In recent years, wearable health technology has emerged as a promising frontier for precision medicine, offering continuous, real-time data streams capable of transforming public health research. Yet, a persistent challenge remains: the demographic skew inherent in most datasets derived from consumers who already possess these devices. Typically, users of wearable technology such as smartwatches and fitness trackers tend to be wealthier, predominantly urban, and disproportionately White, leaving critical gaps in representation. New research led by Ritika Chaturvedi and colleagues confronts this disparity head-on, demonstrating that deploying probability-based sampling coupled with the provision of wearables to participants yields far more equitable and clinically relevant health data.
The landmark study, published in PNAS Nexus on October 7, 2025, details the methodology and results from the “American Life in Realtime” (ALiR) initiative. ALiR recruited a cohort of 1,038 participants from the Understanding America Study, a nationally representative, probability-based sample of adults in the United States. Unlike previous large-scale health studies relying on convenience samples—populations who already own wearable devices—the ALiR study provided Fitbits and tablets directly to participants. This approach effectively eliminated financial and technological barriers to participation, creating a balanced and inclusive cohort stratified across race, education, income, and age.
One of the pivotal outcomes of the ALiR project was its ability to produce health data that genuinely reflect the diverse tapestry of the American population. In stark contrast, data from the National Institutes of Health’s All of Us Research Program—consisting of over 14,000 participants who self-reported owning wearable devices—showed significant demographic skew. The All of Us dataset was heavily weighted toward younger, affluent, White individuals, demonstrating markedly poorer data quality and model performance when applied to minority groups and older women, with detection performance declining by 22 to 40 percent.
This disparity becomes critically salient in the context of COVID-19 detection models. Using wearable sensor data to identify probable infections enables real-time monitoring and intervention; however, model generalizability depends fundamentally on demographic inclusiveness. ALiR’s model exhibited robust performance across all demographic subgroups, underscoring the power of representative sampling and device provision to democratize health data analytics. These findings underscore a vital principle: AI and machine learning models trained on biased datasets inherently encode those biases, perpetuating health inequities unless corrected at the data collection phase.
The implications of this research ripple beyond COVID-19 detection, extending into the broader realm of precision health. Wearable technologies, by capturing physiologic markers such as heart rate variability, activity patterns, and sleep metrics, offer unprecedented granularity. Yet, these advantages can only be fully leveraged when datasets encompass the full population spectrum. ALiR’s methodology exemplifies a scalable, ethically aligned framework, bridging the technology access divide and enhancing the scientific community’s ability to generate valid, actionable insights for all demographic segments.
Technically, ALiR utilized longitudinal data collection techniques, enabling measurement of intra-individual variability over time rather than cross-sectional snapshots. This temporal richness permits advanced modeling algorithms—such as recurrent neural networks and ensemble methods—to detect subtle physiological changes indicative of disease onset. Furthermore, the study integrated multimodal data streams, combining wearable sensor outputs with participant-reported symptoms and demographic metadata to refine model accuracy. This integrative approach exemplifies best practices in person-generated health data analytics, echoing calls from health informatics experts for holistic data capture frameworks.
The logistics of deploying wearables and digital tablets to a representative cohort posed unique challenges. Nevertheless, the research team employed robust protocols to ensure device compliance and minimize attrition, including regular participant engagement via digital platforms and remote technical support. This operational rigor highlights the viability of incorporating technology distribution into large-scale epidemiological studies, pointing toward future initiatives where researchers actively democratize participation by removing socioeconomic hurdles.
Moreover, the ALiR study contributes to the growing movement toward open science and reproducibility. By creating a publicly available benchmark dataset, the authors enable other researchers to validate findings, refine computational models, and innovate upon the foundation of equitable data practices. Such transparency accelerates scientific progress and fosters a collaborative ecosystem where AI-powered health tools can be co-developed with conscientious attention to social determinants of health.
This work also invites reflection on the ethical parameters governing health data collection and AI deployment. Providing devices within a probability sample framework aligns with principles of justice and beneficence, actively redressing the underrepresentation of marginalized groups historically excluded from clinical datasets. It challenges stakeholders—from policymakers to technology companies—to reconsider data acquisition paradigms that inadvertently entrench inequity and calls for structural reforms emphasizing inclusion at the point of data generation.
Looking forward, integrating probability sampling models with large-scale wearable deployments could revolutionize population health surveillance and precision medicine. By empowering diverse communities with access to cutting-edge digital health tools, researchers can garner authentic, real-world evidence across ailments ranging from infectious diseases to chronic conditions such as cardiovascular disease and diabetes. These advancements promise to reduce health disparities by ensuring AI algorithms operate with equitable sensitivity and specificity across all societal segments.
In conclusion, the American Life in Realtime initiative marks a critical turning point in wearable device research and precision health. By dismantling barriers to participation and prioritizing representative data collection, it lays the groundwork for AI-driven health interventions that truly serve everyone—not just the privileged few. As wearable technologies continue to proliferate, adopting inclusive research designs like ALiR’s will be paramount to realizing their full potential as engines of equitable healthcare innovation and public health resilience.
Subject of Research: Equity in precision health through representative wearable data collection.
Article Title: American Life in Realtime: Benchmark, publicly available person-generated health data for equity in precision health
News Publication Date: 7-Oct-2025
References:
Chaturvedi, R., et al. (2025). American Life in Realtime: Benchmark, publicly available person-generated health data for equity in precision health. PNAS Nexus.
Keywords: Public health, wearable technology, precision medicine, health equity, artificial intelligence, longitudinal health data, COVID-19 detection, demographic representation
