Although SO2 pollution levels were overall higher after the Holuhraun eruption began (Table 1), a crude comparison of respiratory health care utilisation in Iceland’s capital area before and after the beginning of the Holuhraun eruption (Table 2) showed that only AMD increased significantly compared to the corresponding months in previous years, as well as compared to the total control period. . However, in time series regression analysis (Table 3), SO2 concentrations were associated with increased AMD, PCMD and HED for respiratory causes at lag 0-2. Analysing the association with SO2 levels which exceeded the 24-hour health limit (125 µg/m3) to quantify the risk associated with high levels yielded similar results, and spline plots suggest a linear dose-response curve (Table 3, see also Figure S3). Stratifying AMD into subgroups, only dispensing of short-acting medications were significantly increased, indicating that dispensing of short-acting drugs are a more sensitive indicator of an immediate need for symptom relief, although the use of short-acting drugs are not specific to asthma disease (Table 4). In PCMD we found that PCMD for respiratory infections and obstructive lung disease were increased and the increase in obstructive lung disease suggested that individuals with this diagnosis were increasingly in need of care following SO2 exposure (Table 4). For COPD diagnoses in HED, our results indicated a positive association, but it was lower than for PCMD, and the results did not reach statistical significance.
Age-specific
Stratifying by age-categories, the results indicated that for AMD, we observed with highest risk estimates with SO2 exposure in individuals under 18, for PCMD, adults had highest risk, and for HED the highest risk was in individuals 65 and older (Table 5), except for lag2-4, where it was higher in individuals under 18 (Table S3). However, the confidence intervals overlapped, and the association between SO2 and HED in elderly was not significant after adjusting for other pollutants than SO2 (Table 5).
Outcomes
Exploring definitions of outcomes in primary care in sensitivity analyses, the tendency towards increasing effect estimates in elderly in analyses of all contacts and recurring contacts (Table S3) rather than non-recurring MD visits suggests that the care burden is higher in this group, but the confidence intervals overlapped with those in the main analysis (Table 5). In HED admissions, a high effect estimate was seen for both continuous SO2 and high SO2 days in children at lag2-4, but the confidence intervals were wide (Table S3).
Age-and lag specific
In the regression analysis, the choice of lags were based on inspection of lag structures generated by lag-splines (Figure S2). For AMD and PCMD visits, the observed effects of SO2 exposure occurred during the same day and up to two days after a peak in SO2 concentration (Figures S2a-b). However, for HED visits in elderly, the observed effects occurred already at lag 0 (Figure S2c-d). A possible explanation for this is that primary care is the first point of contact for most individuals and hospital care is sought only after other care options have been exhausted, but for elderly, hospital care is sought immediately, possibly due to a more severe presentation of symptoms or complications due to underlying diseases, as illustrated by a significant association between SO2 and HED in elderly before adjusting for other pollutants at lag0-2 (Table S3), and significant association only with HED in children at lag 2-4 (Table S3). Unfortunately, we did not have access to information about comorbidities and were unable to adjust for this.
Compliance, other eruptions, longer lags
In the sensitivity analysis to investigate the influence of compliance with official advice, the effect of initial warnings of volcanic gas exposure appeared to be limited, as excluding the first day or week of high SO2 yielded similar estimates for AMD and PCC visits as the main analysis, although there was some loss of statistical power (Table S4). We speculate that since the increases in AMD and PCC persisted after the first exposed day and weeks, this indicates that the observed increases in health care utilisation reflect actual increased respiratory morbidities rather than merely adherence to official advice (Supplemental Table S4). Excluding the years 2010 and 2011 from the analysis to exclude effects of volcanic ash from Eyjafjallajökull and Grímsvötn yielded results within confidence intervals of our main analysis (Table S4). Investigating longer lags, we observed statistically significant increases in total PCMD and HED visits in young and adults at more than 15 days delay, but in both cases, these follow periods of decreases in outcomes, and are challenging to interpret (Figure S5).
