Medical Consequences After a Fume Event in Commercial Airline Crews
INTRODUCTION: Many questions are still being asked about the actual health effects of exposure to a fume event for airline crewmembers. To shed new light on this controversy about so-called aerotoxic syndrome, we undertook a large-scale epidemiological study.
METHODS: We present a retrospective cohort study involving 14,953 crewmembers, including 2577 exposed to a fume event and 12,376 matched controls to estimate the hazard ratio of a subsequent sickness.
RESULTS: Prevalence of diseases that could be related to the fume event based on “possible” or “probable” level and date of occurrence after the fume event was for exposed (controls): neurological 2.9% (2.9%), psychiatric 2.5% (2.4%), vegetative 1.8% (1.5%), irritative 5.1% (4.5%), and functional 2.8% (3.2%). Differences were not significant. Incidences of having any related disease are estimated at 1552 per 100,000 person-years for exposed and 1497 per 100,000 person-years for controls, with a nonsignificant hazard ratio of 1.04 (0.86–1.25) in the Cox model. A subset of 2577 matched pairs exposed/control allowing specific statistical tests for paired data confirmed the lack of difference between exposed and controls: matched-pair risk ratio for any fume event related disease was 1.07 (0.85–1.34).
DISCUSSION: Our results clearly show that fume events are not associated with significant clinical consequences for cabin and cockpit crew. This work does not support the proposal of an “aerotoxic syndrome” in association with exposure to fume events.
Klerlein M, Dubiez L. Medical consequences after a fume event in commercial airline crews. Aerosp Med Hum Perform. 2025; 96(1):12–17.
Health effects of exposure to fume events are a new concern in the aeronautical field.1–3 The first questions were publicly raised after the publication of papers regarding the potential toxic effects of the presence in jet aircraft engine oils of neurotoxic compounds such as organophosphates.4,5 The very first issues involved the BAe146, a four-engine short-haul aircraft designed to be used on small city-based airports.6 Crew working onboard this aircraft frequently complained about poor air quality, leading them to get medical care in emergency care settings immediately after leaving the aircraft. The problem was identified as leaking oil seals in the auxiliary power unit (APU),7 which allowed burnt oil to enter the cabin air supply, and this problem was addressed with a technical modification, but complaints of odor and symptoms persisted, later spreading to different aircraft types. Based on initial inquiries about the Bae146 problem, a growing corpus of scientific reports and articles were published these 25 last years claiming to evaluate the main hypothesis of a toxic etiology for what has been called the “aerotoxic syndrome,” a neologism coined by nonmedical scientists.2,8,9 In brief, the main idea is that crewmembers or passengers could be contaminated by chemical compounds present in the cabin and cockpit air, which is bled from the engines in most jet aircraft, except in the Boeing 787. These toxic compounds could originate from engine jet oils through leakage due to sealing deficiencies, but also from hydraulic fluids leaks, deicing liquids, jet fuel, or any outside pollution during taxiing or low altitude flying.
Despite a huge number of reports and scientific papers with unequal quality due to small sample size, experimental design flaws, or potential bias, there is still no convincing evidence that cabin air is contaminated in sufficient concentration for a great bunch of suspected pollutants to provoke neurological or systemic diseases.10–12 Studies aiming at supporting the concept of a so-called “aerotoxic syndrome” are mainly case reports with a small number of crewmembers affected. Studies looking at symptoms or disease prevalence at a population level are based on postal or onsite self-declaration questionnaires without medical review and are subject to selection bias. The aeromedical specialists are still questioning the toxic origin of the long-lasting symptoms described in the case reports for affected crews.13
The primary objective of the study is to compare the incidence of new pathologies or chronic symptoms in flight crews occurring after a fume event according to whether they have been exposed or not. Secondary objectives were getting an estimation of the prevalence of the main symptoms or pathologies described in the scientific literature and linked to fume event exposure in a vast cohort of crewmembers and providing a breakdown of these health effects according to five main groups of pathologies. We aimed to test the main hypothesis that there is no difference between exposed and unexposed groups. For this study, we present a retrospective analysis of a cohort of almost 15,000 crewmembers of a major airline, active for 6 yr, working on one of the biggest networks in the world combining short, medium, and long haul, and on exclusively Airbus and Boeing jet aircraft.
