JAGS 50:1572–1576, 2002© 2002 by the American Geriatrics Society 0002-8614/02/$15.00

Validity of Divided Attention Tasks In Predicting Falls in Older Individuals: A Preliminary Study

Joe Verghese, MD, MS,* Herman Buschke, MD,* Lisa Viola, DO,* Mindy Katz, MPH,*Charles Hall, PhD,

†

Gail Kuslansky, PhD,* and Richard Lipton, MD*

†‡

OBJECTIVES:

Although cognitive impairment is knownto be a major risk factor for falls in older individuals, therole of cognitive tests in predicting falls has not been es-tablished. Limited attentional resources may increase therisk for falls in older individuals. We examined the reli-ability and validity of divided attention tasks, walkingwhile talking (WWT), in predicting falls.

DESIGN:

A prospective cohort study of 12-months’ dura-tion.

SETTING:

Community-based longitudinal aging study,the Einstein Aging Study.

PARTICIPANTS:

Sixty nondemented community-livingsubjects, aged 65 to 98 (mean age

�

standard deviation

�

79.6

�

6.3).

MEASUREMENTS:

Simple and complex versions of theWWT task in addition to standard balance and cognitiveassessments. The primary outcome was cumulative inci-dence of falls at 12 months.

RESULTS:

Thirteen subjects fell over the 12 months, fourof whom had major injuries. The WWT task had good in-terrater reliability (

r

�

0.602,

P

�

.001). Poor performanceon simple (odds ratio (OR)

�

7.02, 95% confidence inter-val (CI)

�

1.7–29.4) and complex WWT tasks (OR

�

13.7, 95% CI

�

2.3–83.6) was highly predictive of falls.The simple task had a sensitivity of 46% and specificityof 89%. For the complex task, sensitivity was 39%, andspecificity was 96%.

CONCLUSIONS:

The WWT is a reliable and valid test toidentify older individuals at high risk for falls. Future stud-

ies with larger sample sizes and in different settings areneeded to confirm the findings of this study.

J Am GeriatrSoc 50:1572–1576, 2002.Key words: divided attention; walking; older people; falls;

screening

A

s many as one-third of community-living older indi-viduals fall every year, leading to loss of mobility,

early institutionalization, and death.

1,2

Many medical andenvironmental risk factors have been identified for falls.

2,3

Recent guidelines on fall prevention recognize cognitiveimpairment as a leading risk factor for falls.

3

In the ab-sence of previous studies, identification of specific cogni-tive processes most strongly associated with falls and opti-mal methods to assess and treat cognitive risk factorsremain to be determined. Because individuals often engagein other activities while walking, walking in the real worldinvolves divided attention. With aging or dementia, lim-ited attentional resources may decrease the ability of olderpeople to perform two or more simultaneous tasks, in-creasing the risk for falls.

The multiplicity of risk factors makes it unlikely that asingle screening test will identify all at-risk older people.Tinetti et al. showed that interventions based on specificmechanisms reduced rates of falls, highlighting the clinicalrelevance of the etiological heterogeneity of falls.

4,5

Identi-fying high-risk older people, as a prelude to intervention,is an important strategy. Previous studies have used vari-ous gait and mobility tests to predict falls. Some usefulresearch procedures are complex or require specializedequipment and training, limiting their applicability to rou-tine clinical practice.

3–7

Moreover, few tests have been pro-spectively validated, and none focus on the effect of cogni-tive impairment on falls.

3–8

Exploring the influence of cognitive processes on gaitmay help clarify a mechanism of falls and identify a sub-group of older individuals at risk. Divided attention tasksinvolving simultaneous performance of cognitive and man-ual tasks have been used to study attentional mechanisms.

9

Based on the hypothesis that limited attentional resources

From the *Departments of Neurology and

†

Epidemiology and Social Medi-cine, Albert Einstein College of Medicine, Bronx, New York; and

‡

IMR: A Division of Advanced PCS and the Center for Healthy Aging: Advanced PCS, Stamford, Connecticut.

