Association of chronic disease risk and physical activity measured by wearable devices in the All of Us program

Abstract

Background

Physical activity plays an important role in preventing chronic diseases, but most studies rely on self-reported or short-term data that fail to capture habitual behavior. This study utilizes Fitbit data to investigate the relationship between physical activity and various chronic diseases.

Methods

We analyzed data from 22,019 participants in the All of Us Research Program who shared at least six months of Fitbit activity data linked with electronic health records. Various physical activity patterns were evaluated using Cox proportional hazards and logistic regression models, adjusting for age, sex, and body mass index (BMI). To test robustness, sensitivity analyses were conducted using obesity defined by BMI, applying a two-year exclusion window for outcome diagnoses to mitigate potential reverse causation, and incorporating lifestyle covariates (smoking and alcohol use) under a simplified directed acyclic graph (DAG) framework to address residual confounding.

Results

Here, we show that higher physical activity levels are associated with lower risks of multiple chronic conditions. Higher daily step counts were negatively associated with obesity and type 2 diabetes, while greater elevation gains and longer vigorous activity are associated with lower risks of conditions such as morbid obesity, obstructive sleep apnea, and major depressive disorder. All sensitivity analyses yield consistent results, supporting the robustness of findings against reverse causation and lifestyle confounding.

Conclusions

Higher physical activity and lower sedentary time may help prevent diverse chronic diseases. These findings demonstrate the potential of large-scale wearable data to inform personalized prevention and population health strategies.

Plain language summary

This study looked at how physical activity affects the risk of developing long-term health problems. Researchers used data from people who wore Fitbit devices to track their daily activity, including step count, exercise intensity, and how long they were active or sitting. These data were linked with each person’s medical records to understand how activity levels related to chronic diseases such as obesity, diabetes, and depression. The study found that people who were more active—taking more steps, climbing more elevation, and spending more time in intense activity—had a lower chance of developing these diseases. The results suggest that being more active and sitting less can help people stay healthier and may support future public health advice and personal health planning.

Hou et al analyze Fitbit and electronic health record data from over 22,000 participants in the All of Us Research Program to examine links between daily physical activity and chronic disease risk. Higher activity and lower sedentary time are associated with reduced risks of obesity, diabetes, and other chronic conditions.

Introduction

Chronic diseases, including cardiometabolic disorders, obesity, and major depressive disorder (MDD), are the main causes of morbidity and mortality worldwide^1–3. Given the critical role of lifestyle factors in their onset and progression, physical activity has garnered increasing attention as a key modifiable determinant^4,5. Wearable devices enable continuous, noninvasive monitoring of physical activity, providing expanded opportunities to explore its association with chronic disease outcomes. Prior research has established links between physical activity and multiple chronic diseases, such as cardiovascular disease, type 2 diabetes, depression, Parkinson’s disease, and multiple sclerosis^4–8. However, many studies rely on self-reported activity data or short-term objective assessments via accelerometers or pedometers. Self-reported data, though informative, often diverge from objectively measured activity levels^9,10.

The emergence of commercial wearables, such as Fitbit, offers a means to address these limitations by enabling long-term, objective activity tracking. Fitbit-based studies have validated their accuracy in measuring steps, heart rate, and energy expenditure^11,12. These devices generate longitudinal data, providing valuable insights into individual activity patterns and facilitating large-scale epidemiological investigations of activity-related chronic disease risk.

The All of Us Research Program (AoU), a National Institutes of Health (NIH) initiative, seeks to collect health data from over one million individuals across diverse U.S. populations^13,14. The program integrates electronic health records (EHRs), genomic data, physical measurements, and participant surveys. As of version 8, more than 59,000 participants have contributed Fitbit data, encompassing activity and sleep metrics. Together, these data enable large-scale analyses of longitudinal associations between objectively measured activity behaviors and chronic disease outcomes in a diverse U.S. cohort.

In this study, we analyze long-term Fitbit activity data linked with EHRs from the AoU to explore how daily physical activity patterns relate to chronic disease risk. We find that higher physical activity levels, including greater step counts and vigorous activity, are associated with lower risks of obesity, diabetes, depression, and other chronic conditions. These findings highlight the potential of large-scale wearable data to inform personalized health recommendations and preventive strategies in the general population.

Methods

Study participants

All participants provided informed consent to join the AoU Research Program, a large-scale NIH-funded initiative designed to advance research in health and disease^13,14. Eligible participants included all U.S. adults (aged ≥18 years), excluding individuals who were incarcerated or unable to provide independent consent. This study utilized AoU version 8 controlled-tier data (C2024Q3R4) from the AoU Researcher Workbench (https://workbench.researchallofus.org/). At the time of analysis, data were available from 633,540 individuals enrolled between May 2018 and October 1, 2023¹⁵. Data collection included demographic information, survey responses from digital enrollment, physical measurements and vital signs recorded by partner healthcare provider organizations, voluntarily shared EHRs, and Fitbit data linked via participants’ Google Fitbit accounts within the AoU portal^13,14,16.

