The minimum number of days of pedometer monitoring needed to estimate valid average weekly step counts and reactivity was investigated for older adults with intellectual disability. Participants (N = 268) with borderline to severe intellectual disability ages 50 years and older were instructed to wear a pedometer for 14 days. The outcome measure was steps per day. Reactivity was investigated with repeated measures analysis of variance, and monitoring frame was assessed by comparing combinations of days with average weekly step counts (with intraclass correlation coefficients [ICCs] and regression analyses). No reactivity was present. Any combination of 4 days resulted in ICCs of 0.96 or higher and 90% of explained variance. The study concludes that any 4 days of wearing a pedometer is sufficient to validly measure physical activity in older adults with intellectual disability.
Epidemiological research has demonstrated that physically active adults without disabilities, when compared with their sedentary counterparts, demonstrate reduced risk for several chronic diseases, including coronary heart disease, hypertension, and type 2 diabetes (U.S. Department of Health and Human Services, 2008). These findings can also be applied to adults with intellectual disability (ID; Temple, Frey, & Stanish, 2006). Methods of measuring physical activity include interviews and questionnaires, heart rate monitors, accelerometers, and pedometers (Frey, Stanish, & Temple, 2008; Temple et al., 2006). Interviews and questionnaires are often administered by the participants themselves with caregiver assistance or by proxy respondents, such as parents or professional caregivers. Methodological limitations of these methods are the accuracy of the instrument and the accuracy of the respondents, which are both insufficiently described in most studies in adults with ID (Temple et al., 2006). More reliable instruments to assess physical activity are accelerometers and pedometers, of which the pedometer is the most widely used and cost-effective (Tudor-Locke & Myers, 2001).
Daily walking distance or walking time is often underestimated in questionnaires (Bassett, Cureton, & Ainsworth, 2000; Richardson, Leon, Jacobs, Ainsworth, & Serfass, 1994) but is, at the same time, the most commonly reported physical activity by individuals with ID (Draheim, Williams, & McCubbin, 2002; Temple, Anderson, & Walkley, 2000). A pedometer is a movement sensor worn at the ankle or waist, and it measures physical activity by means of the number of steps taken by the participant, which enables a reliable measurement of walking (Tudor-Locke & Myers, 2001). Although the pedometer is easily applicable, some measurement issues are to be considered when using it for research purposes, such as the choice of metric (raw measure like step count or calculated measures like total distance or caloric expenditure), the monitoring frame (number of days measured), data recording and collection procedures (missing data in the absence of a memory function of a pedometer or added data when returning the pedometer by mail), or other sources of error, such as movement during travel by car, slower walking speed, and obesity (Tudor-Locke & Myers, 2001; Tudor-Locke, Williams, Reis, & Pluto, 2002, 2004).
The focus of this study concerns the monitoring frame when measuring activity with pedometers—specifically, the number of days required to obtain a reliable estimate of habitual physical activity. This subject was investigated by Tudor-Locke et al. (2005) to reduce participant burden and study costs. They concluded that any 3 days of measuring is sufficient to get a reliable estimate for weekly physical activity (Tudor-Locke et al., 2005). However, the length of the monitoring frame depends on the population in question (Tudor-Locke & Myers, 2001). Differences between physical activity patterns of the general population and adults with ID could contribute to variability between days. Adults with ID tend to have lower activity levels than the general population (Temple et al., 2006), and walking is their primary type of physical activity (Draheim et al., 2002; Stanish & Draheim, 2005). Therefore, the issue of the monitoring frame was readdressed for the population of adults with ID by Temple and Stanish (2009). From a large pooled data set (n = 154), they concluded that 3 days of monitoring steps per day predicted average weekly physical activity sufficiently in this population, too (Temple & Stanish, 2009).
