**5. Conclusions**

The use of ambient centralized monitoring concentrations as a surrogate for people's exposure may not provide an accurate representation in a population study. In this study we developed a static exposure approach, commonly used in epidemiology, as our baseline metric and by incrementally enhancing the metric we were able to report the potential impact that the application of di fferent metrics would have on a health outcomes assessment. We demonstrated that studies focusing on centralized monitoring ambient concentrations may show reduced ability to detect the true associations between exposure to PM2.5 and health e ffects due to inadequate spatial variability of the concentration and the absence of people's mobility. The magnitude of the misclassification related to the inclusion of indoor MEs and the metric's complexity was large relative to the dynamic nature of human exposure to air pollution.

This analysis illustrates the significance of allowing for population activity and indoor infiltration. The indoor ME showed the highest contribution to the total population exposure (i.e., 83%), while the LU contributed approximately 15%, although people spend only 0.4% of their time there. Consequently, all our models showed lower total exposures than the traditional exposure approach that assumes that the PM2.5 concentrations outside the residence are representative of the total population exposure. Particularly, our most complex and accurate Tier-model estimated an approximately 34% lower mean exposure concentration compared with using simply an outdoor concentration.

The exposure misclassification due to home infiltration and underground ME is likely important in assessing the health burden in an urban area because people in cities spend the majority of their time inside the residence or workplace and the pollution concentrations that occur underground are remarkably high. The misclassification between the traditional exposure approach to estimate health outcomes and our most dynamic metric was found to be 1174 mean predicted mortalities in GLA, with the highest numbers observed in Inner London.

Overall, by quantifying the health burden misclassification we managed to pinpoint the importance of developing a metric that can adequately represent the study population concerned and showed that the use of more dynamic data in epidemiology could significantly increase the accuracy of health impact assessments.

**Author Contributions:** Z.L.: Conceptualization; V.K. and Z.L. Methodology; V.K.: Data analysis; Z.L. and I.E.: Supervision; V.K.: Visualization; V.K.: Original draft preparation; V.K., Z.L. and I.E.: Review and editing. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research received no external funding.

**Conflicts of Interest:** The authors declare no conflict of interest.
