In this study of the association between PM events and dispensing of short- and long-acting asthma drugs, we found that PM events from traffic were associated with increased dispensing of short-acting asthma drugs (Table 3). Dispensing of long-acting asthma drugs, which accounts for one fourth of total asthma drug dispensing and most of the seasonal variation (Table 2) had a higher RR of association for PM events from volcanic sources in the adjusted models than dispensing of the short-acting drugs (RR 1.17 vs 3.13) (Table 3 and Table 4). This association was present in both single- and multi-source models with and without adjusting for other pollutants (correlation matrix for the pollutants in presented in the supplement Table S2).
We found that PM from volcanic eruptions were associated with increased asthma drug dispensing in adults in summer, and furthermore, the dispensing of long-acting drugs were increased following volcanic ash events, suggesting that the individuals affected by this type of pollution tend to be individuals with chronic disease. All volcanic ash events occurred during 2010, and save for one, within six months of the Eyjafjallajökull eruption 2010. Only one volcanic ash event occurred during winter, and thus, the results for that period suffers from very low statistical power (Figure 3 ab).
Perhaps users of long-acting drug are more sensitive to volcanic ash events, or, they are more likely to comply with official advice or increased awareness during the rather short period following the eruption. The results for winter, where there were also very few natural dust and volcanic events, differed somewhat from summer results. However, these differences could also be due to climate factors. Inspecting the effects of temperature from the adjusted model, it is somewhat different in warm and cold temperatures and is associated with reduction in dispensing following low temperatures, and increases after high temperatures (Figure S2).
Although Icelandic dust storm particles, smaller than 10 µm, have properties similar to volcanic ash (28), we found a negative association between dust storms and asthma drug dispensing. All surface particles in Iceland, which are inherently volcanic in origin, contain glassy particles, whereas freshly erupted volcanic particles contain up to 50% glassy particles, but only in particles larger than approximately 20µm is the morphology of Icelandic dust truly volcanic in nature (Butwin et al., 2020) and those particles will be beyond the respirable size fraction. Dust storms typically occur during dry and windy conditions, meaning that simultaneous build-up of gaseous traffic related air pollution and pollen within the study area is unlikely. This could possibly bias the estimated effects of dust towards a lower estimate. However, adjusting for other pollutants, weather and pollen in our models should have accounted for this, although some residual effects may remain. Most dust storms also occur during summer, where there are fewer infectious disease epidemics and asthma drug use is generally lower (Table 1). We can also speculate, that the effects of sand storms on dispensing are transient, as we see in Figure 2 that sand storm has positive (albeit not significant) associations with dispensing at lag 0. This observation is concurrent with the results of Novack and colleagues (12) who found associations between asthma drug dispensing and dust storms at lag 0-2. In our study, PM events from natural dust have lower PM10 concentration than during the volcanic PM events, and it is possible that the PM levels in the capital area are not sufficiently high to induce the adverse mechanical respiratory health effects which are postulated in previous studies (12).
Traffic and local resuspension
Resuspended particles from roads, with contribution from mechanically worn road surface, particles from tires and brake wear, and salt, are generally coarser than particles which are directly associated with exhaust from traffic (19, 20, 32). PM2.5 was available for only some parts of the study period, so we refrained from adjusting for this in our models. Finding mainly effects of PM10 from traffic has been reported in other studies of specific PM sources, where long-term exposure to particles from marine traffic or residential heating were not associated with respiratory health (33). However, when studying health effects of long-term exposure to air pollution, it is not unproblematic to disentangle resuspension and direct traffic exposure in dispersion models as these are highly correlated, giving extra importance to short-term studies with thorough source assignment.
In relation to literature
Dispensing of asthma drug is an indicator of respiratory health in a population (23). Only few studies of asthma drug in association with natural dust exist; (10) reported that in a cohort of children with asthma, drug purchases were increased after desert dust storms, peaking at lag of 3 days. In pre-school age children, asthma emergency department visits nearly doubled (OR 1.8, 95% CI 1.2-2.8) after Asian dust events in Nagasaki, Japan (7). Emergency room visits for respiratory causes increased by 14% after dust events, most for elderly and adult men in Lanzhou, China (8). Admissions for COPD increased during dust storm days in vicinity of Middle Eastern Negev desert (9). Recently, more specific studies have found that the effects on lung function was specific to the source of the particles, so Asian dust in concert with combustion particles had a detrimental effect (11) and furthermore, there are speculations that the detrimental effects of PM from natural sources are due to biological material carried with desert dust particles (1).
