Peaks, sources, and immediate health impacts of PM_2.5 and PM₁ exposure in Indonesia and Taiwan with microsensors

Low-cost microsensors, when appropriately calibrated against research-grade instruments, provide high spatiotemporal resolution measurements that complement regulatory monitoring and enable personal exposure assessment. Ambient PM2.5 is a leading environmental health risk, with levels in many low- and middle-income countries far exceeding WHO guidelines and linked to respiratory, cardiovascular, and all-cause mortality. Accurate exposure assessment with fine spatial resolution improves health effect estimation. Smaller particles such as PM1 may penetrate deeper into the respiratory tract and pose greater toxicity; recent meta-analyses support associations of PM1 with cardiorespiratory morbidity and mortality, but more regional research is needed. High-temporal-resolution sensors can capture short-term peak exposures that may drive acute events such as asthma exacerbations and stroke—exposures often missed by hourly ambient data. Heart rate variability (HRV) reflects autonomic nervous system balance and has been linked to PM exposure; medically certified wearable biosensors now allow less burdensome HRV measurement than traditional Holter monitors. This study aims to assess peak exposure levels, identify sources, and quantify immediate health impacts of PM2.5 and PM1 using integrated environmental and health microsensors in two Asian cities (Bandung, Indonesia; Kaohsiung, Taiwan), where high PM levels, proximity to sources, and dense populations underscore the public health importance.

The paper reviews evidence that PM2.5 is a carcinogen and a major global risk factor, with substantial mortality attributable to exposure and higher-than-guideline levels prevalent in many Asian cities. Improved spatial resolution of exposure enhances estimation of PM–health relationships. Short-term PM2.5 exposure is associated with exacerbations of respiratory disease and HRV changes; peaks may contribute to stroke risk. PM1 may have greater toxicity due to deeper penetration; recent systematic reviews and meta-analyses report associations between PM1 exposure and cardiorespiratory outcomes across age groups, while calling for more geographically diverse studies. HRV serves as a marker of autonomic balance, with reduced HRV linked to myocardial infarction risk. Prior epidemiology has demonstrated PM2.5–HRV associations and outlined mechanistic pathways (inflammation/vasoactive mediators, autonomic perturbation, particle translocation). Advances in wearable, medically certified HRV sensors facilitate assessment of hyperacute physiologic responses to PM exposure. Prior sensor-based studies have documented elevated exposures near community factories, during commuting, and from indoor sources (incense, ETS), but few have quantified immediate HRV impacts at minute-level resolution or contrasted PM2.5 versus PM1 effects.

