(F-240) Using data from the All of Us Research Program for validation of a key digital health study

Introduction:: Recent technological advancements enable individuals to assess and track their physical activity using mobile and wearable devices in free-living conditions. A large-scale digital health study on >700k smartphone users across >100 countries demonstrated a correspondence between activity inequality and obesity, city walkability, and activity level (defined by average daily steps). Such findings, if found to be robust and replicable, could lead to key policy changes impacting public health. Further, insights from these studies can serve as a benchmark of physical activity and well-being that can be leveraged to evaluate the effects of future policy changes. However, the findings of such large digital health studies have yet to be replicated due to a lack of data available from independent studies.


The All of Us (AoU) Research Program is an NIH initiative that seeks to build a robust ecosystem of communities and researchers and strives to deliver the largest and most diverse publicly-available biomedical dataset with a focus on populations usually under-represented in biomedical research [2]. Specifically, the program is accumulating multiple streams of health-related information such as electronic health records, genomics, physical exam measures, participant surveys, and wearables such as Fitbit from 1,000,000 or more Americans. In this research, we seek to replicate the previously-found disparities in physical activity (across geolocation, obesity, and gender) by leveraging longitudinal digital health data measured using Fitbit fitness trackers from AoU.


Materials and Methods:: From the Controlled Tier dataset v6 (C2022Q2R6), we selected 6,525 participants who met the following criteria: provided any Fitbit data between Jan 2016 to Dec 2019, age > 18, completion of the basic survey at the time of enrollment, and in-person physical measurement visit. After excluding participants who did not meet the device adherence criteria and participants with exceedingly large average step counts (i.e., outliers defined by z-score > 4), we performed all analyses with 6,506 participants.

To examine the relationship between activity inequality and various factors such as geographic location, biological sex, and obesity, we used the methods described in Althoff et al.’s paper, which can be interpreted as an indicator of how dispersed the data is–values closer to 1 indicate higher inequality (or variance) within the population.

To investigate the geographic disparity in activity, we created a heat map to visualize the mean of all individuals’ daily steps in each state (Figure 1). The map of the United States was visualized using the GeoPandas package, and states with ≤ 20 participants were excluded as per the data privacy and protection guidelines from the AoU Researcher Workbench. Next, the association between activity inequality and obesity was examined for states with > 30 participants (Figure 2). A greater number of participants per state was used to ensure sufficient data points to compute the activity inequality measure. We applied the Locally Weighted Scatterplot Smoothing to obtain the second-degree polynomial trendline.


Results, Conclusions, and Discussions:: The large-scale study conducted by Althoff et al. revealed variability in daily steps across 111 countries [1]. Our findings concur with this observation, as we have noted variations within the United States (n=18 states) wherein the southern region exhibits lower daily steps than the western or northern regions (Figure 1). The five most active states, as defined by daily steps, were New York (9,541 daily steps, n=188), Colorado (9,383, n=33), Illinois (8,631, n=526), California (8,588, n=988), and Wisconsin (8,587, n=925).

Althoff et al. suggested that activity inequality is associated with higher obesity levels at an international level [1]. Our findings indicate a similar trend in the US among the 12 states with n ≥ 30. The fraction of obese individuals (BMI ≥ 30) in each state increases with activity inequality (LOWESS R2 = 0.722) (Figure 2). While activity inequality exists across the US, the range of calculated Gini coefficients (0.189 to 0.228) was smaller than that of Althoff et al.’s study (0.245 to 0.325), likely due to lower lifestyle and cultural variations between states, as opposed to countries. As a result of this more homogenous lifestyle within the US, the fraction of obese individuals across the 13 states included in our study had a smaller range (0.20 vs. 0.24) and significantly higher mean (0.41 vs 0.14) than the summary statistics for the 46 countries included in the Althoff et al. study.

This study demonstrates the utility and usefulness of the AoU dataset as a large-scale data source for the US population. In consideration of differences in the research methods, the results from this study largely follow outcomes from a key prior study. This suggests that the AoU dataset can be used to both validate future studies and translate results from such studies to the real-world US population. The insights gained from this study and future investigations using the AoU dataset can provide valuable information for understanding physical activity and the associated health outcomes (e.g., obesity) for the US population.


Acknowledgements (Optional): :

References (Optional): : 1. Althoff T, Sosič R, Hicks JL, King AC, Delp SL, Leskovec J. Large-scale physical activity data reveal worldwide activity inequality. Nature. 2017;547(7663):336-339. doi:10.1038/nature23018

2. The All of Us Research Program Investigators. The “All of Us” Research Program. N Engl J Med. 2019;381(7):668-676. doi:10.1056/NEJMsr1809937