Fitbit activity data

This study utilized Fitbit-derived daily activity data to capture detailed physical activity patterns of participants. The dataset included activity calories, basal metabolic rate (BMR) calories, total calories burned, elevation gain, activity minutes at different intensity levels, and total step count. As described in the AoU Fitbit documentation¹⁷, activity intensity levels are assigned based on heart rate zones relative to maximum heart rate, and can be obtained directly from the table in the daily activity summary without recalculating. Fitbit data provides an advantage over traditional self-reported activity measures by offering objective, continuous, and noninvasive tracking. The inclusion of detailed activity metrics, such as light, moderate, and vigorous intensity levels, enables a comprehensive assessment of participants’ activity profiles over time. Fitbit data were collected from participants who linked their Fitbit accounts through the AoU portal, enabling longitudinal tracking of activity behaviors.

We summarized daily activity metrics by averaging activity patterns over each participant’s monitoring period to derive a representative activity profile. This approach facilitated characterization of individual variability and investigation of associations between activity behaviors and health outcomes. Additionally, these metrics enabled differentiation of activity intensities, offering insights into both total activity volume and the composition of activity types and intensities. Wear-time compliance was estimated using minute-level heart rate recordings from the Fitbit “heart rate minute level” table. We classified a minute as wear time if at least one heart rate value was recorded, then computed, for each participant, the average daily wear time and the total number of days with valid heart rate data. A valid monitoring day was defined as a day with at least 600 min of recorded heart rate data, consistent with common practice in wearable-sensor epidemiology¹⁸. Device provenance was inferred from the participant portal flag. Participants were classified into two groups: bring-your-own-device (BYOD) and study-provisioned (WEAR).

Preprocessing

The dataset was accessed via Google BigQuery, extracting all participants with available Fitbit data. SQL queries were used to extract activity summary records, including activity calories, BMR, total calories burned, elevation gain, activity minutes (light, moderate, and vigorous), and total steps.

A cohort filtering process was applied to ensure participants met key inclusion criteria (Supplementary Fig. 1). First, participants aged 18 years or older were included. Age was determined by comparing the first recorded Fitbit activity date with the participant’s date of birth, retaining only those classified as adults at their first activity record. Next, participants were required to have at least six months (≥180 days) of recorded Fitbit activity. This criterion, commonly applied in prior research^19–21, ensured sufficient data quality and excluded participants with limited activity records that could yield unreliable results. Lastly, only participants with linked EHR data were included. This criterion ensured the availability of EHR data for all participants, allowing assessment of associations between Fitbit-derived activity metrics and clinical outcomes.

Outcomes

Study outcomes were derived from diagnostic events recorded in participants’ EHRs. Diagnostic events were identified using International Classification of Diseases (ICD) codes from both ICD-9 and ICD-10 versions. ICD codes were mapped to 1765 distinct phecodes using the PheWAS catalog’s Phecode Maps^22–24, ensuring consistent diagnosis categorization. Each phecode was associated with its corresponding phenotype and diagnostic category. To ensure statistical robustness, phecodes with fewer than 100 cases were excluded.

Statistics and reproducibility

We used multiple logistic regression models to assess the association between physical activity metrics (e.g., steps, activity calories, and minutes at different activity intensities) and health outcomes. Study outcomes were defined as the presence or absence of specific diagnoses, represented by phecodes derived from EHR data. To enhance robustness, participants diagnosed with a given phecode within the first 180 days of Fitbit monitoring were excluded. Analyses were adjusted for age, sex, and BMI at baseline. Statistical significance was determined using two-sided p-values, with values less than 0.05 deemed significant. Bonferroni correction was applied to adjust for multiple comparisons.

We further performed Cox proportional hazards regression models to explore the association between physical activity metrics and selected chronic diseases. Chronic diseases were selected based on prior research linking them to physical activity^25–32. Primary predictor variables were Fitbit-derived activity metrics, including activity calories, elevation, activity minutes at different intensities (e.g., lightly active, moderately active, and vigorously active), and daily step count. Outcomes were defined as EHR-recorded diagnosis events, represented by phecodes. For each phecode, we first calculated baseline activity metrics as the average values over the initial 180 days following each participant’s first recorded Fitbit activity. Participants with fewer than 15 valid days in this baseline period were excluded to ensure reliability. Following the baseline period, activity metrics were aggregated into monthly (30-day) intervals using available Fitbit data, and the time-varying exposure assessment continued until the end of data availability, censoring, or diagnosis, whichever occurred first. These monthly averages were modeled as time-varying covariates in the Cox regression framework. To mitigate reverse causation, participants diagnosed before or within the first 180 days of Fitbit activity monitoring were excluded from analysis for that specific phecode. Cox models were fitted with individual activity metrics as predictors, adjusting for age, sex, and BMI. The event variable indicated whether a diagnosis occurred during follow-up, while the duration variable represented the number of days from baseline to diagnosis or the end of follow-up. To enhance numerical stability and interpretability, daily step counts were scaled in 2000-step increments when assessing their associations with chronic disease outcomes. Elevation gain was analyzed in 100-meter increments to evaluate its effect on disease risk. Durations of different activity intensities (e.g., lightly active, moderately active, and vigorously active) and sedentary time were converted to hours to assess the impact of each additional hour on disease risk. We evaluated Pearson correlations between Fitbit-derived activity measures to assess multicollinearity prior to model fitting. To reduce statistical instability due to high correlation among Fitbit-derived metrics, each activity variable was analyzed in a separate model.