The Einstein Aging Study is supported by National Institute on Aging, Na-tional Institutes of Health Grant AGO3949. Presented in part at the 53rd annual meeting of the American Academy of Neurology, Philadelphia, Penn-sylvania, May 9, 2001. Presented in part as thesis requirement for the Mas-ters Degree in Clinical Research Methods, Albert Einstein College of Medi-cine, New York (JV).

Address correspondence to Joe Verghese, MD, Einstein Aging Study, Albert Einstein College of Medicine, 1165 Morris Park Avenue, Room 338, Bronx, New York 10461. E-mail: jverghes@aecom.yu.edu

JAGS SEPTEMBER 2002–VOL. 50, NO. 9

DIVIDED ATTENTION TASKS AND FALLS

1573

in older persons may increase risk for falls, we devised adivided attention task, walking while talking (WWT), toidentify individuals at high risk for falls. The purpose ofthis preliminary study was to evaluate the validity ofWWT tests in predicting falls in community-dwelling olderindividuals.

METHODS

Enrollment

We enrolled 60 consecutive participants aged 65 and olderfrom a community-based longitudinal aging study, theEinstein Aging Study (EAS). EAS participants were ran-domly recruited from Health Care Finance Administrationpopulation lists of Medicare recipients in the Bronx. EASexclusion criteria include severe visual loss interfering withcompletion of neuropsychological tests, non-English ornon-Spanish speaking, and institutionalization. Subjectshad to be healthy enough to make a clinic visit, but pres-ence of medical illnesses was not used as an exclusion cri-terion. Details of medication use and medical illnesses, in-cluding presence or absence of various risk factors forfalls, were prospectively collected from subjects at the timeof enrollment with structured questionnaires. EAS subjectsreceive detailed neurological and neuropsychological eval-uations at enrollment and annual follow-up visits. Generalcognitive status was assessed with the Blessed Informa-tion-Memory-Concentration test (BIMC).

10

None of thesubjects in this study met dementia criteria of the

Diagnos-tic and Statistical Manual of Mental Disorders, 4th Edi-tion.

Of the 60 subjects, 34 (57%) were women, and 22(37%) lived alone. Seventeen subjects (28%) had had afall in the year preceding enrollment, 12 (20%) reportedneuropathy, and two (3%) had had previous strokes.Other reported risk factors for falls, such as foot deformi-ties, use of prosthetic devices, depression, alcoholism, Par-kinson’s disease, or use of sedatives or antidepressantmedications, were not present in our subjects when as-sessed at the baseline medical interview or clinical exami-nation.

Mobility Assessment

Mobility and balance were assessed in all subjects in thisstudy at enrollment using the following tests.

Tinetti Balance and Mobility Scale

The Tinetti balance and mobility scale (TBMS) was devel-oped to screen for balance and mobility skills in older peo-ple and to determine risk for falls.

4,5

It uses nine tasks as-sessing mobility and balance in sitting and standingpositions, generating scores from 0 to 16. The TBMS issometimes combined with a gait assessment, not usedherein, to improve assessment of fall risk.

5,6

Timed Gait

The timed gait measures the time taken (in seconds) towalk 20 feet, turn, and return (40 feet total), at the sub-ject’s normal walking pace.

Walking While Talking Tasks

Subjects were timed walking the same course as the timedgait (40 feet), while reciting the letters of the alphabet

aloud (WWT-simple). For the WWT-complex task, sub-jects recited alternate letters of the alphabet (a, c, e . . . )while walking. The order of the WWT and TBMS taskswas randomly varied to avoid systematic bias. All subjectsrecited the alphabet in English, including five who were bi-lingual in English and Spanish.

Interrater reliability for the WWT-simple was assessedin eight subjects, comparing evaluations performed by aneurologist and a neuropsychology assistant who wereblinded to the study aims.

Outcome and Follow-Up

Cumulative incidence of falls at 12 months was the primaryoutcome. Falls were defined as sudden, unintentional, un-provoked changes in body posture resulting in the personbeing on a lower level, not due to a major intrinsic event(stroke) or overwhelming hazard.