METHODS
Subjects
In France, the medical follow-up for crewmembers, all exposed to cosmic ionizing radiation, is mandatory every 2 yr and must be conducted by an occupational physician or more recently by an occupational nurse. In our French airline, all the crewmembers based in Paris Charles de Gaulle are followed in the same occupational medicine department, by 8 occupational physicians and 20 occupational nurses. The medical data are collected in an independent digital database whose access is strictly restricted to the occupational medicine team, without any possible access to medical data by nonhealth care workers in the airline. This guarantees strict confidentiality for any medical information collected and is legally protected under French regulations. The medical database is in place since 2002 and is daily updated with a lot of human resources data, including sick leave periods, occupational injuries declared to the employer, or occupational diseases declared by the crewmember. During the mandatory periodic follow-up, the occupational health care provider is informed by the crewmember about his health and the diseases or sick leave medical reason are coded through ICD-10 (International Classification of Diseases - 10th version)14 to record any medical information in the medical database. Since 2017, the occupational medicine department has been informed of any fume event that occurred within 2–4 d and has sent a digital questionnaire to all crewmembers involved in the fume event. This information is also recorded in the occupational medical database, allowing extraction for analysis. The response rate for this questionnaire is 65%. A copy of the questionnaire is provided as supplementary material.
This study protocol was found to conform to generally accepted scientific principles and research ethical standards by the Ethical Review Committee for publications of Cochin University Hospital, Paris, France (CLEP Decision N°: AAA-2024-10,005).
Procedure
We built a cohort with all the crewmembers active from 2017 to 2022 and exclusively followed up by the Roissy CDG occupational health department as an eligibility criterion, leading to an initial cohort of 20,906 crewmembers. We identified all the crewmembers involved in at least one fume event. A fume event is defined by the French Civil Aviation Authority as the release of smoke and/or odors from the air conditioning system. These phenomena take the form of odors, fumes, or mists contaminating the passenger cabin or cockpit.
For each exposed crewmember we identified at least one control nonexposed crewmember paired according to six criteria at the time of the fume event for the exposed crewmember [Sex; Age class (5 yr); Number of flight hours; Number of flights; Working on Short/Medium haul or Long haul flights; Being Cockpit or Cabin crew]. For the whole cohort, we extracted all symptoms and pathology ICD-10 (3 digit) codes whose start date occurred after the date of the fume event. We gave an arbitrary index of biological plausibility regarding the link with a fume event for each symptom or pathology (0 = Impossible, 1 = Possible, 2 = Described in the literature). The symptoms and pathologies with index of biological plausibility > 0 were recoded in five groups to match with the classification of symptoms attributed to fume events in the literature15 (see Table I).
Statistical Analysis
All analyses were performed with Stata SE 16.1, StataCorp, College Station, TX, United States. The dataset was analyzed either with the whole data (14,953 crewmembers with 2577 exposed and 12,372 up to 6 paired non-exposed controls for each exposed), or the 1:1 paired data (5154 crewmembers with 2577 exposed and 2577 non-exposed controls).
Prevalence comparison (whole dataset): we used a two-proportion z-test, two-sided, with an alpha-level at 0.05.
Incidence comparison (1:1 paired data subset): we used the csmatch command developed by P. Cummings for Stata,16 allowing to estimate the matched risk ratio in a convenient way.
Hazard ratio and comparison (whole dataset): we used a Cox proportional hazard model and tested the proportional hazards assumptions using Schoenfeld residuals and graphic log-log plot of survival.17 Postestimations were made to confirm the best model, using the Akaike Information Criterion and a Chi-squared likelihood ratio test.