Although Temple and Stanish used a relatively large data set, the study population consisted mainly of adults younger than 50 years (men: M = 35.5, SD = 11.4; women: M = 37.9, SD = 9.4). Seventeen participants were 50 years or older, and only one participant was older than 65 years. Differences in activity patterns between younger and older adults have been described for the general population (DiPietro, 2001) and are likely to be present in aging adults with ID as well. Retirement from work or day activities may cause differences in activity patterns, since these activities (and commuting to and from these activities) are suggested to contribute to the level of physical activity in this group (Temple & Stanish, 2009). Furthermore, the population assessed by Temple and Stanish consisted of adults with mild and moderate levels of ID, but differences between those subgroups were not investigated. The activity patterns of participants with borderline and severe ID have not been studied at all. It is therefore necessary to analyze the minimal time frame needed to reliably predict weekly step counts for the population of older adults with different levels of ID.
Recently, another issue in measuring physical activity in short-term pedometer studies has emerged, called reactivity, which is the positive influence of wearing the pedometer on ambulatory activity (Clemes, Matchett, & Wane, 2008). In studies with adults with normal cognitive capabilities, reactivity was present in the first 3–4 days of the measurement period and caused validity problems for short-term pedometer studies (Clemes et al., 2008; Clemes & Parker, 2009; Marshall, 2007). This could be of influence in older adults with ID as well, but this has never been a topic of research before, to the best of our knowledge.
The research questions of this study were as follows: (a) How many days of measuring step counts are necessary to predict weekly physical activity in older adults with different levels of ID? (b) Does reactivity influence daily step counts in older adults (50 years and over) with ID?
This study was part of a large cross-sectional study to measure health in older adults with ID. All clients ages 50 years and over of three care providers in the Netherlands were invited to participate, of which 1,050 clients did. Ethical approval was provided by the ethical committee of the Erasmus Medical Center and by the ethical committees of the participating care providers. Informed consent was obtained for all participants, and unusual resistance was reason for aborting measurements at all times (Medical Research Involving Human Subjects Act, 1999). Details about design, recruitment, representativeness of the sample, and communication procedures have been presented elsewhere (Hilgenkamp et al., 2011).
The accuracy and reliability of pedometers have been widely studied, and differences have been presented across different walking speeds, ground surfaces, body composition, and location of wearing. Two of these issues, slow walking speed and being overweight or obese, are likely to be present in older adults with ID (Bohannon, 1997; Lahtinen, Rintala, & Malin, 2007). The Yamax Digi-Walker is the pedometer most frequently used in research but can measure reliably only at minimal walking speeds of around 5 kph (Cyarto, Myers, & Tudor-Locke, 2004; Grant, Dall, Mitchell, & Granat, 2008; Le Masurier & Tudor-Locke, 2003; Pitchford & Yun, 2010). The NL-2000 is able to measure reliably at minimal walking speeds of 3.2 kph (0.9 m/s) and has the advantage of reliably measuring steps in overweight or obese participants as well (Crouter, Schneider, & Bassett, 2005; Grant et al., 2008; Marsh, Vance, Frederick, Hesselmann, & Rejeski, 2007). The other pedometers in this series (NL-1000 and NL-800) have the same piezo-electric measurement mechanism and therefore the same psychometric properties; the NL-1000 and NL-800 differ only in additional information they provide (distance, moderate-to-vigorous physical activity time, or calories). In this study, the NL-1000 (New Lifestyles) was used, which provides steps, distance, and minutes of moderate-intensity physical activity.
As part of a physical fitness assessment, we measured participants' stride length and comfortable walking speed by recording the number of steps and the time necessary to cover a distance of 5 m. This was repeated 3 times (Hilgenkamp et al., 2010).
Inclusion criteria to participate in this study were as follows: The participant (or his or her professional caregiver) showed no resistance to wearing the pedometer, and the participant could reach a comfortable walking speed of 3.2 kph or more in at least one of the three attempts.
