High-throughput geocoding to assess association of short-term air pollution with mortality in a French cancer patient cohort

A retrospective cohort study of 44,268 consecutive cancer patients (2017-2020) in France assessed the relationship of short-term air pollution with mortality. Using a geomatic framework at fine spatial granularity, air quality data (PM_2.5, PM₁₀, NO₂, O₃, temperature) were linked to each patient. Principal component analysis identified exposute groups, and a random forest algorithm was applied to predict mortality. Overall, 9% of patients died during follow-up. Four clusters of patients with distinct air quality profiles were identified. Clusters with the highest particulate matter levels (PM_2.5, PM₁₀, NO₂) showed increased mortality (12%-13%), whereas the cluster with the lowest pollution showed reduced mortality (8%) (chi-square test p<0.001). The prediction algorithm achieved a recall of 80%. The main predictors of death included higher temperature and elevated PM_2.5 within the previous 31 days, as well as older age, male sex, and thoracic cancer. The authors conclude that short-term exposure to lower air quality, captured at an infra-comunal scale, is associated with excess mortality among cancer patients. The identification of a 31-day lag window for predictive algorithn highlights opportunity for targeted prevention.