Step it up for Research

By Briana Cameron, Ph.D.,  and Robert Gentleman, Ph.D.*

 

Regular physical activity is important for good health. Research has shown that exercise can help prevent such things as cardiovascular disease and diabetes (1, 2). But just how much activity does the average person get? And do people in good health have distinctly different activity patterns compared to those in poor health?

 

To answer those questions, 23andMe researchers initiated a study to look at how age, sex, and BMI are related to differences in physical activity, using step data recorded on smartphones. Customers who consented to participate in this research use their 23andMe iOS app to provide step count data collected through the Apple® HealthKit integration (3). HealthKit allows iPhone®  users to collate health information on their phones, smartwatches, or apps in a single location; customers participating in research can then allow 23andMe scientists to access and study the data. (Customers can learn about their own contributions by looking at the Research Overview page.) When combined with aggregated and de-identified customer genetic data from others who are participating in research as well as information from 23andMe’s survey-based research program, researchers can start to understand how physical activity is related to different measures of good health.

 

Steps Data

While these data can be very useful and show us trends in behavior, it is important to acknowledge some of the limitations. The steps data reported on a phone are an imperfect record of a person’s physical activity. There are many different factors that can affect the steps recorded on an iPhone, such as not taking your phone with you every time you walk somewhere or keeping your phone in a backpack or a purse. In addition, the algorithm used to infer steps from the motion detector is not perfect. However, as we show here, the data do align reasonably well with activity levels and, when considered in aggregate over relatively long periods of time, do provide some indications of overall activity level. Many of the concerns affect comparisons between people more than comparisons between time periods for a person.

 

Some of what we learned aligns with reasonable expectations. For example, the data show that daily activity patterns do differ between younger and older customers (Figure 1). On average, customers older than 50 (n=7,692) record fewer steps in a day than those under 50 (n=4,898).

Figure 1. Average number of recorded steps by day of the week, separated by age

 

 

Figure 2 below shows activity level on an average day for both older (n=7,692) and younger (n=4,898) customers across both weekends and weekdays (solid line indicates weekday average, dotted line indicates weekend average). In this figure, the average number of steps taken in a 15 minute interval is shown in both age groups across weekends and weekdays.

 

Figure 2. Average number of recorded steps in 15 minute interval for customers for both weekends and weekdays, separated by age.

The weekday activity pattern in the younger age group also shows a stronger dependence on a typical workday schedule; activity peaks are seen in the morning (8-9am), during lunch (12-1pm), and after work (5-6pm) with dips in activity between these points. This pattern is less pronounced in the older group, perhaps due to customers who have already retired and are no longer tied to a typical work schedule.

 

Neither age group adheres to this schedule on weekends, however. For both groups, the increase in recorded steps typically occurs later on the weekends, evidence of a tendency to sleep in on those days. And while the typical weekend patterns looks highly similar between the age groups, younger customers are more likely to be stepping out after midnight than older customers.

 

Step Differences Among the Sexes

Observed activity habits also differ between men (n=6,081) and women (n=6,509):

 

Figure 3. Average total daily recorded step count by day for males and females

Men record more steps in a day than women across all days of the week (Figure 3). But these differences may have more to do with differences in phone use between sexes, and not necessarily that men are more active than women. Women, for example, may be less likely to have pockets and may be more likely to leave their phones on a desk or table, which would also decrease the number of recorded steps. In general, both sexes tend to follow the same general trend, with Sundays typically being the day with the least amount of activity recorded. Activity tends to be fairly consistent throughout the weekdays. Weekend activity (Friday and Saturday) tends to be highest for both sexes.

 

Steps, BMI and Exercise Frequency

Our researchers also saw distinct differences in activity between customers with different BMI (Figure 4). To study the association between activity levels and BMI, customers were classified as normal weight, overweight, or obese by NIH standards (4). A weekly average for total daily recorded steps was assessed for each week from November 2017 until January 2019 for each BMI category.

 

Figure 4. Average total daily recorded step count by day for each BMI classification

Customers falling into the obese category (n=4,245) tend to record lower daily step counts than customers who are either overweight (n=4,295) or normal weight (n=3,535). While the difference between overweight and normal weight customers is smaller, customers of a normal weight BMI show the highest average recorded step count.

 

Another interesting aspect is the role of seasonality. For all BMI classifications, activity tends to be slightly lower around the holidays (late December – early January) with increased activity during the rest of the year. This increased activity is more evident in both the normal weight and overweight groups; the obese BMI group shows a slightly more stable activity pattern throughout the year. With only one year of data collected, however, more data will be needed to study the consistency of this observed trend and the role that season may play in activity level across different BMI classifications.

 

Finally, the relationship between self-reported exercise frequency and the average total daily step count on weekdays was explored (Figure 5).

 

Figure 5: Average daily steps by self-reported exercise frequency

As expected, customers who report exercising more frequently have a higher average daily step count. This agreement suggests that using phone-based step counts may be a good measurement of general activity level.

 

The integration of self-reported health information and measures of activity, such as step count, provides an interesting opportunity for researchers to understand how physical activity is related to other aspects of health and wellness. Studying trends in activity over time can help researchers pinpoint the activity patterns of healthy people, and move closer to understanding what changes in activity can be made to improve health and well-being. In addition, while studying the activity patterns of a general population may lead to interesting insights, it may be possible to study activity and mobility among people with certain health conditions, such as arthritis, to see how movement may be related to symptom severity.

 

While providing a large amount of near-continuous activity data, phone-based collection of movement is subject to certain limitations that should be addressed. Not all people consistently carry their phone with them, and these differences in phone-use behaviors may limit the comparability of movement patterns between individuals. Furthermore, issues such as battery life and variations in step count algorithms between different devices may create further challenges in making precise statements about activity patterns. Even with these limitations, however, the use of mobile devices to track activity presents a unique opportunity for researchers to observe a much more complete picture of activity than is usually captured via other data collection methods.

 

IOS is a trademark or registered trademark of Cisco in the U.S. and other countries. Apple, iPhone and HealthKit are trademarks of Apple Inc., registered in the U.S. and other countries.

 

*Robert Gentleman, Ph.D., is 23andMe’s vice president of computational biology. Briana Cameron is a scientist in Biostatistics at 23andMe.