**3. Results**

#### *3.1. Exposure Metrics Summary*

The highest annual average exposure concentration was approximately 13.1 μg/m<sup>3</sup> for Tier model 1. Tier model 2 and Tier model 3 indicated that the exposure that occurred indoors was much lower than outdoors due to the infiltration rates of the buildings, resulting in annual average exposure concentrations of 7.18 μg/m<sup>3</sup> and 7.26 μg/m3, respectively. There was an approximately 45% reduction between Tier 1 and Tier 3. This result clearly suggests that spending long periods of time indoors, reduces the exposure to outdoor-generated air pollution. The incorporation of transportation and predominately the highly polluted London Underground in Tier model 4 resulted in an elevated exposure concentration (8.28 μg/m3), pinpointing that even though the time spent in transit is only 2.9%, this microenvironment has a significant contribution to the total exposure. By dividing the

London Underground into subsurface with HVAC and deep line without HVAC, we were able to quantify the impact of the most highly polluted ME on the total exposures (the deep-line underground). Tier 5 showed an approximately 0.30 μg/m<sup>3</sup> higher exposure concentration (8.60 μg/m3) than Tier 4, where an average concentration for the whole underground was used (Table 4).


**Table 4.** Annual exposure calculated in each model stage.

PM2.5 exposure concentration maps for each Tier-model stage were created by BenMap-CE showing how the exposure was distributed across GLA. Figure 3a,b illustrate the spatial distribution of the annual exposures in Tier 1 and 5. The maps of Tiers 2, 3 and 4 are included in the Supplementary Information (Figure A1a–c in the Appendix A.1). The highest exposure concentrations occurred in Inner London for both Tier 1 and Tier 5 (15.4 μg/m<sup>3</sup> and 10.1 μg/m3, respectively), whereas the lowest exposures were observed in Western GLA (less than 10.9 and less than 7.10 μg/m<sup>3</sup> for Tier 1 and 5, respectively). The incorporation of indoor infiltration along with time-activity data led to an overall mitigation of the exposure concentrations in GLA when Tiers 2, 3, 4 and 5 were used. After the utilization of our most complex model, Tier 5 had the highest difference observed at the centre with approximately 37% (Figure 3c), while average reduction in GLA was approximately 34%. Inner London continued to show the highest values (Figure 3b), although the infiltration factors in Inner London were lower than in the outskirts. This could be due to the much higher outdoor concentrations in Inner GLA than in the Outer. In Inner London, the higher number of sources of anthropogenic and traffic-related pollutants, including PM2.5, generate significantly higher ambient pollution levels. Several studies sugges<sup>t</sup> that traffic pollutants are elevated above background concentrations around major roads and highways [13,51]. The percentage of exposure concentration reduction in Tiers 2, 3 and 4 after comparison with our baseline exposure concentration (Tier 1), is illustrated in Figure A2a–c in the Appendix A.2. Apart from proximity to roads, fewer green spaces and the densely constructed city center may also contribute to the higher levels of outdoor particulate pollution [52–54]. Urban populations are subject to daily activity patterns, so that exposure is not a static phenomenon but should be quantified as a function of concentration and time [4]. Therefore, by assigning people's exposure to a single location (e.g., at their residence) and ignoring highly polluted MEs such as the subway, it is unlikely to accurately represent total exposure. Hence, by gradually incorporating time-activity data and indoor MEs, the spatial variability of the exposure concentration across GLA increased. Since we used annual average time-activity data for the entire GLA, time-activity could not change the spatial pattern of the exposure. In our case, the spatial variability of the housing stock and I/O ratios across GLA were the main reasons for any increase in the spatial variability of the exposure concentration.

Figure 4 presents the contribution of each examined microenvironment to the total exposure estimated by Tier-model 5. Indoor exposure concentration is clearly the dominating contributor (approximately 83%) to the total exposure (due to the time that people spent there–95.7%) followed by the deep-line underground ME (14%) albeit people spent on average only 0.31% of their annual time. According to our measurements, the PM2.5 concentration in deep underground lines was around 28 times higher than the outdoor levels, which rationalized the high contribution of that ME to total exposure. In contrast, London population spent only 1.4% of its annual time outside and the outdoor ME contributed only 2% to the total exposure concentration. The findings described above indicate that outdoor PM2.5 levels are unlikely to accurately represent the total exposure of an urban population like in London.

