Using smartphone technology to reduce health impacts from atmospheric environmental hazards

Global environmental change is exacerbating human vulnerability to atmospheric environmental hazards including air pollution, aeroallergens (such as pollen) and extreme weather events. For example, poor air quality and extremes of heat or cold cause exacerbations of heart, lung and other chronic conditions leading to increases in symptoms, hospitalisations and deaths [1–3]. Global environmental change increases the threat from these hazards in a number of ways. A warming climate has brought more frequent heatwaves and other extreme weather events, which interact with air pollution to increase population health risk [4]. For instance, wildfires driven by worsening fire weather, deforestation fires and fuel management burning are increasingly causing prolonged smoke pollution episodes in densely populated airsheds, iconic examples being California and southern Australia [5]. Furthermore, population exposures and resultant sensitization to aeroallergens have increased in many temperate landscapes in both the northern [6], and southern hemispheres [7] due to the global spread of plants known to be allergenic, through the lengthening of pollen seasons, and the increased abundance of pollens in response to anthropogenic climate change, including higher concentrations of carbon dioxide [8].

These environmental health risk factors can work synergistically with each other. For example, interactions between pollen and severe weather events can trigger public health emergencies such as epidemic thunderstorm asthma [9, 10]. Collectively, they translate into substantial economic burdens because of the increased demands on health services and reduced workplace productivity due to absenteeism [11–14]. Further, symptomatic allergic rhinitis, is also associated with impaired learning and reduced examination performance [15, 16]. Allergic diseases overall have been estimated to cost the Australian economy $30 billion per year [17]. Wholesale pharmacy purchases for allergic rhinitis medications, such as oral anti-histamines and nasal corticosteroids, doubled between 2001 and 2010 to $226.8 million [18].

The large geographic areas affected by hazards such as air pollution or airborne allergens, the potentially high number of people exposed, and the difficulty in stemming the exposures present serious public health management challenges. Public information and advisories, typically disseminated via print and electronic media and more recently through electronic notification systems, are a central component of responses to mitigate the human impacts of environmental hazards [19]. There are currently few systems where health advisories are specifically tailored to match the vulnerabilities of individuals; however, individually targeted systems have been shown to be more effective for health interventions than generic broadcast messaging. For example, Licskai et al [20] demonstrated that receipt of personal notifications of real-time and forecast local air quality was associated with substantial reductions in symptom criteria for uncontrolled asthma.

In this context, smartphone technologies offer significant potential to improve public health responses to atmospheric hazards by enabling the delivery of customised, easily accessible health information. Such digital applications are being increasingly used by patients with chronic medical conditions where symptoms, treatments, and test results can be presented in graphical displays either by individual or aggregated population groups as a part of 'quantified self-tracking' [21, 22]. Additionally, simple feedback to app users can be provided in response to reported information. Apps that provide early detection of clinical deterioration can enable personalised adjustment to treatment plans and improve health [23, 24]. Other examples of environmental digital health tools include the mining of social media data to identify smoke affected areas [25], and using crowd-sourced symptom data for the surveillance of influenza-like illness [26]. Marks [27] has argued the case for extending these methods to address the need for systematic syndromic symptom surveillance to enable a rapid response to epidemic thunderstorm asthma or other emerging health threats. Overall, the strong potential for smartphone technologies to reduce the health impacts of environmental and atmospheric health hazards is clear.

The use of smartphone technology in health care is a rapidly expanding field. Boulos et al [28], identified a range of apps for medical providers, speciality or disease-specific apps, medical education and teaching apps, along with apps for patients and the public but none integrated environmental conditions with the symptom data. Some examples of smartphone apps for allergic conditions [29], specific exposures and outcomes such as pollen and allergic rhinitis symptoms [30], or air pollution and asthma symptoms [20], have since been published. However, we could not find other examples where smartphone technology is used to integrate individual health symptoms with multiple environmental conditions in near real-time, in order to characterise individuals' specific sensitivities and generate customised notifications. Here we describe the first 22 months of the operation of AirRater, an integrated online platform that combines environmental monitoring and symptom surveillance via a free smartphone app. In 2015 it was launched in Tasmania, a temperate island state south of mainland Australia, where forest fires and winter wood heating are common episodic causes of poor air quality. AirRater's aims were to improve quality of life for people with asthma, allergic rhinitis, and other conditions affected by smoke, pollen and air temperature. First, we describe the architecture of the AirRater platform, outline the methods used for collecting, integrating and reporting environmental data and describe the smartphone application interface for communication with users. We then briefly summarise environmental conditions during the study period and report user uptake and engagement with the project through time, including perspectives gathered through evaluation surveys. Finally, we discuss the potential utility and limitations of the system for providing easily accessible environmental information, and supporting individual self-management, clinical care and public health practice.

