Cognition and the Risk of Hospitalization for Serious Falls in the Elderly: Results From the Cardiovascular Health Study

Abstract

Background.

Many elderly adults fall every year, sometimes resulting in serious injury and hospitalization. Although impaired cognition is a risk factor for injurious falls, little is known about cognitive decline above the threshold of impairment and risk of serious falls in community-dwelling seniors.

Methods.

In total, 702 of 5,356 older adults participating in the Cardiovascular Health Study experienced an injurious fall between 1990 and 2005, as indicated by hospitalization records. General cognition was measured annually with the Modified Mini-Mental State Examination and processing speed with the Digit Symbol Substitution Test. The Cox regression model was used to calculate hazard ratio and 95% confidence interval with and without time-dependent covariates and adjusted for known risk factors.

Results.

Participants with slightly decreased Digit Symbol Substitution Test scores were at increased risk for a serious fall (hazard ratio = 1.58, 95% confidence interval = 1.15–2.17). The risk continued to increase with each quartile decrease in Digit Symbol Substitution Test score. Participants without prevalent cardiovascular disease at baseline and decreased Modified Mini-Mental State Examination scores (80–89) had a 45% increased risk for a serious fall and those at high risk for dementia (<80) were at twice the risk as participants scoring above 90 (hazard ratio = 2.16, 95% confidence interval = 1.60–2.91).

Conclusions.

Both decreased general cognition and decreased processing speed appear to be potential risk factors for serious falls in the elderly. When assessing the risk of serious falls in elderly patients, clinicians should consider usual factors like gait instability and sensory impairment as well as less obvious ones such as cardiovascular disease and cognitive function in nondemented adults.

MORE than one third of adults aged 65 years and older fall every year (1–3); this proportion increases with age (4). Falls have serious consequences in the elderly population, including injury (2,3), fear of falling again (5), depression (5), admission to a long-term care facility (6), and death (7). Approximately, 5% of all falls result in a serious injury (2,3), leading to substantial medical costs (8). The number of injuries and medical costs associated with falling will likely increase as the U.S. population ages.

Identifying risk factors for falls in the elderly has long been of interest to researchers and clinicians intent on reducing morbidity and mortality in this age group. Impaired cognition is a risk factor for recurrent falls (9,10), injurious falls (11,12), and serious falls requiring hospitalization (13,14).

There is also a strong relationship between falls, cognition, and physical function. Complex measures of physical function can predict cognitive decline (15) and falls (16) in healthy older adults. Specifically, low scores on tests of speed and executive function, including the Digit Symbol Substitution Test (DSST), are associated with increased risk of a fall (17). Incident stroke is associated with a subsequent decrease in cognition (18); many older adults with mild cognitive impairment also present with gait dysfunction (19).

However, it is unknown whether decreased cognition, above the threshold of impairment, is also a risk factor for serious falls. Identifying a relationship between increasing levels of impairment in global cognition or specific cognitive domains may help to identify elderly subpopulations most at risk for serious falls.

The objective of this study was to determine if decreased global cognitive function, processing speed, or both, as measured by standardized cognitive tests, are associated with increased risk of serious falls resulting in hospitalization in older adults. Hospitalization and study records from the Cardiovascular Health Study (CHS), a large longitudinal study population with 15 years of follow-up, were used to examine this association.

METHODS

Study Population

The CHS is a population-based longitudinal study of risk factors for cardiovascular disease and stroke in older adults (20). Participants were recruited from a random sample of people on Health Care Financing Administration Medicare eligibility lists in four U.S. counties: Forsyth County, North Carolina; Sacramento County, California; Washington County, Maryland; and Pittsburgh (Allegheny County), Pennsylvania. The Institutional Review Boards at all institutions involved in this study, including the Coordinating Center at the University of Washington, approved this study. A total of 5,888 participants were enrolled: 5,201 in 1989–1990 and 687 additional African American participants in 1992–1993. Additional details of the study design and recruitment are published elsewhere (20,21).