RESULTS
In the medical database, we selected all the crewmembers (pilot and cabin crew) who were followed up by the Roissy occupational medicine department to ensure the consistency of pathology and symptoms coding practices for the 10 occupational physicians in charge during 2017–2022. That selection allowed an eligibility of 20,906 crewmembers.
Thanks to the systematic recording of the fume event exposure in two databases (the fume events incidents database which is held by the airline and the medical database, which is updated by the medical department after seeking the crewmembers in every flight mentioned in the fume events database), we could easily identify 2577 crewmembers involved in a fume event over the 6-yr study period from 2017–2022.
To control main confounding factors, we matched every exposed crewmember to a nonexposed control and built up the two following control groups: a main group with one to up to six controls, and a randomly selected subset of one-to-one matched control. The 1:1 subset allowed accurate statistical comparison taking matching effect into account, whereas the 1:6 control group enhanced the statistical power of our study, with 12,376 controls that matched the six following criteria: Sex, Age, Function, Flight type, Number of Flights in the career, and Number of Flight Hours in the career. The continuous variables such us age, number of flights, and number of flight hours were categorized to be used in the Cox proportional hazard models. Fig. 1 summarizes the process in a flowchart presentation.
Citation: Aerospace Medicine and Human Performance 96, 1; 10.3357/AMHP.6531.2025
Due to the matching process, the figures for exposed and nonexposed are quite similar and statistically not different for the 1:1 matched subset. For the whole cohort, given that the matching process could not find the same number of matches for exposed crewmembers, differences between the criteria have occurred, and were controlled by the multivariate approach in the Cox proportional hazards model (see Table II).
After estimation of the prevalence of each group of pathologies occurring after the fume event date for the exposed crewmember, the comparison tests could not demonstrate any difference in the two groups. Prevalence figures are all under 5%, indicating the relatively low frequency of the pathologies commonly linked to the fume events when confirmed by a medical doctor during the medical monitoring following exposure or for controls with no exposure. In each group, nearly 85% of the crewmembers included in the study had none of the pathologies of interest.
When dispatching the analysis in each group of pathologies, again no difference was seen between the exposed and the control groups, with the proportion test being nonsignificant for comparison in each group of pathologies. Results and comparison tests are presented in Table III.
Incidence of having at least one ICD-10 coded pathology or symptoms plausibly related to a fume event were estimated after the calculation of the number of person-years for each group with the same following parameters: beginning stated as the fume event date for the exposed crewmember of the pair, with data censored at the first date of occurrence of any of the listed pathology or at December 31, 2003. We found a rate of 1552.4 for 100,000 person-years in the exposed group with a 95% confidence interval (CI) from 1315.4–1832.1, whereas the incidence rate of the control group was 1497.7 for 100,000 person-years in the control group, with a CI of 1385.5–1618.9.
To compare these incidence rates, we used a Cox proportional hazard model firstly without covariates, then with all the confounding factors available. The raw model showed a hazard ratio of 1.04 with a 95% CI from 0.86–1.25 that was statistically not significant (z = 0.44; exact P-value = 0.661). The full model showed a very similar rate of 1.03 (95% CI from 0.85–1.24), with age, sex, flights hours, flights, function, and flight type as covariates controlled for. Three covariates had a significant association with the issue, as shown in Table IV. The hazard rate for exposure was equally nonsignificantly different from 1 (z = 0.34; exact P-value = 0.734).
Comparison of different models with Akaike Information Criterion showed a significantly better fit of the full model (LR χ2 = 131.25; P-value = 0.000). Results and comparison tests are presented in Table IV.
To confirm our first results, we used a statistical approach adapted to paired data which avoided the possibility of overfitting our model. That was possible with the csmatch command for Stata,16 which allows estimation of a cohort matched-pair risk ratio for the outcome of suffering any plausibly related group of symptoms or disease after a fume event for exposed crewmembers compared to their nonexposed pair. We confirmed a nonsignificant risk-ratio of 1.07 [95% CI (0.85–1.34)], z = 0.60 with an exact P-value of 0.547, very close to the results obtained in the whole cohort comparison hazard ratio of 1.04.
