Seven participants were recruited from Queens College, City University of New York. Participants were asked to wear a Fitbit Charge 6 (Google) continuously for 4 weeks from July 2025 to August 2025, except while charging or bathing, and were provided with written setup instructions and technical support during enrollment [20]. Data were collected via the ExpiWell (Expimetrics Inc) platform [21], a mobile app compatible with iOS and Android devices that collects health metrics from Fitbit, captures smartphone GPS data, and allows the administration of EMA surveys.

This protocol was approved by the institutional review board at Queens College, City University of New York (2025-0132-QC). All participants provided written informed consent. Assent and parental consent were obtained for participants aged <18 years. The privacy and confidentiality of research participants’ data were maintained. Participants did not receive compensation for participating in the study.

GPS tracking was performed through participants’ mobile devices at 5- to 15-minute intervals. Hourly meteorological and air pollution data for each GPS time stamp were retrieved from the OpenWeatherMap (OpenWeather Ltd) application programming interface (API), which integrates data from weather stations, satellites, and global climate models [22]. Measures included <2.5 μm and <10 μm particulate matter, ozone, nitrogen dioxide, sulfur dioxide, carbon monoxide, temperature, humidity, and heat index calculated via Rothfusz regression [23]. Environmental exposures were estimated for all GPS time points collected, regardless of participants’ reported indoor and outdoor status on EMA surveys. To preserve spatial resolution while reducing computational load, GPS coordinates were spatially binned to a 0.005° latitude or longitude grid (approximately 556 m; Figure S1 in Multimedia Appendix 2). Daily time-weighted average exposures were calculated for each participant (Multimedia Appendix 2). For a full list of available API options, see Table 1.

EMA surveys were administered 3 times daily (9 AM, 3 PM, and 8 PM) with 30-minute completion windows. Each survey assessed 7 emotional states (anxious, worried, nervous, sad, tired, hopeless, and happy) on a 7-point Likert scale (1=not at all to 7=extremely), as well as the participant’s location (indoors or outdoors).

Fitbit metrics included daily sleep duration and stages (light, deep, and rapid eye movement); resting heart rate; heart rate variability (HRV); breathing rate; and minutes spent in active heart rate zones (fat burn, cardio, and peak). Breathing rate and HRV were available when participants wore the device during ≥3 hours of continuous sleep. Sleep quality was summarized using a composite sleep score based on duration and depth (Multimedia Appendix 2).

Feasibility was evaluated by three predefined criteria: (1) at least 3 GPS measurements per participant per day with corresponding heat and pollution data, (2) at least 1 completed EMA survey per day per participant, and (3) at least 50% of days with complete Fitbit sleep and HRV data.

After the conclusion of the 4-week pilot study, participants were invited to complete feasibility and acceptability surveys to assess their subjective experiences during the study. Participants completed Qualtrics (Qualtrics Inc) surveys evaluating the ease of using the Fitbit and ExpiWell systems (5-point Likert scales) and their comfort, enjoyment, and willingness to continue use (binary responses) [29]. Participants also provided feedback on barriers such as forgetting to wear the device or perceived inconvenience.

To assess agreement between exposure estimates from original and binned GPS coordinates, linear mixed-effects models were used to estimate the pairwise difference between methods with random intercepts. EMA responses were matched to daily time-weighted environmental exposures and same-day Fitbit data. Bivariate Pearson correlations (r) were calculated with false discovery rate correction for multiple testing (Benjamini-Hochberg method).

Due to the small sample size, observed correlations may have been driven disproportionately by a single participant. To account for this, the influence of individual participants on the results was assessed via a sensitivity score calculated as the mean absolute difference in correlation coefficients after sequentially excluding each participant (mean |Δr|), with higher values indicating greater influence of single participants on the overall correlation. Correlations with mean |Δr| >0.1 were considered unstable. Given the small sample, these analyses were primarily exploratory and hypothesis generating. Analyses were conducted in R (version 4.4.1; R Foundation for Statistical Computing).

Seven participants (aged approximately 16 to 24 years; 5 female participants) were followed for 4 weeks (Table 2). One participant was excluded from Fitbit analyses due to a technical issue with data transfer but was retained for GPS analyses and EMA.

Participants recorded a mean of 19.7 (SD 25.8) GPS measurements per day, with 6 participants meeting the feasibility goal of ≥3 daily measurements (Table 2; Figure S2 in Multimedia Appendix 2). Participants completed a mean of 1.4 (SD 0.2) EMA surveys per day. Of the 7 participants, 5 (71%) met the compliance goal of ≥1 survey per day (Table 2). EMA responses are shown in Table 3. Moreover, 5 (71%) of 7 participants met the ≥50% completeness goal for sleep and sleep-associated Fitbit metrics (Table 4).

Compliance data identified several feasibility challenges. One participant experienced a technical issue with Fitbit data transfer, and another showed moderate to high missingness in sleep and sleep-related metrics. Among the 4 participants who completed feasibility questionnaires (Table S3 in Multimedia Appendix 2), 3 reported forgetting to put the Fitbit device back on after removing it and 2 cited perceived annoyance and reluctance to wear the Fitbit for longer periods, consistent with observed gaps in sleep data.

Compliance was further limited by a low EMA response rate in 2 participants and greater missingness on survey items assessing negative emotions, which could introduce bias. One participant with a low EMA response rate also demonstrated poor GPS data compliance. On feasibility surveys, 1 (25%) of 4 participants perceived the system as overly complex and difficult to use without technical support, and 2 participants reported that there was a lot to learn before they could begin using it. These responses may reflect difficulty navigating the survey interface and remembering to keep the ExpiWell app open to allow for continuous GPS tracking.

Across all pollutants and meteorological variables examined, environmental exposures extracted with and without coordinate binning were highly correlated (Pearson r >0.99), and the estimated mean differences between exposures were not statistically significant (Table S1 in Multimedia Appendix 2). Daily time-weighted average exposures for each participant are shown in Figure 1 [30-33] and Table S2 in Multimedia Appendix 2. Mean environmental exposure estimates varied by indoor and outdoor status throughout the study period (Figure S3 in Multimedia Appendix 2).

Bivariate correlations indicated potential relationships between environmental exposures and physiologic or emotional measures; however, directionality and causation cannot be inferred. Most correlations were stable to the exclusion of single participants (sensitivity index <0.1; Figure 2). Unstable correlations were excluded from interpretation due to the potential for single participants to bias the results in a small sample. Higher heat index (r=−0.22; P=.02) and nitrogen dioxide (r=−0.24; P=.01) levels were associated with lower HRV, suggesting reduced parasympathetic tone. Sulfur dioxide exposure correlated with higher “nervous” (r=0.19; P=.04) and “hopeless” (r=0.21; P=.04) scores. Conversely, higher heat index exposure was associated with lower “sad” (r=−0.21; P=.03) scores. See correlation coefficients, P values, and sensitivity indices in Tables S4 to S6 in Multimedia Appendix 2.