Susceptibility to short-term ozone exposure and cardiovascular and respiratory mortality by previous hospitalizations

Background Ozone (O3) has been associated with cardiorespiratory mortality although few studies have explored susceptible populations based on prior disease. We aimed to investigate the role of previous hospitalization on the association between short-term exposure to O3 and cardiovascular (CV) and respiratory mortality. Methods We performed time series analyses using generalized additive models and case-crossover on 136,624 CV and 23,281 respiratory deaths in Stockholm County (1990–2010). Deaths were linked to hospital admissions data. We constructed 2-day and 7-day averages using daily 8-h maximum for O3 and hourly values for PM2.5, PM10, NO2, and NOx from a fixed monitor. Results We observed a 0.7% (95% CI: 0.1%, 1.3%) and 2.7% (95% CI: 0.8%, 4.6%) higher risk of CV and respiratory death per 10 μg/m3 higher 2-day and 7-day average O3 respectively. Individuals previously hospitalized for myocardial infarction demonstrated 1.8% (95% CI: 0.4%, 3.4%) higher risk of CV death per 10 μg/m3 higher 2-day average O3 and similar associations were observed in individuals with no previous hospitalization for any cause. Individuals with previous hospitalizations did not show susceptibility towards O3-related risk of respiratory mortality. We observed no associations for other pollutants. Conclusion Short-term ozone exposure is associated with CV and respiratory mortality and our results may suggest higher susceptibility to CV mortality following O3 exposure in individuals previously hospitalized for myocardial infarction. Higher risks were also observed in individuals with cardiovascular death as their first presentation of disease. Electronic supplementary material The online version of this article (10.1186/s12940-018-0384-z) contains supplementary material, which is available to authorized users.


Background
Higher levels of ambient Ozone (O 3 ) have been associated with cardiovascular (CV) and respiratory mortality [1,2]. Severity of health effects of O 3 experienced by individuals with pre-existing diseases or other determinants may be larger than among other individuals [3]. Knowledge on susceptible populations is important for providing relevant guidelines for air quality standards protecting significant susceptible groups within the general population. In an attempt to identify such groups, some studies have explored age differences and reported elderly [3][4][5][6] and children [7] as populations with higher risk. Susceptibility according to sex or other sociodemographic characteristics [3,4,8] have also been explored but with mixed results. Previous studies that have explored the impact of pre-existing diseases on the susceptibility to O 3 mortality have for example used surrogates for pre-existing diseases using secondary diagnosis of death [3] or have considered total mortality rather than cause-specific mortality, both factors affecting accurate identification of susceptible populations [3,6,8,9]. The pre-existing diseases that conferred increased susceptibility in these studies were not fully consistent from study to study but centered around diabetes, atrial fibrillation and atherosclerotic diseases. In addition, there is some evidence to suggest that susceptibility factors to O 3 mortality are more important in cities with low air pollution concentrations [3].
Illness related to CV or respiratory diseases resulting in hospitalizations such as ischemic heart disease events, exacerbations of chronic obstructive lung disease (COPD), and diabetes are likely to be harbingers of CV or respiratory mortality and may be markers of susceptibility to the effects of O 3 . In a previous study from Stockholm, we observed association between 2-day levels of O 3 and risk of non-traumatic mortality and this risk was higher among individuals with previous hospitalizations for acute myocardial infarction (AMI) [8]. We therefore hypothesized that individuals with previous hospitalizations for AMI, other CV diseases, respiratory diseases, and diabetes would carry higher risks of CV mortality per incremental increase in short-term O 3 exposure compared to the risk in the overall population. Likewise, we hypothesized that individuals with previous hospitalizations for COPD, pneumonia, or other respiratory diseases would carry higher risks of respiratory mortality per incremental increase of short-term O 3 exposure compared to the risk in the overall population. Consequently, in the present study we aimed to investigate the role of previous hospitalizations on the associations between short-term exposure to O 3 and CV and respiratory mortality. In the setting of the low-level pollution environment of Stockholm, Sweden, we made use of the well-validated Swedish death registry and inpatient registry [10] to include all deaths in Stockholm and all hospitalizations in Sweden to study these associations using continuous monitoring of ambient air pollution concentrations.

