Abstract
Objectives
To identify longitudinal trajectories of medical fitness among licensed aviation personnel and examine factors associated with trajectory membership.
Design
Retrospective cohort study.
Setting
Aviation medical examination records maintained by the Korean Society of Aerospace and Environmental Medicine, Republic of Korea, 2015–2022.
Participants
7740 licensed aviation personnel with at least five examinations, drawn from 14 237 licensed individuals.
Primary and secondary outcome measures
Latent classes of final fitness classifications across the first five examinations were identified using latent class analysis. Associated factors were examined using multinomial logistic regression. Model 2 additionally included body mass index change, fifth-examination smoking status and licence change.
Results
Three trajectories were identified: sustained fit (86.3%), persistent conditional fit (10.2%) and progressive deterioration (3.6%). Older age was associated with both persistent conditional fit (OR per year 1.10, 95% CI 1.09 to 1.12) and progressive deterioration (OR per year 1.06, 95% CI 1.04 to 1.08), as was first-examination obesity (OR 1.58, 95% CI 1.29 to 1.93 and OR 1.78, 95% CI 1.30 to 2.44, respectively). Compared with current smoking, former smoking was associated with higher odds of persistent conditional fit (OR 1.62, 95% CI 1.27 to 2.06) but lower odds of progressive deterioration (OR 0.71, 95% CI 0.52 to 0.98). Commercial pilot licence holders had lower odds of persistent conditional fit than airline transport pilot licence holders (OR 0.69, 95% CI 0.52 to 0.91). Additional variables showed no clear associations and did not significantly improve model fit; Nagelkerke pseudo R² values were 0.181 and 0.182.
Conclusions
Repeated examination records identified distinct fitness trajectories. Older age and first-examination obesity were associated with both adverse trajectories. Potential selection bias, overlapping measurement periods and limited model explanatory performance restrict causal interpretation and generalisability.
Keywords: Occupational Health, Longitudinal studies, Workplace, EPIDEMIOLOGY
STRENGTHS AND LIMITATIONS OF THIS STUDY.
Linked individual records from a centrally managed aviation medical examination database spanning 8 years enabled analysis of medical fitness classifications across five examinations.
Regression models incorporated first-examination body mass index and smoking status alongside selected changes during follow-up and fifth-examination measurements.
Restricting the analytic sample to personnel with at least five examinations may introduce selection bias, including the healthy worker survivor effect.
Covariates measured at the fifth examination or over the period used to define the trajectories, together with variable intervals between examinations, limit temporal and causal interpretation.
Unmeasured occupational exposures, self-reported smoking status and a predominantly male, single-country sample limit interpretation and generalisability.
Introduction
Aviation safety is fundamentally dependent on the physical and mental health of aviation personnel. Deterioration in the health of flight crew operating aircraft carrying hundreds of passengers, as well as air traffic controllers (ATCs) managing complex airspace, represents not merely an individual concern but a critical threat to public safety.1 In the early history of aviation, mechanical failure was the predominant cause of accidents; however, advances in technology have reduced mechanically caused incidents to near-negligible levels. As aircraft have become more reliable, human factors have become central to aviation accident prevention; an analysis of recent US accident data found that pilot error was the most frequent human factor in general aviation accidents and crew error in commercial jet accidents.2 Furthermore, the March 2015 Germanwings Flight 4U9525 accident—in which a co-pilot with a history of depression deliberately crashed the aircraft into the French Alps, killing all 150 people on board—heightened global awareness of the importance of mental health management among aviation personnel.3
Aviation personnel operate in a complex occupational environment that may structurally increase the risk of specific diseases. Both pilots and ATCs are exposed to circadian rhythm disruption resulting from shift work and night operations, while pilots additionally experience recurrent jet lag. Such disruption has been associated not only with sleep disorders but also with an increased risk of metabolic conditions, including hypertension and type 2 diabetes.4 5 A large-scale cohort study further reported that misalignment between sleep timing and work schedules is associated with an increased risk of type 2 diabetes.6 Among pilots, short rest intervals between flights have been associated with elevated stress hormone levels,7 and increased mental workload has generally been associated with elevated heart rate in simulated and actual flight, although findings for heart rate variability have varied across studies.8 Among ATCs, sustained concentration and high workload in high-density traffic environments are primary contributors to fatigue,9 which has been associated with reduced situational awareness and increased performance errors, thereby representing a latent threat to aviation safety.10 Although flight crew and ATCs are exposed to distinct occupational hazards, they share overlapping risks across cardiovascular and metabolic domains, underscoring the need for systematic health management.
In response to these occupational health risks, the International Civil Aviation Organization stipulates in its Personnel Licensing standards11 that licence holders who recognise deterioration in their medical fitness that may compromise the safe exercise of their privileges must refrain from exercising those privileges and seek aeromedical advice as necessary. In the Republic of Korea, Article 42 of the Aviation Safety Act was amended in January 2022 to require flight crew and ATCs holding aviation medical certificates to report any deterioration in their physical or mental condition to the Minister of Land, Infrastructure and Transport; these provisions entered into force on 19 January 2025.12 This regulatory framework indicates that aviation medical examinations function not merely as administrative procedures but also as an occupational health surveillance system for the early detection of health risks among aviation personnel. However, the extent to which serial examination data accumulated within this system are used to characterise health deterioration trajectories remains unclear.
A review of the existing literature indicates that several studies have used aviation medical examination data in the Korean context. Han et al13 analysed data from 2015 to 2017 and reported that hypertension, cardiovascular disease and ophthalmological conditions were the leading causes of conditional fit classifications, collectively accounting for more than 50% of such determinations. Kwon14 examined pilots at a national carrier from 2015 to 2019 and found that disqualification rates increased among individuals aged 50 years and older, with cardiovascular disease identified as the primary cause. International studies have reported similar findings: a permanent disqualification rate of 1.3% was observed among Turkish civil aviation pilots,15 and a Norwegian study found that the disease profile underlying medical licence revocation shifted over time, with cardiovascular causes declining and otolaryngological and neurological conditions increasing in relative proportion.16 However, most of these studies used cross-sectional designs or short-term follow-up periods of 3 to 5 years, limiting their ability to characterise longitudinal changes in individual-level medical fitness.
