Quantitative gait dysfunction and risk of cognitive decline and
Joe Verghese, Cuiling Wang, Richard B Lipton, Roee Holtzer, Xiaonan Xue
............................................................... ............................................................... .....
See end of article for
Dr Joe Verghese, Einstein
Ageing Study, Albert
Einstein College of Medicine,
1165 Morris Park Avenue,
Room 338, Bronx, New York
10461, USA; jverghes@
Revised 26 December 2006
Accepted 9 January 2007
Published Online First
19 January 2007
J Neurol Neurosurg Psychiatry 2007;78:929–935. doi: 10.1136/jnnp.2006.106914
Background: Identifying quantitative gait markers of preclinical dementia may lead to new insights into early
disease stages, improve diagnostic assessments and identify new preventive strategies.
Objective: To examine the relationship of quantitative gait parameters to decline in specific cognitive domains
as well as the risk of developing dementia in older adults.
Methods: We conducted a prospective cohort study nested within a community based ageing study. Of the
427 subjects aged 70 years and older with quantitative gait assessments, 399 were dementia-free at
Results: Over 5 years of follow-up (median 2 years), 33 subjects developed dementia. Factor analysis was
used to reduce eight baseline quantitative gait parameters to three independent factors representing pace,
rhythm and variability. In linear models, a 1 point increase on the rhythm factor was associated with further
memory decline (by 107%), whereas the pace factor was associated with decline on executive function
measured by the digit symbol substitution (by 29%) and letter fluency (by 92%) tests. In Cox models adjusted
for age, sex and education, a 1 point increase on baseline rhythm (hazard ratio (HR) 1.48; 95% CI 1.03 to
2.14) and variability factor scores (HR 1.37; 95% CI 1.05 to 1.78) was associated with increased risk of
dementia. The pace factor predicted the risk of developing vascular dementia (HR 1.60; 95% CI 1.06 to
Conclusion: Our findings indicate that quantitative gait measures predict future risk of cognitive decline and
dementia in initially non-demented older adults.
dementia by many years.1
dementia may help identify high risk elderly patients for
further evaluation and interventions. We previously reported in
another cohort (Bronx Ageing Study)2that clinical gait
abnormalities predicted risk of non-Alzheimer’s dementia in
older adults. While an integral aspect of patient evaluation,
clinical gait assessments have several limitations. Most assess-
ment protocols are not standardised or validated. Most gait
abnormalities are mild,3and detection is dependent on the
examiner’s expertise. Clinicians may use the presence of gait
abnormalities to assign dementia subtypes raising issues of
diagnostic circularity.2 4Quantitative gait assessments, inde-
pendent of clinical diagnosis, may help avoid these short-
It has been reported that slowing of gait may precede
development of cognitive impairment.5–8However, gait is a
complex motor behaviour with many measurable facets besides
velocity, and with an intricate relationship to different aspects
of cognition.9Moreover, single gait variables are often highly
correlated with one another so that their independent effects on
risk of cognitive decline and dementia may be hard to observe
while adjusting for other gait variables. To address this issue,
we used factor analysis to identify independent gait domains
derived from quantitative assessments.9 10Gait variability has
been linked to multiple adverse outcomes in older adults,
including Alzheimer’s disease.11 12However, its role in predict-
ing dementia is not established. Hence we included gait
variability measures in our analyses.
Based on our and other studies2 5–9we hypothesised that
quantitative gait assessments may help reveal subtle alterations
in brain function early in the course of the dementia. Our aim
ementia is widely recognised as a global public health
problem. There is increasing evidence that subtle clinical
and physiological abnormalities precede the diagnosis of
Identifying early markers of
was to examine the relationship of quantitative gait parameters
with decline in general and specific cognitive domains in a
population of non-demented older adults. We also studied
whether quantitative gait parameters could predict risk of
incident dementia. Identifying quantitative gait markers of
preclinical dementia may provide new insights into early
biological stages of dementia, improve diagnosis and risk
assessment procedures, and facilitate development of novel
We undertook a prospective cohort study nested within the
Einstein Ageing Study.3 13 14The primary aim of the Einstein
Ageing Study was to identify risk factors for dementia. Study
design and methods have been previously reported.3 13 14In
brief, potential subjects (age 70 years and over) identified from
population lists of Bronx County were contacted by letter
explaining the purpose and nature of the study, and then by
telephone. The telephone interview included verbal consent,
medical history questionnaire and cognitive screening tests.13
Following the interview, an age stratified sample of subjects
who matched on a computerised randomisation procedure were
invited for further evaluation at our research centre. Informed
consent was obtained at enrolment according to protocols
approved by the local institutional review board. Subjects
returned at yearly intervals. Between 1993 and 2005, 1148
subjects were enrolled. Mean age at entry was 77.4 (5.2) years.