Zoom In Zoom Out Reset image size

Structure of the AirRater platform

Data gathering and spatial integration of environmental exposures

AirRater draws environmental data from a number of sources. These include PM_2.5 data from the Environment Protection Authority (EPA) Tasmania, pollen data from AirRater-operated sites around Tasmania, and weather data from EPA Tasmania and the Australian Bureau of Meteorology.

Particle monitoring

AirRater draws particle (PM_2.5) data from the Baseline Air Network of the EPA Tasmania (BLANkET), see figure 2(d) [31]. These data are available at 10 minute intervals from 34 locations, including all major population centres. AirRater uses these data to spatially model 10 minute and hourly average PM_2.5 concentrations​ across Tasmania, using ordinary Kriging [32] to interpolate values between monitoring sites.

Zoom In Zoom Out Reset image size

Pollen monitoring

Pollen and spore data are collected from six AirRater-operated sites (figure 2(c)) using a Burkard volumetric sampler (Burkard Manufacturing Co. Ltd, Hertfordshire) with either a daily (two sites) or weekly (four sites) sampling head [33].

Twenty seven pollen and one fungal taxon are counted individually (table S1 available at stacks.iop.org/ERL/13/044019/mmedia), with remaining taxa contributing towards the total pollen count. Within the app, users have access to the total pollen count along with six priority taxa: Poaceae (grasses); Myrtaceace (Eucalypt family); Cupressaceae (cypress family); Betula (birch); Acacia (wattle); Alternaria (an allergenic fungal spore) and Plantago (plantain). These taxa were selected based on their allergenic potential and local abundance [33–37]. Pollen counts are converted to atmospheric concentrations (grains m⁻³) and expressed as a daily mean value [38]. State-wide map surfaces for the total pollen concentration and key taxa of interest are interpolated using inverse distance weighting.

Weather monitoring

The EPA Tasmania operate automatic weather stations at each BLANkET station, which provide air temperature, wind speed, relative humidity, precipitation and atmospheric pressure to the AirRater environmental database at 10 minute intervals. In addition, we acquire meteorological data for the same variables at half-hour intervals from the Australian Bureau of Meteorology (figure 2(d)). Temperature data is spatially interpolated across Tasmania using regression-Kriging with elevation as a covariate, while other variables are assigned to user symptom reports based on proximity to the closest station.

User interaction

User evaluation

We evaluated the use of the AirRater app using information from the following sources: (a) Apple and Google store data, which report the daily number of app downloads, the AirRater server which logs the number of unique users, daily usage, and symptom reports; (b) the registration questionnaire (described above), which includes health and demographic information; and (c) user evaluation surveys. All registered users with at least one month of experience were invited to participate in online evaluation surveys. Respondents were included in a draw to win a gift voucher in an attempt to increase uptake. These surveys sought feedback on the usefulness and functionality of the app, and how and when it had been used to manage symptoms. The surveys were conducted 3, 9 and 12 months after launching the app. In the third evaluation survey, we compared the recalled presence of symptoms in the previous four weeks with the number of symptom reports to the app by the same user in the same four-week period.

The project was approved by the Human Research Ethics Committee of the University of Tasmania H0015006.

Zoom In Zoom Out Reset image size

Our results cover two main areas: the environmental exposures experienced during the study period for particulate matter, pollen and temperature; and user uptake and engagement with the app as measured by recruitment levels, symptom reporting, and the results of user evaluation surveys.

Environmental exposures during the study period

User uptake and engagement

Recruitment

During the reported study period, Apple and Google store data reported 3,443 downloads of the AirRater app. AirRater server data reported 1,959 (57%) of downloads as registered users. Of those who registered, 48% reported having a medical diagnosis of asthma, 65% reported allergic rhinitis, 34% reported both asthma and allergic rhinitis, and 11% reported that they cared for a child or other person with one of these conditions. Most registered users were female (72%) with ages ranging from less than 1 to 73 years. Overall, the proportion of users who elected to register stayed relatively constant through time (figure 4(a)). Of these, 48% also elected to be study participants.