Statistical analyses were performed using Python (v3.10, https://www.python.org/) on the AoU Researcher Workbench, a secure cloud-based platform. Descriptive statistics were used to summarize participant characteristics and Fitbit activity metrics, reported as means ± standard deviations or medians, interquartile ranges (IQR) as appropriate. The analyses were conducted by the libraries including Pandas (version 2.2.3)³³, NumPy (version 1.24.4)³⁴, Statsmodels (version 0.14.4)³⁵, scikit-learn (version 1.6.0)³⁶ and Lifelines (version 0.30.0)³⁷. No repeated experimental measurements or biological replicates were involved, as analyses were based on observational longitudinal data from the AoU. All results were reproducible using the same inclusion criteria, preprocessing procedures, and analysis pipelines described herein.

Sensitivity analysis using BMI for obesity classification

To evaluate the robustness of our findings, we conducted a sensitivity analysis by defining obesity based on BMI rather than EHR-based ICD codes. This alternative classification aimed to capture cases that might not have been explicitly diagnosed in the EHR but met the clinical criteria for obesity (BMI ≥ 30 kg/m²). Height and weight records were extracted from the AoU. Participants were classified as having BMI-defined obesity if they had at least one recorded BMI ≥ 30 kg/m² at any point. To reduce the risk of reverse causation, we excluded participants who had a BMI ≥ 30 kg/m² before the start of Fitbit monitoring or within the first 180 days of Fitbit activity tracking. We repeated the logistic regression and Cox proportional hazards regression models analyses using BMI-defined obesity as the outcome variable.

Sensitivity analysis based on two-year outcome exclusion

To address concerns of reverse causation, our primary analysis excluded participants diagnosed with each condition before or within the initial 180 days of Fitbit monitoring. In a sensitivity analysis, we applied a more stringent exclusion criterion by removing participants diagnosed within the first 730 days (2 years) of monitoring. This allowed us to examine the robustness of our findings to stricter temporal separation between exposure and outcome.

Sensitivity analysis with lifestyle covariates

To address potential lifestyle confounding, we leveraged the Lifestyle survey within the AoU, which collects self-reported information on tobacco, alcohol, and other substance use behaviors. We extracted relevant items corresponding to smoking and alcohol consumption (e.g., “smoking: 100 cigs lifetime,” “smoking: current daily cigarette number,” “alcohol: alcohol participant,” and “alcohol: 6 or more drinks occurrence”). Participants were classified as current smokers if they reported current cigarette, cigar, hookah, e-cigarette, or smokeless tobacco use, and as nonsmokers otherwise. Similarly, participants were classified as alcohol users if they reported current or past-year alcohol consumption, regardless of frequency, and as nonusers if they reported no alcohol use. Logistic and Cox proportional hazards models were re-estimated, including these two lifestyle covariates as additional confounders, alongside baseline age, sex, and BMI. This covariate structure reflects a simplified directed acyclic graph (DAG) framework, representing a minimal sufficient adjustment set of demographic and lifestyle factors commonly associated with both physical activity and chronic disease outcomes (Supplementary Fig. 14).

Ethics declarations

The AoU Research Program has received approval from the AoU Institutional Review Board (IRB), and all participants provided informed consent for broad data sharing and secondary research use. Access to the data is restricted to registered and authorized researchers who have completed ethics training and agreed to the AoU data user code of conduct.

The current study is a secondary analysis of de-identified data obtained from the AoU Research Program. Analyses of de-identified data provided through approved data use agreements do not constitute human subjects research and therefore do not require additional IRB approval.

The AoU Research Program was not involved in the design and conduct of the study; collection, management, and analysis of the data; preparation, review, or approval of the manuscript; or the decision to submit the manuscript for publication. The other authors declare no competing interests.