The present study aimed to identify distinct trajectories of medical fitness deterioration among licensed aviation personnel using serial examination data and to examine the determinants of trajectory.
Methods
Study design
This retrospective longitudinal cohort study was conducted among aviation personnel in Korea using examination records collected from January 2015 to December 2022. Repeated-measures data from the same individuals were used to analyse changes in medical fitness and to characterise latent patterns of medical fitness.
Study population and data collection
This study used the aviation medical examination database maintained by the Korean Society of Aerospace and Environmental Medicine (KSAEM). Established in 1989 under the Aviation Safety Act, KSAEM centrally manages data generated through the review of aviation medical certificates delegated by the Ministry of Land, Infrastructure and Transport. The research team submitted a data use plan, obtained approval from the KSAEM Executive Committee, and was granted access to the raw data.
The study population comprised licensed aviation personnel who underwent aviation medical examinations between 2015 and 2022. The initial dataset included 99 388 examination records from 30 369 individuals. After 28 758 records from individuals without a valid licence were excluded, 70 630 records from 14 237 individuals remained. Of these, 7740 individuals with five or more examinations were included in the analytic sample to allow repeated observation of changes in medical fitness during the study period. The first five examinations were used for each participant. The remaining 6497 individuals with fewer than five examinations were excluded from the analytic sample.
Variables
The variables collected in this study included general characteristics, health-related variables, disease classifications and final fitness classifications.
General characteristics
General characteristics comprised sex, age, licence type and total flight hours. Licence type was categorised into four groups: airline transport pilot licence (ATPL), air traffic controller licence (ATCL), private pilot licence (PPL) and commercial pilot licence (CPL). Total flight hours were obtained from the examination records and were coded as 0 for ATCL holders. For the comparison between participants with five or more examinations and those with fewer than five examinations, all characteristics were taken from the first examination (table 1); in this comparison, age was also presented in categories (≤29, 30–39, 40–49, 50–59 and ≥60 years), and licence types not classified into the four groups were categorised as ‘other’; no participant held an ‘other’ licence at the fifth examination.
Table 1.
Comparison of general characteristics at the first examination between participants with ≥5 and <5 examinations (n=14 237)
| Variable | Category | ≥5 (n=7740) |
<5 (n=6497) |
χ2 (p value) |
| Sex | Male | 7579 (97.9) | 5880 (90.5) | 376.087 (<0.001) |
| Female | 161 (2.1) | 617 (9.5) | ||
| Age (years) | ≤29 | 877 (11.3) | 2371 (36.5) | 1469.214 (<0.001) |
| 30–39 | 3038 (39.3) | 2357 (36.3) | ||
| 40–49 | 2383 (30.8) | 973 (15.0) | ||
| 50–59 | 1166 (15.1) | 586 (9.0) | ||
| ≥60 | 276 (3.6) | 210 (3.2) | ||
| Mean±SD | 40.61±9.51 | 35.03±10.67 | −32.627 (<0.001) | |
| Licence type | ATPL | 3126 (40.4) | 1143 (17.6) | 2466.455 (<0.001) |
| ATCL | 62 (0.8) | 830 (12.8) | ||
| PPL | 639 (8.3) | 1902 (29.3) | ||
| CPL | 3885 (50.2) | 2510 (38.6) | ||
| Other | 28 (0.4) | 112 (1.7) | ||
| Total flight hours | <1000 | 1919 (24.8) | 4310 (66.3) | 2532.468 (<0.001) |
| (hours) | 1000–4999 | 3009 (38.9) | 1126 (17.3) | |
| 5000–9999 | 1499 (19.4) | 407 (6.3) | ||
| ≥10 000 | 1313 (17.0) | 654 (10.1) | ||
| Mean±SD | 8377.50±69445.07 | 3020.41±23143.08 | −6.378 (<0.001) | |
| BMI (kg/m²) | Underweight | 42 (0.5) | 112 (1.7) | 81.202 (<0.001) |
| Normal | 2389 (30.9) | 2043 (31.5) | ||
| Overweight | 2629 (34.0) | 1894 (29.2) | ||
| Obese | 2680 (34.6) | 2448 (37.7) | ||
| Mean±SD | 24.25±2.48 | 24.35±3.01 | 2.097 (0.036) | |
| Smoking status | Never | 3253 (42.0) | 3810 (58.6) | 527.889 (<0.001) |
| Former | 3048 (39.4) | 1442 (22.2) | ||
| Current | 1439 (18.6) | 1245 (19.2) | ||
| First-round fitness classification | Fit | 6926 (89.5) | 5765 (88.7) | 13.958 (<0.001) |
| Conditional fit | 809 (10.5) | 711 (10.9) | ||
| Unfit | 5 (0.1) | 21 (0.3) |
Other (licence types not classified above); BMI categories: underweight (<18.5), normal (18.5 to <23), overweight ≥23 to <25, obese ≥25 kg/m2.
ATCL, air traffic controller licence; ATPL, airline transport pilot licence; BMI, body mass index; CPL, commercial pilot licence; PPL, private pilot licence.
Health-related variables
Health-related variables included body mass index (BMI) and smoking status. BMI was calculated as weight in kilograms divided by height in metres squared using measured height and weight, and categorised according to the Asia-Pacific criteria17 as underweight (<18.5 kg/m²), normal weight (≥18.5 to <23 kg/m²), overweight (≥23 to <25 kg/m²) or obese (≥25 kg/m²). For the regression analyses, underweight and normal weight were combined into a single category (<23 kg/m²). Smoking status was obtained from a self-administered checklist completed at the examination and categorised as never, former or current smoking.
Record linkage and measurement timing
The raw data were structured at the individual examination-record level. To link repeated examination records for the same individual, an anonymised unique identifier that precluded direct personal identification was assigned to each participant, enabling longitudinal tracking by year and examination round. For the regression analyses, BMI and smoking status were taken from the first examination, at the start of the observation period, whereas age, sex and licence type were taken from the fifth examination and therefore describe participants at the end of the observation period. Age was analysed continuously in years. Total flight hours were summarised as the maximum recorded value across examinations 1–5, representing cumulative flying experience, and categorised as <1000, ≥1000 to <5000, ≥5000 to <10 000 or ≥10 000 hours. To capture changes during follow-up, BMI change was calculated as fifth-examination BMI minus first-examination BMI and analysed continuously; fifth-examination smoking status was also included; and licence change was categorised as changed or unchanged according to whether licence type differed between the first and fifth examinations. Reference categories were female sex, ATPL, ≥10 000 flight hours, BMI <23 kg/m², current smoking, and changed licence.