Zoom In Zoom Out Reset image size

App downloads spiked in January 2016, coinciding with severe bushfires and associated episodes of poor air quality across the state, and again in late December 2016 coinciding with many media reports of hay fever and media promotion by the Department of Health and Human Services (figure 4(a)). Once this event passed (end of February 2016) the number of active users reduced.

Symptom reporting

Registered users provided the majority of symptom reports (94%). Almost 1,300 users made at least one symptom report during the study period, with 294 users making between 10 and 100 reports, and 27 reporting more than 100 times. Most reports related to nasal symptoms (n = 5,251 or 42%) followed by eyes (n = 3,578 or 29%), lungs (n = 2,247 or 18%) and throat (n = 904 or 7%). Large spikes in symptom reports were associated with spikes in downloads following publicity in local media (figure 4(b)). Because there was considerable variation in the number of people who used the app through time, we defined active users as the number of individuals that had used the app in the preceding week regardless of making a symptom report. Among active users, the rate of symptom reporting fluctuated but there were some clear spikes that coincided with severe smoke episodes (figure 5). Mean daily symptom reports per 100 active users varied seasonally, with the highest proportion in spring (72%), followed by winter (42%), summer (40%) and then autumn (37%).

Zoom In Zoom Out Reset image size

Evaluation

The average response rate in the three evaluation surveys over the course of the study was 25%. Respondents found out about the app through social media (31%), traditional media (21%) and word of mouth (17%). Based on a five point Likert scale, from 'no use' to 'extremely useful', just 6% of respondents said the app was of no use, while 50% rated it as either 'very' or 'extremely' useful for their needs. The feature of the app rated as the most useful was access to daily pollen data (63%), followed by notification of reduced air quality in saved locations (39%). After eight or more months of use, most respondents (65%) reported that they were more aware of how their environment influenced their health. Around a third (30%–36%) had used information from the app to support decisions about their medications, outdoor activities or home environment, and a few (11%) had discussed their results with health professionals (figure 6). These results suggest that the app was prompting positive changes in behaviour to improve health. Free text comments were provided by 34% of respondents. These were overwhelmingly positive and many suggestions were subsequently incorporated into later versions of the app.

Zoom In Zoom Out Reset image size

Comparison of symptom reporting between the app and survey

Only 25% of respondents who reported symptoms in the previous four weeks in the survey also logged symptoms via the app during the same period. However, 97% of respondents who reported 'no symptoms in the previous 4 weeks' in response to the survey, also did not log symptoms via the app in the same time period (table 1). This confirmed our expectation that symptoms, if present, were only reported some of the time, but we did not have further information to determine what influenced a user's decision to report.

Table 1. Reported symptoms in the last 4 weeks via the app compared with survey responses for the same period.

Symptoms (last 4 weeks)	Survey response that symptoms were not present in the past 4 weeks	Survey response that symptoms were present in the past 4 weeks	Total
No Airrater symptom reports in 4 weeks prior to survey	29	185	214
Any AirRater symptom reports in 4 weeks prior to survey	1	60	61
Total	30	245	275

AirRater is an integrated monitoring, modelling and communication tool that provides individuals and agencies with near real-time, spatially-resolved information on exposure to pollen, air pollution (including smoke) and climatic stressors, and allows individuals to enter symptom data (figure 1). Access to such data from a single source in a user-friendly format has not been previously available.

This study demonstrates the value of this functionality to participants, especially the capacity for generating individual-level information about associations between health symptoms and a range of different environmental hazards. Another clear benefit of the AirRater platform is that it is building an integrated epidemiological dataset for investigating the association of multiple environmental exposures and symptoms among a population of high-risk individuals. This will be potentially useful for public health agencies in understanding the epidemiology of environmental exposures, and developing strategies for managing the impact of environmental health events [41]. Overall, the system is providing a combination of individual health management support and epidemiological intelligence.

However, a number of important knowledge gaps and technical challenges must be addressed to enable translation of AirRater's demonstrated potential into a sustainable risk management tool applicable in multiple settings as discussed below. These include objective verification of health benefits, and the provision of accurate exposure estimation where robust monitoring data are not available through dense monitoring networks.