Data Collection

The study protocol consisted of an extensive baseline clinic visit, followed by alternating clinic visits and telephone contacts semiannually. The baseline clinic visit included standardized questionnaires, a physical examination, and laboratory testing (20).

At the initial clinic visit, the following participant information was collected by interview: age, race, education level, marital status, current treatment for arthritis, current diagnosis of osteoporosis, and any vision problems. Participants completed questionnaires on depressive symptoms (Center for Epidemiological Studies—Depression scale) (22), Activities of Daily Living (ADLs), and Independent Activities of Daily Living. Physical and performance measures were also collected, including time to complete five chair stands and gait speed (m/s). Body mass index was calculated from measurement of each participant's height and weight. Participants were instructed to bring all current prescription and over-the-counter medications to every clinic visit, where a medical inventory was completed (23).

At the initial examination, participants reported if they had ever been diagnosed with myocardial infarction, angina, congestive heart failure, claudication, stroke, or transient ischemic attack. Self-reports were then validated through medical record review or from answers to a physician questionnaire. During follow-up, participants provided information for all hospitalizations between contacts. All hospitalizations were validated in a similar fashion and then adjudicated by the CHS Mortality and Morbidity Review Subcommittees. Classification of prevalent and incident morbid events for these conditions has been described previously (24,25). In this report, prevalent cardiovascular disease (CHD) is defined as previous myocardial infarction, angina, angioplasty, or bypass surgery. Complete ascertainment of hospitalization records was later verified by cross-checking hospitalization records with the Medicare database.

The Mini-Mental State Examination (26) and the DSST (27) were used to assess cognition at the initial clinic visit. Beginning at the first follow-up clinic visit, the Modified Mini-Mental State Examination (3MS) (28) replaced the Mini-Mental State Examination. The 3MS screens for dementia by testing five cognitive functions (orientation, registration, attention and calculation, recall, language) (26). The 3MS differs from the Mini-Mental State Examination in three ways: four items were added, the scoring range expanded to 30–100 points, and administrative materials were developed to improve the reliability and validity of the exam (28). Participants enrolled in the new cohort were never tested with the Mini-Mental State Examination. This analysis begins at the first follow-up clinic visit to ensure consistency in the cognitive measures used; henceforth, this article will refer to the first follow-up visit for the original cohort and the initial visit for the African American cohort, collectively, as the baseline clinic visit. The DSST is a measure of processing speed (26). The participant is given a code with paired symbols and digits and then allotted 90 seconds to complete a test grid by matching the appropriate symbols to the corresponding letters. The score is the number of symbols completed correctly. Lower scores on the 3MS and DSST indicate lower cognition.

Outcome Identification

A serious fall was defined as a fall causing an injury requiring admission to the hospital; events were identified from hospitalization records. Initially, participants were excluded from this study if they died before the baseline clinic visit (n = 100) or were hospitalized for a serious fall before the baseline clinic visit (n = 17). Figure 1 shows the process taken to identify participants who required hospitalization after experiencing a serious fall. A participant was classified as experiencing a serious fall if the following three conditions were met in at least one hospitalization record: (i) presence of an injury code (International Classification of Diseases—Clinical Modification codes 800–959) in any of the first three diagnosis codes, indicating that the injury was the primary cause of hospitalization; (ii) presence of an E-code for falling (International Classification of Diseases—Clinical Modification codes: E880–E886, E888); and (iii) 3MS and DSST scores available for the baseline clinic visit. Only information from the incident fall hospitalization event was considered; subsequent falls were not included in the analysis. A total of 5,356 CHS participants were eligible for this study and were observed from 1990 to 2005.

Figure 1.

Flow diagram of outcome identification.

Statistical Analyses

Baseline characteristics for the fallers and nonfallers were compared using the chi-squared test of homogeneity. The Kaplan–Meier product limit estimator was used to calculate and plot survival curves as a function of time since the baseline clinic visit date. Survival curves were compared using the log-rank test. All statistical analyses were conducted using Stata10 (StataCorp, College Station, TX).

