Determination of Personalized Asthma Triggers from Evidence based on Multimodal Sensing and Mobile Application

Objective: Asthma is a chronic pulmonary disease with multiple triggers manifesting as symptoms with various intensities. This paper evaluates the suitability of long-term monitoring of pediatric asthma using diverse data to qualify and quantify triggers that contribute to the asthma symptoms and control to enable a personalized management plan. Materials and Methods: Asthma condition, environment, and adherence to the prescribed care plan were continuously tracked for 97 pediatric patients using kHealth-Asthma technology for one or three months. Result: At the cohort level, among 21% of the patients deployed in spring, 63% and 19% indicated pollen and Particulate Matter (PM2.5), respectively, as the major asthma contributors. Of the 18% of the patients deployed in fall, 29% and 21% found pollen and PM2.5 respectively, to be the contributors. For the 28% of the patients deployed in winter, PM2.5 was identified as the major contributor for 80% of them. One patient across each season has been chosen to explain the determination of personalized triggers by observing correlations between triggers and asthma symptoms gathered from anecdotal evidence. Discussion and Conclusion: Both public and personal health signals including compliance to prescribed care plan have been captured through continuous monitoring using the kHealth-Asthma technology which generated insights on causes of asthma symptoms across different seasons. Collectively, they can form the underlying basis for personalized management plan and intervention. KEYWORDS: Personalized Digital Health, Medical Internet of Things, Pediatric Asthma Management, Patient Generated Health Data, Personalized Triggers, Telehealth,

💡 Research Summary

This paper presents a comprehensive evaluation of a multimodal digital health platform, kHealth‑Asthma, for personalized monitoring of pediatric asthma. The authors recruited 97 children aged 5–17 receiving care at Dayton Children’s Hospital and equipped them with a “kHealth kit” consisting of a mobile questionnaire, a Fitbit activity/sleep tracker, a Microlife peak‑flow meter, and a Foobot indoor‑air sensor. In addition, outdoor environmental data—pollen (12‑hour updates), PM2.5, ozone, temperature, and humidity (hourly updates)—were automatically fetched based on each participant’s ZIP code from public APIs (pollen.com, EPA AIRNow, Weather Underground). All data streams were synchronized in real time to a secure Firebase‑backed cloud, anonymized, and made available through a clinician‑oriented dashboard for visualization and analysis.

The study excluded 21 participants due to insufficient active sensing (<20% questionnaire completion), leaving 76 subjects for analysis. Participants were monitored for either one month (80 subjects) or three months (17 subjects), providing a total of over 1.2 million data points. The authors defined a “asthma episode” as any day with reported symptoms (cough, wheeze, chest tightness, rapid breathing, speech limitation, nasal flaring), night‑time awakenings, activity limitation, rescue inhaler use, or a drop in PEF/FEV1 beyond one standard deviation of the individual’s mean.

Seasonal cohort analysis revealed distinct trigger patterns: in the spring cohort (21% of participants), 63% identified pollen and 19% identified PM2.5 as major contributors; in the fall cohort (18%), 29% cited pollen and 21% cited PM2.5; in the winter cohort (28%), a striking 80% pointed to PM2.5 as the primary trigger. To illustrate personalized trigger detection, the authors selected one poorly controlled patient from each season and divided each patient’s monitoring period into a learning phase and a prediction phase. Correlations were assessed by comparing the maximum outdoor pollutant values on the day of, or the day before, an asthma episode with the healthy thresholds (pollen 0–2.4, ozone/PM2.5 0–50).

Key case findings include: Patient‑A (winter‑to‑spring) showed PM2.5 as the dominant trigger during pollen‑free weeks, while pollen became the primary driver once the season changed; combined high pollen and PM2.5 levels were associated with more severe symptoms (chest tightness, night awakenings). Patient‑B (winter) experienced frequent episodes despite relatively stable outdoor conditions, suggesting that even modest PM2.5 elevations can provoke symptoms when controller medication adherence was only 50%. The authors noted that indoor Foobot data and Fitbit activity/sleep metrics did not provide reliable signals, likely due to placement issues, power interruptions, and confounding factors, reinforcing the importance of high‑quality outdoor environmental monitoring.

The discussion emphasizes that continuous, multimodal data capture enables clinicians to identify patient‑specific triggers, assess medication adherence, and tailor interventions such as real‑time alerts for high pollen or PM2.5 forecasts and reminders to improve controller medication compliance. Limitations include data loss from poor questionnaire compliance, sensor deployment challenges, and the need for larger, longer‑term cohorts to validate predictive models.

In conclusion, the kHealth‑Asthma framework successfully integrates patient‑generated health data with environmental sensing to uncover personalized asthma triggers across seasons. This approach offers a scalable pathway toward remote, data‑driven asthma management, supporting proactive, individualized care plans and informing future telehealth interventions.