5

Falls were classified asmajor if they resulted in injury. History of falls in the yearpreceding enrollment was obtained. To prospectively ascer-tain presence and circumstances of falls, research staff whowere blinded to test performance conducted structuredphone interviews with subjects at 6 and 12 months after en-rollment. If subjects were unable to give precise date of fall,1-week windows were used. Prospectively identified fallswere reconfirmed with a significant other when available(one-third of subjects lived alone), with medical records, orat EAS follow-up visits. The subjects were reliable; all sixsubjects who reported falls at the 6-month interview re-called the fall at 12 months. Seven additional subjects re-ported a fall at 12 months, but not at 6 months.

Statistical Analysis

Chi-square tests were used to compare categorical vari-ables, with Fisher exact tests applied as appropriate.Mann-Whitney

U

tests and independent-samples

t

testswere used for continuous variables. All tests were two-tailed, with a criterion level of .05. Validity was comparedby computing sensitivity, specificity, and positive predic-tive value (PPV) for the tests. (Sensitivity is the proportionof individuals with falls who have a positive test result.Specificity is the proportion of individuals without fallsthat have a negative test result. Positive predictive value isthe proportion of individuals who have a positive test re-sult and have the disease.) Because of our small sample, weused an a priori strategy to identify a subgroup of individ-uals with poor WWT performance that would be largeenough for statistical analysis. Cutoff test scores were cho-sen at one standard deviation from the group mean. Re-sultant scores were 18 seconds or longer for timed gait(mean

�

standard deviation

�

14.8

�

3.6), 20 seconds orlonger for WWT-simple (16.3

�

4), 33 seconds or longerfor WWT-complex (22

�

11.4), and 10 or fewer pointsfor the TBMS (11.2

�

1.4). To assess the influence of thischoice, we performed post hoc analyses to examine a rangeof cutoff scores on these tests identified by receiver operat-ing characteristic (ROC) techniques. Odds ratios (ORs) with95% confidence intervals (CIs) were computed to study as-sociation between tests and falls. Survival analysis for WWTtests at selected cutoff scores was performed using theKaplan-Meier method, and survival curves compared us-ing log rank tests. Subjects were censored when they fell orwhen they completed the 12-month follow-up.

1574

VERGHESE ET AL.

SEPTEMBER 2002–VOL. 50, NO. 9 JAGS

RESULTS

Baseline

There were no significant differences in sex (58.8% vs55.8% women) or living arrangement (47% vs 33% livingalone) between the 17 subjects with a history of falls in theyear preceding enrollment and the 43 subjects withoutfalls. Subjects with previous falls were slightly older (82

�

6.9 vs 78.7

�

5.9,

P

�

.07) and had higher BIMC scores(3.7

�

1.6 vs 2.4

�

1.8,

P

�

.02). Older people with neur-opathy were more likely to have had falls in the year be-fore enrollment (47% vs 9%,

P

�

.002). Older peoplewith previous falls walked more slowly at baseline (17.1

�

3.7 vs 13.9

�

3.4 seconds,

P

�

.002). They took longer tocomplete the WWT-simple (20.2

�

8 vs 15.2

�

3.9 sec-onds,

P

�

.002) and the WWT-complex (28.9

�

11.1 vs20.1

�

10.9 seconds,

P

�

.001) and had lower scores onthe TBMS (10.1

�

1.9 vs 12.3

�

2.3,

P

�

.001).

Follow-Up

One subject died during the follow-up period and was ex-cluded from the univariate but not the survival analysis.Of the remaining 59 subjects, six (10.2%) fell by 6 monthsand 13 (22%) by 12 months. Environmental hazards, suchas uneven flooring or icy pavements, were cited as thecause of falls in most cases (62%). Seven falls (54%) oc-curred at home. Four (31%) falls were major, resulting infractures of the pelvis (n

�

1) and foot (n

�

1) and exten-sive bruising (n

�

2). Two subjects had syncopal episodesnot recorded as falls. Table 1 shows that there were no sig-nificant differences in age, sex, or self-reported height be-tween fallers and nonfallers. Older people with neuropa-thy did not have more prospectively identified falls. Fallershad higher baseline BIMC scores (3.6

�

1.9 vs 2.5

�

1.8,

P

�

.06) and walked slower at baseline and during WWTtasks. Baseline TBMS scores were not significantly differ-ent (

P

�

.06).