Cox regression was used to estimate the association between experiencing a serious fall and cognition and is presented as a hazard ratio with 95% confidence interval (CI). Participants were at risk for a serious fall beginning on the date of their baseline clinic visit and were censored on the date of hospitalization admission for a serious fall, the date of death, or June 30, 2005, whichever came first. Models were constructed separately for DSST and 3MS score. DSST scores were grouped into quartiles based on the distribution of baseline DSST scores in the study population (Quartile 1 = 48–90, Quartile 2 = 40–47, Quartile 3 = 29–39, Quartile 4 = 0–28). As there is no agreed-upon cutoff to indicate impaired cognition for this test (29), the lowest quartile was assumed to represent participants at high risk for dementia. A cutoff score of 79/80 on the 3MS has been used to indicate a need for further screening for dementia (18,30); thus, scores were grouped accordingly (90–100, 85–89, 80–84, <80). Initially, baseline cognition was considered as the predictor of interest. However, cognition is likely to change over time and subsequent models considered cognition as a time-dependent covariate. Scores on the 3MS and DSST were updated to the most recently available score at the time of each study clinic visit. A test for trend was performed using the log-rank statistic.

Both the baseline and time-dependent models were also adjusted for four groups of covariates suspected to influence the relationship between cognition and having a serious fall. Model 1 was adjusted for demographic characteristics, including age, education (less than high school, high school graduate, vocational school/some college, college graduate), clinic location, race/ethnicity (white, non-white), and marital status (married, previously married, never married). Model 2 was additionally adjusted for medical comorbidities, including prevalent CHD, stroke, osteoporosis (yes, no, unknown), receiving treatment for arthritis (yes, no, unknown), vision problems (yes, no, unknown), body mass index category (<18.5, 18.5–24.9, 25.0–29.9, ≥30.0), and depressive symptoms. Model 3 was additionally adjusted for prescription drug use (benzodiazepines, anticholinergics, insulin). Finally, Model 4, the “fully adjusted model,” was adjusted for all aforementioned variables and measures of functional status, including difficulties with ADLs (0, ≥1), difficulties with Independent Activities of Daily Living (0, ≥1), gait speed, completed five chair stands, and quartiles of physical activity (kcal; <540, 540–1226.24, 1225.25–2549.9, ≥2550). The baseline model was adjusted for the baseline values of all confounders. The time-dependent model included time-dependent covariates for variables in models 2–4. Covariate values and cognitive test scores were carried forward if missing at a subsequent visit. Interactions between test score and age, sex, education, benzodiazepine use, anticholinergic use, and CHD were considered. Only the interaction between 3MS score and CHD status at baseline was significant.

The proportional hazards assumptions were checked using Schoenfeld residuals, and all models were stratified by gender (p < .01). Separate results are not reported for males and females because stratification merely assumes a different baseline hazard function; the hazard ratio remains the same for comparisons between males or between females.

RESULTS

Over 15 years of follow-up (mean 10.3 years, median 11.8 years), 702 participants experienced at least one serious fall requiring hospitalization. Hospitalized fallers were more likely to be older, white, and female and have a normal body mass index (Table 1). They were also more likely to be taking benzodiazepines, to have a diagnosis of osteoporosis, or to have vision problems. Additionally, hospitalized fallers were more likely to have difficulties with one or more ADLs or Independent Activities of Daily Living and they took longer to complete five chair stands. Whereas hospitalized fallers and participants not hospitalized for a fall reported similar baseline scores on the 3MS, hospitalized fallers scored worse on the DSST at baseline. Participants scoring in DSST quartiles 2–4 had shorter time to a serious fall than participants with the best performance in DSST Quartile 1 (Figure 2; p < .001).

Table 1.

