Physical activity (PA) benefits older people’s health: preventing disease,1 reducing disability2 and improving well-being.1 Adults, including older adults, are advised to perform at least moderate-intensity PA (leading to increases in breathing, heart rate and temperature) for ⩾30 minutes (in ⩾10 minute bouts) on at least 5 days weekly.1 3 However, recent self-reported findings suggest that only 18% of men and 14% of women aged 65–74 and 8% of men and 4% of women aged 75 or over achieve this.4 Benefits may occur from lower-intensity PA in older age; walking (light-intensity if strolling, 2 mph, moderate-intensity if faster) remains important for maintaining activities.1 Factors associated with decreased PA levels in older people include: increasing age; female gender; obesity; medical problems; disability; pain; depression; smoking; reduced education; social isolation; low exercise self-efficacy; attitudinal barriers; bad weather; and unsafe neighbourhoods.5^–7 These findings are from self-reported activity; however, the predominant activity, walking, is unreliably recalled.8 Questionnaires also suffer from recall bias and floor effects, with the baseline too high for most respondents.8 Motion sensors (pedometers and accelerometers) are sensitive to walking, objectively quantify PA as a continuous variable and are unrestricted by floor values.8 9

Pedometers are cheap and easy to wear; they measure step-count but not intensity and therefore cannot distinguish walking speeds. They are superior to questionnaires10 and logs,11 but may under-report at slower speeds,12 including frail elderly,13 and in the obese.14 The media-promoted 10 000 daily steps target ignores the guidelines’ intensity component15 and may be unsustainable for older adults with chronic disease.16 Older people’s pedometer studies are limited to those with specific disorders,17^–19 volunteers18 20 21 or cohort survivors.22 Several population-based adult studies include small numbers of older people (<200).9 23^–26 Those examining PA determinants found that step-counts were inversely associated with age,9 18 20 23^–26 female gender,9 23^–25 body mass index,9 18 21 24^–26 waist circumference21 and lower income.9 26 Most lacked health, disability or psychological variables, but Ashe found that mobility, depression and chronic disease self-efficacy predicted step-counts.20

Accelerometers record activity counts, the product of movement frequency and intensity, with results downloaded for computer analysis. They have been validated in older people27 and objectively measure activity.8 Accelerometers are worn like pedometers and similarly cannot measure swimming or cycling activities, but can record continuously for up to 21 days.27 Accelerometers give participants no PA feedback; pedometers do, which could lead to participants increasing their PA whilst monitoring it, advantageous in intervention studies19 28 but problematic for observational studies. This has not been tested empirically. Accelerometers are expensive, approximately $350/£175/€230. Dual-mode accelerometers (e.g. Actigraph: Actigraph, LLC Florida 32502 USA. Actigraph. 2008. www.theactigraph.com) provide simultaneous, time-stamped activity count and step-count measurements. Those using accelerometers have reported step-counts as a simple stable metric to monitor PA.29 The step-count function exactly counts observed steps at all treadmill speeds, whereas pedometers may under-record at low speeds.12 Most PA in older people is walking-based and hence activity counts and step-counts are likely to be highly correlated. Accelerometers have been used successfully to describe PA levels in small groups of older patients.17 19 27 A larger study of 163 volunteers aged 70 or over showed low PA levels and weak associations with quality of life30 31 and a further study of 184 volunteers aged 65–85 showed that depressive mood was associated with accelerometer-assessed step-count.32 However, no physical health, disability or anthropometric variables were presented in either study.31 32 There are no large population-based studies of older people using accelerometers or pedometers, including physical health, disability, psychological and anthropometric measures.

Our main objectives were to assess customary PA levels measured objectively, using accelerometers, in a population-based sample of older people and to examine associations with other factors. A further objective was to assess whether wearing pedometers increases PA in those asked to maintain usual levels.

METHODS

Target population

Community-dwelling older people at least 65 years old, able to walk outside and registered with a general practice (primary care centre) in Oxfordshire, United Kingdom.

Exclusion criteria

One thousand, five hundred and twenty-nine patients aged at least 65 were registered. Two hundred and seventy-three (18%) were excluded by computer record search and by general practitioner and district nurse examination of patient lists for: dementia (50), care home resident (28), terminally ill plus spouses (32), housebound (86), uncontrolled cardiac failure/unstable angina/recent myocardial infarction/angioplasty/coronary artery bypass (29), in another research study (48).