Relate to literature
Although it has been suspected that the complex mixture of a volcanic plume (7) could have different health effects than merely SO2, we found that our results were consistent with concentration-response functions found in studies from urban settings; e.g. respiratory mortality rates were estimated to increase by 2.4% per 27 µg/m3 SO2 (Li et al. 2017), corresponding to a point estimate of 0.88% per 10 µg/m3 SO2, lower, but within the confidence interval of the estimated increases of HED found in our study, namely 1.02% (95% CI 0.02 – 2.03%) per 10 µg/m3. In previous studies of SO2 exposure during volcanic eruptions, the SO2 concentration on Miyakejima Island increased after Mount Oyama erupted in 2000 meaning that residents and aid workers returning to the island from 2005 and onwards were exposed. Children living on the island who were exposed to daily mean concentration of 125 µg/m3 SO2, had increased rates of wheezing.(36) In follow-up studies, permanent residents of Miyakejima Island (n=168) who lived in areas with high SO2 exposure reported increased rates of cough and wheeze(13), both symptoms of asthma. None of the studies of people exposed at Miyakejima Island found adverse effects on lung function.(12, 14, 36)
Regarding individual susceptibility, we observed the highest effect estimates for PCMD visits for asthma and COPD (Table 4), indicating that individuals with these diseases are at increased risk, which is. Hyper-responsiveness to SO2 has previously been reported as common (20-25%) in individuals with positive asthma test (methacholine test), indicating that they may be particularly vulnerable to severe SO2 exposure.(4) Non-smokers have previously been reported as being at risk of symptoms after volcanic SO2 exposure,(11) but unfortunately, smoking status was not available in our data.
Respiratory infections were increased in our study and similarly rates of cough and acute pharyngitis diagnosed at a clinic in SO2 –exposed communities near The Kilauea Volcano in Hawaii increased after a eruption activity increased in 2008.(18) Respiratory infection diagnoses in PCMD were statistically significantly associated with SO2, a similar association has previously been reported for hospital visits for upper respiratory tract infections.(37) Volcanic eruptions have been associated with altered rates of other health outcomes such as accidents and mortality,(38) but these fell outside the scope of this study.
Methodology
Our study employs a time series design were individual level risk factors are not time-dependent and thus should not confound the association between short term exposure to air SO2 and health outcomes, leaving bias due to unmeasured confounders, seasonal variation, and other intermittent exposures as main concerns. Previous studies of health effects of air pollution in Iceland have yielded lower effect estimates of daily air pollution on morbidity than the current study,(39, 40) other air pollution exposure types are thus not likely to bias the results. This includes H2S, which has a low correlation (<0.1) with the exposure of interest during the study period (data not shown). During the reference period, there were two ash-rich volcanic eruptions, in Eyjafjallajökull 2010 and Grímsvötn 2011. While ash from Eyjafjallajökull had local respiratory health effects(24) and there may have been adverse health effects in the capital area,(40) neither eruption had significant SO2 emissions and sensitivity analysis excluding this period yielded results within the confidence interval of the main study results (Supplemental Table S5). The exposed period October and November of 2014 did not coincide with any viral respiratory illnesses (e.g. influenza and RS-virus) epidemic,( 41, 42) but it did coincide with the MD labour conflict which resulted in lower PCMD and HED attendance during those days. Hence, results from the analysis comparing the eruption period with the time before may have underestimated of true effect of the SO2 from the eruption.
Strengths and limitations
SO2, PM10 and NO2 data was missing for a number of days which were excluded from the analysis. As the volcanic plume effectively changed the chemical composition of the atmosphere during the eruption period (Supplemental Tables S1a-b), the correlations of SO2 with other air pollutants were altered after the eruption. Results that were adjusted for other pollutants were nearly identical to the main analysis (Table 3), indicating that the effect of the very high concentrations of SO2 was unambiguous.
It is a strength of the current study that health data were collected prospectively from population-wide registers, which minimizes the risk of information bias from individuals knowing their exposure status. Although the exposure would have been known to the public for at least part of the exposed period but we attempt to address this source of bias in the unadjusted analysis and found only moderate changes to the results (Table S3). As a study outcome, we use dispensing of asthma-medication, a novel and more sensitive proxy for respiratory health in a population (28) than primary care attendance and hospital visits. It targets individuals who are already sensitive to poor air quality (26, 29) who in most cases have contact with the health care system. A limitation with register-data is that the data are not collected for research purposes and diagnoses given in the health care system could be biased and lead to overestimation as medical professionals assume respiratory outcomes to be more likely during the eruption. However, as the estimated effect are similar across all respiratory outcomes we conclude that this source of bias is not likely to explain our findings. By defining exposure status based on residential postcode we make several assumptions, firstly, that the whole area is equally exposed, and secondly, that the study participants are physically near their homes. However, these sources of bias would result in wider confidence intervals or bias the results toward the null. Reykjavík and the Icelandic capital area was exposed to SO2 concentrations above 125µg/m3 during a total of ten days which occurred mostly during October and November 2014. This limits the statistical power of the study and our options for further analysis, as does the fact that our study period did not extend until after the eruption, meaning that we cannot fully assess the complete impact of the health impact from the eruption on the population from our results.