Selected Characteristics of Participants Hospitalized for a Fall and Participants Never Hospitalized for a Fall^*

	Ever Hospitalized for All
	Yes (N = 702)		No (N = 4,654)
Characteristics	n	%	n	%	p value
3MS score					.12
90–100 (best)	477	68.0	2,956	63.5
85–89	110	15.7	789	17.0
80–84	54	7.7	452	9.7
<80 (worst)	61	8.7	457	9.8
DSST score					.006
Quartile 1: 48–90 (best)	153	21.8	1,285	27.6
Quartile 2: 40–47	178	25.4	1,069	23.0
Quartile 3: 29–39	209	29.8	1,212	26.0
Quartile 4: 0–28 (worst)	162	23.1	1,088	23.4
Gender: male	207	29.5	2,053	44.1	<.001
Age, years					<.001
65–69	126	18.0	1,398	30.1
70–74	245	34.9	1,638	35.2
75–79	184	26.2	1,014	21.8
≥80	147	20.9	600	12.9
Education					.12
Less than high school	168	24.0	1,297	28.0
High school graduate	197	28.1	1,306	28.1
Some college/vocational school	174	24.8	1,068	23.0
College graduate	162	23.1	969	20.9
Clinic					.16
Wake Forest University	151	21.5	1,173	25.2
University of California, Davis	196	27.9	1,235	26.5
Johns Hopkins University	169	24.1	1,018	21.9
University of Pittsburgh	186	26.5	1,228	26.4
Race/ethnicity: white	653	93.0	3,849	82.7	<.001
Marital status					.12
Married	449	64.1	3,153	67.8
Previously married	224	32.0	1,312	28.2
Never married	28	4.0	186	4.0
Body mass index					.001
Underweight (<18.5)	10	1.4	70	1.5
Normal (18.5–24.9)	301	43.0	1,660	35.8
Overweight (25.0–29.9)	279	39.9	1,939	41.8
Obese (≥30.0)	110	15.7	971	20.9
Cardiovascular heart disease	128	18.2	945	20.3	.20
Stroke	27	3.8	205	4.4	.50
Osteoporosis	84	12.7	300	6.9	<.001
Vision problems	67	9.8	303	6.6	.002
Arthritis: on treatment	186	27.0	1,176	25.7	.49
Using benzodiazepines	86	12.3	399	8.6	.002
Using anticholinergics	3	0.4	25	0.5	.71
Using insulin	12	1.7	109	2.3	.29
ADL difficulties: ≥1	91	13.0	466	10.0	.017
IADL difficulties: ≥1	224	31.9	1,155	24.8	<.001
Gait speed: ≥1.0 m/s	215	30.8	1,554	33.6	.14
Time to complete five chair stands, s					.006
≤10	135	23.6	1,196	30.2
11–12	150	26.2	992	25.1
13–15	177	30.9	1,026	25.9
≥16	110	19.2	746	18.8
Physical activity (kcal)					.54
Quartile 1: <540	169	26.5	1,105	25.9
Quartile 2: 540–1226.24	155	24.3	1,022	24.0
Quartile 3: 1226.25–2549.9	145	22.8	1,081	25.4
Quartile 4: ≥2,550	168	26.4	1,056	24.8

Notes: ADL = Activities of Daily Living; DSST = Digit Symbol Substitution Test; IADL = Independent Activities of Daily Living; and 3MS = Modified Mini-Mental State Examination.

^*Numbers may not add to total due to missing data.

Figure 2.

Kaplan–Meier curve using baseline Digit Symbol Substitution Test quartiles (Quartile 1 [best performance], Quartile 2, Quartile 3, and Quartile 4 [worst performance]).

Lower baseline DSST scores were associated with increased risk of experiencing a serious fall (Table 2). The risk of a serious fall was 50% greater (hazard ratio = 1.48, 95% CI = 1.17–1.86) in participants scoring in DSST Quartile 2 compared with participants in DSST Quartile 1. The risk of a serious fall continued to increase with each quartile decrease in DSST score (p < .001). However, baseline 3MS score was not associated with an increased risk of a serious fall even in the lowest score range (hazard ratio = 1.28, 95% CI = 0.93–1.77). No trend in risk was observed as 3MS score decreased (p = .29). The mean time between the baseline cognitive measurement and incident fall was 7.4 years (standard deviation [SD] = 3.9 years).

Table 2.