Study population
We included residents of Stockholm County older than 30 years with CV or respiratory deaths occurring in the County from 1990 to 2010. We obtained age, sex, date and cause of death, and all dates and diagnoses of hospitalizations from the Swedish Board of Health and Welfare with mortality and hospitalizations data available from 1990 and 1987, respectively. We linked all CV and respiratory deaths to all previous hospitalizations in Sweden up to 3 years before death to create susceptible groups based on the principal diagnosis code (International Classification of Diseases, Ninth Revision (ICD-9) and Tenth Revision (ICD-10) at discharge (Table 1). In a previous study, hospitalizations prior to 3 years before death did not identify a population sensitive to air pollution related overall mortality [8] and therefore we chose a 3 year period for previous hospitalizations in order to have a uniform hospitalization period for all individuals. Table 1 Characteristics and categorization of cardiovascular (ICD-9 390-459; ICD-10 I00-I99) and respiratory (ICD-9 460-519; ICD-10 J00-J99) deaths by previous admissions for diseases among
In an effort to avoid double counting of individuals with multiple hospitalizations in different hospitalization categories we followed the above-mentioned hierarchies to ensure mutually exclusive groups. Only one hospitalization was considered for each individual and category. Count data was created per day for each mortality outcome and further stratified by categories of previous hospitalizations.

Air pollution and meteorological data
Exposure data was provided by the Stockholm-Uppsala County Air Quality Management Association. We identified a priori O 3 as our main exposure of interest based on our previous study and included exposure data for Nitrogen dioxide (NO 2 ), Nitrogen oxides (NO x ), particles with aerodynamic diameter smaller than 2.5 and 10 μm (PM 2.5 and PM 10 ) for comparison with other studies. Hourly means of O 3 , NO 2 , NO x , PM 2.5 , PM 10 , temperature and relative humidity were obtained from a single centrally located urban background monitoring station.
For O 3 , we first calculated an 8-h running mean over 24-h period to compute an 8-h maximum. We calculated 24-h averages of NO 2 , NO x , PM 2.5 , PM 10 , temperature, and relative humidity from hourly values.

Statistical analysis
We performed both time series and case-crossover analyses to analyze associations between air pollutants and the risk of cause-specific mortality in those with and without previous hospitalizations. In time series, we employed generalized additive models based on quasi-Poisson distribution. These models were adjusted for long-term and seasonal trends, temperature, relative humidity, daily influenza hospital admissions, day of the week, and public holidays. We modeled seasonal and long-term time trend by penalized splines with 5 degrees of freedom (df) per year; holidays and day of week as indicator variables; the mean values of the relative humidity during the day of death and previous day as linear terms; the mean values of the temperature during the day of death and previous day as cubic splines with 3 degrees of freedom; and a 7-day running average of daily influenza admissions in Stockholm as cubic spline with 4 df.
For comparison we also employed the case-crossover analyses for both the main effects assessing associations between O 3 exposure and CV or respiratory mortality and to estimate the presence of effect modification by season, age, sex, and susceptibility by previous hospitalizations using multiplicative terms including O 3 and the specific effect modifier in separate models. We used a time stratified referent selection strategy with conditional logistic regression. Control periods were matched with case periods on same day of the week, within the same calendar month, and year as the day of death. This way, time trend and slowly changing individual characteristics were adjusted by the design. We adjusted for temperature and daily influenza hospital admissions as a restricted cubic spline with 2 degrees of freedom (df) and relative humidity as a linear term. Age was dichotomized by median age at the time of death and season by warm (April to September) and cold (October to March) season. For susceptibility by previous hospitalization, we performed a likelihood ratio test to compare the model with an interaction term including hospitalization categories (using the largest category as the reference) with the reduced model without the covariate containing the hospitalization categories.
We used a 0-1 day lag (2 day average) and 0-6 day lag (7-day average) in single-pollutant models in our initial analysis based on evidence from previous studies indicating most apparent effects both within the immediately preceding days and up to a week [5,6]. We then selected a lag for subsequent analyses of CV and respiratory mortality that demonstrated the strongest association, in separate models, assessed by largest effect estimates with statistical significance. We also evaluated two-pollutant models to adjust for other pollutants. We then proceeded with stratified analyses of CV and respiratory mortality according to previous hospitalizations in time series analyses and separately employed the whole dataset in case-crossover analyses using a multiplicative term for O 3 and hospitalization category.
In sensitivity analyses, we controlled for the effects of temperature over longer periods using averaging times of 0-6 days for temperature. Our cause-specific mortality data starts from 1990 and previous hospitalization data starts from 1987. We relaxed our criterion of a three-year uniform hospitalizations prior to death to include up to five-year previous hospitalizations for the purpose of comparing results to our previous study [8].