While app users strongly endorsed the AirRater app because of perceived health benefits, objective verification of health benefits is required to demonstrate the health management value of these types of tools [20, 22, 42]. Measurable health benefits are challenging to demonstrate because of the nature of the data collected through smartphone applications. Key issues include inconsistent usage by individuals through time, retention of users, and the difficulty of evaluating reporting bias such as increases in reporting associated with obvious environmental events. Our experience during periods of extreme events suggests that downloads of the app increased substantially when media reports suggested that the AirRater app is a helpful tool, however when these extreme events had passed continued engagement was challenging. Evaluations also need to address the accessibility of the technology to better understand its role in public health planning and responses: uptake can be uneven because of limited access to smartphone technology, local internet access, or familiarity with digital technology in older age groups or groups from language backgrounds different to that used in the app. AirRater had strong uptake and active use, with similar uptake across all geographical regions including regions of low population density and contrasting socio-economic status (see figures 2(a) and (b)). This is in contrast to other studies which have shown that health apps are more often used by those from higher socioeconomic and health literate groups [43]. Noting that evaluation of app data alone cannot provide information about non-users, such limitations must be addressed in population level epidemiological studies based on crowd-sourced data [44].

Robust environmental exposure estimates are fundamental to the validity of any such tool. Access to environmental information was one of the most useful features of AirRater reported in the user evaluations. In Tasmania, real-time air quality data are available for all populated regions and this is especially useful during planned burns and bushfires when smoke pollution can rise and fall rapidly [45]. However, such comprehensive monitoring networks are unavailable in many other regions or nations. The development and incorporation of modelled forecasts based on remote sensing of active fires and other indices of air quality could address this limitation for poorly monitored locations, as could the availability of low cost air quality sensors [46].

Providing reliable estimates of concentrations of airborne allergens is another challenge for broadening the use of AirRater or similar surveillance networks. Pollen apps are common in North America and Europe, and are available for parts of Australia. Although many provide forecasts, few are based on accurately measured aeroallergens or validated forecasting methods [47]. In Australia, availability of this information is largely dependent upon sparsely located, university-based research teams [48]. Some monitors operate for a few months of the year, are restricted in the range of taxa that they report and most display either grass or total pollen counts rather than separate taxa. The advantage of AirRater, like many European pollen apps, is that multiple common taxa are reported, and participants can then use the correlation with symptoms to make behavioural changes to reduce their health impacts [30]. While only 11% of participants in this study reported that they discussed results with their health professional, there is potential for the data generated by AirRater to inform more evidence-based decisions around pollen treatment including immunotherapy [49]. There is a clear case for providing multi-taxon, geographically specific aeroallergen exposure estimates. However, identifying and counting pollen is labour intensive and costly, and providing such data is often cost-prohibitive. Finding ways to sustainably provide accurate, current and forecast pollen conditions for large or geographically dispersed populations is a high priority for environmental health research. While there have been promising advances in automated fluorescence and machine-learning pollen image classification [50, 51], as yet there are no cost-effective, scientifically-validated automated pollen monitors available for deployment. An alternative approach is phenological modelling combined with remote sensing to estimate the timing, magnitude and spatial distribution of pollen release from allergenic taxa. This has the virtue of providing spatially explicit pollen predictions. Studies from the northern hemisphere demonstrate that such an approach has strong potential [52, 53].

Monitoring and managing environmental hazards will become increasingly important as exposures continue to increase with global climate change. While there is potential for considerable individual, clinical and wider public health benefits from smartphone apps such as AirRater, it is essential that environmental data are reliable and validated. The way in which individual users interact with and report data need to be better understood before conclusions about individual or public health benefits can be drawn. Improved monitoring of air quality and pollen, especially the incorporation of short-term forecasting, will increase the functionality of such systems. Future approaches for incorporating highly spatially resolved data could include a tiered monitoring approach that integrates gold standard regulatory monitoring with automated sensor technology, remote sensing and modelling estimates. Nonetheless, we conclude that digital technologies offer considerable promise in supporting the public health management of environmental health hazards and in supporting individuals to manage health conditions sensitive to changes in the environment.

The authors would like to acknowledge input to the AirRater project from Dr Fabienne Reisen and Dr Martin Cope at CSIRO, Dr John Innis at EPA Tasmania, Professor Simon Haberle at the Australian National University, and Professor Haydn Walters and Dr Jennifer Sprent at the University of Tasmania. The team would also like to acknowledge the hard work of the pollen technicians and Dr Anne Watson, along with the support received at each of the pollen monitoring sites. Funding for the AirRater project was provided by Sense-T through support from the Australian Government. Current funding is provided by the Menzies Institute for Medical Research. Funding for Dr Amanda Wheeler and Dr Grant Williamson's research fellowships was provided by The Centre for Air quality and health Research and evaluation (CAR). The research team would like to acknowledge the support of the AirRater Stakeholder Group.

Conflict of Interest: The authors declare they have no conflict of interest.