Article Text
Abstract
Objectives The objective of this study was to use health administrative and environmental data to quantify the effects of ambient air pollution on health service use among those with chronic diseases. We hypothesised that health service use would be higher among those with more exposure to air pollution as measured by the Air Quality Health Index (AQHI).
Setting Health administrative data was used to quantify health service use at the primary (physician office visits) and secondary (emergency department visits, hospitalisations) level of care in Ontario, Canada.
Participants We included individuals who resided in Ontario, Canada, from 2003 to 2010, who were ever diagnosed with one of 11 major chronic diseases.
Outcome measures Rate ratios (RR) from Poisson regression models were used to estimate the short-term impact of incremental unit increases in AQHI, nitrogen dioxide (NO2; 10 ppb), fine particulate matter (PM2.5; 10 µg/m3) and ozone (O3; 10 ppb) on health services use among individuals with each disease. We adjusted for age, sex, day of the week, temperature, season, year, socioeconomic status and region of residence.
Results Increases in outpatient visits ranged from 1% to 5% for every unit increase in the 10-point AQHI scale, corresponding to an increase of about 15 000 outpatient visits on a day with poor versus good air quality. The greatest increases in outpatient visits were for individuals with non-lung cancers (AQHI:RR=1.05; NO2:RR=1.14; p<0.0001) and COPD (AQHI:RR=1.05; NO2:RR=1.12; p<0.0001) and in hospitalisations, for individuals with diabetes (AQHI:RR=1.04; NO2:RR=1.07; p<0.0001) and COPD (AQHI:RR=1.03; NO2:RR=1.09; p<1.001). The impact remained 2 days after peak AQHI levels.
Conclusions Among individuals with chronic diseases, health service use increased with higher levels of exposure to air pollution, as measured by the AQHI. Future research would do well to measure the utility of targeted air quality advisories based on the AQHI to reduce associated health service use.
- PREVENTIVE MEDICINE
- PUBLIC HEALTH
This is an Open Access article distributed in accordance with the Creative Commons Attribution Non Commercial (CC BY-NC 4.0) license, which permits others to distribute, remix, adapt, build upon this work non-commercially, and license their derivative works on different terms, provided the original work is properly cited and the use is non-commercial. See: http://creativecommons.org/licenses/by-nc/4.0/
Statistics from Altmetric.com
Strengths and limitations of this study
Use of a large population-based data set and inclusion of multiple major chronic diseases and conditions over an 8-year period.
Exposure measurements were captured by multiple monitoring stations on an hourly basis, providing a thorough data set of air quality and pollutants.
Use of fixed exposure monitoring sites to infer exposures of individuals across relatively large geographic areas (Local Health Integration Networks) may lead to potential misclassification of exposure.
Lack of individual level data may have overestimated or ‘double counted’ health service use among individuals with more than one chronic disease or when individuals were admitted to hospital through the emergency department and were counted as both the emergency department and hospital admission encounters.
We have not analysed other pollutants, such as carbon monoxide and sulfur dioxide, which may be important morbidity factors in certain regions.
Introduction
Air pollution poses measurable adverse health effects across populations and thus, is increasingly identified as a significant public health issue. The Air Quality Health Index (AQHI) is a composite measure of air quality (nitrogen dioxide (NO2), fine particulate matter (PM2.5) and ozone (O3)), specifically developed to measure the impact of air pollution on human health.1–3 Previous research has used population-based data to examine the short-term impact of air pollution on asthma-related morbidity and has shown an association between increased daily maximum AQHI and increased asthma health services use on the same day and on the following 2 days.1
The impact of air pollution on health service use is relatively well defined for pulmonary and cardiovascular disease; however, the acute effects in individuals with other chronic conditions are less well known. Recent studies have suggested that other chronic diseases may also be aggravated by air pollution, including chronic obstructive pulmonary disease (COPD), cardiovascular disease (including heart attack and stroke), cancer and diabetes.4–12 The objective of this study was to use the AQHI to quantify the impact of exposures to poor air quality on short-term morbidity measured by health service use for individuals with 11 major chronic diseases. Quantifying effects of ambient air pollution on health service use in vulnerable populations with chronic diseases will assist in informing public policy on air quality standards.
Methods
Study population
We used health administrative data from Ontario, Canada—a province with a diverse multicultural population of approximately 13 million in 2010 and universal health system access for all residents. We included individuals aged 0–99 years if they had ever been diagnosed with any of these 11 chronic diseases: asthma, COPD, diabetes, hypertension, angina, acute myocardial infarction, ischaemic heart disease, congestive heart failure, stroke, lung cancer and non-lung cancers. We used previously validated algorithms with International Classification of Disease codes to identify these diseases and conditions (see online supplementary material, S1).
