%). In addition, ≥8 points (the Medium and High categories) corresponds to a >50% 10-year mortality rate [29].

DISCUSSION

In a nationally-representative sample, we found that integrating a multi-domain cognitive screen, the MoCA-SA, or individual items such as the clock-draw or 5-word delayed recall, significantly improved the modified Lee Index’s 5-year mortality predictions. To our knowledge, this is the first study to examine whether a screening tool for mild cognitive impairment predicts mortality in a nationally-representative population of older adults and the first to integrate such a cognitive screening tool into a prognostic index.

Results show that cognition is an independent risk factor for mortality in older adults with a wide spectrum of health statuses. Prior prognostic indices had been limited in their inclusion of clinically-relevant cognitive screening tools.⁵ Only one prior index in community-dwelling older adults, the Gagne index, found that self-reported dementia improved mortality predictions.³⁷ Our findings were further differentiated from indices developed among nursing home residents which focused on 6-month or 1-year mortality and utilized cognitive measures available in the Minimum Dataset which assess moderate-to-severe cognitive impairment.^38–41 Using the MoCA-SA, at Lee Index scores of <4 points there was a 7% absolute mortality difference between cognitive groups, and within the 4–7 point and 8–12 point Lee index groups, there was a 15% absolute mortality difference between cognitive groups. This likely reflects the central role that early cognitive changes play in mortality risk even in the absence of unhealthy behaviors, loss of physical function, and/or comorbidities.⁷ This is consistent with prior studies, including a 2019 HRS study by Aliberti et al., which showed that mild cognitive impairment was predictive of mortality even in the absence of physical frailty.⁵ Mechanisms for the effect of early cognitive decline on mortality include biological pathways (e.g. reflective of vascular changes), reduced health literacy (e.g. ability to access and synthesize health information), lower use of preventive health services, reduced ability to mobilize social and financial resources, and others.⁸ Higher or maintained cognition may reflect a reduced mortality risk from traditional health risk factors through similar biological and behavioral pathways. At the highest Lee index scores (>12 points), there was less differentiation in mortality rates between cognitive groups. However, these comparisons were limited by smaller sample sizes at the highest Lee index categories.

There is significant variation across clinical practices in preferences for specific cognitive assessments and the available time to administer cognitive screens. Moreover, some clinicians have indicated their preference for alternative cognitive assessments to the MoCA in light of recent training requirements.¹⁵ Our results indicate that the delayed-recall item contributed as much to mortality predictive accuracy for the Lee Index as the longer MoCA-SA scale. Delayed recall assesses working memory, and similar assessments are contained in other cognitive screening tests, including the Mini-Cog, Saint Louis University Mental Status (SLUMS) exam, the MMSE, or as part of the mental status physical exam. In addition, the clock-draw item, which can be conducted as a stand-alone cognitive test or within cognitive screening tools such as the Mini-Cog and SLUMS exam, improved mortality predictions of the Lee Index. Both findings are consistent with prior literature showing the clock draw and working memory assessments as strongly predictive of mortality.^{42, 43}

Clinicians can integrate scores into the Lee Prognostic Index (available online) by adding or subtracting points (<18 MoCA Score: +2 points, 24+ MoCA Score: −2 points) (Table 3).²¹ Alternatively, the clock-draw or delayed-recall items can be integrated into the Lee Index (0 words or 0–1 points on clock: +2 points; 3–5 words or 3 points on clock: −2 points).^{42, 43} While this method is simple, it prioritizes accessibility to prognostic estimates for busy clinicians and captures the majority of information provided by the objective cognitive screen. Alternatively, some clinicians may prefer to use reference tables or figures with mortality rates.⁷ In addition, external validation of the Lee index and inclusion of cognition has relevance to risk stratification of older adults to better estimate heterogeneity of treatment effects in randomized controlled trials.⁴⁴ To this end, the full Lee index can be used as both a continuous measure of risk or be divided into quartiles of mortality risk.

This study has several limitations. First, the MoCA-SA is a cognitive screening tool and is not a substitute for clinical diagnoses of MCI or dementia. However, it is a well-tested survey adaptation of the widely-used MoCA which is directly applicable to screening for cognitive impairment in general clinical practice. Second, results are specific both to the 5-year mortality interval and a modified Lee Prognostic Index. Two other mortality indices for 5- or 10-year mortality, the Schonberg index and the Sueomoto index, have similar items as the Lee index and minimal inclusion of cognitive measures.^{23, 24} Therefore, we anticipate cognition, as measured by the MoCA-SA, would be relevant to similar prognostic indices or mortality for different timeframes, but future studies are required for this validation. Third, the study involves one cross-sectional assessment of mortality risk and does not account for how risk factors, including cognition, might change over time and subsequently impact mortality risk. Fourth, the data are focused on community-dwelling older adults able to participate in a 2-hour survey, and therefore do not assess the association of mortality or the Lee index with more profound cognitive impairment or in different living situations. Consequently, use of this prognostic model in clinical settings should be limited to community-dwelling older adults, and the model will require further validation in nursing home or assisted living facility residents.

In conclusion, we find that a brief, clinically-relevant cognitive assessment, the MoCA-SA, predicts 5-year mortality and can be integrated into the modified Lee Prognostic Index to improve its performance. The delayed recall and clock draw items can provide similar prognostic information when a full MoCA is unavailable or in time-limited settings. Study results show the importance of considering cognition as an independent mortality risk factor and provide a model for how to incorporate cognition into clinical mortality risk stratification tools in older adults.