The inception cohort was mostly women (59%) and 67% were
white subjects (black 27%).
Abbreviation: HR, hazard ratios
Quantitative gait evaluations were introduced in 2001. Of the
510 subjects seen between 2001 and 2005, 427 (84%) had
quantitative assessments. Reasons for not obtaining assess-
ments included tester unavailability (n=53), subject illness
(n=10) or refusal (n=20). Subjects who did and did not
receive quantitative assessments were similar in terms of age,
sex, education and cognitive status at enrolment. We report
follow-up until March 2006.
Research assistants conducted quantitative studies, indepen-
dent of clinician’s evaluations, using a computerised walkway
(180635.560.25 inches) with embedded
(GAITRite; CIR systems, Havertown, PA, USA).14Subjects were
asked to walk on the mat at their ‘‘normal pace’’ for two trials
in a quiet well-lit hallway wearing comfortable footwear and
without any attached monitoring devices. Start and stop points
were marked by white lines on the floor, and included three
feet from the walkway edge for initial acceleration and terminal
deceleration. Based on footfalls recorded on the walkway, the
software automatically computes gait parameters (see table 1)
as the mean of two trials. The GAITRite system is widely used in
clinical and research settings, and excellent reliability has been
reported in our and other centres.14 15
An extensive neuropsychological test battery validated in our
and other ageing populations was administered at all visits to
all subjects to assess cognition and assign dementia diagno-
sis.2 13We examined performance on the following tests based
on associations between gait and cognition noted in our and
other studies5–9: general cognition (Blessed Information–
Memory–Concentration Test16), memory (Free and Cued
Selective Reminding Test17), executive function (Digit Symbol
Substitution18and Letter Fluency Tests),19and attention (Digit
Data collected at each visit from subjects and caregivers
included sociodemographic variables (age, sex and education),
medications and depressive symptoms.3 13 20Research assistants
administered the Lawton–Brody scale21to assess limitations on
activities of daily living. Study clinicians also obtained a history
of functional decline during the clinical evaluation.2 9 13The
presence of diabetes, heart failure, hypertension, angina,
myocardial infarction, depression, stroke, Parkinson disease,
chronic obstructive lung disease and arthritis was used to
calculate a summary comorbidity index, as previously described.9
Additional sources consulted included medical records and
primary care providers. Study neurologists conducted gait
evaluations at each visit using previously described methods.2 3 14
Abnormal gaits were classified as non-neurological (eg, arthritis)
or neurological (hemiparetic, unsteady, ataxic, spastic, neuro-
pathic, parkinsonian and frontal).2 3Gait abnormalities were
clinically graded as mild (walks without assistance), moderate
(uses walking aids) or severe (wheelchair bound or stands with
assistance).2 3Clinicians computed the Hachinski ischaemic score
as a cerebrovascular disease risk score based on medical history
During follow-up, subjects with suspected dementia received a
diagnostic workup, including imaging studies and blood
tests.2 13Triggers included new cognitive complaints by subjects
or caregivers, study staffs’ observations and a pattern of
worsening neuropsychological test scores.2 13All available
clinical and neuropsychological information on all subjects
was reviewed at consensus case conferences attended by study
neurologists, neuropsychologist and a social worker, irrespec-
tive of whether subjects had triggers or whether or not they
were evaluated for dementia. Dementia diagnosis was assigned
using the Diagnostic and Statistical Manual, fourth edition,23
and subtyped using established criteria for Alzheimer’s
disease,24vascular dementia4and other dementias. In subjects
diagnosed with dementia, neuroimaging was used to help
allocate the diagnosis of ‘‘probable’’ Alzheimer’s disease or
‘‘probable’’ vascular dementia. We have reported good agree-
ment between clinical diagnoses of Alzheimer’s disease,25
vascular dementia2and dementia with Lewy bodies,26and
pathological findings in our study. Study clinicians did not have
access to quantitative gait parameters during evaluations or at
the diagnostic conferences.
Baseline characteristics were compared with descriptive statis-
tics, applying non-parametric tests as appropriate. Factor
analysis using the principal component method was performed
on baseline scores on eight individual gait variables in 399 non-
demented subjects.10The initial factors were then subjected to
an orthogonal varimax rotation to reduce the larger number of
highly correlated variables to a smaller number of uncorrelated
independent predictors to be used in the final analysis. To
identify clinical correlates of quantitative gait dysfunction, the
prevalence of neurological gait abnormalities2among subjects
in the lowest tertile of each factor was examined.