Exposure measures: air quality
The Ontario Ministry of the Environment and Climate Change provided hourly AQHI and air pollutant measures (NO2, PM2.5, O3) from 1 January 2003 to 31 December 2010; these air quality measurements were collected by 42 fixed-site air quality monitors located across 33 cities in 14 Local Health Integration Networks in Ontario, administered by the Ministry of the Environment and Climate Change. These monitors made continuous measurements of common air pollutants near real time. Online supplementary material, S2 provides details about quality control conducted by the Ministry of the Environment and Climate Change. The AQHI is calculated using the following formula:Units are parts per billion (ppb) for NO2 and O3, and micrograms per cubic metre (µg/m3) for PM2.5.13 The AQHI presents a simple scale ranging from 1 to 10+ where 1 represents good air quality and a low health risk, and 10 represents poor air quality and a high-health risk (see online supplementary material, S2).We calculated daily 1 h maximum values of AQHI, NO2, PM2.5 and O3 for each Local Health Integration Network. One-hour maximum values were used, as it has been suggested in the literature that these provide the most significant results when predicting short-term health effects, compared to mean and median values.14
Outcome measures: health services use
We used total daily all-cause health service use to measure acute health morbidities for individuals with chronic diseases. We obtained these data from four health administrative databases housed at the Institute for Clinical Evaluative Sciences: the Ontario Health Insurance Plan Database contains information on fee-for-service billings for physician services rendered; the National Ambulatory Care Reporting System contains data for emergency department visits; the Canadian Institute for Health Information Discharge Abstract Database records the primary and secondary diagnoses for all hospitalisations; the Ontario Registered Persons Database includes information on sex, birth date and region of residence. We calculated all-cause rates of health service use per 1000 disease-specific populations.
Definition of covariates
We adjusted for potential confounding variables in our multivariable analysis. These variables included age group (10-year age groups), sex, region of residence, socioeconomic status, day of the week, temperature, season and year of data (2003–2010). We examined individuals aged 20+ years in all analyses; however, we also included individuals aged 0–19 years in the analysis of asthma and diabetes, excluded those aged under 30 years in the analysis of COPD and excluded those aged under 40 years in the analysis of congestive heart failure (see details in online supplementary materials, S1). We categorised region of residence according to Local Health Integration Network. We inferred socioeconomic status from neighbourhood income quintiles, based on census data. We categorised seasons as follows: spring (March to May), summer (June to August), fall (September to November) and winter (December to February). There was less than 0.5% missing data in demographic variables and 10% in income quintiles; missing categories were excluded from the regression analyses. Missing daily temperatures (less than 1%) were substituted by the mean temperature of the respective month.
Statistical analysis
We calculated yearly distributions of the daily maximum values of the AQHI and pollutants, as well as prevalence and health service use for the chronic diseases from 2003 to 2010. We used Poisson regressions to examine the associations between air quality and health service use; we examined immediate effects with a time lag of 0, 1 and 2 days. The unit of analysis for health service use was the number of outpatient claims or emergency department or hospital admissions on day 0. We included offset terms for the disease-specific populations and adjusted for confounders. We used rate ratios (RR) with p values to calculate the risk of short-term disease-specific health service use associated with acute exposure to poor air quality.15 We checked for overdispersion and violations of Poisson assumptions by plotting residuals versus mean exposure level.15 We used Pearson’s χ2 test to assess model goodness-of-fit. To calculate the expected increase in health service use on days with an AQHI of 3, 6 and 10 (chosen based on AQHI health advisory category cut-points for low risk, medium risk and high risk, respectively), we multiplied the disease-specific daily average health service use rates from the Poisson models by the average disease prevalence. We used SAS V.9.2 (SAS Institute Inc, Cary, North Carolina, USA) to perform the analyses.16
Results
Air quality trend
The distributions of the mean daily maximum AQHI, NO2, PM2.5 and O3 by year and season are shown in online supplementary material, table S3. For 2003–2010, average measures±SDs were AQHI: 3.45±1.19, NO2: 24.39±12.41 ppb, PM2.5: 13.93±9.53 µg/m3 and O3: 39.57±14.15 ppb, with a general improvement in all air quality metrics over time. Air quality and pollutant values varied seasonally: AQHI was highest in the spring (3.79±1.09) and lowest in the fall (3.02±1.13); NO2 was highest in winter (27.98±12.58 ppb) and lowest in summer (19.71±10.14 ppb); PM2.5 and O3 were highest in summer (PM2.5: 18.06±11.31 µg/m3; O3: 48.63±15.53 ppb) and lowest in winter (PM2.5: 11.12±6.77 µg/m3; O3: 30.76±7.12 ppb). Air quality and air pollutant values also differed by geographic region, with Ontario's smallest and most populous Local Health Integration Network having the highest average AQHI (4.05±1.27) and NO2 (33.18±12.18 ppb), while Ontario's largest and least populous Local Health Integration Network had the lowest average AQHI (2.87±0.92), PM2.5 (10.57±7.04 µg/m3) and O3 (35.98±9.85 ppb).