Disease classification
Disease classification was performed through a systematic review and recoding of free-text clinical findings recorded in the comprehensive assessment and classification-rationale fields of the medical examination forms. Hypertension, diabetes mellitus, hyperlipidaemia, sleep disorders and thyroid disease were classified as individual conditions. All remaining conditions were grouped into the following categories: neurological; cardiovascular, respiratory; gastrointestinal; musculoskeletal; haematological; ear, hearing and vestibular; ophthalmological; genitourinary; dermatological/immunological; psychiatric; malignancy and other conditions. The presence or absence of each condition was coded as a binary variable. For table 2, a condition was coded as present if it was recorded at least once across examinations 1–5, and participants could have more than one disease category.
Table 2.
Recorded diseases across examinations 1–5 by latent class (N=7740)
| Disease | Total | Class 1 (n=786) | Class 2 (n=277) | Class 3 (n=6677) |
| Hypertension | 375 (4.8) | 238 (30.3) | 80 (28.9) | 57 (0.9) |
| Ophthalmological disease | 302 (3.9) | 170 (21.6) | 53 (19.1) | 79 (1.2) |
| Gastrointestinal disease | 188 (2.4) | 74 (9.4) | 33 (11.9) | 81 (1.2) |
| Genitourinary disease | 181 (2.3) | 114 (14.5) | 17 (6.1) | 50 (0.7) |
| Diabetes mellitus | 142 (1.8) | 88 (11.2) | 28 (10.1) | 26 (0.4) |
| Ear, hearing and vestibular disorders | 140 (1.8) | 93 (11.8) | 19 (6.9) | 28 (0.4) |
| Cardiovascular disease | 136 (1.8) | 87 (11.1) | 31 (11.2) | 18 (0.3) |
| Thyroid disease | 79 (1.0) | 59 (7.5) | 10 (3.6) | 10 (0.1) |
| Musculoskeletal disease | 60 (0.8) | 30 (3.8) | 15 (5.4) | 15 (0.2) |
| Respiratory disease | 40 (0.5) | 21 (2.7) | 8 (2.9) | 11 (0.2) |
| Malignancy | 32 (0.4) | 14 (1.8) | 3 (1.1) | 15 (0.2) |
| Haematological disease | 19 (0.2) | 11 (1.4) | 4 (1.4) | 4 (0.1) |
| Neurological disease | 19 (0.2) | 10 (1.3) | 3 (1.1) | 6 (0.1) |
| Dermatological/immunological disease | 19 (0.2) | 10 (1.3) | 4 (1.4) | 5 (0.1) |
| Psychiatric disorder | 9 (0.1) | 5 (0.6) | 0 (0.0) | 4 (0.1) |
| Hyperlipidaemia | 8 (0.1) | 4 (0.5) | 0 (0.0) | 4 (0.1) |
| Sleep disorder | 6 (0.1) | 1 (0.1) | 2 (0.7) | 3 (0.0) |
| Other diseases | 4 (0.1) | 1 (0.1) | 0 (0.0) | 3 (0.0) |
Disease presence was defined as at least one recorded occurrence across examinations 1–5. Diseases are listed in descending order of total frequency. Percentages were calculated within each column and rounded to one decimal place. Individuals could have more than one disease category. Results are descriptive and unadjusted for age or sex.
Final fitness classification was determined by a physician and coded as fit (0), conditional fit (1) or unfit (2). Classifications from the first to the fifth examinations served as indicators for LCA.
Statistical analysis
Python was used to assist with data preparation. All analyses were conducted using SPSS V.29.0 (IBM) and Mplus V.8.5 (Muthén & Muthén, Los Angeles, California, USA). General characteristics, disease classifications and final fitness classifications were summarised as frequencies and percentages, and continuous variables were summarised as means and SD. To characterise selection into the analytic sample, first-examination characteristics were compared between participants with five or more examinations and those with fewer than five examinations using chi-square tests for categorical variables and independent-samples t-tests for continuous variables. This comparison described group differences relevant to potential selection bias, including the healthy worker survivor effect (HWSE), but did not control for such bias.
LCA was performed based on the time-series pattern of final fitness classifications across the first through fifth examination rounds. One-class to four-class solutions were estimated. To determine the number of latent classes, the Akaike information criterion (AIC), Bayesian information criterion (BIC), sample-size-adjusted BIC (aBIC), entropy, adjusted Lo-Mendell-Rubin likelihood ratio test (LMR) and bootstrap likelihood ratio test (BLRT) were applied.18 19 Lower AIC, BIC and aBIC values indicated better relative model fit, and higher entropy indicated greater classification separation. The LMR and BLRT compare a k-class model with a (k−1)-class model, with p<0.05 indicating that the additional class significantly improved model fit. Class selection also considered class size, parsimony and interpretability. Each participant was assigned to the class with the highest posterior probability. To facilitate visual representation of health deterioration trajectories by latent class, a growth mixture model was additionally applied,20 estimating the trajectory of fitness classification scores for each class, where 0 indicated fit, 1 indicated conditional fit and 2 indicated unfit.
Multinomial logistic regression was then performed to identify factors associated with latent class membership, using most likely class membership as the outcome and class 3 (sustained fit) as the reference category. Model 1 included fifth-examination age, sex and licence type; maximum recorded total flight hours across examinations 1–5; and first-examination BMI and smoking status. Model 2 additionally included BMI change between the first and fifth examinations, fifth-examination smoking status and licence change. All covariates were entered simultaneously within each model, and results were reported as adjusted ORs with 95% CIs. The two models were compared using a likelihood ratio test, and model fit was also described using Nagelkerke pseudo R². Interpretation emphasised the magnitude and precision of associations rather than statistical significance alone.