Risk Estimates Measuring the Association Between Baseline Cognitive Test Score and a Serious Fall Requiring Hospitalization^*

	Unadjusted (N = 5,356)		Model 1† (N = 5,337)		Model 2‡ (N = 5,333)		Model 3§ (N = 5,333)		Model 4‖ (N = 5,102)
Category	HR	95% CI	HR	95% CI	HR	95% CI	HR	95% CI	HR	95% CI
DSST
Quartile 1¶	1.00	Ref	1.00	Ref	1.00	Ref	1.00	Ref	1.00	Ref
Quartile 2	1.53	1.24–1.90	1.54	1.24–1.92	1.51	1.21–1.88	1.50	1.20–1.87	1.48	1.17–1.86
Quartile 3	1.76	1.42–2.16	1.81	1.45–2.26	1.74	1.39–2.18	1.74	1.39–2.17	1.72	1.37–2.18
Quartile 4	1.96	1.57–2.44	2.25	1.74–2.92	2.09	1.61–2.72	2.07	1.59–2.69	1.91	1.44–2.53
3MS
90–100	1.00	Ref	1.00	Ref	1.00	Ref	1.00	Ref	1.00	Ref
85–89	0.97	0.79–1.19	0.97	0.78–1.20	0.94	0.76–1.17	0.93	0.76–1.16	0.96	0.76–1.20
80–84	0.99	0.75–1.32	0.92	0.69–1.24	0.88	0.65–1.18	0.88	0.65–1.18	0.91	0.67–1.25
<80	1.28	0.98–1.68	1.40	1.04–1.88	1.30	0.96–1.74	1.30	0.97–1.76	1.28	0.93–1.77

Notes: CI = confidence interval; DSST = Digit Symbol Substitution Test; HR = hazard ratio; and 3MS = Modified Mini-Mental State Examination.

^*Results are stratified by gender.

^†Model 1 is adjusted for demographic characteristics.

^‡Model 2 is adjusted for demographic characteristics and medical comorbidities.

^§Model 3 is adjusted for demographic characteristics, medical comorbidities, and prescription drug use.

^‖Model 4 is adjusted for demographic characteristics, medical comorbidities, prescription drug use and measures of physical function.

^¶DSST scores: Quartile 1 = 48–90 (best), Quartile 2 = 40–47, Quartile 3 = 29–39, and Quartile 4 = 0–28 (worst).

In the time-dependent model, every quartile decrease in DSST score was associated with an increased risk of serious falls (p < .001; Table 3). Compared with participants with the best performance, DSST Quartile 1, participants with a DSST score in Quartile 2 were at 1.6 times the risk for a serious fall, participants in Quartile 3 were at 2.5 times the risk, and participants in Quartile 4 were at 2.9 times the risk in the fully adjusted model. As cognition worsened, in DSST quartiles 2–4, adjustment for covariates had a greater effect on the risk estimates. The mean time between final DSST measurement and incident fall was 2.5 years (SD = 2.4 years).

Table 3.

Risk Estimates Measuring the Association Between Cognitive Test Score and a Serious Fall Requiring Hospitalization Using Time-Dependent Covariates^*