Interrater Reliability

There was good correlation (

r

�

0.602,

P

�

.001) betweenthe assessments done by the clinician and an independentneuropsychology assistant on the WWT-simple test.

Validity

Baseline walking speed had good specificity (84.7%) butmodest sensitivity (38.4%) and PPV (41.7%). WWT-com-plex had the highest specificity (95.6%) but modest sensi-tivity (38.5%) at the selected cutoff scores (Table 2).WWT-simple had better sensitivity (46.1%) but lowerspecificity (89.4%) and PPV (54.5%). The TBMS hadmodest sensitivity (61.5%) and specificity (69.5%) butpoor PPV (36.4%). Validity varies as a function of the cut-off score; for instance, selecting extreme scores (

�

26 sec-onds) on WWT-complex improves sensitivity (54%) at theexpense of specificity (85%).

Table 2 presents validity of a range of cutoff scoresidentified by the ROC technique. We assessed baselineWWT performance as a predictor of falls over 12 months.Poor performance on WWT-simple (OR

�

7.02, 95% CI

�

1.7–29.4) and WWT-complex tasks (OR

�

13.7, 95% CI

�

2.3–83.6) was strongly predictive of falls. Kaplan-Meierplots of the WWT-simple (Figure 1) and WWT-complex(Figure 2) show significant separation by the end of thestudy (

P

�

.001). Fifty-five percent of subjects failing theWWT-simple fell, compared with the 15% of older peoplewho did not exceed this score. Of subjects scoring over thecutoff on the WWT-complex, 71% fell, compared with15% of controls.

Clinical Observations

All subjects completed the WWT tasks. The subjects foundthe instructions easy to follow. In general, all subjectsslowed down while doing the WWT task, especially thecomplex version. No one fell during the task. Four sub-jects stopped walking during the WWT-complex of whomthree fell. Many subjects who slowed down were alsonoted to make errors in reciting alternate alphabets.

DISCUSSION

Our preliminary study shows that WWT tests combiningcognitive and gait tasks powerfully predict falls in nonde-mented, community-living older people. There were strik-ing mean differences between older subjects who did anddid not have prospectively identified falls in baseline walk-ing time and in the performance of WWT tasks. In con-

Table 1. Comparison of Baseline Characteristics in Prospectively Identified Older Subjects with and without Falls over 12 Months

VariableFallers

(n

�

13)Nonfallers(n

�

46)

P

-value

Age, mean

�

SD 79.4

�

5.7 79.7 � 6.6 .9Female, n (%) 9 (69.2) 25 (54.3) .5Height, inches, mean � SD 63.6 � 4.7 64.2 � 4.4 .7Neuropathy, n (%) 3 (23) 8 (17.4) .7Prior falls at baseline, n (%) 9 (69.2) 7 (15.2) �.001BIMC score at entry, mean � SD 3.6 � 1.9 2.5 � 1.8 .08*Timed gait (40 feet), sec, mean � SD 17.5 � 4.1 14.1 � 3.4 �.001*WWT-simple time, sec, mean � SD 20.2 � 8 15.2 � 3.9 .037*WWT-complex time, sec, mean � SD 28.9 � 11.1 20.1 � 10.9 �.001*TBMS, points, mean � SD 10.6 � 2 11.9 � 2.5 .064*

*Mann-Whitney U test.BIMC: Blessed Information-Memory-Concentration test (worst score 32); WWT: walking while talking test; TBMS: Tinetti balance and mobility scale (range 0–16).

JAGS SEPTEMBER 2002–VOL. 50, NO. 9 DIVIDED ATTENTION TASKS AND FALLS 1575

study designs and tests. A recent functional imaging studyreported that total brain activation measured during adual task was significantly less than the sum of activationwhen the tasks were performed individually, suggesting abiological limit to the total activation possible at a giventime.11 Hence, walking while talking may limit resourcesavailable for either activity.

