Reply to Zhuang et al. and to Rui and Chen

From the Authors: We thank Zhuang and colleagues and Rui and Chen for their endorsement and constructive suggestions on our paper titled “Differential Effects of Wildfire Smoke PM2.5 Exposure on Respiratory Disease Emergency Department Visits in the Western United States.”1 Specifically, we appreciate their discussions on the complex mixture of wildfire smoke, effect modification, infection-mediated pathways of wildfire smoke PM2.5 on respiratory health, the value of negative control checks to enhance causality, and the advocacy of a wildfire smoke PM2.5 threshold for policy implications. Our exposure assessment was designed specifically to estimate smoke and nonsmoke PM2.5, and its primary predictors are total and nonsmoke PM2.5 simulated by US Environmental Protection Agency’s Community Multiscale Air Quality model.2 While gaseous air pollutants and volatile organic compounds3 in fire smoke may display similar spatial patterns, our exposure modeling framework enables us to focus on PM2.5 in fire smoke specifically rather than treating fire smoke as a mixture of PM2.5 with other air pollutants. Our case-crossover design eliminates time-invariant confounding by using each case as their own control. Consequently, genetics and socioeconomic status are unlikely to affect our findings. While interannual variability of wildfires may affect the annual disease burden, it is unlikely to affect our epidemiological results based on acute smoke PM2.5 exposure. Nonetheless, we agree with Rui and Chen that the true disease burden due to acute exposure to smoke PM2.5 may be underestimated as our health data only capture severe illnesses warranting a visit to the emergency department. The addition of communicable disease surveillance information proposed by Zhuang et al. helps understand infection-mediated mechanisms. As we discussed in our study, the ­composition of wildfire smoke PM2.5 can dampen antiviral defense and alter airway inflammation.4 Beyond toxicology, behavior change during wildfire events is another possible explanation for effects of smoke PM2.5 on communicable respiratory disease. With the presence of visible smoke, people tend to stay indoors and reduce air ventilation. This shift can increase close contact and the risk of disease transmission. Future studies are warranted to pair daily smoke exposure with virology surveillance. As Zhuang et al. mentioned, there is evidence of a positive association between smoke season PM2.5 and influenza in the following months.5 While our studies with a case crossover design focus on short-term exposure within a week, future work should consider a broader temporal relationship not only through lag structures but also by examining seasonal patterns. We agree that negative exposure controls can strengthen causal inference in time-series and case-crossover analyses. Many previous studies of ambient PM2.5 have used leads of future exposure that cannot cause a current event and expect estimates near the null at those leads.6,7 Future studies on wildfire smoke can use this technique to strengthen causality. However, negative outcome controls are harder to use in our setting because wildfire episodes can disrupt access to care and disrupt care seeking. Outcomes that appear unrelated may not share the same confounding ­structure as respiratory endpoints. We therefore favor future ­studies that prespecify negative exposure controls alongside ­primary estimates. Regarding setting a threshold concentration for wildfire smoke PM2.5 for health systems and public agencies, we support this idea but have concerns about feasibility. There is evidence that PM2.5 in wildfire smoke can be more harmful than PM2.5 from other sources, highlighting the need for early detection.8-10 However, it is difficult to measure smoke-specific concentration with ground monitors because source separation is not directly observable. This is why we designed a carefully constructed modeling system to hindcast smoke-specific PM2.5 concentration using a fusion of multiple data products and model simulations. Current forecast and nowcast ­systems provide valuable situational awareness, yet multimodel evaluations show substantial discrepancies and periods of underprediction of surface concentrations during major fire events.11,12 Building an operation-ready system may require extensive resources and advances in data assimilation, plume dynamics, and automated validation and model updates. While such a system is desirable for the long run, we urge public interventions during wildfire events at this moment, since our analysis showed no threshold for respiratory effects, and we aim to follow best public health practice. In addition, we call for future studies to improve wildfire smoke PM2.5 forecasting and to evaluate thresholds for intervention. Supplementary material is available at American Journal of Respiratory and Critical Care Medicine online. Please see the ICMJE disclosure forms, which have been provided as supplementary material.