Firstly, to determine whether gait was related to decline on
specific cognitive domains, regardless of development of
dementia, linear mixed effects models controlled for age, sex
and education were applied to the 399 initially non-demented
subjects.27A random intercept was included in the model to
allow the entry point to vary across individuals. The ‘‘factor’’
term in the model (see table 5) represents the association
between gait factors and selected tests at baseline. ‘‘Time’’
below are automatically calculated as the mean of two trials by the gait software
Definition of quantitative gait parameters. All quantitative parameters described
Variable Unit Definition
Distance covered on two trials by the ambulation time
Distance between heel points of two consecutive footfalls of the same foot.
Variability in length between strides is reported as standard deviation.
No of steps taken in a minute
Time elapsed between first contact of current footfall and the last contact of
previous footfall, added to the time elapsed between the last contact of current
footfall and the first contact of next footfall
Duration when the foot is in the air and is the time taken from toe off to heel strike
of the same foot. Variability in swing time is reported as standard deviation.
Duration when the foot is on the ground and is the time taken from heel strike to
toe off of the same foot
930 Verghese, Wang, Lipton, et al
represents average rate of change in test performance over time.
An interaction between ‘‘factor’’ and ‘‘time’’ was included to
model the effect of baseline gait factors on rate of change in
cognitive function. We analysed performance at baseline and
yearly follow-up visits on the Blessed Test16and specific
cognitive domains, including memory,17executive function18 19
and attention.18Model assumptions were examined graphically
and analytically, and were adequately met.
Cox proportional hazards models28were used to compute
hazard ratios (HR) with 95% confidence intervals (CI) for
developing dementia19based on baseline gait factors in 399
non-demented subjects.3We also studied Alzheimer’s disease24
and vascular dementia4as outcomes.28Given the low number of
incident Alzheimer’s disease and vascular dementia cases, these
secondary analyses are intended to support and complement
the linear models examining decline in individual cognitive
domains implicated in the early stages of these dementia
All analyses reported are adjusted for age, sex and education.
We conducted analyses using both follow-up time and age as
the time scale, and the results were not materially different.
Using age as the time scale in Cox models is considered more
appropriate than follow-up time in cohort studies.29When age
is the time scale, the hazard function can be directly interpreted
as the age specific incidence function and age is accounted for
in the non-parametric term of the hazard function providing a
more flexible and effective control of age.29Time to event was
from age at assessment, which accounts for the left truncation
occurring at study inclusion, to age at dementia or to final study
contact, whichever came first. Proportional hazards assump-
tions of the models were examined analytically and graphically
and were adequately met.
Finally, to corroborate our findings and facilitate compar-
isons with other studies, we derived a reduced set of predictors
selected from the highest loading variable on each factor (see
table 4). We also included velocity, which has been reported to
predict dementia.5–8The association of these single variables
with incident dementia were examined individually as well as
entered together in Cox models, adjusted for age, sex and
Of the 427 subjects, 28 with dementia diagnosed at or before
the visit they received the quantitative assessments were
excluded. In the remaining 399 subjects over 798 person years
(median 2 years), 33 developed dementia.23Of these, 12 were
subtyped as Alzheimer’s disease,2417 vascular dementia4and 4
other dementias. Of the 399 subjects (median 3 visits), 384 had
one or more yearly follow-ups, 16 were active but had not had
their first follow-up visit and 6 died.
Table 2 shows the characteristics of the participants at
enrolment. Subjects with incident dementia were older at entry.
There were no significant group differences in abnormal gait
subtypes.2 3Most gait abnormalities were of mild severity
(61.9%). Moderate (35.4%) and severe (2.7%) gait abnormal-
ities were less common in this community based sample as
Baseline characteristics by final cognitive status
Clinical gait abnormalities (%)
ADL scale limitations* (%)
Physical self maintenance
Illness index (0–10)
Hachinski ischaemic score (0–15)
Blessed Test score (0–32)
FCSRT, total recall (0–48)
Digit Symbol Substitution Test, total
Letter Fluency Test, total
Digit Span Test, total
Geriatric Depression Scale (0–15)
ADL, activities of daily living; FCSRT, Free and Cued Selective Reminding Test.
Values are mean (SD) unless otherwise stated.
*Subjects with impairment in one or more physical self maintenance or instrumental activities on the Lawton–Brody
Quantitative gait parameters at baseline by final cognitive status.
Stride length (cm)
Stride length variability (SD)
Swing time (s)
Swing time variability (SD)
Stance time (s)
Double support time (s)
Values are mean (SD).