Results

Participant characteristics

The AoU Research Program included 58,785 participants with available Fitbit activity data. Among them, 22,019 participants had both corresponding EHR data and at least 6 months of Fitbit monitoring (Supplementary Fig. 1). The cohort included both BYOD and WEAR participants (57% and 43%, respectively). The median participants age was 52 years ((IQR) 25th–75th percentile: 38– 63 years). The majority of participants were female (percentage (%): 71.4%), followed by males (28.2%), with 0.4% having unknown sex information. Race distribution was as follows: white (77.5%), black (6.3%), Asian (3.4%), other (7.3%), and unknown (5.5%). Hispanic or Latino participants comprised 7.8%, while 90.4% were non-Hispanic, and 1.9% had unknown ethnicity. The median Fitbit monitoring duration was 2.53 years (IQR: 1.42, 4.32). Participants had a median daily step count of 6166 steps (IQR: 3698, 8600). Participants had a median daily step count of 6166 steps (IQR: 484.3, 1050.2), with a median BMR of 1474.0 kcal calories (IQR: 1287.9, 1686.9). The median total daily calories burned was 2101.6 kcal (IQR: 1787.4, 2507.4). Median daily elevation gain was 33.9 meters (IQR: 4.3, 91.0). Daily median activity intensity metrics included sedentary time (738 min; IQR: 12.1–931.3), lightly active time (177.0 min; IQR: 118.3–227.5), moderately active time (9.6 min; IQR: 2.6–19.0), and vigorously active time (13.0 min; IQR: 6.0–24.7). The median average daily wear time, estimated from minute-level heart rate data, was 1335 min (IQR 1220–1397), indicating high device adherence with expected non-wear periods for charging or removal. Most days met established validity thresholds, with 99.2% classified as valid monitoring days. The median average daily active time was 199 min (IQR, 135–258). Participants contributed a median of 794 days (IQR 411–1574) of valid Fitbit records, consistent with a median monitoring duration of 2.53 years. (Table 1).

Table 1.

Baseline characteristics of study participants

Variable	Fitbit activity participants (N = 22,019)
Age	52.00 (IQR: 38.00, 63.00)
Sex
Female	15,724 (71.41%)
Male	6214 (28.22%)
Race
White	17,060 (77.48%)
Black	1382 (6.28%)
Asian	743 (3.37%)
Others	1615 (7.33%)
Unknown	1219 (5.54%)
Ethnicity
Hispanic or Latino	1716 (7.79%)
Non-Hispanic or Latino	19,896 (90.36%)
Unknown	407 (1.85%)
Fitbit monitoring length	2.53 (IQR: 1.42, 4.32) years
Daily steps	6166 (IQR: 3698, 8600)
Daily elevation	33.92 (IQR: 4.30, 91.00) meters
Activity intensity (minutes)
Lightly active	177.02 (IQR: 118.28, 227.45)
Moderately active	9.61 (IQR: 2.62, 18.97)
Vigorously active	12.96 (IQR: 6.03, 24.74)
Sedentary	737.98 (IQR: 12.05, 931.29)
Calories
Daily activity caloriesa	765.68 (IQR: 484.26, 1123.62)
Daily BMR caloriesb	1474.02 (IQR: 1287.86, 1686.94)
Daily calories burnedc	2101.64 (IQR: 1787.42, 2507.41)

Values are presented as median (IQR: 25th–75th percentile) for continuous variables, and count (percentage) for categorical variables.

^aThe number of calories burned for the day during periods the user was active above the sedentary level. This includes both activities burned calories and BMR.

^bTotal BMR calories burned for the day.

^cTotal calories burned for the day.

To evaluate potential multicollinearity among activity behaviors, we calculated pairwise Pearson correlation coefficients between all major Fitbit-derived metrics, including steps, activity calories, calories out, elevation gain, and minutes spent in different activity intensities and sedentary time. As shown in Supplementary Fig. 2, most activity variables exhibited weak-to-moderate correlations. Notably, the correlations between sedentary time and physical activity intensities were generally weak. For example, sedentary minutes showed only modest negative correlations with vigorous active minutes (r = −0.23) and negligible correlations with lightly active (r = −0.06) and moderately active minutes (r ≈ 0.00).

Associations between physical activity patterns and chronic disease incidence

After Bonferroni correction, 43 significant associations were identified between physical activity metrics and health outcomes. Daily step count was significantly associated with 27 diseases, including MDD and obstructive sleep apnea (OSA), while activity calories correlated with seven diseases, including sleep apnea and chronic fatigue syndrome. Morbid obesity and obesity were significantly associated with BMR-calories burned, with additional associations observed for total daily energy expenditure. Elevation gain was significantly linked to 11 diseases, including obesity and OSA. Lightly active minutes were significantly associated with torsion dystonia and inflammatory neuropathy, moderately active minutes with three diseases (including joint pain), and vigorously active minutes with 16 diseases (including type 2 diabetes and obesity) (Fig. 1).

Fig. 1. Associations between physical activity patterns and chronic disease incidence.

a Major depressive disorder, obstructive sleep apnea, etc., are associated with daily steps. b Type II diabetes, obesity, etc., are associated with vigorously active minutes. c Morbid obesity is associated with calories out. d Obesity, obstructive sleep apnea, etc., are associated with elevation. e No significant associations were observed between sedentary minutes and chronic disease phenotypes. f Torsion dystonia, etc., are associated with lightly active minutes. g Morbid obesity and obesity are associated with calories BMR. h Pain in joint, etc., are associated with moderately active minutes. The analyses are based on 22,019 participants.