Disease variables were not entered into the multinomial logistic regression models. Within the aviation medical examination system, diagnosed conditions inform the final fitness classification; including disease variables as explanatory variables would therefore create overlap between information contributing to the fitness classifications used to define latent classes and information included in the regression models. Instead, recorded diseases across examinations 1–5 were summarised by latent class as frequencies and percentages (table 2). These comparisons were descriptive and were not adjusted for age or sex.
Patient and public involvement
None.
Results
Comparison of general characteristics between repeated examinees and short-term examinees
To characterise selection into the analytic sample, characteristics at the first examination were compared between participants with ≥5 examinations (analytic sample, n=7740) and those with <5 examinations (n=6497) (table 1). Participants with ≥5 examinations were more likely to be male (97.9% vs 90.5%) and were older (mean age 40.6 vs 35.0 years). They more often held an ATPL (40.4% vs 17.6%) or CPL (50.2% vs 38.6%), whereas PPL (8.3% vs 29.3%) and ATCL (0.8% vs 12.8%) holders were more common among those with <5 examinations. Total flight hours of ≥5000 hours were also more frequent among participants with ≥5 examinations (36.4% vs 16.4%). Former smoking was more common (39.4% vs 22.2%) and never smoking less common (42.0% vs 58.6%) in this group, whereas the proportion of current smokers was similar (18.6% vs 19.2%). Differences in mean BMI were small ((24.25) vs (24.35) kg/m²). The proportions classified as fit or conditionally fit at the first examination showed small absolute differences between groups; however, unfit classification was less frequent among participants with ≥5 examinations (0.1% (n=5) vs 0.3% (n=21)).
Latent class analysis results
LCA was conducted based on the time-series pattern of final fitness classifications across the first to the fifth examination rounds to characterise trajectories among personnel. Model fit indices were compared across one-class to four-class solutions (table 3).
Table 3.
Model fit indices by number of latent classes (n=7740)
| Number of classes | AIC | BIC | aBIC | Entropy | LMR | BLRT | Class distribution (%) | |||
| 1 | 2 | 3 | 4 | |||||||
| 1 | 29 424.98 | 29 494.52 | 29 462.74 | – | – | – | 100.0 | |||
| 2 | 15 701.29 | 15 847.33 | 15 780.60 | 0.979 | <0.001 | <0.001 | 13.6 | 86.4 | ||
| 3 | 14 739.36 | 14 961.89 | 14 860.20 | 0.986 | <0.001 | -* | 10.1 | 3.9 | 85.9 | |
| 4 | 14 282.38 | 14 581.41 | 14 444.76 | 0.986 | 0.017 | -* | 9.4 | 4.0 | 1.5 | 85.1 |
Lower AIC, BIC and aBIC values indicate better relative fit. Entropy ranges from 0 to 1, with higher values indicating more accurate classification. LMR and BLRT compare the kclass model with the (k−1)-class model; p<0.05 favours the k-class model. Class proportions are model-estimated proportions. Bold indicates the selected model.
*Not reported because none of the bootstrap draws converged.
aBIC, sample-size-adjusted BIC; AIC, Akaike information criterion; BIC, Bayesian information criterion; BLRT, bootstrap likelihood ratio test; LMR, adjusted Lo-Mendell-Rubin likelihood ratio test.
The AIC, BIC and aBIC decreased consistently as the number of classes increased. The LMR test indicated statistically significant improvement for the two-class (p<0.001), three-class (p<0.001) and four-class (p=0.017) solutions. Entropy values were 0.979 for the two-class solution and 0.986 for both the three-class and four-class solutions, indicating high classification separation.
Although the four-class solution yielded lower AIC, BIC and aBIC values and a significant LMR test compared with the three-class solution, its smallest class had a model-estimated proportion of 1.5% and included 100 participants (1.3%) based on most likely class membership. Considering class size, parsimony and interpretability, the three-class solution was selected.
Based on most likely class membership, the final three-class solution yielded the following distribution: Class 1 (persistent conditional fit), 10.2% (n=786); Class 2 (progressive deterioration), 3.6% (n=277) and class 3 (sustained fit), 86.3% (n=6677). Figure 1 presents the mean fitness classification scores across the five examination rounds for each latent class.
Figure 1.
Medical fitness trajectories by latent class. Medical fitness classification (0, fit; 1, conditional fit; 2, unfit) at examinations 1–5 for the three latent classes identified among 7740 licensed aviation personnel: sustained fit, persistent conditional fit and progressive deterioration.
Factors associated with latent class membership
Multinomial logistic regression was performed with most likely class membership as the outcome and class 3 (sustained fit) as the reference category (table 4). Model 1 included fifth-examination age, sex and licence type; maximum total flight hours across examinations 1–5; and first-examination BMI and smoking status. Model 2 additionally included BMI change, fifth-examination smoking status and licence change between the first and fifth examinations.
Table 4.