Sample size and selection

This was a pilot study assessing objective PA levels in older people. The pilot was also designed as a randomised controlled trial to test the effect of two interventions on recruitment (questionnaire inclusion with the study invitation and telephone contact a week later). Random selection and randomisation were performed at household level, to avoid contamination of partners receiving different interventions. The required sample size was 560. Full details of these interventions and their recruitment effects have been reported.33

Recruitment

All 560 patients were invited to participate in a study measuring customary PA levels for a week using accelerometers and pedometers. Invitations were sent out over 20 weeks from September 2006, allowing time for recruitment. 280 patients selected at random also received a 12 page questionnaire with their invitation, assessing self-reported PA, health, disability and psychosocial measures (table 1). A questionnaire is published online only. Subjects were encouraged to reply, even if not participating in the PA study, allowing comparisons between participants and non-participants.

Baseline and objective PA assessment

Participants completed the questionnaire if it had not been randomly sent with their invitation. Weight was assessed using calibrated, sensitive scales, height measured with a wall-mounted tape measure and waist circumference assessed using a constant-tension, spring-loaded tape and a standard technique. Participants were given an accelerometer (Actigraph GT1M, Fl, USA) to wear over the hip on a belt, all day from waking, for 7 days, only removing it for bathing or swimming. They were asked to maintain usual activities and record them in a log. Half of participants, selected at random, were additionally given a Yamax Digi-walker SW-200 pedometer to wear over the other hip, and recorded daily step-counts on their logs.

Follow-up

Participants were seen 7 or more days later, to allow 7 full days’ recording. Accelerometer traces were checked using the Actigraph-provided ActiLife Monitoring System, alongside activity logs.

Data management

Activity was recorded using 5 second epochs. Participants with less than 5 days of data were excluded. Cut-points were used to distinguish between different PA levels: sedentary <200 count/min, light 200–1999 count/min, moderate 2000–3999 count/min, vigorous ⩾4000 count/min.30

Outcome measures

Average accelerometer daily step-count was chosen as the main objective PA outcome, as walking is the predominant PA in this age group and step-counts give readily understandable effect estimates. Average daily activity counts were also analysed, as was the proportion of time spent in different PA levels and the number of minutes spent weekly in at least moderate PA, in at least 10 minute bouts.

Analysis

The effects of age and sex on study participation were examined using logistic regression (STATA 9) while allowing for household clustering using the “cluster” option. The correlation between average daily step-counts and activity counts was estimated. Associations between average daily step-counts and other factors were examined using linear regression, giving crude estimates and then adjusting for structural factors (age, sex, pedometer use and household clustering). General health and disability were strongly associated with daily step-counts, so adjustment for these factors was performed. Backward stepwise linear regression (p<0.05) identified those factors independently associated with average daily step-counts. (Structural factors were locked into the model.) The analyses were repeated using average daily activity count as the outcome. The associations between self-reported PA measures and average daily step-count were also analysed using linear regression.

Step-count validation

Activity logs of individuals recording average daily step-counts lower than 2500 were examined alongside hourly accelerometer data to see whether low step-counts were due to low PA levels or to not wearing the accelerometer.

Ethical approval

Oxfordshire REC A (reference no. 06/Q1604/94).

RESULTS

Response rate and participant/non-participant comparisons

Of 560 individuals randomly invited, 240 (43%) participated. Overall comparison of participants (n = 240) and non-participants (n = 320), adjusted for age, sex and household clustering, found that men participated more (OR 1.4 (1.1 to 1.8)), but found no association with age (baseline age 65–69 OR 1, age 70–79 OR 0.8 (0.5 to 1.1), age 80 or over OR 0.8 (0.5 to 1.4)). A comparison of non-participants who returned a questionnaire (n = 76) with participants (n = 240) found that participants reported more PA than non-participants.43

Feasibility

All participants wore the accelerometers and completed follow-up. Two records with less than 5 days’ data were excluded. Results are therefore based on 238 participants.

Validation

Twenty activity logs of participants averaging daily step-counts lower than 2500 were examined. All reported wearing the accelerometer and very low PA levels were recorded (e.g. resting, watching television), consistent with the low step-counts reported.

Distributions of outcome variables

Figure 1 shows the distributions of: (1) average daily step-counts, mean 6443 (95% CI 6032 to 6853); and (2) average daily activity counts, mean 224 963 (95% CI 209 419 to 240 506). These accelerometer outcomes were highly correlated (r = 0.95).