Exploratory analyses
All exploratory analyses were performed post hoc in the subgroup of CV mortalities in individuals hospitalized for AMI. Since the timing of the hospitalization in relation to date of death may be important in modulating the susceptibility to short-term O 3 exposure, we explored patterns of association by stratifying the time lapse between the last AMI hospitalization before death and death into four oneweek categories and six six-month categories before death and employed models for each time-category. Each timecategory included a separate model of daily CV deaths for individuals with AMI hospitalizations occurring within in specified time-period before date of death. For example, the first model included daily counts of CV deaths in a dataset in which CV deaths had been preceded with last AMI hospitalizations within 0 to 6 days (week 0) before death. In this manner, we created datasets for each stratum and performed stratified analysis.
All estimates are expressed as percent increase risk of death risk with 95% confidence intervals per 10 μg/m 3 increase in air pollution levels. All case-crossover and time series analyses were performed using Stata and R, respectively.

Results
A total of 159, 905 CV and respiratory deaths were observed in Stockholm County from 1990 to 2010. We observed 6 times more CV deaths compared to respiratory deaths with a slight majority occurring in women ( Table 1). The average age at the time of CV death was 77 years in men and 84 years in women and at the time of respiratory death 79 years in men and 82 years in women.
Average concentrations of all measured pollutants were low by international comparison (Table 2). Ozone measurements showed a near-normal distribution with sizable temporal variability whereas other pollutants demonstrated right-skewed distributions. Ozone was moderately positively correlated with PM 10 , and negatively correlated with NO 2 and NO x . Ozone and temperature were also moderately correlated (Table 3).
We observed associations between 8-h maximum O 3 levels and CV mortality for both averaging periods with narrower confidence intervals for 8-h maximum O 3 averaged over 2 days before death and similar estimates in both in time series analysis and case-crossover analyses (Table 4). No associations were observed between average PM 2.5 , PM 10 , and NO 2 levels and CV mortality regardless of averaging period or analyses method. In two-pollutant models, the association between O 3 and CV mortality was slightly increased while estimates for PM 2.5 , NO 2 and NO x remained non-significant (Additional file 1: Table S1). In the two-pollutant model with PM 10 , we observed a slightly stronger estimate for O 3 (0.9% increased risk per 10 μg/m 3 ; 95% CI 0.2%, 1.6%) and unexpectedly a decreased risk of CV mortality related to PM 10 (-1.3%; 95% CI: -2.4%, − 0.2%, per 10 μg/m 3 increment).
In further analyses, 2-day average O 3 was associated with a higher CV mortality risk in individuals older than the median age (82 years old, 1.7% (95% CI 0.9%, 2.4%) compared with individuals equal to or below the median age (0.6%; 95% CI -0.2%, 1.4%) with a p-value for interaction of 0.04 (Additional file 1: Table S2). Differences in O 3 associations between men and women or deaths occurring in the warm or cold season were not as clear (p-values for interaction 0. 7 and 0.9 respectively).
Associations between 2-day O 3 and CV mortality by categories of hospitalization demonstrated positive We further stratified individuals with previous AMI hospitalization based on the admission date of the last AMI in time series analyses. The risk of CV mortality following 2-day average O 3 did not demonstrate a clear trend for hospitalization occurring the month immediate preceding death. Risk estimates among individuals with hospitalizations occurring 1-36 months were consistently positive (Fig. 2).
We observed stronger associations between 8-h maximum O 3 levels and respiratory mortality in 0-6 day lag compared to 0-1 day lag irrespective of analysis method (Table 4). We did not observe associations between average PM 2.5 , PM 10 , and NO 2 levels and respiratory mortality for any averaging period. In two-pollutant models, associations between O 3 exposure and respiratory mortality remained positive in the co-presence of NO 2 , and NO x (Additional file 1: Table S1). In the two-pollutant model including PM 2. 5 and PM 10 , the association for O 3 became weaker and non-significant and the estimates for both particulate matter fractions were reduced.
In further exploratory analyses associations between 7day average O 3 and respiratory mortality remained unchanged in models including a multiplicative term for O 3 and sex, age, or season (p-values for interaction ≥0.1, Additional file 1: Table S3).
Associations between 7-day O 3 exposure and respiratory mortality by previous hospitalization categories were similar in time series and case-crossover methods (Fig. 3). Positive associations were observed in individuals with previous hospitalization for COPD and other diseases, however; there was little support for effect modification   Associations were similar when modeling the same averaging periods for temperature as for exposure as well as the use of 5-year rather than 3-year previous hospitalizations data for vulnerable population subgroups.