Associations of participant characteristics with latent class membership: models 1 and 2 (N=7740; reference: class 3)
| Variable | Category | M1 | M2 | ||||||
| Class 1 | Class 2 | Class 1 | Class 2 | ||||||
| OR | 95% CI | OR | 95% CI | OR | 95% CI | OR | 95% CI | ||
| Age (5th) | Per year | 1.104 | 1.090 to 1.119 | 1.056 | 1.036 to 1.077 | 1.102 | 1.088 to 1.117 | 1.056 | 1.035 to 1.077 |
| Sex (5th; ref=Female) | Male | 1.196 | 0.366 to 3.913 | 1.609 | 0.385 to 6.718 | 1.206 | 0.370 to 3.929 | 1.615 | 0.386 to 6.756 |
| Licence type (5th; ref=ATPL) | ATCL | 1.363 | 0.576 to 3.225 | 2.062 | 0.616 to 6.905 | 1.141 | 0.466 to 2.791 | 2.009 | 0.570 to 7.082 |
| PPL | 0.594 | 0.102 to 3.466 | 3.388 | 0.644 to 17.830 | 0.490 | 0.083 to 2.906 | 3.470 | 0.637 to 18.897 | |
| CPL | 0.685 | 0.518 to 0.906 | 1.426 | 0.978 to 2.081 | 0.626 | 0.462 to 0.849 | 1.434 | 0.945 to 2.176 | |
| Total flight hours (maximum, 1st–5th; ref≥10 000) | <1000 | 1.397 | 0.724 to 2.697 | 0.955 | 0.425 to 2.146 | 1.682 | 0.838 to 3.379 | 0.963 | 0.399 to 2.324 |
| ≥1000 to <5000 | 0.801 | 0.560 to 1.145 | 0.824 | 0.488 to 1.391 | 0.896 | 0.610 to 1.315 | 0.824 | 0.474 to 1.430 | |
| ≥5000 to <10 000 | 0.853 | 0.684 to 1.063 | 0.769 | 0.532 to 1.112 | 0.896 | 0.714 to 1.123 | 0.765 | 0.523 to 1.119 | |
| BMI (1st; ref<23 kg/m²) | Obese | 1.576 | 1.287 to 1.931 | 1.780 | 1.300 to 2.437 | 1.588 | 1.293 to 1.950 | 1.756 | 1.278 to 2.413 |
| Overweight | 1.095 | 0.885 to 1.355 | 1.236 | 0.885 to 1.727 | 1.101 | 0.888 to 1.364 | 1.226 | 0.876 to 1.715 | |
| Smoking status (1st; ref=current) | Never | 1.162 | 0.898 to 1.503 | 0.748 | 0.544 to 1.030 | 0.868 | 0.461 to 1.633 | 0.713 | 0.306 to 1.658 |
| Former | 1.618 | 1.270 to 2.060 | 0.709 | 0.515 to 0.977 | 1.554 | 1.122 to 2.153 | 0.557 | 0.369 to 0.843 | |
| BMI change (5th−1st) | Per 1 kg/m² increase | 1.001 | 0.935 to 1.071 | 0.974 | 0.886 to 1.071 | ||||
| Smoking status (5th; ref=current) |
Never | 1.371 | 0.730 to 2.575 | 1.167 | 0.487 to 2.796 | ||||
| Former | 1.053 | 0.759 to 1.460 | 1.474 | 0.953 to 2.280 | |||||
| Licence change (1st−5th; ref=changed) |
Unchanged | 1.285 | 0.957 to 1.725 | 0.995 | 0.674 to 1.469 | ||||
| Likelihood ratio χ² (24)=912.825, p≤0.001 Cox and Snell R2=0.111, Nagelkerke R2=0.181 |
Likelihood ratio χ² (32)=920.297, p≤0.001 Cox and Snell R2=0.112, Nagelkerke R2=0.182 |
||||||||
Class 3 (sustained fit) was the reference outcome category; class 1 indicates persistent conditional fit and class 2 progressive deterioration. Class membership was based on the most likely latent class (class 1, n=786; class 2, n=277; class 3, n=6677). Model 1 included fifth-examination age, sex and licence type; maximum total flight hours across examinations 1–5; and first-examination BMI and smoking status. Model 2 additionally included BMI change (fifth minus first examination), fifth-examination smoking status and licence change between the first and fifth examinations. All ORs are mutually adjusted within each model; in model 2, ORs for first-examination smoking status are conditional on fifth-examination smoking status. BMI categories: <23 kg/m² (underweight and normal weight), overweight (≥23 to <25 kg/m²) and obese (≥25 kg/m²). Likelihood ratio test comparing model 2 with model 1: χ²(8)=7.47, p=0.49.
ATCL, air traffic controller licence; ATPL, airline transport pilot licence; BMI, body mass index; CPL, commercial pilot licence; PPL, private pilot licence.
In model 1, older age was associated with membership in both class 1 (OR per year 1.104, 95% CI 1.090 to 1.119) and class 2 (OR 1.056, 95% CI 1.036 to 1.077). Obesity at the first examination (vs BMI <23 kg/m²) was associated with both class 1 (OR 1.576, 95% CI 1.287 to 1.931) and class 2 (OR 1.780, 95% CI 1.300 to 2.437), whereas overweight showed no clear evidence of an association with either class. Compared with current smoking, former smoking at the first examination was associated with higher odds of class 1 membership (OR 1.618, 95% CI 1.270 to 2.060) but lower odds of class 2 membership (OR 0.709, 95% CI 0.515 to 0.977). CPL holders had lower odds of class 1 membership than ATPL holders (OR 0.685, 95% CI 0.518 to 0.906), whereas no clear association was observed for class 2 (OR 1.426, 95% CI 0.978 to 2.081). Sex, total flight hours and ATCL or PPL licensure showed no clear evidence of an association with either class; estimates for sex and for ATCL and PPL licensure had wide CIs.
In model 2, BMI change between the first and fifth examinations showed no clear evidence of an association with class 1 (OR per 1 kg/m² increase 1.001, 95% CI 0.935 to 1.071) or class 2 (OR 0.974, 95% CI 0.886 to 1.071). Fifth-examination smoking status and licence change also showed no clear evidence of an association with class membership. The associations of age, first-examination obesity and CPL licensure with class membership remained similar in magnitude; the ORs for first-examination obesity were 1.588 (95% CI 1.293 to 1.950) for class 1 and 1.756 (95% CI 1.278 to 2.413) for class 2. Adding these variables did not significantly improve model fit (likelihood ratio test, χ²(8)=7.47, p=0.49). Nagelkerke pseudo R² changed little, from 0.181 in model 1 to 0.182 in model 2.
Recorded diseases by latent class
Table 2 presents the proportion of participants in each latent class with at least one recorded occurrence of each disease category across examinations 1–5. Hypertension was the most frequently recorded condition (class 1, 30.3%; class 2, 28.9%; class 3, 0.9%), followed by ophthalmological disease (21.6%, 19.1% and 1.2%, respectively). Diabetes mellitus (11.2%, 10.1% and 0.4%) and cardiovascular disease (11.1%, 11.2% and 0.3%) were recorded at similar proportions in classes 1 and 2. Genitourinary disease (14.5% vs 6.1%), ear, hearing and vestibular disorders (11.8% vs 6.9%) and thyroid disease (7.5% vs 3.6%) were more frequently recorded in class 1 than in class 2, whereas gastrointestinal disease (9.4% vs 11.9%) and musculoskeletal disease (3.8% vs 5.4%) were more frequently recorded in class 2. In class 3, the proportion was 1.2% or lower for all disease categories. Psychiatric disorders, hyperlipidaemia and sleep disorders were each recorded in fewer than 10 participants overall. These comparisons were not adjusted for age or other covariates and are presented descriptively.