DISCUSSION
We present here the first retrospective cohort study on medical consequences of a fume event, to our knowledge. Until now, the scientific articles describing medical effects of exposure to fume events were mainly case reports or experimental studies with a limited number of subjects.18–20 Moreover, all of the pathologies and symptoms analyzed in our study were confirmed by a medical doctor, and therefore digitally coded following the ICD-10 classification, which is used in our medical database since its very beginning in 2002. This medical confirmation of the pathologies or symptoms is quite rare in the epidemiological studies involving mainly healthy workers such as crewmembers and highlights a possible discrepancy between perceived health status that is explored by self-questionnaires and more medically reliable information when based on personal contact with a medical doctor and frequently supported by medical documents such as hospital or other medical reports.21 This strength of this study is linked to the mandatory occupational health follow-up in a single medical department in charge of about 20,000 crewmembers, with a medical consultation every 2 yr due to the ionizing radiation exposure legally imposed by the French labor code, and a mandatory fitness-for-duty examination after any 30-d sick leave for occupational injury.
Our results show that there is no difference between exposed and nonexposed crewmembers, neither in prevalence nor in incidence of pathologies or symptoms possibly related to a fume event. The number of fume events, including exposed crewmembers in the cohort, is huge (357 events) as is the number of crewmembers included (2577 exposed and 12,376 controls). These figures give a high statistical power to our study, which is able to objectivize small differences of prevalence or incidence between the two groups.
The paired study was an important feature, allowing that the main confounding factors were a priori controlled in the analysis, even though we used a multivariate analysis for the whole cohort analysis. As these results may be surprising, it is important to examine several limitations of the study:
Diagnostics recordings: This monocentric study used medical data recorded by several medical doctors whose pathology coding habits could differ from one another. Moreover, for every crewmember exposed, the fume event exposure was available to the doctor and could have led to some differential bias due to possible specific inquiries about consequences of the fume event on health.
Underreporting: For pilots there is a known underreporting behavior that could lead to lower prevalence or incidence of recorded pathologies.22
Confounding factors: Because the study was retrospective, some confounding factors could have been missed or not have been available.
Censored data: Finally, a question could be raised about the possible absence of the most serious pathology in the database due to resignation for medical reasons without a fit-to-work examination after sick leave. This specific question has been studied by a retrospective census of all loss of medical certificate, which did not retrieve any case of loss following a fume event exposure.
All these limitations have a weak consequence on the analysis: the multiple investigator effect and the pilot underreporting of diseases or symptoms are nondifferential biases and may just lead to some power limitations which are largely compensated by the cohort size. The possible differential bias due to investigator knowledge of exposure status would, in any case, cause an “over”-recording of medical consequences.
This study focused on the medically confirmed potential health effects of fume events. It clearly contradicts the “aerotoxic” hypothesis regarding the potential occurrence of a delayed neuropathy due to organophosphate compounds contaminating cabin air after a fume event. However, this toxic hypothesis still needs scientific answers since most cabin air measurements were done outside of a fume event. For this purpose, a new prospective cohort study has begun in France in 2024 which aims at sampling air during a fume event for specific chemical analysis (all the organophosphorus compounds present in jet engine oils and their pyrolysis derivates plus carbon monoxide) and is combined with an immediate (72 h) and a delayed (3 mo) neurocognitive medical evaluation.
This retrospective cohort study involving nearly 15,000 crewmembers over 7 yr, with a matched pairs design, gives for the first time new answers about the medical consequences of an exposure to a fume event. The absence of significant differences in prevalence of subsequent diseases or symptoms between the exposed and the control group, and the hazard ratio being nonsignificantly different from 1 bring strong arguments against intermediate and long-term neurological effects of an exposure to a fume event. Because the toxicological data during the fume events themselves are still lacking, further investigation must be carried out, and we believe that the ongoing French AVISAN study23 could bring important results for a better understanding of this problem.
Process flowchart.
Contributor Notes