Distributions of chronic diseases and morbidities
Table 1 shows the distributions of the chronic diseases and associated all-cause health service use per 1000 disease population. Hypertension (24.9%) and asthma (13.3%) had the highest disease prevalence. Individuals with lung cancer had the highest emergency department visits and hospitalisation rates (209.6 and 330.6 per 1000, respectively), while individuals with diabetes had the highest outpatient visit rate (2805.9 per 1000). Chronic disease prevalence rates from 2003 to 2010 are presented in online supplementary material, figure S5.
Associations of air quality measures and disease morbidities
Table 2 and figure 1 show the results of the disease-specific multivariable Poisson regressions that modelled the association between AQHI, air pollutants and health service use on day 0 of exposure (see day 1 and 2 of exposure in online supplementary material, table S4a and S4b). The risk of individual pollutant was adjusted for the presence of other pollutants. None of the regression models violated model assumptions. Overall, the increase in the rate of outpatient visits for each unit increase in the same-day AQHI and each unit (10-ppb) increase in daily maximum NO2 (adjusted for PM2.5 and O3) varied from 1% to 14%. Those with non-lung cancers (AQHI:RR=1.05; NO2:RR=1.14) or COPD (AQHI:RR=1.05; NO2:RR=1.12) had the greatest increase in same-day outpatient visits. Increase in rate of same-day hospitalisation was highest among those with diabetes (AQHI RR=1.04; NO2:RR=1.07) and COPD (AQHI RR=1.03; NO2:RR=1.09). The increase in rate of same-day emergency department visits was relatively small. The impact on health service use remained 1 and 2 days after the measured peak AQHI and NO2.
Table 3 shows a translation of the population data into predicted daily rates and daily expected all-cause health service use among populations with each of the chronic conditions on days with a maximum AQHI of 6 (medium health risk) or 10 (high), as compared to days with a maximum AQHI of 3 (low health risk). Given that the increase in the rate of emergency department visits was small (see table 2), we focused on the per cent change in outpatient visits and hospital admissions. The highest per cent changes in outpatient visits on a day with a maximum AQHI of 10 were predicted in COPD, non-lung cancers and angina (43.9%, 39.6% and 32.4%, respectively). The highest per cent changes in hospitalisations were predicted in diabetes, COPD and angina (29.7%, 26.7% and 19.4%, respectively). The greatest increase in expected counts of outpatient visits and hospitalisations on a day with a maximum AQHI of 10 was predicted in hypertension (3862 more expected outpatient visits) and non-lung cancers (49 more expected hospitalisations), respectively. Overall, the increase in air pollution-attributable outpatient visits comparing moderate and high-risk air quality days to low-risk days amounted to a total of 6052, and 15 301 visits per day, respectively.
Discussion
To date, in magnitude and breadth, this is one of the largest time series analyses examining the impact of air quality on health. We examined the effects of health service use across an entire health system and we analysed all major chronic diseases over 8 years. Strategically, this study focused on the AQHI, an important national translational tool that distils the complexity of air quality science and health risk into a simple visual linear scale for individuals with chronic disease, healthcare providers and public health and environmental policymakers. This study demonstrated a consistent and quantifiable health risk based on the AQHI that may inform public health and environmental policy.
The WHO describes a pyramid of air pollution-attributable health impacts that includes subclinical effects, symptoms, health services utilisation and mortality,17 ,18 and the US Environmental Protection Agency has also concluded that PM2.5, O3 and NO2 are causal or likely causal factors for numerous cardiac and pulmonary health outcomes.10–12 Plausible mechanistic pathways have been described, including enhanced coagulation/thrombosis, a propensity for arrhythmias, acute arterial vasoconstriction, systemic inflammatory responses and the chronic promotion of atherosclerosis.19 Our findings strengthen and extend an understanding of the acute impact of air pollution on chronic diseases for which there is limited published evidence, in particular for diabetes, hypertension and malignancy.
Hypertension remains the worldwide leading risk factor for mortality. A number of studies have reported associations between air pollutants and hypertension, as well as between air pollutants and diabetes incidence and prevalence;20–24 however, few have demonstrated its impact on health-related morbidity. Szyszkowicz25 used a large administrative database in Canada, and measured the association between ambient air pollution and hypertension presentations to emergency department visits during a 10-year period. They reported significantly elevated risks of emergency department visits (4–6%) up to 3–6 days after peaked exposures to PM10, PM2.5, NO2 and SO2. Similarly, our study showed an 8% increase in same-day physician visits per 10 ppb increase in same-day peak NO2, the fifth highest ratio observed after non-lung cancers, angina, COPD and congestive heart failure. This finding may indicate that modest elevations in air pollutants worsen blood pressure control, consequently leading to higher outpatient health service use or alternatively, that hypertension is a sensitive surrogate marker for multiple comorbidities.