Discussion
This study identified three longitudinal trajectories of medical fitness among licensed aviation personnel and examined factors associated with membership in the two adverse trajectories. Obesity at the first examination and older age were associated with both adverse trajectories, whereas BMI change, fifth-examination smoking status and licence change showed no clear association. The association with age appeared stronger for persistent conditional fit than for progressive deterioration. This is consistent with a study of pilots at a Japanese airline, in which pilots aged ≥60 years had a higher incidence of cardiovascular disease and a lower rate of return to active duty than younger pilots.21 At the same time, a longitudinal study of European helicopter emergency medical services pilots found considerable interindividual variability in physiological risk markers among older pilots,22 indicating that older age does not uniformly lead to adverse fitness trajectories. In addition, age in this study was measured at the fifth examination and therefore reflects both age at entry and the time elapsed between examinations; the observed associations should not be interpreted as effects of ageing itself.
Among licence types, CPL holders had lower odds of persistent conditional fit than ATPL holders, whereas no clear association with licence type was observed for progressive deterioration. This association remained after adjustment for the other covariates. Studies of airline pilots have reported health effects of shift work and a high prevalence of cardiometabolic risk factors.4 23 However, the present examination records did not directly measure work schedules, occupational exposures or health surveillance practices and therefore cannot determine whether such conditions explain the observed difference between CPL and ATPL holders. Among ATCs, fatigue has been reported to be higher during night shifts,24 and shift type has been associated with body fat percentage and sleep duration.25 In the present study, however, estimates for ATCL and PPL holders had wide CIs, and flight hours were coded as 0 for ATCL holders and therefore did not represent flying experience in this group. The findings should thus be interpreted as associations with licence type rather than evidence of specific occupational mechanisms or protective effects of health management systems.
Obesity at the first examination was associated with both adverse trajectories, whereas overweight showed no clear evidence of an association. The BMI categories used in this study follow the Korean obesity criterion, which defines obesity as a BMI of 25 kg/m² or higher.17 Among commercial drivers undergoing mandatory medical examinations, obese drivers were less likely to receive full-length certification and more likely to report hypertension, diabetes mellitus and sleep disorders,26 and among oilfield workers, obesity was associated with hypertension and diabetes.27 In aviation, overweight and obesity have been reported in more than half of airline pilots.23 These findings suggest that obesity-related chronic conditions may provide one possible pathway linking obesity to adverse medical fitness patterns. However, disease variables were not included in the regression models, and this pathway was not tested in the present study. The association with first-examination obesity remained after BMI change was included in model 2, whereas BMI change itself showed no clear association with either trajectory. A longitudinal study of British Army pilots also reported increasing BMI over time.28 Thus, the revised analyses distinguish initial weight status from subsequent BMI change, without establishing that initial weight status is more important or that weight change is unrelated to medical fitness. BMI change was measured over the same period used to define the trajectories, which further limits its temporal interpretation. Taken together, the findings identify obesity as one factor associated with adverse medical fitness patterns, while the modest Nagelkerke pseudo R² values indicate that the models provide only a limited account of class membership. Obesity should therefore not be regarded as a sufficient explanation or a principal determinant of these trajectories.
Smoking status at the first examination showed opposite associations with the two adverse trajectories. Compared with current smoking, former smoking was associated with higher odds of persistent conditional fit but lower odds of progressive deterioration, whereas never smoking showed no clear association with either trajectory. These associations persisted after fifth-examination smoking status was included in model 2, and fifth-examination smoking status itself was not clearly associated with either trajectory. The opposite directions are compatible with two possible explanations described in studies of smoking cessation. First, people who develop health problems are more likely to stop smoking, which can make former smokers appear to be at higher risk; this has been referred to as the ‘sick quitter’ effect.29 In class 1, defined by persistent conditional fit classifications, former smoking may therefore partly reflect cessation following health problems rather than a direct effect of past smoking. In addition, former smokers have been reported to retain elevated cardiovascular risk relative to never smokers for many years after cessation,29 and among Korean adults, metabolic syndrome and inflammatory markers have been associated with the duration of smoking cessation and cumulative smoking exposure.30 31 Second, for progressive deterioration, current smoking at the first examination may have carried a higher risk of subsequent decline than former smoking; in large cohort studies, former smokers had lower cardiovascular risk than current smokers, although their risk remained elevated relative to never smokers.29 32 However, the timing of cessation relative to the examinations was not recorded, and neither explanation could be tested in the present study.
Psychiatric disorders and sleep disorders were infrequently recorded across examinations 1–5. In contrast, a study of commercial pilots during the COVID-19 pandemic reported that more than one-third exhibited symptoms of diagnosable mental health conditions.33 These findings are not directly comparable, because the populations, study periods and ascertainment methods differ; symptom-based assessments and conditions recorded in medical certification examinations measure different things. A narrative review also identified concerns about licence loss, stigma and confidentiality as barriers to help-seeking among pilots,34 providing a possible context for the low frequency of recorded psychiatric conditions. However, the present study did not assess mental health symptoms, help-seeking or nondisclosure and therefore cannot determine whether the low recorded frequency reflects underreporting. The findings should be interpreted as recorded conditions within the medical examination system rather than estimates of the underlying prevalence of psychiatric or sleep disorders.
The findings of this study may also be affected by selection bias, including the HWSE. As shown in table 1, participants with five or more examinations differed from those with fewer than five examinations at the first examination: they were older, more often held an ATPL or CPL and had greater total flight hours, whereas differences in the distribution of first-examination fitness classifications were small and unfit classifications were uncommon in both groups. Requiring at least five examinations may preferentially retain individuals able to continue undergoing medical examinations, a concern consistent with the HWSE.35–37 However, examination counts alone cannot distinguish health-related withdrawal from late entry or other reasons for fewer examinations. Prior studies have reported decreases in unfit classification rates and physiological risk markers among older pilots,22 38 providing context for possible health-related selection, although such selection was not established in the present study. If personnel with deteriorating health left before completing five examinations, the proportion of adverse trajectories may have been underestimated, and the estimated associations may also have been affected. Reasons for discontinuing examinations were not recorded, and the present analyses did not correct for potential selection bias. The findings should therefore be interpreted as medical fitness patterns among personnel with at least five recorded examinations, rather than as patterns representative of all aviation personnel.