Two panel studies were conducted: Bandung, Indonesia (2018–2019) and Kaohsiung, Taiwan (2019–2020). Inclusion criteria: Indonesia—age 18–65, >1 h daily commute, non-smoker/non-drinker, no cardiopulmonary disease/medication; Taiwan—age 40–75 living in a randomly selected community, non-smoker or quit >1 year, no history of hypertension or heart-related diseases. Final datasets included 49 Indonesian and 51 Taiwanese subjects after cleaning. Monitoring seasons: Indonesia wet (Nov 2018–Jan 2019) and dry (Jul–Sep 2019); Taiwan summer (Jul–Aug 2019) and winter (Feb–Mar 2020). Personal PM2.5 and PM1 were measured with calibrated AS-LUNG-P devices (PMS3003; 15-s sampling, with temperature/humidity, GPS, accelerometer). Each device was calibrated against a GRIMM EDM-180 (FEM) in a custom chamber; correction equations (R2 ~0.97–0.99) converted raw readings to research-grade values. Ambient PM was monitored with AS-LUNG-O (1-min) installed in each city (calibrated similarly); ambient data in Indonesia from the outdoor unit were not retained for analysis in the presented results, whereas a community-site outdoor unit was operated in Taiwan. HRV was measured using the medically certified RootiRx single-lead ECG patch (500 Hz), previously validated against 12-lead Holter. Derived indices included time domain (SDNN, RMSSD), frequency domain (HF, LF, VLF, TP, LF/HF), and heart rate (HR). A noise filter excluded 5-min segments with unrealistic SDNN or HR. Subjects carried AS-LUNG-P continuously (except shower/sleep) and wore Rooti continuously; Indonesia: AS-LUNG-P 1–6 days, Rooti 1–3 days (working days). Taiwan: AS-LUNG-P seven 24-h days; Rooti one 48-h (some two) periods, including non-working days. Time-activity diaries (TADs) recorded, every 30 min, activities, locations, indoor ventilation, and up to two visible nearby sources (e.g., cooking, ETS, community factories, mosquito coil, traffic by mode). Data processing and cleaning: PM <1 µg/m3 and ghost peaks (15-s spikes >10× neighborhood) removed; rainy hours excluded. PM 15-s data averaged to 5-min to align with HRV; 5-min intervals with >50% missing were removed. Missing rates: 4.4% (Indonesia) and 1.9% (Taiwan). TADs were reprocessed to 5-min. Statistical analyses: - Source contributions: Generalized additive mixed models (GAMM) modeled PMpersonal − PMambient as outcome with source indicators (primary source per 30-min window), adjusting for season and relative humidity via thin-plate splines, subject random effects, and AR(1) correlation. Sources with <10 occurrences and periods outside the studied city were excluded. Traffic was disaggregated by mode (walking, biking, scooter, car, bus/minibus). - PM–health: HRV indices and HR were log10-transformed. GAMM regressed log(HRV/HR) on PMpersonal (5-min), adjusting for age category, gender, BMI category (country-appropriate cutoffs), outdoor status, season, activity intensity (accelerometer-derived), humidity (spline), time of day (penalized cubic spline), subject random effects, and AR(1) correlation. Analyses excluded sleep periods. Lag effects up to 6 h (non-sleep) were explored. Software: R 4.0.2 / RStudio 1.1.456. Ethical approvals were obtained, and informed consent was secured.

• Exposure levels: Mean 5-min personal PM2.5/PM1 in Indonesia: 30.4 ± 20.0 / 27.0 ± 15.7 µg/m3; in Taiwan: 14.9 ± 11.2 / 13.9 ± 9.8 µg/m3. Peak 5-min PM2.5/PM1: Indonesia 473.6 / 154.0 µg/m3; Taiwan 467.4 / 217.7 µg/m3. PM1/PM2.5 ratios were ~0.9 in both countries. Personal peak exposures far exceeded concurrent ambient maxima (e.g., Taiwan ambient PM2.5 max 59.7 µg/m3 vs personal peak 467.4 µg/m3), highlighting the importance of personal monitoring for peak capture. - Source exposure patterns: Traffic was the most frequently encountered category. Highest PM2.5 distributions by source: Indonesia—mosquito coil burning and community factory emissions; Taiwan—community factory emissions. - Source contributions (GAMM, PMpersonal − PMambient): Indonesia—significant PM2.5 increments for mosquito coil burning (5.82 µg/m3), community factories (4.73), cooking (2.73), and traffic by mode (e.g., biking 2.38, bus/minibus 2.21, scootering 1.68, walking 1.54). Taiwan—community factories (PM2.5 10.1; PM1 6.66 µg/m3), mosquito coil burning (PM2.5 9.82; PM1 7.23), agricultural waste burning (PM2.5 7.17; PM1 6.75; winter stronger), and incense burning (PM2.5 2.59; PM1 1.84) were significant; car travel showed lower exposure than non-commuting. Seasonal differences were evident (e.g., higher summer factory contributions; higher winter agricultural burning). - Peak-related metrics: Maximum PM2.5 exposures by source frequently involved cooking and mosquito coils (Indonesia) and community factories and mosquito coils (Taiwan). Some sources accounted for large fractions of 24-h exposure (e.g., Indonesia aromatic products 42%, mosquito coil 18%; Taiwan ETS 39.8%, factory emissions 37.1%). - Immediate health effects: 5-min PM exposure was associated with immediate changes in HRV indices and HR. Taiwan (all subjects): significant changes in multiple HRV indices and HR with PM2.5 and PM1; effects generally larger in summer; HR increased with PM. Indonesia (scooter subgroup, n=13): significant decreases in HRV indices with PM, especially in the dry season; changes for a 10 µg/m3 increase typically ranged from about -3% to -6% across SDNN, RMSSD (PM1), LF/HF, LF, VLF, TP in dry season analyses. Across countries, estimated percent changes per 10 µg/m3 generally fell within or slightly above ranges reported in prior meta-analyses for short-term PM–HRV associations. No significant lag effects up to 6 h were identified. - Comparative insights: PM1 tended to show greater immediate HRV impacts than PM2.5. Outdoor status was associated with larger HRV impacts. Despite frequent traffic exposures, community factories and mosquito coil burning contributed the largest incremental exposures and highest peak metrics in both countries.