Instructions concerning the use of the pedometer were given to the professional caregiver and the participant. The pedometer was attached to the belt of the participant, midway between the iliac crest and the umbilicus. Participants wore the pedometers from morning until bedtime during the monitoring period. Pedometers were removed during water-related activities (e.g., swimming, bathing, showering). The monitoring period of 2 weeks was chosen to compensate for missing days, thus ensuring that a full week of monitoring (i.e., 7 consecutive days) took place. The professional caregiver was instructed to record in a diary every evening the number of steps, distance, and activity minutes as measured by the pedometer. After 14 days, the pedometers were retrieved by the research team, and the diaries were checked for completeness. If data of any of the last 7 days were missing in the diary while the participant was wearing the pedometer, the 7-day memory of the pedometer was checked to retrieve missing results.
We computed descriptive characteristics of participants. If the 1st days of the measurement period were influenced by reactivity, they needed to be deleted in the analyses for determining the monitoring frame, so we analyzed reactivity first.
Reactivity was analyzed by comparing the results of the consecutive days with a repeated measures analysis of variance (ANOVA), with gender and level of ID as a between-subjects factor, following the analysis methods of previous research (Clemes & Parker, 2009). Days with steps per day that were significantly different from other days of the consecutive measurement period were omitted from further analyses. The Greenhouse–Geisser epsilon F was interpreted in the case of violation of the assumption of sphericity. If an overall significant F level was indicated, differences between days were examined with pairwise analyses with Bonferroni correction, for the complete group and for subgroups of gender and different levels of ID. Only participants with a complete set of data were included in this part of the analyses, because the consecutiveness of the measuring days was the point of interest. The day at which the pedometer measurement started was determined by the day of the physical fitness assessment, which could be any weekday. Because of this mix of days, this data set was suitable to assess the differences between the consecutive days. All days influenced by reactivity were deleted from further analyses.
To determine the necessary monitoring frame, the data of the consecutive 14 days were reorganized according to the days of the week. The measurement period was 14 days, resulting in two step counts per day of the week.
An extensive analysis method was used, similar to previous research about this subject (Temple & Stanish, 2009; Tudor-Locke et al., 2005). This method consisted of an analysis of the intraindividual variability compared with the interindividual variability and of three statistical analyses: ANOVA, intraclass correlation coefficients (ICC), and regression analysis to determine the suitable monitoring frame.
First, we calculated the coefficient of variation (CoV = SD/mean*100). We calculated the coefficient of the intraindividual variability (variability within an individual, CoVw) by dividing the individual standard deviation of the steps per day by the individual mean steps per day of the complete data set times 100. We calculated the interindividual variability (variability between individuals) by dividing the standard deviation of the steps per day of the complete study population by the mean steps per day of the complete study population times 100. The assumption of the coefficient of variation was a linear relationship between the mean and the standard deviation, which was checked graphically and with Pearson correlation.
To determine whether patterns were present in physical activity across the days of the week, we calculated the average steps per day (and standard deviations) of both weeks for every day of the week. This provided a data set that incorporated both weeks of data.
A repeated measures ANOVA was used to determine whether the days of the week had different mean steps per day. To investigate whether these patterns were different for men or women, participants with different levels of ID, participants with or without Down syndrome, or participants below 65 years or 65 years and over, we added these variables as between-subjects factors to study their interaction effects with the within-subjects factor days of the week. The Greenhouse–Geisser epsilon F was interpreted in the case of violation of the assumption of sphericity. If an overall significant F level was indicated, we examined differences between days with pairwise analyses with Bonferroni correction. We used ICCs to analyze how combinations of days compared with the criterion score.
To determine how many days of wearing the pedometer are necessary to predict weekly step counts, we calculated ICCs. For the analysis of these ICCs, mean steps per day (and standard deviations) were calculated for each day separately and for any combination of any number of days of the 1st week of measuring, resulting in 119 combinations of 2, 3, 4, 5, and 6 days. We examined these unique combinations to enable conclusions about combining data from nonconsecutive days, as in the case of data sets characterized by missing data patterns, as did Tudor-Locke et al. (2005). The mean steps per day of the 1st week and the mean steps per day of the complete measuring period were computed and used as the two criterion scores. For comparison with the mean of the 1st week, we included only participants with a complete 1st week of data; for comparison with the mean of the complete measuring period, we included only participants with a complete measuring period of data. Then, we compared all computed combinations with these criterion scores with ICCs for mixed effects, looking at absolute agreement between single measures (which are the mean scores of the combinations). According to the minimum level of consistency needed for multiple day observation of physical activity, an ICC of 0.8 was considered sufficient (Baranowski & de Moor, 2000).
