Data from the All of Us research program reinforces existence of activity inequality

Abstract

Large-scale and detailed analyses of activity in the United States (US) remain limited. In this work, we leveraged the comprehensive wearable, demographic, and survey data from the All of Us Research Program, the largest and most diverse population health study in the US to date, to apply and extend the previous global findings on activity inequality within the context of the US. We found that daily steps differed by sex at birth, age, body characteristics, geography, and built environment. Quantifying activity inequality using the modified Gini index, we found a strong correlation with obesity prevalence (R² = 0.804) and a moderate correlation with perceived walkability (R² = 0.426) and the activity gender gap (R² = 0.385). This study demonstrates the value of digital health technologies in exploring and understanding public health practices while highlighting the need to examine complexities, including biopsychosocial factors that may contribute to activity inequality.

Fig. 1. Demonstration of activity differences across US states using the AoU cohort.

a The forest plot illustrates the variation in activity levels across the 22 states included in the analysis. Lower levels of daily steps are typically found in states with higher average temperatures. The gray vertical region indicates the average daily steps of all states (including states with n ≤ 20). Each error bar indicates the 90% confidence interval. b (Figure adapted from Althoff et al. Figure 1) The distribution of two states with high activity (New York and Illinois) and two states with low activity (Texas and Arizona). New York has the highest mode (9336), and Texas has the lowest (6915). The mode of each state is indicated by dots, and dashed vertical lines correspond to the mean of each state. c (Figure adapted from Althoff et al. Figure 1) The distribution plot, normalized by the population mode of steps, reveals that New York and Arizona have similar distributions. The dark gray vertical dashed line indicates the distribution plot normalized to each state’s mode, and the light gray vertical dashed lines are placed at 50% below and above the mode.

Fig. 2. Relationship between gender and daily steps, activity inequality, and activity level.

(Figure adapted from Althoff et al. Extended Data Figure 7). a Probability density curve by sex assigned at birth of two states with high activity (New York and Illinois) and two states with low activity (Texas and Arizona). The vertical dashed lines represent the mode steps of each state. The number in parentheses represents the activity inequality measure of each state. b Activity inequality is explained by the gender gap calculated by dividing the difference in daily steps between males and females by the daily steps of males (R² = 0.385). The 95% confidence interval is estimated using bootstrapping and is indicated as the light gray bands around the regression line. c Association between average steps per day and activity quantile. On average, males take more steps per day than females in all activity quantiles. The 95% confidence interval (i.e., error bar) is constructed using bootstrapping.

Fig. 3. Obesity prevalence is associated with an increase in activity inequality.

(Figure adapted from Althoff et al. Figure 2). a Two linear regression lines were fit: with all states (R² = 0.639) and without the two outliers that were identified visually—Washington and Louisiana (R² = 0.804). b As daily steps decrease, obesity prevalence increases. There is no significant difference in this association between males and females, as indicated by the overlapping 95% confidence intervals (i.e., error bars), which is estimated through bootstrapping.

Fig. 4. Relationship between perceived walkability and activity inequality and average daily steps.

a (Figure adapted from Althoff et al. Figure 3) Perceived Walkability and activity level measured by Fitbit. Analysis with the AoU dataset displays a linear relationship between activity inequality and walkability score as indicated in dark gray (R² = 0.426, p < 0.01). b (Figure adapted from Althoff et al. Figure 3) Average steps were taken over the course of weekdays and weekends in the five most (indicated in green) and five least (orange) perceived walkability scores. c The walkability-activity gradient (WAG), which is measured by the slope of the linear regression model (y = average steps per day, x = perceived walkability) for each subgroup, represents the predicted change in daily steps for a one-unit change in perceived walkability. The gray dotted line indicates the average perceived walkability score. Individuals with normal BMI consistently recorded the highest average steps, followed by those with overweight and obese BMI. WAG was also highest for normal BMI individuals when perceived walkability exceeded 10. d For all age groups, the highest WAG was observed beyond a perceived walkability score of 10. However, the increase in steps was less pronounced in older adults compared to younger groups.