Quantitative gait dysfunction and risk of cognitive decline and dementia931
previously reported.3A higher proportion of subjects who
developed dementia reported limitations on instrumental but
not basic activities of daily living. While the mean scores on the
cognitive test scores were in the normal range, subjects who
developed dementia had worse scores compared with controls
Table 3 shows impairment in multiple baseline gait variables
compared with controls in initially non-demented subjects who
went on to develop dementia.
Factor analysis with varimax rotation yielded exactly three
orthogonal factors that accounted for 87% of the variance in
baseline quantitative gait performance (table 4).10The factor
with the highest variance had strong loadings by velocity and
length measures, and was termed ‘‘pace’’ factor. The second
loaded on variables reflecting gait rhythm such as cadence and
timing, and was termed ‘‘rhythm’’ factor. The final factor
loaded heavily on gait ‘‘variability’’ measures.11 12Mean factor
score was 0 (SD 1). The factors can be conceptualised as
summary risk scores with higher scores denoting worse
Table 5 summarises the results of our primary analyses using
linear models.27Only the pace factor was associated with global
cognitive decline measured using the Blessed Test. At the
average level of the pace factor, Blessed scores worsened by
0.23 points per year. This rate increased annually by 0.15 points
(by 65%) for every additional point on the baseline pace score.
To determine whether quantitative gait dysfunction was
related to decline in some cognitive domains but not others, we
examined selected tests (table 5). Episodic memory declined by
0.15 units (by 107%) for each 1 point increase in rhythm factor
scores. Each 1 point increase in the pace factor was associated
with an increased annual rate of decline in executive function
measured on the Digit Symbol Substitution Test by 0.73 points
(by 29%) and on the Letter Fluency Test by 0.46 points (by
92%). Variability was not associated with decline on general or
specific cognitive measures.
Modelled as a continuous variable (table 6), baseline rhythm
(HR 1.48; 95% CI 1.03 to 2.14; p=0.03) and variability factors
(HR 1.37, 95% CI 1.05 to 1.78; p=0.02) predicted future risk of
dementia. The incident dementias were diagnosed early and at
mild severity; mean Blessed score was 10.5 (3.9) (worst score
32, .7 abnormal)16at diagnosis.
When we excluded nine subjects who developed dementia in
the first year following quantitative assessment from the
analysis to account for any diagnostic misclassification and
the possibility that quantitative abnormalities may occur only
close to the time of dementia diagnosis, the significant
associations noted between gait factors and dementia were
unchanged. Both slow and fast gait velocity may be associated
with increased gait variability. When the analyses were
repeated using coefficient of variation (SD/mean6100) instead
of SD to define variability, the rhythm (HR 1.53; 95% CI 1.01 to
2.21) and variability factors (HR 1.46, 1.14 to 1.86) still
predicted incident dementia.
Baseline neuropsychological test performance was worse in
subjects who went on to develop dementia, although most
rotated and extracted by factor analysis
Factor loading of eight quantitative variables on the three independent gait factors
Gait variablePace factorRhythm factor Variability factor
Stride length (cm)
Double support time (s)
Swing time (s)
Stance time (s)
Stride length variability (SD)
Swing time variability (SD)
Variance explained (%)
Higher factor scores denote worse performance.
domains assessed by linear mixed models, controlled for age, sex and education
Summary of association of the three gait factors with baseline and rate of change on general and specific cognitive
Cognitive domainGeneralMemory Executive functionAttention
Cognitive testBlessedFCSRT Digit SymbolLetter Fluency Digit Span
0.40 (0.14, 0.66)
0.23 (0.11, 0.34)
0.15 (0.03, 0.27)**
–0.0002 (–0.15, 0.15)
–0.14 (–0.27, –0.02)
–0.09 (–0.22, 0.04)
–3.57 (–4.86, –2.29)
2.48 (2.08, 2.88)
–0.73 (–1.15, –0.31)***
–3.02 (–4.28, –1.77)
0.50 (0.16, 0.84)
–0.46 (–0.82, –0.11)***
–0.53 (–0.89, –0.16)
0.71 (0.58, 0.83)
–0.09 (–0.23, 0.04)
0.34 (0.08, 0.59)
0.23 (0.11, 0.34)
0.05 (–0.07, 0.17)
–0.05 (–0.19, 0.08)
–0.14 (–0.27, –0.02)
–0.15 (–0.28, –0.02)*
–1.05 (–2.30, 0.19)
2.48 (2.08, 2.88)
–0.09 (–0.51, 0.33)
–0.92 (–2.14, 0.29)
0.50 (0.16, 0.84)
–0.01 (–0.37, 0.34)
–0.18 (–0.53, 0.18)
0.71 (0.58, 0.83)
–0.09 (–0.22, 0.04)
0.16 (–0.08, 0.41)
0.23 (0.11, 0.34)
0.08 (–0.04, 0.20)
–0.11 (–0.24, 0.02)
–0.14 (–0.27, –0.02)
0.04 (–0.09, 0.16)
–0.61 (–1.81, 0.58)
2.48 (2.08, 2.88)
–0.11 (–0.53, 0.32)
–0.20 (–1.36, 0.96)
0.50 (0.16, 0.84)
–0.21 (–0.57, 0.15)
–0.14 (–0.48, 0.20)
0.71 (0.58, 0.83)
0.07 (–0.05, 0.20)
FCSRT, Free and Cued Selective Reminding Test.
