The fire that occurred on September 26, 2019, at the Lubrizol and NL Logistique sites caused significant environmental impacts. As part of the "Post-Fire Health 76" study, a health survey was conducted by Santé publique France to assess the type and frequency of the population’s exposure to various health impacts resulting from the fire, as well as the symptoms and health issues experienced during and after the event. Within the exposed area (Rouen and its surroundings) and a non-exposed area (Le Havre and its surroundings), a sample of 10,777 people was randomly selected by the INSEE Survey Division. Based on this survey, it was estimated that 92% of the adult population studied experienced at least one of the nuisances or forms of pollution generated by the accident, most commonly odors. Two-thirds of the population reported at least one symptom they attributed to the accident: stress, anxiety, distress, eye and throat irritation, coughing, shortness of breath, headaches, sleep disturbances, etc. One year after the fire, the perception of the nuisances and pollution caused by the fire had a negative effect on the health-related quality of life of those exposed. These initial analyses were conducted using data weighted by survey weights adjusted for total non-response, utilizing INSEE’s Fidéli survey frame (Demographic File of Tax Origin on Housing and Individuals). This database provided variables for both respondents and non-respondents, particularly the distance from residence to the fire and socioeconomic data. Non-response-adjusted weights were used to produce the study’s first report. However, the existence of endogenous selection bias was suspected, particularly if those most affected by the fire’s consequences were more likely to respond than others. The lack of health variables for non-respondents made it impossible to verify the hypothesis that survey participation was independent of health status. Therefore, in parallel with the survey, a request for access to data from the National Health Data System (SNDS) for individuals randomly selected for the survey was made to perform a matching procedure. The objective was to conduct a sensitivity analysis of non-response treatment by constructing indicators of healthcare utilization reflecting health status, available for both respondents and non-respondents. Thus, a significant association was identified between certain health indicators and survey participation, particularly in exposed areas. Individuals who had consulted healthcare professionals in the 12 months preceding the survey were more likely to participate in the survey than others. Conversely, individuals whose indicators suggested poor health or economic vulnerability were less likely to participate. A significant association was identified between indicators derived from the SNDS (doctor visits, long-term conditions, hospitalization, etc.) and certain key variables measured by the survey, particularly the physical health score, the mental health score, as well as post-fire physical and mental health symptoms in exposed areas. The identification of these simultaneous associations between SNDS indicators, survey participation, and variables of interest illustrates the existence of endogenous selection that is partly differential between exposed and unexposed areas. Santé publique France / Contribution of the SNDS to the study of non-response to a health survey: "Post-Fire Health 76: a study focused on your health." However, after including these health indicators in the statistical models used to adjust for survey non-response, the estimates remain highly stable. Minor changes in these estimates were observed, consistent with the selection bias previously identified, but these changes remain limited in magnitude. The impact on the health of those closest to the fire site would thus actually be slightly lower than what was shown in the previous estimates produced without using the SNDS. In conclusion, the correction for non-response is improved when variables from the SNDS are used in the statistical model. However, the difference in the resulting estimates is very small, calling into question the relevance of the effort required to implement and process these matched data.