Walking Speed, Cognitive Function, and Dementia Risk in the English Longitudinal Study of Ageing

Ruth A. Hackett, PhD; Hilary Davies-Kershaw, PhD; Dorina Cadar, PhD; Martin Orrell, PhD; Andrew Steptoe, DSc, DPhil


J Am Geriatr Soc. 2018;66(9):1670-1675. 

In This Article


Study Population

Our data were from the English Longitudinal Study of Ageing (ELSA): a representative study of community-dwelling adults aged 50 and older in England.[21] Data collection began in 2002-03, with follow-up every 2 years. We used data from Waves 1 (2002–03) to 7 (2014–15). All participants provided informed consent. Ethical approval was obtained from the National Research Ethics Committee.


The outcome was incident dementia from Wave 3 (2006–07) to Wave 7 (2014–15). We used 2 information sources to define dementia, as in previous work.[22–24] The primary indicator was self-reported physician-diagnosed dementia. The second criterion was applied to participants who were not able to respond directly and was a caregiver's comparison of functional performance with that from 2 years before. We used a 16-item adapted short-form version of the Informant Questionnaire on Cognitive Decline in the Elderly (IQCODE)[25] and defined individuals with an average score of 3.5 or greater as having dementia, based on previous work suggesting high sensitivity and specificity at this cut-point.[26] There were 289 incident cases of dementia between Waves 3 and 7, 240 of which were based on physician-diagnosed dementia and 49 on IQCODE scores.

Walking Speed

Walking speed was assessed in participants aged 60 and older. Participants were asked to walk a distance of 8 feet (2.43 m) from a standing start on even ground at their usual pace, and the time taken was recorded. The average time of two walks was calculated. We used walking speed at Wave 1 (2002–03) as a predictor of dementia.

Cognitive Function

We aggregated information from 4 cognitive tests (memory (immediate and delayed), time orientation, verbal fluency, processing speed) to obtain cognitive function scores at Waves 1 and 2. To compute an overall score, we transformed each of the measures into a z score and derived average total scores in 2002–03 and 2004–05. More information about these tasks is provided elsewhere.[27]


We classified age into 3 categories (60–69, 60–79, ≥80). Socioeconomic status was defined using deciles of nonpension wealth (1=low, 10=high). We divided education into 3 categories (<junior high school, high school, university). Mobility impairment at baseline was ascertained by asking participants whether they had difficulty with one or more common arm and leg functions (e.g., getting in or out of bed). Activity of daily living (ADL) impairment was indexed by asking participants whether they had difficulty with 6 activities (e.g., dressing, including putting on shoes and socks,). Physician diagnoses of coronary heart disease, stroke, diabetes, cancer and hypertension were entered as binary (yes/no) variables, because these conditions may affect dementia risk. Because depression is associated with dementia onset, we included the 8-item Center for Epidemiologic Studies Depression Scale (CES-D) score. Baseline cognitive function was included in all analyses. Baseline walking speed was controlled for in analyses examining change in walking speed as a predictor of future dementia.

Statistical Analysis

We compared characteristics at Wave 1 of participants who developed dementia with characteristics of those who did not using logistic regression and univariate analysis of variance. Because age is an important factor in dementia risk, we controlled for age. We used Cox proportional hazards regression to model the association between walking speed at Wave 1 (2002–03) and cumulative dementia from Wave 3 (2006–07) to Wave 7 (2014–15). Participants who had dementia at Waves 1 and 2 were excluded. If the precise date of dementia diagnosis was unknown, we used the midpoint date between the waves of data collection. We censored individuals who dropped out of the study or died. We used the last ELSA interview date as the censor date. We calculated change scores in walking speed and cognition by subtracting values in 2004–05 (Wave 2) from values in 2002–03 (Wave 1). We then computed an interaction term between changes in walking speed and cognitive function. Preliminary analyses removing those who died did not change the pattern of results. We conducted 3 sensitivity analyses. First, we modelled the association between walking speed at Wave 2 (2004–05) and cumulative dementia risk (Waves 3–7). Second, we used only physician-diagnosed dementia as the outcome. Third, we conducted the analyses including new dementia cases from Wave 4 to 7 (2008–2015) rather than from Wave 3 to 7. Results are presented as hazard ratios (HRs) and 95% confidence intervals (95% CI). All analyses were conducted using SPSS version 24 (IBM Corp., Armonk, NY).