See methods for explanation of model terms.
Values are estimates with 95% CI.
Significant interactions are in bold.
*p=0.02, **p=0.01, ***p,0.001.
932 Verghese, Wang, Lipton, et al
scores were within the normal range (table 1). To examine
whether quantitative gait abnormalities predicted dementia
independent of baseline cognitive test performance, we
repeated the full models with additional adjustments for
memory and executive function. Adjusting for baseline memory
scores17in the full model made the association of the rhythm
factor with incident dementia non-significant (HR 1.36, 95% CI
0.86 to 2.13), but not the variability factor (HR 1.56, 95% CI
1.10 to 2.23). Adjusting for digit symbol scores18made the
association of the variability factor with dementia non-
significant (HR 1.29, 95% CI 0.99 to 1.67), but not the rhythm
factor (HR 1.48, 95% CI 1.01 to 2.15).
Neurological gaits, which we reported to predict risk of
vascular dementia in another cohort,2were diagnosed in 94
subjects at baseline. Prevalence of neurological gaits among
subjects with scores in the worst tertile on the rhythm factor
was 27%, variability 31% and pace 41%. The association of the
rhythm (HR 1.55; 95% CI 1.06 to 2.27) and variability factors
(HR 1.35; 95% CI 1.03 to 1.76) and dementia remained
significant, even after adjustments for neurological gaits,2
Exclusion of subjects with Parkinson disease and strokes did
not materially change the results.
None of the factors predicted Alzheimer’s disease (table 6).
Only the pace factor (HR 1.60; 95% CI 1.06 to 2.41; p=0.02)
predicted the risk of vascular dementia.
Single gait variables
Swing time (1/10 s), stride length (cm) and stride length
variability (SD) were the highest loading individual variables
on the three factors. Swing time (HR 3.11, 95% CI 1.43 to 6.78),
stride length (cm) (HR 0.98, 95% CI 0.96 to 1.00), stride length
variability (SD) (HR 1.71, 95% CI 1.19 to 2.45) and velocity (cm/
s) (HR 0.98, 95% CI 0.96 to 0.99) predicted dementia when each
variable was individually fitted into the Cox model adjusted for
age, sex and education. However, these individual variables
were highly correlated, and did not show independent
associations when entered together in the same model.
Our findings show that quantitative gait dysfunction predicts
risk of cognitive decline in initially non-demented older adults.
The pace factor predicted decline in executive function, the
cognitive domain primarily involved in vascular dementia.30A
1 point increase in the pace factor predicted decline on
Substitution18(by 29%) and the Letter Fluency tests19(by
92%). A 1 point increase in the rhythm factor was associated
with episodic memory17decline (by 107%). Quantitative gait
dysfunction also predicted risk of dementia. A 1 point increase
in both the rhythm (by 48%) and variability factors (by 37%)
predicted risk of developing dementia, adjusted for age, sex and
by theDigit Symbol
education. The associations remained robust after additional
adjustments for other potential confounders, such as medical
illnesses, cerebrovascular disease, baseline cognitive status,
preclinical dementia and neurological gaits.