The study demonstrates that calibrated low-cost environmental microsensors coupled with a medically certified HRV wearable can quantify minute-level peak PM2.5 and PM1 exposures, identify proximate sources, and capture immediate cardiovascular autonomic responses. The findings directly address the objectives by (1) documenting large discrepancies between personal peak and ambient levels, underscoring the inadequacy of ambient hourly surrogates for peak-related health assessments; (2) quantifying source-specific contributions after controlling for concurrent ambient levels, revealing community factories and mosquito coil burning as dominant contributors in both countries, with additional roles for agricultural waste and incense burning in Taiwan; and (3) demonstrating statistically significant, immediate HRV and HR responses to 5-min PM exposures in Taiwan and among Indonesian scooter riders, with stronger effects for PM1 and seasonal modulation. These results are significant for exposure science and environmental epidemiology in high-PM regions, showing that near-source exposures drive peaks and immediate physiologic perturbations consistent with autonomic pathway mechanisms. The work supports targeted community-level interventions (e.g., regulating small/mid-scale factories, reducing coil/incense/agricultural waste burning) and behavior change strategies to reduce peak exposures and acute risk. The approach is scalable and particularly relevant for resource-limited settings where traditional monitoring is sparse. Differences across seasons, countries, and microenvironments (indoor/outdoor) point to context-specific mitigation opportunities. Overall, integrating personal sensing with activity/source logging and appropriate mixed modeling advances exposure-health assessment beyond conventional station-based designs.

This study shows that high-resolution microsensors can capture peak PM2.5 and PM1 exposures, pinpoint dominant proximate sources, and detect immediate HRV and HR responses at minute-level resolution in Indonesia and Taiwan. Community factories and mosquito coil burning emerged as major contributors to peak and incremental exposures in both settings, exceeding commonly emphasized sources such as traffic and ETS in impact metrics. Immediate autonomic effects were observed, with stronger associations for PM1 and during drier/warmer seasons. The methodology provides a practical template for exposure assessment, source attribution, and acute health effect evaluation in resource-limited regions. Future work should: expand panels and settings to enhance generalizability; refine time-activity/source logging for finer temporal alignment; incorporate multi-pollutant sensing to assess confounding and composition; evaluate intervention impacts (e.g., zoning enforcement, emission controls on community factories, alternatives to mosquito coils/incense); and continue validating medically certified wearables for cardiorespiratory endpoints. These actions can inform targeted policies to reduce peak exposures and protect cardiovascular health.

• Temporal aggregation and alignment: PM peaks (15 s) and TADs (30 min) were aggregated to 5-min segments to match HRV resolution, likely underestimating true peak magnitudes and diluting source attribution. The assumption that all 5-min segments within a 30-min TAD share the same primary source may bias source contributions downward; only primary sources were modeled, undercounting multi-source periods. - Potential confounding: Unmeasured co-pollutants and environmental factors may confound PM–HRV associations; multi-pollutant measurements were not analyzed. - Sample size and representativeness: Moderate panel sizes and specific inclusion criteria limit generalizability; occupational distributions may not match the broader population. - Ambient data constraints: Ambient corrections relied on outdoor sensor data available within the study communities (with limitations noted for Indonesia), which may not capture broader background variability. - Conservative bias: Many limitations (temporal aggregation, primary-source only) likely bias effects and source contributions toward underestimation rather than overestimation.