	Unadjusted		Model 1†		Model 2‡		Model 3§		Model 4‖
Category	HR	95% CI	HR	95% CI	HR	95% CI	HR	95% CI	HR	95% CI
DSST
Quartile 1¶	1.00	Ref	1.00	Ref	1.00	Ref	1.00	Ref	1.00	Ref
Quartile 2	1.75	1.32–2.31	1.80	1.35–2.38	1.74	1.32–2.31	1.73	1.31–2.31	1.58	1.15–2.17
Quartile 3	2.72	2.12–3.51	2.87	2.21–3.74	2.74	2.10–3.57	2.73	2.09–3.55	2.53	1.88–3.41
Quartile 4	3.57	2.79–4.58	4.01	3.04–5.28	3.64	2.75–4.82	3.60	2.71–4.76	2.90	2.08–4.03
Sample size		5,356		5,337		5,333		5,333		5,102
No prevalent CHD at baseline
3MS
90–100	1.00	Ref	1.00	Ref	1.00	Ref	1.00	Ref	1.00	Ref
85–89	1.46	1.15–1.87	1.53	1.19–1.97	1.47	1.14–1.89	1.47	1.14–1.89	1.49	1.12–1.98
80–84	1.53	1.12–2.10	1.66	1.20–2.31	1.57	1.13–2.19	1.56	1.12–2.17	1.39	0.94–2.08
<80	2.88	2.35–3.55	2.90	2.30–3.66	2.67	2.09–3.40	2.65	2.08–3.38	2.16	1.60–2.91
Sample size		4,277		4,264		4,260		4,260		4,105
Prevalent CHD at baseline
3MS
90–100	1.00	Ref	1.00	Ref	1.00	Ref	1.00	Ref	1.00	Ref
85–89	1.32	0.82–2.14	1.12	0.73–1.96	1.17	0.71–1.92	1.17	0.71–1.93	1.04	0.53–2.05
80–84	2.21	1.35–3.61	1.93	1.14–3.24	1.76	1.03–2.99	1.77	1.04–3.02	1.96	1.00–3.81
<80	1.30	0.79–2.18	1.12	0.64–1.96	1.01	0.57–1.79	1.01	0.57–1.79	1.23	0.60–2.53
Sample size		1,079		1,073		1,073		1,073		997

Notes: CI = confidence interval; DSST = Digit Symbol Substitution Test; HR = hazard ratio; and 3MS = Modified Mini-Mental State Examination.

^*Results are stratified by gender.

^†Model 1 is adjusted for demographic characteristics.

^‡Model 2 is adjusted for demographic characteristics and medical comorbidities.

^§Model 3 is adjusted for demographic characteristics, medical comorbidities, and prescription drug use.

^‖Model 4 is adjusted for demographic characteristics, medical comorbidities, prescription drug use, and measures of physical function.

^¶DSST scores: Quartile 1 = 48–90 (best), Quartile 2 = 40–47, Quartile 3 = 29–39, and Quartile 4 = 0–28 (worst).

When considering 3MS score as a time-dependent covariate, the risk of experiencing a serious fall differed for those with and without prevalent CHD at study enrollment (p < .01). Among participants without prevalent CHD at baseline, the risk of a serious fall increased about 45% for participants scoring between 80 and 89 when compared with participants with the best 3MS performance (≥90); the risk was 2.16 times (95% CI = 1.60–2.91) greater for participants scoring below 80. However, in participants with prevalent CHD at baseline, 3MS score was not associated at any score level with an increased risk of serious falls. The mean time between final 3MS measurement and incident fall was 2.3 years (SD = 2.4 years).

DISCUSSION

A decrease in cognition was associated with increased risk of being hospitalized for a serious fall. In both baseline and time-dependent analyses, as processing speed (DSST score) decreased, the risk of experiencing a serious fall increased. The risk for a serious fall was greater among participants with lower cognition in the time-dependent analysis compared with the baseline analysis. This suggests that regular measurements of cognition are more important than a single measurement and may indicate those on the path to worsening cognitive outcomes, including imminent serious falls and dementia.

A similar, albeit weaker, relationship was observed between global cognition (3MS score) and the risk of serious falls in the time-dependent analysis of individuals without prevalent CHD only. It is possible that decreased cognition was not a risk factor for serious falls among individuals with CHD because these participants were receiving more frequent medical care, leading practitioners to identify risk factors for falling earlier, and prompting interventions to prevent serious injury. Another possibility is that participants in this sicker group were taking medications that also increase fall risk (31,32), subsequently blunting the importance of decreased cognition as a risk factor for serious falls.

Previous studies examining the relationship between cognition and serious falls have focused on cognitive impairment. No previous studies have considered the effects of decreased cognition on serious fall risk. In a population of community-dwelling elderly adults, cognitive impairment was observed to be a risk factor for injurious falls (odds ratio [OR] = 2.2, 95% CI = 1.5–3.2) (14). Impaired executive function was associated with major injurious falls (OR = 1.9, 95% CI = 1.1–3.2) but not minor injurious falls in another elderly population (12). A Norwegian study in elderly women observed an association between cognitive impairment and fracture but not serious falls (11). However, this relationship may be modified by functional status: an increased likelihood of fall-related injuries has been observed among individuals with no ADL difficulties (OR = 2.0, 95% CI = 1.3–3.3), but not among those with ADL difficulties (13).