This study has a number of limitations that need to be
considered. While our gait variables were selected based on our
and prior studies,5–9all possible aspects of gait were not
measured or analysed. However, most other gait variables (such
as step length) either can be derived or are highly correlated
with our selected variables. Subjects walked at their normal
pace. In future studies, walking at different speeds, or using
stressors such as the walking while talking test31could be
studied as predictors of cognitive decline. This nested cohort
study was necessarily restricted to subjects who received
quantitative assessments since 2001 in our study, but subjects
seen previously in our cohort were not differentially excluded
and our analyses accounted for staggered entry.29Our small
sample size, short follow-up and multiple analyses necessitate
caution. However, to our knowledge, this is among the first and
largest study to report longitudinal associations between
multivariate quantitative gait measures and cognition.5–9The
confidence intervals for significant associations in our analyses
were narrow, but we may have seen stronger associations with
additional follow-up. While some diagnostic misclassification is
inevitable, we have reported good reliability for our dementia
diagnostic procedures using pathology as the gold stan-
dard.2 25 26
The upper brainstem contains neurons that control muscu-
lature for gait through their projections to the lower brainstem
and spinal cord.32These brainstem nuclei are, in turn, under the
influence of descending inputs from the basal ganglia,
cerebellum and cortical motor areas.32 33It is assumed that
stride length and velocity (pace) are controlled supraspinally by
phasic output from the basal ganglia to the supplementary
motor area, whereas spinal and brainstem mechanisms may
determine cadence (rhythm).32 33Neural substrates of gait
variability are less understood.11It has been suggested that
regulation of gait variability is automated and requires minimal
cognitive input in healthy adults, but may be perturbed in the
presence of disease.11 12
The specificity of the association of pace with executive
function and vascular dementia (and not incident dementia
overall) favours a vascular aetiology. We have reported a higher
prevalence of mixed vascular pathology with advancing age in
subjects with dementia in this cohort.25It is possible that some
of our associations reflect mixed pathology although neuro-
imaging was used to assign dementia subtypes. Decreased
velocity and stride length (which load on the pace factor) in
older adults were associated with white matter disease and
strokes on neuroimaging in a prior study.34Then again, the
association of gait rhythm and variability with dementia in our
study was significant even when controlled for cerebrovascular
disease or excluding subjects with strokes. The presence of
temporal lobe atrophy on imaging studies has been previously
vascular dementia as a function of baseline quantitative gait factors adjusted for age, sex and
years of education
Hazard ratios with 95% CI for developing any dementia, Alzheimer’s disease and
Hazard ratio (95% CI)
1.30 (0.95 to 1.78)
1.48 (1.03 to 2.14)
1.37 (1.05 to 1.78)
0.95 (0.48 to 1.88)
1.55 (0.81 to 2.99)
1.18 (0.67 to 2.00)
1.60 (1.06 to 2.41)
1.59 (0.95 to 2.67)
1.22 (0.78 to 1.9)
Higher factor scores denote worse performance.
Quantitative gait dysfunction and risk of cognitive decline and dementia933
correlated with poor mobility, independent of cerebrovascular
Gait rhythm predicted memory decline, which can start
many years before Alzheimer’s disease is diagnosed.1Increased
gait variabilityhas been reported
Alzheimer’s disease.11 12However, Alzheimer involvement of
the motor cortex is said to occur only late in the disease process
and gait disturbances early in the disease is considered an
exclusion criterion.2 36–38The low prevalence of neurological
gaits among subjects in the worst factor score tertiles suggests
that quantitative gait abnormalities may not have clinical
correlates. Alternatively, non-specific gait patterns associated
with quantitative gait abnormalities may be under-recognised
by clinicians. Our study protocol2 3required clinicians to judge
whether gait is normal or abnormal (including non-specific
unsteady gaits) before assigning subtypes, minimising this
possibility. A recent clinicopathological study reported that
Parkinsonian gait in the absence of idiopathic Parkinson
disease was correlated with substantia nigra neurofibrillary
tangles, even in cases without clinical Alzheimer’s disease or
with minimal Alzheimer pathology.39These findings raise the
possibility that Alzheimer pathology may involve brain regions
regulating gait early but the subtle quantitative gait abnorm-
alities may be overshadowed by behavioural symptoms. The
unique relationships of individual gait factors with specific
cognitive domains suggest aetiological roles for both vascular
and neurodegenerative mechanisms. These findings also
identify potential gait pathways, which may be amenable to
intervention if our results are validated.
The strengths of this study include our validated diagnostic
procedures,2 13 25 26systematic quantitative assessments2 3 9and
minimal attrition. We conducted a number of sensitivity
analyses to account for potential confounders such as pre-
clinical dementia and neurological gaits. Given the limitations
of categorical clinical gait classifications, the continuous
quantitative gait measures may better help characterise motoric
manifestations of the preclinical stages of dementia. Clinicians
were blinded to quantitative gait data before and at the
diagnostic conferences, minimising bias and diagnostic circu-
larity. The quantitative measures are easily collected, do not
require attached monitoring devices and do not require
extensive training or specialised personnel.
Gait factors may be conceptualised as summary risk scores
that can be easily derived for use in clinical or research settings.
A 1 point (1 SD) increase in the factor score is associated with
SD unit changes in all component variables with a larger
contribution from higher loading variables. Factor analysis has
been used to explain other complex physiological phenom-
enon,40and appears to be well suited for studying gait, as
suggested by our findings and others.41This approach enabled
us to identify empirically defined and statistically independent
factors representing distinct gait domains. The single gait
variables identified from the three factors support our main
findings, and will facilitate comparisons with other studies.5–8
We also identified single variables other than velocity,5–8such as
swing time, that predicted dementia.