Several limitations should be considered when interpreting these findings. First, the observational design and timing of covariate measurements limit causal interpretation. Although BMI and smoking status were measured at the first examination, age and licence type were measured at the fifth examination, and fifth-examination age also reflects the time elapsed between examinations, which varied across participants. Maximum flight hours, BMI change and fifth-examination smoking status overlapped with the period used to define the latent classes. The analyses incorporated selected changes and fifth-examination measurements rather than modelling covariates as time-varying exposures at each examination, and temporal precedence cannot be established for these overlapping measurements. Second, the available variables were limited to medical examination records. Sleep, occupational stress, detailed lifestyle behaviours and other potentially relevant factors were not sufficiently captured, and residual confounding by unmeasured factors cannot be excluded. Nagelkerke pseudo R² values were modest (0.181 and 0.182), indicating that the models provided only a limited account of class membership. Third, recorded diseases were excluded from the regression models because they inform the fitness classifications used to define the latent classes. Alternative analyses of baseline disease burden, incident diseases and diagnostic transitions were not undertaken, and the analyses did not establish how underlying disease processes contributed to the observed fitness patterns. Disease comparisons across classes were descriptive and unadjusted for age or sex; because the classes differed in age distribution, these differences may partly reflect age rather than trajectory membership. Fourth, smoking status was self-reported, and possible misclassification could not be assessed. Finally, the analytic sample was predominantly male, limiting exploration of sex-based differences, and the use of single-country data requires caution when generalising the findings to aviation medical systems in other national contexts.
Conclusions
This study used serial medical examination records collected from 2015 to 2022 to identify three longitudinal patterns of medical fitness among licensed aviation personnel with at least five examinations: sustained fit, persistent conditional fit and progressive deterioration. Older age and obesity at the first examination were associated with both adverse patterns, while associations with licence type and former smoking differed between the patterns. BMI change, fifth-examination smoking status and licence change showed no clear additional associations.
These findings illustrate how repeated medical examination records can characterise distinct fitness patterns. However, potential selection bias, overlapping measurement periods and the limited explanatory performance of the models restrict causal interpretation and generalisability. Further longitudinal research incorporating disease timing, occupational exposures and repeated health-related measurements is needed to clarify the processes underlying these patterns and inform preventive strategies.
Supplementary Material
Footnotes
Contributors: J-HC, E-HC and YYC conceived and designed the study. E-HC obtained the data, analysed and interpreted the data and drafted the manuscript. J-HC, E-HC, YYC and H-SJ critically revised the manuscript for important intellectual content. All authors approved the final version and agree to be accountable for all aspects of the work. E-HC is the guarantor. Claude (Anthropic) was used to assist with reducing the manuscript word count to meet journal requirements. All content was reviewed and approved by the authors, who take full responsibility for the accuracy and integrity of the work.
Funding: The authors have not declared a specific grant for this research from any funding agency in the public, commercial or not-for-profit sectors.
Competing interests: None declared.
Patient and public involvement: Patients and/or the public were not involved in the design, or conduct, or reporting, or dissemination plans of this research.
Provenance and peer review: Not commissioned; externally peer reviewed.
Data availability statement
Data may be obtained from a third party and are not publicly available. The data were obtained from the aviation medical examination database maintained by the Korean Society of Aerospace and Environmental Medicine (KSAEM) under a data use agreement and are not publicly available. Researchers wishing to access the data may apply to KSAEM with a data use plan, subject to approval by the KSAEM Executive Committee. The authors are not permitted to transfer or disclose the data to third parties.
Ethics statements
Patient consent for publication
Not applicable.
Ethics approval
This study was reviewed by the Institutional Review Board (IRB) of Eulji University and was determined to be exempt from full review (EUIRB2023-082). The data were limited to records collected through informed consent procedures at the time of examination and were provided in anonymised form, precluding individual identification. The research team pledged not to use the received data for commercial purposes and not to transfer or disclose them to third parties.
References
- 1.Minoretti P, Emanuele E. Health in the Skies: A Narrative Review of the Issues Faced by Commercial Airline Pilots. Cureus 2023;15:e38000. 10.7759/cureus.38000 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2.Yilmaz AA. Critical Connections: Network Analysis of Human Errors in Aviation Accidents. The International Journal of Aerospace Psychology 2025;35:203–29. 10.1080/24721840.2025.2531741 [DOI] [Google Scholar]
- 3.Bureau d’Enquêtes et d’Analyses pour la sécurité de l’aviation civile . Final report on the accident on flight 4u9525. BEA; 2016. Available: https://bea.aero/docspa/2015/f-cp150324.en.pdf [Google Scholar]
- 4.Stefańska O, Barbarska O, Minkiewicz-Zochniak A. Health Implications of Shift Work in Airline Pilots and Cabin Crew: A Narrative Review and Pilot Study Findings. Nutrients 2025;17:3906. 10.3390/nu17243906 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5.EASA, European Union Aviation Safety Agenc . Study on the analysis, prevention and management of air traffic controller fatigue. final report d4c1-v4.1. Cologne, 2024. Available: https://www.easa.europa.eu/sites/default/files/dfu/ATCO_Fatigue_Study_Final_report_D4C1_-_V4.1_final_21_May_2024.pdf [Google Scholar]