Over a year, poor performance on the WWT-simplecaptured 55% of falls with high specificity (89%). Of theseven high-risk older people identified by WWT-complex,five fell in the following year. The progressive increase inPPV, from 42% for timed gait (OR � 4.3) to 55% forWWT-simple (OR � 7), and 71% for WWT-complex

Table 2. Validity of Timed Gait, Tinetti Balance and Mobility Scale (TBMS), Walking While Talking (WWT)-Simple, andWWT-Complex Tests for Identifying Falls at 12 Months for a Range of Scores

Test Sensitivity Specificity Positive Predictive Value Odds Ratio (95% confidence interval) P-value

TBMS�8 7.6 91.3 20.0 .87 (.09–8.7) 1�9 23.0 80.4 25.0 1.2 (.28–5.43) .7�10* 61.5 69.5 36.4 3.5 (1.01–13.2) .055

Timed gait�12 sec 100.0 23.9 27.1 1.37 (1.15–1.63) .1�14 sec 76.9 56.5 33.3 3.6 (.87–13.8) .1�18 sec* 38.4 84.7 41.7 4.3 (1.05–17.85) .057

WWT-simple�16 sec 53.8 60.0 28.0 1.7 (.58–6.06) .53�18 sec 46.2 76.1 35.3 2.7 (.76–9.84) .16�20 sec* 46.1 89.4 54.5 7.02 (1.7–29.4) .009

WWT-complex�26 sec 53.8 84.7 50.0 6.5 (1.67–25.2) .008�30 sec 38.5 91.3 55.5 6.6 (1.4–29.9) .019�33 sec* 38.5 95.6 71.4 13.7 (2.3–83.6) �.001

*Preselected cutoff score.

Figure 1. Kaplan-Meier survival plot for the probability of nothaving had a fall after 1 year (52 weeks) based on performanceon the walking while talking (WWT)-simple task. The cutoffscore was 20 seconds or longer (see Methods).

trast, performance on cognitive (BIMC) and balance (TBMS)tasks did not show significant group differences at base-line. The TBMS had higher sensitivity but lower specificitythan either form of WWT in our sample. A previous studyreported a sensitivity:specificity ratio of 70:52 for thecombined Tinetti gait and balance scale in predicting fallsover a year in 225 community-living older people.6

Prior research on cognitive/motor tasks and falls issparse. Shumway-Cook et al. reported that adding a cog-nitive task (counting backwards by three) to a balance as-sessment (timed up-and-go test) did not improve the abil-ity to identify fallers at cross-section.8 A direct comparisonwith our study is not possible because of the differences in

Figure 2. Kaplan-Meier survival plot for the probability of nothaving had a fall after 1 year (52 weeks) based on performanceon the walking while talking (WWT)-complex task. The cutoffscore was 33 seconds or longer (see Methods).

1576 VERGHESE ET AL. SEPTEMBER 2002–VOL. 50, NO. 9 JAGS

(OR � 13.7), demonstrates the incremental validity ofWWT tests over timed gait. These findings suggest that theprocesses involved by the WWT are etiologically relevant inthe causation of falls. Both WWT tests identified high-riskgroups for possible intervention. Because falls are etiologi-cally heterogeneous, modest sensitivity is expected. TheWWT tests are unlikely to identify people who fall becauseof unrelated mechanisms.