Figure 2 illustrates the associations between physical activity metrics and chronic disease incidence. An increase in daily step count was significantly associated with a lower risk of type 2 diabetes with neurological manifestations (odds ratio (OR) = 0.25; 95% confidence interval (CI) = 0.14–0.44; p = 0.01), MDD (OR = 0.63; 0.54–0.73; p < 0.01), sleep apnea (OR = 0.49; 0.37–0.64; p < 0.01), chronic fatigue syndrome (OR = 0.46; 0.35–0.59; p < 0.01), chronic pain syndrome (OR = 0.20; 0.12–0.35; p < 0.01), and OSA (OR = 0.57; 0.47–0.69; p < 0.01). Neurological disorders (OR = 0.35; 0.25–0.51; p < 0.01) and shortness of breath (OR = 0.49; 0.37–0.64; p < 0.01) were inversely associated with step count. Higher step count was also significantly associated with a lower risk of tachycardia NOS (OR = 0.45; 0.33–0.60; p < 0.01) and acute renal failure (OR = 0.27; 0.18–0.42; p < 0.01).

Fig. 2. Heatmap of significant phenotypes across the physical activity patterns.

Heatmap showing significant associations between daily physical activity metrics and chronic disease phenotypes. Associations were estimated using a multivariable logistic regression model adjusted for age, sex, and baseline BMI. Statistical significance was determined using two-sided P values, with Bonferroni correction applied for multiple comparisons (α = 0.05). The analyses are based on 22,019 participants.

Higher BMR calories showed a significant positive association with the presence of obesity (OR = 1.62; 1.44–1.82; p < 0.01) and morbid obesity (OR = 2.22; 1.90–2.59; p < 0.01). Higher activity calories showed a negative association with sleep apnea (OR = 0.76; 0.68–0.85; p = 0.02), OSA (OR = 0.83; 0.77–0.90; p = 0.03), and chronic fatigue syndrome (OR = 0.70; 0.61–0.81; p < 0.01).

Greater daily elevation gain showed significant negative associations with multiple conditions. Higher elevation gain was linked to lower risk of abdominal pain (OR = 0.83; 0.77–0.90; p = 0.01), urinary tract infection (OR = 0.65; 0.55–0.75; p < 0.01), gastroesophageal reflux disease (OR = 0.81; 0.75–0.88; p < 0.01), and OSA (OR = 0.79; 0.72–0.87; p < 0.01). Elevation gain was also inversely associated with obesity risk (OR = 0.78; 0.71–0.86; p < 0.01). MDD was also significantly associated with higher elevation gain (OR = 0.80; 0.73–0.88; p = 0.02), as were sleep apnea (OR = 0.68; 0.59–0.79; p < 0.01). Other conditions significantly associated with higher elevation gain included other dyspnea (OR = 0.81; 0.74–0.88; p < 0.01) and hypovolemia (OR = 0.52; 0.40–0.66; p < 0.01).

Greater time spent in vigorously active minutes was significantly associated with a lower risk of OSA (OR = 0.79; 0.73–0.86; p < 0.01), obesity (OR = 0.79; 0.73–0.87; p < 0.01), and dyspnea (OR = 0.84; 0.77–0.90; p = 0.02). Moderately active minutes were significantly and positively associated with bursitis (OR = 1.31; 1.19–1.45; p < 0.01), joint pain (OR = 1.13; 1.08–1.19; p < 0.01), and osteoarthrosis (OR = 1.14; 1.08–1.20; p = 0.02). Lightly active minutes were inversely associated with torsion dystonia risk (OR = 0.53; 0.40–0.69; p = 0.02).

Cox model results for physical activity and chronic diseases

Cox proportional hazards regression models identified significant associations between physical activity metrics and chronic disease risk (Table 2 and Supplementary Figs. 3–10). A 2000-step increase in daily steps showed significant negative associations with obesity (hazard ratios (HR) = 0.85; 95% CI: 0.80–0.90), morbid obesity (HR = 0.79; 0.73–0.86), and type 2 diabetes (HR = 0.78; 0.72–0.84). Higher step counts were also showed significant negative associations with MDD (HR = 0.83; 0.77–0.90). The risk of OSA is lower rather than higher step count (HR = 0.84; 0.79–0.89). Other dyspnea (HR = 0.84; 0.80–0.89) and essential hypertension (HR = 0.88; 0.84–0.92) also showed lower HR associated with higher step counts.

Table 2.