Despite a compelling body of literature confirming the health risk of various air pollutants on individuals with chronic disease, these risks are underappreciated by patients and seldom acted on by clinicians.26 The complexities of communicating air quality, its associated health risks, and actionable risk-reduction messages may be contributing to low physician and patient uptake. Our study extends prior research by evaluating a new method of reporting and advising clinicians and patients about air quality and its risks: the AQHI. The AQHI was designed to create a more direct link between air quality and health, and to present the index in a format that could easily communicate risk reduction to the general population.27 Using a format similar to the ultraviolet ray index that the general public is familiar with, the AQHI ‘translates’ air quality on a scale of 1–10 and attaches specific risk-reduction health messages. Hong Kong, with the number of air pollution-related deaths ranked eighth highest out of 193 countries in the world,28 has recently adopted AQHI for real-time monitoring and reporting. In December 2013, AQHI was commissioned by the Hong Kong government and a free smartphone AQHI application was made available from iTunes.29 Similar scalable communication strategies are being executed in Canada. The effectiveness of these strategies remains to be determined.
This study has several strengths, namely the use of a large population-based data set and inclusion of multiple major chronic diseases and conditions over an 8-year period. To date, very few published studies are of comparable magnitude. The use of large sample sizes allowed Poisson regression models to be fully adjusted for multiple potential confounders. Furthermore, exposure measurements were captured by multiple monitoring stations on an hourly basis, providing a thorough data set of air quality and pollutants. Our study also calculated the predicted daily rate and expected volume of health services use. These rates can be used by others to generate their respective expected health service use counts based on their local disease-specific prevalence population.
A few limitations are worth noting. First, the use of fixed exposure monitoring sites to infer exposures of individuals across relatively large geographic areas (Local Health Integration Networks) may lead to potential misclassification of exposure—namely, the Berkson error. However, it has been shown that with exposure measured on a numerical scale, it leads to minimal bias in regression.30 Our study used large population data; the CIs remained narrow and the loss of power was minimal. Second, the lack of individual level data may have overestimated or ‘double counted’ health service use among individuals with more than one chronic disease or when individuals were admitted to hospital through the emergency department and were counted as both the emergency department and hospital admission encounters. The low proportionate rate of hospitalisation in this study was small, suggesting that this impact was minimal. Finally, in our analysis of the AQHI, we have not analysed other pollutants, such as carbon monoxide and sulfur dioxide, which may be important morbidity factors in certain regions.
Future directions would do well to explore the role that comorbidities play in the relationship between exposure to air pollution and health service use. Further, it would be interesting to study the effects of air pollution on health service use among those who are revisiting the healthcare system for a disease or condition that they already have, compared to those who are being diagnosed for a new disease or condition.
Conclusion
This study suggests that air pollution exposure should be considered a risk factor for acute health morbidities among people with certain common chronic diseases. Timely AQHI air quality advisories with integrated risk reduction messages may lead to reduced morbidity associated with acute exposures to air pollution in individuals living with chronic diseases; the AQHI may also be useful in guiding public health and environmental health policy by accurately and consistently quantifying the health impacts of air pollution.
References
Supplementary materials
Supplementary Data
This web only file has been produced by the BMJ Publishing Group from an electronic file supplied by the author(s) and has not been edited for content.
- Data supplement 1 - Online supplement
Footnotes
Contributors TT initiated, designed the study, conducted the statistical analysis and drafted the manuscript. LF prepared all tables and graphs. JS prepared data sets and descriptive statistics. JS, ASG, SD, CL and RF commented on drafts. All authors have seen and approved the final version.
Funding Funding for this project was made available through the Best in Science Programme, provided by Ontario's Ministry of the Environment and Climate Change. This study was also supported by the Institute for Clinical Evaluative Sciences, which is funded by an annual grant from the Ontario Ministry of Health and Long-Term Care. The Ministry of the Environment and Climate Change, and the Institute for Clinical Evaluative Sciences had no role in study design, analysis, interpretation of data or writing of the report. The opinions, results and conclusions presented in this report are those of the authors and are independent from the funding sources. No endorsement by the Institute for Clinical Evaluative Sciences or the Ministry of the Environment and Climate Change is intended or should be inferred.
Competing interests None declared.
Provenance and peer review Not commissioned; externally peer reviewed.
Data sharing statement No additional data are available.