For other respiratory outcomes, a number of studies have found increases in association with PM from natural sources or dust storms. In East Asia, dust episodes are associated with respiratory health outcomes, but researchers have speculated that Asian dust modified the effect of PM2.5 to make it less harmful to health (34). Also, researchers from Asia indicate that PM events should be considered as outliers in analyses of PM health effects (35); (34). In our analyses, dust storms were also associated with decreased RR, which could be reflective of a similar phenomenon. In a Southern European meta-analysis it was estimated that mortality increased by 0.51 % per 10 3 desert PM10 and observed similar estimates for cardiorespiratory causes (5) and mortality (2).
Effect estimates for volcanic ash were positive in either model, but did not reach statistical significance, probably due to the low number of events. Estimated risks were higher in the adjusted models for some sources, indicating that accounting for the contribution from other pollutants strengthened the results. However, the confidence intervals overlapped. In the graphical representation of the lag-association (Fig. 2), the CI bands indicate that the statistical power for traffic and resuspension was superior, as these events were the most common.
In a sensitivity analysis excluding the year 2010, the effect estimates for PM events from traffic remained statistically significant, but other estimates were not significant (data not shown).
NO2 was mainly associated with traffic pollution, and in all models, there was a positive association between NO2 and dispensing. For SO2, there was a special case in fall-winter 2014 when there was substantial SO2 contribution from the volcanic eruption in Holuhraun (36), but none of the PM events occurred during days where the volcanic plume was over the capital area. However, we performed sensitivity analysis of the data without the days with volcanic plume as well as excluding the whole eruption period, and the results were only marginally altered (data not shown). Thus, we are confident in the robustness of our results.
In the main analysis (Table 3 and Table 4), we did not adjust for PM10 to avoid bias, as our exposure metric, the pre-defined PM events, are inherently strongly associated with PM10, however, we report results adjusted for PM10 in the supplement (Table S3). In these models, the estimates of PM10 then corresponds to the effect of PM10 under 50 µg/m3.
There was a lot of missing data in the time series for O3 and H2S, also during PM event days, so full results cannot be reported with adjustment for these pollutants. After adjusting for O3 and H2S, the associations with traffic PM events did not reach statistical significance and there were no volcanic PM events left in the time series, so results cannot be reported for those (Supplementary Table S4).
In the analysis, we used the generalized non-linear model which were adjusted for time trends using standard methods to estimate the effects of PM events, but it can be argued that a case-crossover design could be more reasonable as adjustment for season and time trend are inherent in the model specifications. However, tests of these models revealed that they performed poorly, perhaps as the PM events were both rather rare, and, in the case of volcanic events, occurred within the span of a few months.
Limitations and strengths
As with all pharmaco-epidemiological studies, there is some uncertainty associated with the causal chain from the onset of the biological effect (asthma exacerbation due to PM exposure) and the measured outcome (individuals filling prescription in a pharmacy). This chain is modified by availability (perhaps the individual has recently filled his prescription the week prior to the event, perhaps the event occurs during the weekend where pharmacies have limited opening hours. These factors mean that the uncertainty of the observed association must be treated with some caution, even more so than in a study with a direct clinically measured outcome.
A strength of this study is the long time series of data and the presence of many environmental covariates. Despite the extent of the study period, there were only 9 days with contribution from volcanic eruptions, all from the Eyjafjallajökull eruption. The data has been collected prospectively and subjects would in must studies be unaware of their exposure status. It is a limitation that there is no information about individual-level covariates of interest such as smoking status and co-morbidities, as these might be of interest for subset analysis, to further investigate groups with respiratory susceptibilities.
During the exposure period, there was also the Holuhraun volcanic eruption 2014-2015, but the days with high exposure (either SO2 over the health limit, or mature volcanic plume (37) from that eruption did not coincide with any of the specific source PM events in the current study. The capital area is scarcely the most affected area, but studies of more affected areas are hampered by a severe lack of measurements (data not shown).