Cox proportional hazards models for the association between Fitbit activity metric and chronic diseases

	Stepsa	Elevationb	Sedentaryc	Lightly activec	Moderately activec	Vigorously activec
Chronic disease	HR (95% CI)
Obesity	0.85 (0.80–0.90)	0.86 (0.78–0.95)	1.08 (1.06–1.10)	0.85 (0.79–0.91)	0.68 (0.44–1.05)	0.71 (0.50–1.02)
Morbid obesity	0.79 (0.73–0.86)	0.79 (0.67–0.92)	1.05 (1.03–1.08)	0.87 (0.79–0.95)	0.60 (0.32–1.12)	0.45 (0.24–0.83)
Type 2 diabetes	0.78 (0.72–0.84)	0.65 (0.53–0.80)	1.04 (1.01–1.07)	0.82 (0.74–0.91)	0.35 (0.18–0.69)	0.36 (0.20–0.64)
Major depressive disorder	0.83 (0.77–0.90)	0.82 (0.71–0.95)	1.03 (1.00–1.06)	0.84 (0.77–0.92)	0.72 (0.40–1.29)	0.57 (0.34–0.94)
Obstructive sleep apnea	0.84 (0.79–0.89)	0.83 (0.75–0.92)	1.03 (1.01–1.06)	0.87 (0.81–0.94)	0.31 (0.18–0.52)	0.47 (0.32–0.70)
Other dyspnea	0.84 (0.80–0.89)	0.87 (0.79–0.95)	1.04 (1.02–1.06)	0.91 (0.85–0.97)	0.51 (0.33–0.78)	0.32 (0.22–0.47)
Essential hypertension	0.88 (0.84–0.92)	0.92 (0.86–0.99)	1.05 (1.04–1.07)	0.88 (0.83–0.93)	0.67 (0.46–0.96)	0.67 (0.51–0.88)
Chronic pain syndrome	0.93 (0.80–1.09)	0.89 (0.68–1.16)	1.05 (1.00–1.12)	0.78 (0.64–0.95)	0.20 (0.04–1.02)	0.19 (0.04–0.77)

Bold indicate statistically significant associations.

^a2000 steps.

^b100 m.

^cHours.

Each 100-meter increase in elevation gain was associated with a lower risk of obesity (HR = 0.86; 0.78–0.95), morbid obesity (HR = 0.79; 0.67–0.92), type 2 diabetes (HR = 0.65; 0.53–0.80), and MDD (HR = 0.82; 0.71–0.95). OSA (HR = 0.83; 0.75–0.92), other dyspnea (HR = 0.87; 0.79–0.95), and essential hypertension (HR = 0.92; 0.86–0.99) were also inversely associated with elevation gain.

Greater sedentary time was positively associated with multiple conditions. Each additional hour of sedentary time was associated with a higher hazard of obesity (HR = 1.08; 1.06–1.10), morbid obesity (HR = 1.05; 1.03–1.08), type 2 diabetes (HR = 1.04; 1.01–1.07), and OSA (HR = 1.03; 1.01–1.06). Other dyspnea (HR = 1.04; 1.02–1.06) and essential hypertension (HR = 1.05; 1.04–1.07) also showed positive associations with excessive sedentary time.

Greater time spent in light activity was associated with a lower risk of several conditions. Each additional hour of light activity showed significant negative associations with obesity (HR = 0.85; 0.79–0.91), morbid obesity (HR = 0.87; 0.79–0.95), type 2 diabetes (HR = 0.82; 0.74–0.91), MDD (HR = 0.84; 0.77–0.92), OSA (HR = 0.87; 0.81–0.94), other dyspnea (HR = 0.91; 0.85–0.97), essential hypertension (HR = 0.88; 0.83–0.93), and chronic pain syndrome (HR = 0.78; 0.64–0.95). Each additional hour of vigorously active time was associated with a significantly lower risk of morbid obesity (HR = 0.45; 0.24–0.83), type 2 diabetes (HR = 0.34; 0.20–0.64), OSA (HR = 0.47; 0.32–0.70), other dyspnea (HR = 0.32; 0.22–0.47), essential hypertension (HR = 0.67; 0.51–0.88), and MDD (HR = 0.19; 0.44–0.77).

Sensitivity analysis using body mass index (BMI) for obesity classification

To validate our findings, we performed a sensitivity analysis using BMI-defined obesity as an alternative outcome measure. Logistic regression models assessing associations between physical activity metrics and BMI-defined obesity confirmed the primary findings. Higher daily step counts were associated with a significantly lower risk of obesity (OR = 0.79; 95% CI: 0.71–0.88; p < 0.01), as were higher elevation gain (OR = 0.77; 0.69–0.86; p < 0.01) and time spent in vigorously active intensity (OR = 0.80; 0.72–0.88; p < 0.01). Conversely, BMR calories were positively associated with obesity (OR = 2.17; 1.82–2.60; p < 0.01), suggesting a potential compensatory metabolic effect. Cox proportional hazards models further corroborated these findings, demonstrating consistent associations between physical activity metrics and the risk of developing BMI-defined obesity over time (Supplementary Fig. 11). Higher daily step counts, higher elevation gain, and greater time spent in vigorously active intensity were significantly associated with a lower hazard of obesity onset. These results underscore the robustness of our primary analysis and emphasize the protective role of higher physical activity against obesity, regardless of diagnostic criteria.

