In our study, 66,162 hospital admissions for diseases of the circulatory system were recorded during 110 cycles of 10 days each (Table 2). The mean concentration of air pollutants per cycle was 53.7, 44.0, 31.1, 29.9, 20.0, 1.7, and 69.0 μg/m3 for CO, NO2, PM10, NO, PM2.5, SO2, and O3, respectively. The average air temperature was 11.9 °C. The NAO index ranges between 1.6 and 1.5, with an average of 0.13.
North Atlantic Oscillation and hospital admissions
The association between the NAO index and hospital admissions through diseases of the circulatory system was not statistically significant (coeff.: −0.80, std. error: 6.18, p value: 0.89).
Thermal comfort and hospital morbidity
Four sets of thermal indexes were calculated: WC uses temperature and wind, DI and ET are built using temperature and relative humidity, and ETw adds wind. To identify the index that best expresses vulnerability to thermal conditions, linear models associating hospital admissions and the comfort index were built (Table 3).
The comfort indexes are significantly associated with hospital morbidity, and the coefficients tend to be higher when the discomfort indexes do not include wind. Although only winter months are being analyzed, thermal comfort indexes built with average temperature are strongly associated with hospital morbidity due to diseases of the circulatory system than those indexes using minimum temperature.
Considering the results, the DI was selected as the comfort index that best characterizes vulnerability to thermal conditions (being also the thermal comfort index with lower p value in the models).
Air quality and hospital admissions
The relationship between the selected air pollutants and hospital admissions was analyzed. Excepting O3, all pollutants are positively associated with hospital admissions for diseases of the circulatory system and only NO2 is not significantly associated (Table 4).
North Atlantic Oscillation, thermal comfort, air quality, and hospital admissions
No significant association was found between the NAO index and DI (coeff.: 0.28, std. error: 0.18, p value: 0.12) (the comfort indexes that included wind are significantly associated with NAO in their average and maximum component, but not in the minimum). As mentioned before, the DI combines temperature and relative humidity, and none of them is significantly associated with the NAO index. In fact, both high and low NAO indexes can be related to different synoptic situations that are linked with similar temperature and relative humidity but completely different precipitation levels (Table 5).
NAO index is significantly associated with CO, PM2.5, NO, NO2, and PM10 (Table 6). Similarly, the air pollutant level significantly decreased while the wind speed and precipitation increased, except O3 which has a different behavior, associated with fair weather, no wind, and strong sunlight. NAO is also significantly associated with precipitation (coeff.: −17.7, std. error, 3.69, p value: 0.00) and wind speed (coeff.: −0.31, std. error: 0.09, p value: 0.00) (Fig. 2).
As previously mentioned, no significant linear association between the NAO index and morbidity was found. The absence of direct association was also supported by the results of the causal mediation analyses: the average direct effect and total effect of NAO index in hospital admissions are not significant. However, NAO influences air quality and, through this indirect effect, can also affect human health (Fig. 2). The estimated average causal mediation effect is significantly different from zero in some pollutants (CO, PM2.5, NO, and SO2). The results suggest that the NAO is associated with CO, PM2.5, NO, and SO2 levels, which in turn increase the probability of hospitalization due to diseases of the circulatory system. The estimated average increase in hospitalizations resulting from the indirect effect of NAO is 6.84 (p value <0.01) through CO, 5.86 (p value <0.01) through PM2.5, 4.92 (p value <0.01) through NO, and 5.27 through SO2 (Table 7).
The results from sensitivity analyses supported the robustness of our findings, by checking ACME estimation for several referential NAO index values; it was possible to conclude that ACME increases as NAO index increases, and there is no change on the statistical significance of ACME and ADE.