• Download figure
• Open in new tab
• Download powerpoint

Figure 1

Distributions of average daily total step-counts and average daily total activity counts (both from accelerometers).

Mean percentage time spent in different PA levels

Sedentary (including sleeping) 89.5% (89.0% to 89.9%); light 8.1% (7.8% to 8.4%); moderate 1.8% (1.7% to 2.0%); vigorous 0.6% (0.5% to 0.7%). Only 2.5% (6/238) participants achieved recommended levels of at least 150 minutes/week (in at least 10 minute bouts) and 61.8% (147/238) achieved none.

Step-count and structural variables (table 2)

Men achieved higher average daily step-counts than women, after adjusting for structural variables, but not after adjusting for disability and general health. There was a strong, graded association with increasing age and decreased PA, which was reduced by adjustment for disability and general health, but remained significant. Those randomised to a pedometer achieved a higher daily step-count, but this did not reach statistical significance at p = 0.05.

Step-count and self-reported health (table 2)

Poorer health in the following variables was associated with a reduced step-count: general health; disability; long-standing illness; pain; medication use; chronic disease; falls; and walking aid use. Several variables showed a dose–response effect. Some individual chronic diseases showed a negative association with daily step-count (stroke, diabetes and arthritis), but only for diabetes did this persist after adjustment. No association was seen with smoking.

Step-count and anthropometric measures (table 2)

Body mass index and waist circumference were strongly negatively associated with daily step-count in a dose–response manner.

Step-count and psychological and social measures (table 2)

Depression score was inversely associated with step-count, but not after controlling for general health and disability. Attitudes to activity, exercise self-efficacy and belief in control over exercise were all positively related to step-count in a dose–response manner. Living alone and loneliness were not related.

Model of factors predicting average daily step-count (table 3)

Step-count was lower with increasing age, greater disability, poorer general health, a high body mass index and diabetes. Higher levels of exercise self-efficacy and stronger exercise control beliefs independently predicted higher step-counts. In the model, after controlling for other factors, wearing a pedometer had a borderline effect on increasing step-count.

Model of factors predicting average daily activity count (supplementary table)

Analyses using average daily activity count as the outcome showed very similar associations in direction and strength as for step-counts (data not shown). Modelling showed that the same factors predicted both outcomes, except that self-efficacy and pedometer use were not significant at p = 0.05 for activity count.

Comparison of objective (accelerometer) and self-reported PA measures (table 4)

Most self-reported PA measures were associated with average daily step-counts in the predicted direction; several showed dose–response relationships. The strongest associations were with number of long walks, comparative activity level and dog-walking.

DISCUSSION

Main study findings

This is the first moderately sized population-based study of older people published to date with objective PA measures and a broad range of health, psychological and anthropometric variables. It was feasible to monitor PA levels in community-dwelling older people for a week using accelerometers. We found low PA levels; only 2.5% achieved recommended levels,1 3 considerably lower than self-report.4 The following factors independently predicted lower accelerometer step-counts: increasing age, poor general health, disability, diabetes, higher body mass index, low exercise self-efficacy and low perceived exercise control.

Participants randomised to wearing a pedometer (giving PA feedback) recorded slightly higher accelerometer step-counts, despite being asked to continue customary PA. This difference was not statistically significant in univariate models, but in our best-fit model after adjustment for other variables it was of borderline significance. Although the evidence is not strong, this suggests that wearing pedometers may have at least a short-term impact on increasing PA levels.

Study strengths

We recruited a population-based sample of older people and related a broad range of physical health, anthropometric and psychological variables to objective PA measures. We focused on step-counts as our main outcome, as the majority of time is spent in sedentary or light PA, with walking the major PA in this age group, and step-counts therefore summarise the activity that people do, in a readily understandable measure. Moreover, we found that step-counts and activity counts were highly correlated (r = 0.95) and that repeating our analyses with activity count as the outcome produced similar findings. We examined the activity logs of participants with low step-counts and found that readings were consistent with reported activities. Average daily step-count over 1 week was correlated with several baseline self-reported PA measures undertaken in the month prior to accelerometry, suggesting that 7 days’ recording provides a good measure of usual PA levels.