Sensitivity analysis based on two-year outcome exclusion

To further evaluate the potential impact of reverse causation, we performed a sensitivity analysis excluding participants who developed target conditions within the first 2 years (730 days) of their Fitbit monitoring period. While this approach reduced the number of included events, many key associations remained consistent. For example, each 2000-step increment in daily steps continued to show significant inverse associations with several chronic conditions, including obesity (HR = 0.83, 95% CI: 0.78–0.89), morbid obesity (HR = 0.76, 0.69–0.84), type 2 diabetes (HR = 0.79, 0.72–0.87), MDD (HR = 0.83, 0.76–0.91), and OSA (HR = 0.78, 0.73–0.84). Similarly, vigorous physical activity remained strongly associated with lower risks of type 2 diabetes (HR = 0.41, 0.22–0.79), morbid obesity (HR = 0.35, 0.16–0.77), and OSA (HR = 0.48, 0.31–0.76). Compared to the primary analysis, which identified 43 significant associations after Bonferroni correction, the 2-year exclusion reduced this number to 25. Nonetheless, the consistency of effect sizes for key outcomes suggests that reverse causation is unlikely to fully explain the observed associations. (Supplementary Fig. 12 and Supplementary Table 1).

Sensitivity analysis with lifestyle covariates

To assess the potential influence of lifestyle-related confounding, we repeated all logistic and Cox regression models with additional adjustment for two binary covariates derived from the AoU Lifestyle survey: current smoking status and alcohol use. The inclusion of these lifestyle covariates produced results that were highly consistent with the primary analyses. Across logistic regression models, 42 significant associations were identified between physical activity metrics and chronic disease outcomes, and all associations retained the same directionality and similar effect magnitudes. Likewise, Cox proportional hazards models showed negligible differences in HR and significance patterns, further supporting the robustness of the findings (Supplementary Fig. 13 and Supplementary Table 2). This sensitivity analysis aligns with a simplified DAG framework, representing a minimal sufficient adjustment set of demographic and lifestyle factors that plausibly confound the relationship between physical activity and chronic disease.

Discussion

This study examined the associations between Fitbit-derived physical activity metrics and chronic disease incidence using data from the AoU Research Program. We utilized the latest AoU curated data repository (CDR) version 8, released in February 2025, incorporating participant data up to October 1, 2023. With over 59,000 participants contributing Fitbit data, this dataset uniquely enables the exploration of long-term associations between objectively measured physical activity and chronic disease risk in a diverse cohort¹⁵. Utilizing this comprehensive dataset enhances the robustness and generalizability of our findings, ensuring their relevance to public health efforts and evidence-based physical activity interventions for disease prevention. Our findings suggest that higher physical activity levels—measured by daily step count, elevation gain, and activity intensity—are statistically associated with chronic diseases, including obesity, type 2 diabetes, MDD, and OSA. Sensitivity analyses additionally adjusting for smoking and alcohol use produced results that were consistent with the main models, indicating that lifestyle factors had a limited influence on the observed associations. Conversely, excessive sedentary time was positively associated with these conditions. Compared with the work by Master et al.³⁸, which linked long-term, time-varying step counts from consumer wearables with incident chronic disease risk, our study highlighted additional values of other Fitbit-derived activity dimensions, including caloric expenditure, elevation gain, sedentary time, and fine-grained activity-intensity measures. These findings align with growing evidence emphasizing the role of regular physical activity in improving health and mitigating disease risk.

The inverse association between daily step count and the risk of obesity, type 2 diabetes, and major depressive disorder is consistent with Master et al.'s AoU analysis and other prior research demonstrating protective associations between higher step counts and cardiometabolic and mental health outcomes^5,39–41. Our results indicate that higher step counts are significantly associated with a lower risk of these conditions. This aligns with the concept that higher physical activity enhances energy expenditure, improves metabolic function, and alleviates depressive symptoms^4,42. These findings reinforce public health recommendations advocating higher daily step counts as a practical strategy for chronic disease prevention.

Our study identified elevation gain, a less frequently examined metric, as significantly associated with a lower risk of several chronic diseases, including type 2 diabetes, morbid obesity, and OSA. Prior research suggests that elevation-related activities, such as stair climbing, may confer health benefits, though further evidence is needed to confirm their effects on cardiovascular and metabolic health^41,43. These findings imply that integrating elevation-related activities into daily routines may offer additional health benefits. Further studies are warranted to elucidate the role of elevation-related activities in chronic disease prevention.

This study further highlighted the association between activity intensity and health outcomes. Greater time in vigorously active intensity was significantly associated with a lower risk of obesity, type 2 diabetes, and OSA, while higher moderately active and lightly active time correlated with reduced risks of several chronic conditions, including MDD. These findings align with prior research demonstrating the benefits of varying physical activity intensities on cardiometabolic and mental health outcomes^44,45. However, moderately active intensity was also positively associated with bursitis, joint pain, and osteoarthrosis, suggesting that while high-intensity activity is beneficial for chronic disease prevention, it may increase musculoskeletal strain. These findings underscore the need for personalized physical activity recommendations, particularly for individuals at risk of musculoskeletal injuries⁴⁶.