In conclusion, quantitative gait dysfunction is a marker for
preclinical stages of dementia. The quantitative gait abnorm-
alities and neuropsychological impairments at study onset can
be interpreted as the consequence of accumulating dementia
pathology. Quantitative gait performance predicts dementia
even after accounting for baseline cognitive performance.
However, our intent is not to replace neuropsychological tests
in dementia assessment but to understand motoric manifesta-
tions of early dementia stages.2 5–9Vascular and non-vascular
substrates, neuroanatomical pathways and physiological pro-
cesses underlying quantitative gait dysfunctions should be
further explored to develop effective preventive interventions to
reduce the burden of dementia.
This paper was orally presented in part at the Fourth International
Congress on Vascular Dementia, Porto, Portugal, in October 2005, and
at the First International Congress on Gait and Mental Function,
Madrid, Spain, in February 2006.
Joe Verghese, Richard B Lipton, Department of Neurology, Albert Einstein
College of Medicine, Yeshiva University, Bronx, New York, USA
Cuiling Wang, Xiaonan Xue, Department of Epidemiology and Population
Health, Albert Einstein College of Medicine, Yeshiva University, Bronx,
New York, USA
Roee Holtzer, Ferkauf Graduate School of Psychology, Yeshiva University,
Bronx, New York, USA
Funding support: The Einstein Ageing Study is funded by the National
Institute on Ageing (grant AG03949). Dr Verghese is funded by the
National Institute on Ageing (grants K23 AG024848 and RO1
Competing interests: None.
1 Hall CB, Lipton RB, Sliwinski M, et al. A change point model for estimating the
onset of cognitive decline in preclinical Alzheimer’s disease. Stat Med
2 Verghese J, Lipton RB, Hall CB, et al. Abnormality of gait as a predictor of non-
Alzheimer’s dementia. N Engl J Med 2002;347:1761–8.
3 Verghese J, Levalley A, Hall CB, et al. Epidemiology of gait disorders in
community-residing older adults. J Am Geriatr Soc 2006;54:255–61.
4 Chui HC, Victoroff JI, Margolin D, et al. Criteria for the diagnosis of ischemic
vascular disease proposed by the State of California Alzheimer’s disease
Diagnostic and Treatment Centers. Neurology 1992;42:473–80.
5 van Iersel MB, Hoefsloot W, Munneke M, et al. Systematic review of quantitative
clinical gait analysis in patients with dementia. Z Gerontol Geriatr
6 Wang L, Larson EB, Bowen JD, et al. Performance-based physical function and
future dementia in older people. Arch Intern Med 2006;166:1115–20.
7 Waite LM, Grayson DA, Piguet O, et al. Gait slowing as a predictor of incident
dementia: 6-year longitudinal data from the Sydney Older Persons Study.
J Neurol Sci 2005;229–230:89–93.
8 Marquis S, Moore MM, Howieson DB, et al. Independent predictors of cognitive
decline in healthy elderly persons. Arch Neurol 2002;59:601–6.
9 Holtzer R, Verghese J, Xue X, et al. Cognitive processes related to gait velocity:
results from the Einstein Aging Study. Neuropsychology 2006;20:215–23.
10 Johnson RA, Wichern DW. Applied multivariate statistical analysis, 5th edn.
Upper Saddle River, NJ: Prentice-Hall, 2002.
11 Hausdorff JM. Gait variability: methods, modeling and meaning.
J Neuroengineering Rehabil 2005;2:19.
12 Sheridan PL, Solomont J, Kowall N, et al. Influence of executive function on
locomotor function: divided attention increases gait variability in Alzheimer’s
disease. J Am Geriatr Soc 2003;51:1633–7.
13 Lipton RB, Katz MJ, Kuslansky G, et al. Screening for dementia by telephone
using the memory impairment screen. J Am Geriatr Soc 2003;51:1382–90.
14 Verghese J, Katz MJ, Derby CA, et al. Reliability and validity of a telephone-
based mobility assessment questionnaire. Age Ageing 2004;33:628–32.
15 Bilney B, Morris M, Webster K. Concurrent related validity of the GAITRite
walkway system for quantification of the spatial and temporal parameters of gait.
Gait Posture 2003;17:68–74.
16 Blessed G, Tomlinson E, Roth M. The association between quantitative measures
of dementia and of senile change in the cerebral grey matter of elderly subjects.
Br J Psychiatry 1968;114:797–811.
17 Buschke H. Selective reminding for analysis of memory and learning. J Verbal
Learn Verbal Behav 1973;12:543–50.