Discussion
Among all deaths in Stockholm County between 1990 and 2010 we observed associations between higher short-term concentrations of O 3 and CV and respiratory mortality. Estimates were higher for respiratory than for CV mortality. Associations between O 3 and CV mortality were relatively stronger in individuals hospitalized for AMI as well as for those individuals who had not been hospitalized for any disease. Associations between O 3 and respiratory mortality were not affected by individuals' previous hospitalizations. We did not observe associations between PM 2.5 , PM 10 , NO 2 and NO x levels and CV or respiratory mortality.
Associations for CV mortality were similar for both averaging periods of 8-h maximum O 3 . In contrast, we observed higher estimates for the association between 7day average 8-h maximum O 3 and respiratory mortality compared to the shorter averaging period of 2-day. This pattern is largely consistent with the The Air Pollution and Health: a European Approach study that also reported stronger estimates for the association between O 3 and respiratory mortality in longer lags (0-20 day) as compared to shorter lags (0 and 0-1 day), while the opposite was observed for CV mortality [11]. This implies that exposure to high O 3 levels has a more immediate effect on triggering CV mortality, whereas the effect on respiratory mortality might require longer O 3 exposure and other biological processes [11].
Previous studies have reported positive associations between PM 2.5 , PM 10 , NO 2 and NO x and CV and respiratory mortality [12,13]. We found no evidence of associations for these pollutants in our study. Possible explanations may include low concentrations and the use of single monitors to assess pollutants with more spatial heterogeneity than O 3 . Little research has been conducted on identifying individuals susceptible to the harmful effects of air pollution based on previous CVD, however, our results are generally concordant with results from the existing studies. In our previous study from Stockholm [8], we observed higher total mortality in association with 8-h maximum O 3 exposure in individuals with previous AMI compared to the full study population. Risk estimates are similar in the current study investigating CV mortality (1.7% vs 1.8%). A study from Italy [6], also investigating total mortality, reported non-significant associations between 0 and 5 day 8-h maximum O 3 and total mortality in individuals with previous AMI hospitalization (2.6% increase risk per 10 μg/m 3 95% CIs: − 1.0%, 6.4%). However, diabetes was the pre-existing condition that conferred the greatest susceptibility to ozone, results which were not confidently confirmed in our analyses. Another study from Italy [9] investigating a more specific outcome, reported a borderline significant association between O 3 exposure and out-ofhospital coronary deaths in individuals who had previous hospitalizations for ischemic heart disease (p-value 0.059) which lends some support to our finding that suggest susceptibility to CV mortality in individuals with previous hospitalization to AMI. In contrast to our results, a large multicity study from the US using secondary cause of death as a proxy for pre-existing disease reported no increased susceptibility between O 3 and mortality for atherosclerotic conditions, however additional susceptibility was observed for atrial fibrillation [3].
Our results suggest that association between O 3 exposure and CV death in individuals with AMI hospitalizations appeared to be driven mainly by AMI hospitalizations more than 1 month preceding death. Studies have reported the highest post AMI mortality within the first 28 days [14]. During this time patients are typically under intensive medication and the increased antithrombotic medication may potentially reduce the harmful effects of O 3 . We may also speculate that other risks are more dominant within the first month post-AMI compared to the effect of O 3 . In addition, individuals are plausibly less frequently outdoors during the first month after an AMI and they may be less exposed to outdoor O 3 concentrations. Hence, the increased risk observed in individuals with AMIs more than 1 month preceding death may reflect an increased exposure to O 3 or less exposure misclassification.
Out-of-hospital CV deaths with no previous hospitalization for any causes demonstrated an indication of association with O 3 exposure. It is likely that many of these individuals had an undiagnosed (and untreated) underlying condition that made them susceptible to O 3 exposure. Epidemiological studies as well as autopsy series reported presence of structural heart disease, including coronary artery disease, in 50-95% of adults who suffered sudden cardiac death without previously known heart disease [15,16]. In our study, CV death in individuals not hospitalized since 1987 were relatively younger (mean age 73 years) compared to the rest of the individuals. Younger individuals are more likely to be exposed to higher O 3 levels through outdoor physical activity. In addition, in a previous study in Stockholm based on a cardiac arrest registry, we reported associations between higher short-term exposure to O 3 and risk of out-of-hospital cardiac arrest [17], largely supporting the theory that the previously non-hospitalized deaths may reflect cardiac arrests in individuals with latent heart disease.