Study weaknesses

As our study was cross-sectional, we cannot assume causality for any associations. PA levels are both an outcome and a determinant of health, and longitudinal or intervention studies are required for greater understanding of these associations. Our study was based on a single practice in a semi-rural, middle-income area with a response rate of 43%. Whilst there was no age bias for recruitment, participants were significantly more likely to be male and physically active than non-participants.43 Our findings are only generalisable to older people living in the community who can walk outside the home, and the low levels of PA that we report may overestimate actual PA levels in this group. Swimming is not recorded, which could lead to underestimation of PA levels; however, the diaries suggested that this occurred infrequently. Within those recruited, the effect estimates are likely to be true associations, not affected by selection bias. The self-reported physical activity measure used was the Zutphen PA Questionnaire,34 but with additional questions on dog-walking (self-constructed) and housework from the Physical Activity Scale for the Elderly.35 Results for these additional items were presented separately and showed associations with average daily step-count, confirming their value in the composite instrument. Although this is the largest published study to date of older people and objective PA measures, it may have been underpowered to detect whether wearing pedometers to monitor PA increases PA in those asked to maintain usual levels.

Comparison with other studies

Our average daily accelerometer step-counts are consistent with a review suggesting 6000–8500 steps/day for healthy older adults and 3500–5550 for those with chronic illness.44 Recent studies show only older people from Canada and Japan reaching the higher levels,20 32 with American seniors averaging lower levels.9 25 26 Although our findings need to be taken in the context of a 43% response rate and may overestimate actual PA levels, this also holds for the other population-based studies with response rates of 17%,9 40%25 and 45%.23 24 Our average daily activity count of 224 962 (SD 121 722) is lower than that of young American adults (357 601 (138 425)),45 as expected. The only data on older people is reported as counts/minute/day; our value for people aged 70 and over (267 (133)) is consistent with Davis and Fox’s findings (246 (92)).30 Only 2.5% (6/238) of our participants achieved the recommended level of at least 150 minutes/week of at least moderate PA (in at least 10 minute bouts), consistent with 1.8% (3/163) reported using accelerometry by Davis and Fox30 and much lower than the self-reported levels detailed above.4

We confirmed associations from other studies between lower average daily step-counts and female gender,9 23^–25 increasing age,9 18 20 23^–26 body mass index9 18 21 24^–26 and waist circumference.21 We extended this work by examining health variables, disability and psychological factors and modelling their effects on step-counts. The associations of both gender and depression score with PA were explained by poor health and disability. The strong association between depression and disability in older people has been well described,46 but the only other study using objective PA data and able to control for disability found that depression score independently predicted step-count.20 Despite the study only including 15 subjects with diabetes, this independently predicted decreased step-count, even after controlling for body mass index, disability and general health. Ashe et al20 did not find this and it has not been examined in most other pedometer studies,9 21^–23 25 26 but it fits with the wide body of work showing the importance of PA for both primary prevention and management of diabetes47 and with the low step-counts reported in diabetes subjects.48 Psychological variables were important: exercise control (perceived control over your exercise levels) and exercise-related self-efficacy (belief in your ability to exercise) both predicted step-count. This fits with other work showing that control predicts exercise adherence in older people42 and that the Chronic Disease self-efficacy scale predicted daily step-counts.20

Other studies have highlighted the importance of dog-walking for maintaining mobility in older people.49 We believe this is the first study objectively showing the effect that dog-walking can have, resulting in approximately 1700 further daily steps, even after adjusting for confounders such as age, poor health and disability.

Implications for future work

Objective PA measurement using accelerometers and pedometers is feasible and acceptable to older people and provides researchers with a significant improvement in the characterisation of PA. Although accelerometers have the advantage of providing time-stamped activity records for up to 21 days without needing manual recording, they are expensive. Our findings show that in older people step-counts and activity counts are highly correlated and therefore cheaper pedometers may be adequate for most observational studies in this age group. The strong independent relationships seen between both diabetes and obesity and decreased step-counts underline the importance of continuing work in this area to try to increase PA in these groups.50 The associations between objective PA measures and psychological factors such as exercise self-efficacy and control suggest that their potential as possible mediators of PA should be explored further in intervention studies to increase PA.

CONCLUSIONS

This is the first population-based sample of older people with objective PA and anthropometric measures. PA levels in older people are well below recommended levels, emphasising the need to increase PA levels in this age group, particularly in those who are overweight/obese or have diabetes. The independent effects of exercise self-efficacy and control on PA levels highlight their role as potential mediators for intervention studies.