The relationship between cognition, physical function, and serious fall risk is complicated. Dementia may lead to decreases in physical functioning (33,34), and functional decline is associated with an increased fall risk (35). However, physical impairment has also been identified as a mediator between cognitive impairment and falls (36), but not specifically serious falls. We attempted to examine the role of physical function in this association by controlling for ADLs, Independent Activities of Daily Living, gait speed, completed repeated chair stands, and physical activity. Our risk estimates were slightly attenuated after the inclusion of these variables in the model (Model 4), but a significant association remained in the DSST model. However, other studies have shown that functional decline may precede decreases in cognitive function in the nondemented elderly (15,37). We identified decreased cognition as a risk factor for falling; decreased processing speed, a specific cognitive domain, appears to be an especially strong risk factor.

We suggest that cognitive function and physical function are interrelated, with each having a strong effect on the other; declining cognition incites declines in physical function and vice versa. In addition, cognition and physical function are each independent risk factors for serious falls. Impairments in physical function may increase the probability of a fall; coupled with increasing frailty in the elderly, falls may more often end with serious consequences. A small study observed that cognitive function is a better predictor than physical function of ability to complete complex tasks (38). We speculate that with the development of dementia, however, the role of physical function as a mediator may become more important, leading physical impairment to be a stronger risk factor than cognitive impairment. Prevention strategies targeting either one of these risk factors may also prevent declines in the other. Studies have shown that older adult participants with higher levels of physical activity have a reduced risk of cognitive decline (39,40) and dementia (41). A large, behavior-based, randomized control trial reported encouraging, but limited, results that preserving cognitive function may also prevent functional decline (42).

This study was subject to several limitations. First, by selecting only the most serious falls, we addressed a very specific subset of the elderly who experience a fall. However, we were able to characterize this subset very well with detailed review of their medical records. Second, our outcome definition and data sources may be responsible for slight misclassification of study participants. Because we identified serious falls based on the first three International Classification of Diseases—Clinical Modification diagnosis codes in the hospitalization records, we may have included or excluded a small number of events based on erroneous coding. Third, other studies have linked decreases in executive function (43) and memory (36) with noninjurious falls in the elderly. However, the CHS cohort was evaluated longitudinally with measures for global cognition and processing speed only; thus, we were unable to completely explore all pathways by which falls and cognition may be connected. Finally, it has been suggested that simple measures of physical function (eg, gait speed) are less sensitive to small differences in physical functioning than more complex measures, such as walking through an obstacle course (15,44). In addition, we were unable to adjust for balance, a known risk factor for falls in the elderly, as it was not measured at the baseline clinic visit. However, the small change in our risk estimates after adjustment for physical function suggests that even the inclusion of more complicated measures is unlikely to completely attenuate the association observed.

The participants of the CHS are a large group, who represent a diverse segment of our nation's community-dwelling elderly. They have been well characterized on hundreds of variables and have demonstrated high retention for more than 15 years. In addition, great care has been taken to ensure quality data on medication use, hospitalization, and prevalent disease status.

Falling results in substantial morbidity and mortality in older adults. Early identification and prevention efforts in this population have the potential to significantly reduce pain, suffering, medical costs, and comorbidities. This study suggests that the inclusion of simple cognitive tests, in addition to commonly collected health information, helps to identify persons at increased risk for serious falls. Whether cognitive function is a mediator between physical function and serious falls or part of a completely separate pathway remains to be determined. Additional studies with more complex measures of physical function and cognitive function may help to elucidate this relationship.

FUNDING

National Heart, Lung, and Blood Institute (N01-HC-85079, N01-HC-85080, N01-HC-85081, N01-HC-85082, N01-HC-85083, N01-HC-85084, N01-HC-85085, N01-HC-85086, N01-HC-35129, N01 HC-15103, N01 HC-55222, N01-HC-75150, N01-HC-45133, U01 HL080295).