Limitations mainly relate to the preliminary nature ofour study. Our findings need to be validated by futurestudies with larger samples and in other settings. A higherfall rate will be expected in nursing homes or clinics. Thecutoff scores were chosen to optimally differentiate perfor-mance stratified by fall risk, and the values need to be val-idated in larger samples. Our post hoc analysis shows thatchanging scores alters test characteristics. The WWT cut-off scores reflect the nature of our sample, and differentscores and distances might be used and validated in othersettings. We had a lower incidence of falls than previousstudies, which have used fall diaries and mail-back monthlypostcard calendars.5 Even in these studies, recall bias is aproblem because cognitively impaired older people mayunderreport falls. Our nondemented subjects had minimalrecall bias, as suggested by the correspondence betweenthe 6- and 12-month falls data. Our small sample did notinclude older people with dementia, who have a higherrate of falls and more difficulty completing dual tasks.3,9,12

Six subjects had mild cognitive abnormalities as defined bya BIMC score of 6 or higher. Subjects who fell had nonsig-nificantly higher BIMC scores (P � .08). This study fo-cused on the influence of the cognitive task on gait and notthe effect of gait on the cognitive task. Previous studiesand our own observations indicate that the simultaneousperformance of two tasks leads to deterioration in bothsubtasks.9,11,12 Nursing home residents who stopped totalk were reported to have more falls.13 It may be useful toinclude qualitative differences in task performance in fu-ture refinements of the WWT.

This is the first study to show that performance on di-vided attention tasks can predict falls in older people. Ourintention was not to replace current screening tests such asthe TBMS but to extend the scope of current fall assess-

ments by adding cognitive parameters.3 For instance, com-bining the TBMS with WWT-simple improves the sensitiv-ity to 71% with fair specificity (70%). Future strategies inclinical or community settings might include fall question-naires to detect at-risk older people with previous falls or ahistory of dementia followed by cognitive (WWT) or bal-ance tests (TBMS) to identify modifiable risk factors. TheWWT offers the chance to introduce specific interventionssuch as cognitive rehabilitation or pharmacological treat-ments. WWT tasks are simple, reliable, and take less thana minute to perform. If our results are confirmed, WWTtasks can be easily incorporated into standard clinical ex-aminations to identify older people at high risk for falls.

REFERENCES1. Sattin RW. Falls among older persons: A public health perspective. Annu Rev

Public Health 1992;13:489–508.2. Tinetti M, Speechley M, Ginter S. Risk factors for falls among elderly per-

sons living in the community. N Engl J Med 1988;319:1701–1707.3. American Geriatrics Society, British Geriatrics Society, and American Acad-

emy of Orthopedic Surgeons Panel on Falls Prevention. Guideline for the pre-vention of falls in older persons. J Am Geriatr Soc 2001;49:664–672.

4. Tinetti ME, Williams TF, Mayeweski R. A fall risk index for elderly patientsbased on number of chronic diseases. Am J Med 1986;80:429–434.

5. Tinetti ME, Baker DI, McAvay G et al. A multifactorial intervention to re-duce the risk of falling among elderly people living in the community. N EnglJ Med 1994;331:821–827.

6. Raiche M, Hebert R, Prince F et al. Screening older adults at risk of fallingwith the Tinetti balance scale. Lancet 2000;356:1001–1002.

7. Brauer SG, Burns YR, Galley P. A prospective study of laboratory and clini-cal measures of postural stability to predict community-dwelling fallers. JGerontol A Biol Sci Med Sci 2000;55A:469–476.

8. Shumway-Cook A, Brauer S, Woollacott M. Predicting the probability forfalls in community-dwelling older adults using the Timed Up & Go Test.Phys Ther 2000;80:896–903.

9. Della Sala S, Baddeley A, Papagno C et al. Dual-task paradigm: A means toexamine the central executive. Ann N Y Acad Sci 1995;769:161–171.

10. Blessed G, Tomlinson E, Roth M. The association between quantitative mea-sures of dementia and of senile change in the cerebral grey matter of elderlysubjects. Br J Psychiatry 1968;114:797–811.

11. Just MA, Carpenter PA, Keller TA et al. Interdependence of nonoverlappingcortical systems in dual cognitive tasks. Neuroimage 2001;14:417–426.

12. Camicioli R, Howeison D, Lehman S et al. Talking while walking: The effectof a dual task in aging and Alzheimer’s disease. Neurology 1997;48:955–958.

13. Lundin-Olsson L, Nyberg L, Gustafson Y. “Stops walking when talking” asa predictor of falls in elderly people. Lancet 1997;349:617.