Third, we used stepwise regression analysis to determine both how many and what types of days were necessary to predict both the criterion scores. We evaluated the total predictive value of the model by the adjusted R2, which represents the explained variance in outcome by the model. First, only one single day mean was entered in a model; second, the consecutive day was entered; then the next consecutive day was entered; and so on until the model explained 90% of the variance or over. We used SPSS Version 15.0 for all analyses, and the significance level was set at p = .05.
Descriptive characteristics of participants are presented in Table 1. Of the 1,050 older adults who participated in the large epidemiological study, 268 complied with the inclusion criteria for the assessment of physical activity with a pedometer.
A complete data set of 14 consecutive days of measuring was available for 135 participants, and these were included in this part of the analyses. The assumption of sphericity was violated, so Greenhouse–Geisser's F was used and proved to be significant (F = 8.031, df = 9.491, p < .001). Pairwise comparisons revealed that only the 1st day was significantly different from all other days, which was due to different moments of the day the participant was given the pedometer and started wearing it. The data from this day were omitted in further analyses. No other differences between the means of the days of measuring were noticed (see Figure 1), neither for the complete group nor for subgroups of gender or different levels of ID.
A linear relationship between the mean and standard deviation of steps per days was confirmed graphically (Pearson r = .69, p < .001). CoVw (intraindividual variability) was 42.76%, and CoVb (interindividual variability) was 54.49%.
Steps per day of every weekday of the measurement period and the mean per day of both weeks are presented in Table 2. When performing the repeated measures ANOVA, the assumption of sphericity was violated. Therefore, Greenhouse–Geisser's F was used. The results of this analysis revealed a significant effect for the within-subjects factor days of the week (F = 15.413, df = 5.286, p < .001). All weekdays did not differ significantly from the other weekdays, Sunday had significantly lower step counts than any other day of the week, and Saturday had significantly different lower counts than Tuesday and Thursday. No interaction effects were found between days of the week with gender, level of ID, Down syndrome, or age younger than 65 years, which means that the patterns of physical activity across the days of the week were not different for any of these groups. Consequently, data of all participants were combined for subsequent analyses.
The mean, median, and range of ICCs for any single and any combination of days with both the criterion scores (1-week mean and mean of total data set) are presented in Table 3, including the number of used combinations and the range of the numbers of participants used to calculate the ICC for every combination. Data of the complete 1st week were available for 183 participants, whereas data of the complete measuring period (13 days) were available for 136 participants.
The adjusted R2s of models with all possible consecutive combinations of 1, 2, 3, and 4 days are presented in Table 4. Explained variance of 90% (adjusted R2 = .90) was reached in four of the models with 3 consecutive days (if starting at Monday, Tuesday, Saturday, or Sunday), whereas if starting on Wednesday, Thursday, or Friday, 4 consecutive days of wearing the pedometer were necessary to reach 90% variance explained.
Reactivity was not present in the pedometer data of older adults with ID. Steps per day in the weekend were lower than during the week: Sunday had significantly lower steps per day than any other day of the week, and Saturday also had significantly lower steps per day than Tuesday and Thursday. A monitoring frame of any 4 days was sufficient to predict weekly step counts, with both an agreement of 0.8 or higher and explained variance of 90%.
This conclusion is somewhat different from conclusions drawn in previous research (Temple & Stanish, 2009; Tudor-Locke et al., 2005). These earlier studies concluded that 3 days of measuring is sufficient to predict weekly step counts in adults with and without ID, on the basis of ICCs and regression analyses. The difference in the current study is that regression analysis was performed not only for models with consecutive days starting with the single day that explained the most of the variance (such as in previous research) but for models with consecutive days that could start on any one of the 7 days of the week. In the current study, 3 days were also sufficient to explain 90% of the variance in the best models, but in 3 out of 7 starting days, 90% of the variance was explained only when 4 consecutive days were included.