Conversely, excessive sedentary time was consistently associated with a higher risk of multiple chronic diseases, including obesity, type 2 diabetes, and essential hypertension. These findings align with previous studies highlighting the adverse associations between excessive sedentary behavior and metabolic health^47–49. The strong association between sedentary behavior and chronic disease risk underscores the importance of reducing sedentary time as a complementary strategy to increasing physical activity for improving health outcomes.

To evaluate the robustness of our findings, we performed a sensitivity analysis using BMI-defined obesity as an alternative outcome measure. Logistic regression analyses corroborated our primary findings, showing that higher daily step counts, higher elevation gain, and greater time spent in vigorously active intensity were significantly associated with a lower risk of obesity. Cox proportional hazards models further reinforced these associations, demonstrating that physical activity metrics were consistently linked to a lower hazard of obesity development over time. These findings enhance the validity of our primary results, indicating that the observed associations persist regardless of whether obesity is defined by BMI criteria or clinical diagnosis. The sensitivity analysis underscores the robustness of the association between physical activity and obesity risk, suggesting that increasing daily movement and engaging in higher-intensity activities may confer benefits across varying obesity definitions. While we mitigated reverse causality in the primary analysis by excluding events in the initial 180 days, we further examined this issue through a sensitivity analysis that removed all events occurring within the first 2 years of monitoring. Despite a reduction in statistical power, the persistence of major findings (e.g., inverse associations between step counts and obesity/type 2 diabetes) suggests that our results are not solely driven by reverse causation. These findings bolster confidence in the temporal directionality of the observed associations.

A key strength of this study is the use of real-world wearable data, offering objective and granular insights into long-term physical activity patterns. Unlike traditional epidemiological studies relying on self-reported physical activity, wearable-derived metrics provide greater accuracy and allow for detailed assessments of activity intensity, duration, and elevation. However, several limitations should be considered. First, important confounders such as diet and socioeconomic status were not included due to data limitations, which may impact the observed associations. Although we mitigated reverse causation by excluding participants diagnosed within the initial 180 days, residual confounding remains possible. Our reliance on Fitbit users may introduce selection bias, as they are likely more health-conscious than the general population⁵⁰. The AoU Fitbit data lack minute-level wear-time flags, so adherence was approximated using minute-level heart rate availability rather than directly observed wear indicators, which may lead to some misclassification of partial-wear days. Additionally, consumer-grade devices like Fitbit have documented limitations in accurately measuring activity intensity, heart rate, and energy expenditure, particularly during high-intensity movement^51,52. Race and ethnicity were not included as covariates due to their highly imbalanced distributions in our sample, which caused model instability and convergence issues. Finally, when examining each activity intensity separately, we did not adjust for sedentary time in the same models, which limits our ability to assess their joint effects. Future work with richer covariate data, more granular device wear-time tracking, and demographically balanced samples will help address these limitations and yield more robust insights.

Conclusions

In conclusion, our findings highlight the importance of increasing daily steps, incorporating elevation-related activities, and engaging in diverse physical activity intensities to lower chronic disease risk. Equally important is reducing sedentary behavior to mitigate chronic disease risk. These findings offer valuable insights for designing targeted public health interventions that promote physical activity and reduce sedentary behavior, ultimately enhancing population health outcomes. Future research should investigate the underlying mechanisms of these associations and develop personalized activity recommendations to optimize health benefits while minimizing risks. With the growing application of machine learning in precision medicine^53–58, integrating these approaches into future research could enhance the development of tailored physical activity interventions, enabling more precise health outcome predictions and personalized recommendations.

Author contributions

Y.H. and R.Z. conceived the study. Y.H. served as the principal author, extracting data, designing and conducting the entire experiment and writing the manuscript. E.C. assisted in conceiving the study design and reviewed the manuscript. S.I., K.L., L.S.C., and M.H. provided clinical expertise and reviewed the manuscript. All authors contributed to the production of the final manuscript.

Peer review

Peer review information

Communications Medicine thanks Aiden Doherty and the other anonymous reviewer(s) for their contribution to the peer review of this work.

Data availability

This study used data from the All of Us Research Program’s Control Tier Dataset v8, available to authorized users on the Researcher Workbench, which can be accessed via https://workbench.researchallofus.org/. The source data for the main figures are provided as Supplementary Data files. Specifically, the source data for Figs. 1, 2 are provided in Supplementary Data 1. Source data for Supplementary Figs. 3–10 are identical to the Cox proportional hazards model results reported in Table 2 and are therefore not provided separately.

Code availability

Code used for this study is available to approved researchers upon request. Researchers can access the code on the All of Us Research Workbench platform by contacting our study team.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at https://doi.org/10.1038/s43856-025-01372-x.