**Figure 3.** Maps of annual distributions across the Greater London Area (GLA): (**a**) Tier-model 1 annual mean PM2.5 exposure concentration (μg/m3), (**b**) Tier-model 5 annual mean PM2.5 exposure concentration (μg/m3), and (**c**) percentage of the PM2.5 exposure concentration difference between Tier-model 1 and Tier-model 5.

**Figure 4.** Contribution of each microenvironment (ME) to the total exposure. Indoor exposure shows the greater contribution followed by the deep underground lines.

#### *3.2. Epidemiological Implications and Health Impact Misclassification*

Because in epidemiology the concentration from central-site monitors is used as a proxy for the exposure to air pollution, we selected Tier 1 as our reference and compared it with the estimates of Tiers 2, 3, 4 and 5. The mean change in the estimates of all-cause mortality when applying Tier model 2 was predicted to be 1541 (95% CI: (427–2633)) deaths, while when using Tier 3 exposure concentration estimates the death cases were reduced to 1521 (95% CI: (421– 2598)). The impact on mortality when applying the 4th Tier model, which included the transportation microenvironments (tMEs), was estimated to be 1257 (95% CI: (347–2151)) cases. Due to the significance of the deep-line underground, the most complex Tier model 5 presented the lowest number of cases compared with the other 3 metrics (Tiers 2, 3 and 4). Namely, once Tier 5 was applied the prediction for the estimated avoided mortalities were 1174 (95% CI: (324 – 2010)). We can assume that the calculated change in mortality represents the potential health burden misclassification that might occur when changing the exposure metrics to assess the population exposure. Subsequently, we were able to estimate the percentage decrease in predicted mortalities when altering the exposure metric's complexity. The substantial changes in avoided mortality predictions indicate that using a static exposure approach in a study might lead to significant uncertainty in a health burden assessment. As anticipated, the predicted mortality was significantly reduced when increasing the model complexity. The highest changes were observed in Tier-model 2 and 3, due to the time that people spent indoors in urban areas, the big difference between outdoor and indoor exposure and the absence of highly polluted transportation MEs, pinpointing the importance of taking into serious consideration the exposure that occurs inside buildings when estimating health effects. The model predicted most avoided cases when Tier 2 was applied and while increasing complexity the cases showed a decrease of 1.95%, 18.4% and 23.8% for Tier 3, 4 and 5, respectively. As explained above, the London Underground contributes significantly to the total average exposure concentration of the study population by increasing the estimates. Therefore, we can securely presume that this is the main reason for the high decrease in avoided mortalities when Tier 4 and, predominantly, Tier 5 were used in BenMap-CE.

All results are summarized in Table 5.


**Table 5.** Change in the annual mean estimates of mortality (predicted avoided mortalities) between the di fferent exposure metrics and decrease between the estimated change in mortality predictions.

Looking at the spatial distribution of the predicted change in mortalities shown in Figure 5 (Tier 1–Tier 5) we can notice that the biggest change in mortality occurred in central GLA. Several factors could explain this result such as the outdoor PM2.5 concentration, the housing stock (I/O ratios) and the population. As described above, after the inclusion of the time-activity data there was an overall reduction in exposure concentration because people usually spend most of their time (>95%) in indoor MEs (excluding transportation), where the concentration of outdoor PM2.5 is lower than the measured ambient levels. Because the health impact function used by BenMap-CE is a concentration response function, the amount of the reduced exposure concentration determines the fraction of the mortality reduction. In our case, knowing that moving from Tier 1 to Tier 5 would result in a greater reduction of exposure concentration that appeared in central GLA (Figure 3c), we could presume that the high outdoor PM2.5 concentration and the building type of that area, were largely responsible for the mortality change. The similar distribution patterns between Figures 3c and 5 also supported this argument. As already shown in Figure 2, the infiltration factors of the buildings there were lower than the rest of the GLA, leading to higher mitigation of the exposure concentration. In the Appendix A.3, Figure A3a–c show the spatial distribution of the predicted change in mortality between Tier 1 and Tiers 2, 3 and 4, respectively.

Overall, these outcomes demonstrate the importance of the complexity of an exposure metric when incorporated into an epidemiological study. Here, we proved that indoor MEs such as the home and the subway are governing human exposure to air pollution and any possible absence in a metric is likely to cause considerable misclassification of the magnitude of mortality.