This short monitoring frame, which is valid for consecutive days as well as randomly distributed days, is useful in case of missing data. For the population studied here, it means that the data for 11 participants can be included in further analyses—that of those who had successfully worn the pedometer for 4–6 days and would have been excluded if using a monitoring frame of 7 days.
The intraindividual variability was smaller than the interindividual variability (42.76% vs. 54.49%) but still very large. In such heterogeneous study populations, the ICC tends to be higher because it is a ratio of the true variance between subjects divided by the sum of the true variance and other causes of variance (variance due to measurement error, interaction, or random error). When the true variance is high, the relative contribution of the other causes of variance is automatically lower, resulting in high ICCs. In heterogeneous populations, such as the current study population, the ICC is therefore not the recommended method of analyzing agreement, and the conclusion of a monitoring frame of 4 days in this study was based on the regression analyses.
Other potential threats to an accurate, reliable, and valid measurement of physical activity with a pedometer were mentioned by Tudor-Locke & Myers, 2001, such as the used metric, procedures for data recording and collecting, slower walking speeds, and automobile travel.
This study followed the recommendations of these authors regarding the used metric and used raw step counts, in contrast with calculated metrics like the total distance covered, which could introduce possible error, for example, for individuals with smaller stride lengths.
Procedures for data recording and collecting were set up to minimize missing data (memory function in the pedometer) and error through the logistics of transportation or mailing of the pedometers after the data collection period back to the research team (the pedometer was picked up at the same day by the research team).
Slower walking speeds can cause an error in the estimated number of steps because pedometers have a minimum speed at which they can measure steps reliably. To minimize the influence of this error, we measured comfortable walking speed in advance during the physical fitness assessment, and only participants with sufficiently fast comfortable walking speeds were included. We realize that these included participants might in reality sometimes walk too slowly as well (as in the general population), but most of their walking would have been at or above their own comfortable walking speed and therefore reliably recorded.
Obesity was mentioned as a possible source of error, too, but the pedometer used in this study was tested as reliable in measuring overweight and obese participants as well (Crouter et al., 2005).
A limitation of this study is the lack of information on the amount of automobile travel during the data collection period, which could have influenced the step counts (Tudor-Locke & Myers, 2001).
The large number of participants is a strength of this study, which enabled a detailed analysis of patterns across the days of the week and the minimal monitoring frame. Although the study population of 268 participants out of a total 1,050 participants is a small, selected group, this bias does not interfere with the results in this article because no interaction effects were present for gender, level of ID, Down syndrome, or age younger than 65 years on the basis of the repeated measures analyses. In further analyses about the actual results of physical activity, this bias should be reckoned with when drawing conclusions.
Another advantage in this study is the initial monitoring frame of 14 days. Patterns across days of the week were analyzed with means of the days of both these weeks, which is more stable than data for 1 week only. Furthermore, the predictive value of different monitoring frames could be compared not only with the 1-week average but also with the average of the complete set of days, which gives more insight into the validity of the monitoring frame.
In conclusion, if measuring physical activity in older adults with ID, it is important to use an appropriate pedometer, which can measure steps reliably at slow walking speeds and in overweight participants. To predict average weekly step counts, participants need to wear the pedometer successfully for 4 days (and not necessarily consecutively). This reduces the burden for participants as well as for caregivers when participating in research or intervention projects, and the costs and duration of research will be further minimized.
Editor-in-Charge: Glenn Fujiura
Thessa Hilgenkamp (e-mail: firstname.lastname@example.org), Erasmus Medical Center, Department of General Practice, Intellectual Disability Medicine, P.O. Box 2040, Rotterdam, Rotterdam 3000 CA, the Netherlands; Ruud Van Wijck, University of Groningen, Groningen, the Netherlands; Heleen Evenhuis, Erasmus University Medical Center, Rotterdam, the Netherlands.