Ozone exposure has been consistently associated with respiratory mortality in several studies including large multi-city studies [1,4,11,[18][19][20]. Associations have generally been reported from exposure periods between 0 and 1 and 0-5 day lags and our estimates of 2.7% higher risk per 10 μg/m 3 7-day exposure are in line with previously reported estimates. In addition to studies investigating O 3 exposure and respiratory mortality, several studies have reported associations between O 3 exposure and hospital admission for COPD [21][22][23][24][25][26].
Few previous studies have investigated the role of previous hospitalizations for respiratory disease on association between O 3 and respiratory mortality. None of these have reported increased risk of death in individuals with previous respiratory hospitalizations [3,6,9]. In individuals with COPD hospitalizations, the timing of hospitalizations did not seem important in determining the susceptibility to O 3 in relation to respiratory mortality.
We observed an increased risk of CV mortality among elderly following short-term exposure to O 3 . Other studies have also reported elderly as susceptible to the detrimental effects of O 3 exposure [6,8,9,27]. We did not observe risk difference in men or women, while, a majority of studies have reported stronger associations between O 3 exposure and CV outcomes in women [3,6], although some inconsistencies remain [8] and studies investigating specific pathophysiological mechanisms are lacking [3]. Furthermore, we observed no evidence to support differing risk to O 3 -related risk of respiratory mortality by age and sex.
Ozone has a strong seasonal trend with high levels in warm period of the year and low level in cold period. In previous studies stronger associations have been demonstrated for the warm period of the year between O 3 exposure and CV and respiratory mortality [6,11,19,28], while, we did not observe seasonal influence on our risk estimates.
Inflammation and oxidative stress are integral components in the pathogenic process of respiratory [29] but also of atherosclerotic heart disease [30]. Ozone has a high oxidative potential and has in animal studies been shown to adversely react with airway epithelium inducing smooth muscle hyperplasia and dysfunction, and contribute to subsequent bronchial hyperresponsiveness [31]. An experimental study in healthy young adults reported increased inflammatory markers and decreased lung function measures following exposure to O 3 [32]. Other experimental studies have reported increase in levels of inflammatory markers such as pro-inflammatory cytokines and C-reactive protein after exposure to O 3 [33][34][35][36]. Ozone exposure has also been associated with alterations in the autonomic nervous system [37][38][39][40][41] and changes in blood pressure in individuals with [42] and without [43] previous cardiovascular disease. In addition to possibly contributing to the risk of developing atherosclerotic heart disease, O 3 exposure may be involved in the exacerbation of these disease processes in post-AMI patients.
This study has several limitations. Like most epidemiological studies, we did not have information on individualspecific O 3 exposure, instead we used ambient concentrations from a single monitoring station as exposure estimates for the entire population. Moreover, we did not have data regarding the time spent indoors, activity pattern, and location of residents, which could introduce potential misclassification; however, this type of misclassification is likely non-differential and would bias towards the null. Although our dataset included possibility for multiple hospitalizations for several diseases we created mutually exclusive subpopulations for comparison purposes and could not take into account the full complexity of presence of multiple diseases. Our study was based exclusively on fatalities occurring during the study period and therefore precluded studying associations of non-fatal disease incidence. The major strength of our study is the full coverage of the mortality and previous hospitalization registries. These registries have been shown to have a high validity [10]. Although some level of inaccuracy is still expected in the registry data, it is not expected to vary with our exposure and thus the net result would be an underestimation of the observed estimates.

Conclusion
Short-term O 3 exposure was associated with CV and respiratory mortality. We found suggestive evidence of increased of susceptibility for CV mortality following shortterm exposure to O 3 in individuals previously hospitalized for AMI and observed an indication of higher risks in individuals with out-of-hospital CV death as their first presentation of disease. These results may support the consideration of increased risk in large susceptible subpopulations in health impact assessments of O 3 exposure.

Additional file
Additional file 1: Table S1. Two pollutant models. Percent change in cardiovascular and respiratory mortality associated with a 10 μg/m 3 increase in 2-day and 7-day average air pollution concentrations, respectively in time series analyses. Table S2. Percent change in cardiovascular mortality associated with a 10 μg/m 3 increase in 2-day average 8-h maximum O 3 concentrations by season, age, and sex in case-crossover analyses. Table S3. Percent change in respiratory mortality associated with a 10 μg/m 3 increase in 7-day average 8-h maximum O 3 concentrations by season, age, and sex in case-crossover analyses. (DOCX 25 kb)