Researchers Use Fitbit Distribution to Collect Representative Health Data

In a recent groundbreaking study, researchers demonstrated that distributing wearable devices like Fitbits to a diverse group of participants results in more equitable and accurate health data collection compared to relying solely on individuals who already own such devices. Traditional data collection methods often overrepresent wealthy, urban, white, and physically active populations, which can lead to biases in health research. To address this, Ritika Chaturvedi and her team recruited 1,038 participants for the American Life in Realtime (ALiR) project, a longitudinal health study using probability sampling techniques. Participants received Fitbits and tablets, ensuring broad demographic representation across race, education, and income levels, including underrepresented groups like minorities and older adults. The study compared COVID-19 detection models trained on ALiR data versus the NIH's All of Us program, which includes over 14,000 individuals who already owned wearable devices. Results showed that models developed with ALiR data performed consistently across different demographic subgroups, whereas models based on existing wearable owners performed significantly worse among older women and non-white populations. The findings suggest that providing devices and using probability sampling effectively removes participation barriers, leading to high-quality, inclusive health data. Such approaches hold promise for developing AI tools that work equally well across all populations, ultimately advancing health equity. The full study is published in PNAS Nexus.