18 Wechsler D. Wechsler adult intelligence scale—revised (WAIS-R). New York:
Psychological Corporation, 1981.
19 Monsch AU, Bondi MW, Butters N, et al. Comparisons of verbal fluency tasks in
the detection of dementia of the Alzheimer type. Arch Neurol 1992;49:1253–8.
20 Yesavage JA, Brink TL, Rose TL, et al. Development and validation of a geriatric
depression rating scale: a preliminary report. J Psychiatr Res 1982–
21 Lawton MP, Brody EM. Assessment of older people: self-maintaining and
instrumental activities of daily living. Gerontologist 1969;9:179–86.
22 Rosen WG, Terry RD, Fuld PA, et al. Pathological verification of ischemic score in
the differentiation of dementias. Ann Neurol 1980;7:486–8.
23 American Psychiatric Association. Diagnostic and statistical manual of mental
disorders, 4th edn. Washington DC: American Psychiatric Association, 1994.
24 McKhann G, Drachman D, Folstein M, et al. Clinical diagnosis of Alzheimer’s
disease: report of the NINCDS-ADRDA Work Group under the auspices of
934Verghese, Wang, Lipton, et al
Department of Health and Human Services Task Force on Alzheimer’s Disease. Download full-text
25 Crystal HA, Dickson D, Davies P, et al. The relative frequency of ‘‘dementia of
unknown etiology’’ increases with age and is nearly 50% in nonagenarians. Arch
26 Verghese J, Crystal HA, Dickson DW, et al. Validity of clinical criteria for the
diagnosis of dementia with Lewy bodies. Neurology 1999;53:1974–82.
27 Laird NM, Ware JH. Random-effects models for longitudinal data. Biometrics
28 Cox DR. Regression models and life tables. J R Stat Soc [B] 1972;34:187–220.
29 Thiebaut AC, Benichou J. Choice of time-scale in Cox’s model analysis of
epidemiologic cohort data: a simulation study. Stat Med 2004;23:3803–20.
30 Roman GC, Sachdev P, Royall DR, et al. Vascular cognitive disorder: a new
diagnostic category updating vascular cognitive impairment and vascular
dementia. J Neurol Sci 2004;226:81–7.
31 Verghese J, Buschke H, Viola LC, et al. Validity of divided attention tasks in
predicting falls in older individuals—a preliminary study. J Am Geriatr Soc
32 Drew T, Prentice S, Schepens B. Cortical and brainstem control of locomotion.
Prog Brain Res 2004;143:251–61.
33 Morris ME, Iansek R, Matyas TA, et al. Ability to modulate walking cadence remains
intact in Parkinson’s disease. J Neurol Neurosurg Psychiatry 1994;57:1532–4.
34 Rosano C, Brach J, Longstreth WT Jr, et al. Quantitative measures of gait
characteristics indicate prevalence of underlying subclinical structural brain
abnormalities in high-functioning older adults. Neuroepidemiology
35 Guo X, Steen B, Matousek M, et al. A population-based study on brain atrophy
and motor performance in elderly women. J Gerontol A Biol Sci Med Sci
36 Ala TA, Frey II WH. Validation of the NINCDS-ADRDA criteria regarding gait in
the clinical diagnosis of Alzheimer disease. Alzheimer Dis Assoc Disord
37 Braak H, Braak E. Evolution of the neuropathology of Alzheimer’s disease. Acta
Neurol Scand 1996;165(Suppl):3–12.
38 Pearson RC, Esiri MM, Hiorns RW, et al. Anatomical correlates of the distribution
of the pathological changes in the neocortex in Alzheimer’s disease. Proc Natl
Acad Sci USA 1985;82:4531–4.
39 Schneider JA, Li J-L, Li Y, et al. Substantia nigra tangles are related to gait
impairment in older persons. Ann Neurol 2006;59:166–73.
40 Lakka HM, Laaksonen DE, Lakka TA, et al. The metabolic syndrome and total and
cardiovascular disease mortality in middle-aged men. JAMA
41 Chau T. A review of analytical techniques for gait data. Part 1: Fuzzy, statistical
and fractal methods. Gait Posture 2001;13:49–66.
Access a vast information database with Toll-Free linking
‘‘Toll-free’’ linking gives you immediate access to the full text of many of the cited articles in a
paper’s reference list—FOR FREE. With the support of HighWire’s vast journal catalogue, a huge
reference library is now open to you. If HighWire hosts the journal, you can view the full text of the
referenced article, completely free of charge by following the Free Full Text links in the references.
Quantitative gait dysfunction and risk of cognitive decline and dementia935