- 6.Vetter C, Devore EE, Ramin CA, et al. Mismatch of Sleep and Work Timing and Risk of Type 2 Diabetes. Diabetes Care 2015;38:1707–13. 10.2337/dc15-0302 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7.Minoretti P, D’Acquino D, Lavdas K, et al. Short Rest Between Flights Is Associated With Increased Serum Stress Hormone Levels in Airline Pilots: A Cross-Sectional Study. Cureus 2024;16:e69505. 10.7759/cureus.69505 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8.Wang P, Houghton R, Majumdar A. Detecting and Predicting Pilot Mental Workload Using Heart Rate Variability: A Systematic Review. Sensors 2024;24:3723. 10.3390/s24123723 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9.Peukert M, Claus L, Meyer L. Subjective and objective fatigue dynamics in air traffic control. Ind Health 2025;63:431–42. 10.2486/indhealth.2024-0206 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10.Bongo PP, Seva RR. Impact of fatigue on air traffic controllers’ performance. Int J Ind Ergon 2021;85:103228. 10.1016/j.ergon.2021.103228 [DOI] [Google Scholar]
- 11.International Civil Aviation Organization . Annex 1 to the convention on international civil aviation: Personnel licensing. 14th edn. Montreal: ICAO, 2022. [Google Scholar]
- 12.Ministry of Land, Infrastructure and Transport . Aviation safety act, article 42. National Law Information Center; 2022. Available: https://www.law.go.kr/법령/항공안전법 [Google Scholar]
- 13.Han BS, Kwon YH, Shin YY. Analysis of Three Years of Airmen Medical Certificate for Preparing Aviation-related Health Promotion Plan. Korean J Aerosp Environ Med 2020;30:40–9. 10.46246/KJAsEM.30.1.40 [DOI] [Google Scholar]
- 14.Kwon YH. Survey of Medically Disqualified Airline Pilots in Korea: 2015~2019. Korean J Aerosp Environ Med 2020;30:70–4. 10.46246/KJAsEM.30.2.70 [DOI] [Google Scholar]
- 15.Gunduz SH, Metin S. Medical reasons for permanent and temporary disqualification of Turkish civil aviation pilots. Arch Environ Occup Health 2024;79:113–20. 10.1080/19338244.2024.2359416 [DOI] [PubMed] [Google Scholar]
- 16.Høva JK, Thorheim L, Wagstaff AS. Medical Reasons for Loss of License in Norwegian Professional Pilots. Aerosp Med Hum Perform 2017;88:146–9. 10.3357/AMHP.4551.2017 [DOI] [PubMed] [Google Scholar]
- 17.Haam J-H, Kim BT, Kim EM, et al. Diagnosis of Obesity: 2022 Update of Clinical Practice Guidelines for Obesity by the Korean Society for the Study of Obesity. J Obes Metab Syndr 2023;32:121–9. 10.7570/jome23031 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18.Lo Y, Mendell NR, Rubin DB. Testing the number of components in a normal mixture. Biometrika 2001;88:767–78. 10.1093/biomet/88.3.767 [DOI] [Google Scholar]
- 19.Nylund KL, Asparouhov T, Muthén BO. Deciding on the Number of Classes in Latent Class Analysis and Growth Mixture Modeling: A Monte Carlo Simulation Study. Structural Equation Modeling: A Multidisciplinary Journal 2007;14:535–69. 10.1080/10705510701575396 [DOI] [Google Scholar]
- 20.Jung T, Wickrama KAS. An Introduction to Latent Class Growth Analysis and Growth Mixture Modeling. Social & Personality Psych 2008;2:302–17. 10.1111/j.1751-9004.2007.00054.x [DOI] [Google Scholar]
- 21.Takazoe K, Gomi H. The Health Status of Pilots Over Age 60 at a Japanese Airline. Aerosp Med Hum Perform 2025;96:392–8. 10.3357/AMHP.6582.2025 [DOI] [PubMed] [Google Scholar]
- 22.Bauer H, Nowak D, Herbig B. Aging and Cardiometabolic Risk in European HEMS Pilots: An Assessment of Occupational Old-Age Limits as a Regulatory Risk Management Strategy. Risk Anal 2018;38:1332–47. 10.1111/risa.12951 [DOI] [PubMed] [Google Scholar]
- 23.Wilson D, Driller M, Johnston B, et al. The Prevalence of Cardiometabolic Health Risk Factors among Airline Pilots: A Systematic Review. Int J Environ Res Public Health 2022;19:4848. 10.3390/ijerph19084848 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 24.Chang YH, Yang HH, Hsu WJ. Effects of work shifts on fatigue levels of air traffic controllers. Journal of Air Transport Management 2019;76:1–9. 10.1016/j.jairtraman.2019.01.013 [DOI] [Google Scholar]
- 25.Sonati J, De Martino M, Vilarta R, et al. Quality of Life, Health, and Sleep of Air Traffic Controllers with Different Shift Systems. Aerosp Med Hum Perform 2015;86:895–900. 10.3357/AMHP.4325.2015 [DOI] [PubMed] [Google Scholar]
- 26.Thiese MS, Moffitt G, Hanowski RJ, et al. Commercial Driver Medical Examinations: Prevalence of Obesity, Comorbidities, and Certification Outcomes. J Occup Environ Med 2015;57:659–65. 10.1097/JOM.0000000000000422 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 27.Yu X, Zhang S, Liu H, et al. Overweight or obesity and their association with cardiometabolic risk factors among oilfield workers in Chinese population: a cross-sectional study. Front Public Health 2026;14:1658235. 10.3389/fpubh.2026.1658235 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 28.Porter WD, Wilde GD, Jeffery NP, et al. Longitudinal Changes in the Body Mass Index of British Army Pilots. Aerosp Med Hum Perform 2024;95:118–22. 10.3357/AMHP.6356.2024 [DOI] [PubMed] [Google Scholar]
- 29.Tasdighi E, Yao Z, Dardari ZA, et al. Association between cigarette smoking status, intensity, and cessation duration with long-term incidence of nine cardiovascular and mortality outcomes: The Cross-Cohort Collaboration (CCC). PLoS Med 2025;22:e1004561. 10.1371/journal.pmed.1004561 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 30.Shin HS, Oh JE, Cho YJ. The Association Between Smoking Cessation Period and Metabolic Syndrome in Korean Men. Asia Pac J Public Health 2018;30:415–24. 10.1177/1010539518786517 [DOI] [PubMed] [Google Scholar]
- 31.Cho JH, Lee DC, Lee HJ. Association of Duration of Smoking Cessation or Cumulative Smoking Amount with Serum hs-CRP Level in Korean Adults: A Nationwide Population-Based Cross-Sectional Study. Toxics 2022;10:533. 10.3390/toxics10090533 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 32.Duncan MS, Freiberg MS, Greevy RA, et al. Association of Smoking Cessation With Subsequent Risk of Cardiovascular Disease. JAMA 2019;322:642–50. 10.1001/jama.2019.10298 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 33.Ackland CA, Molesworth BRC, Grisham JR. Pilot mental health during the COVID-19 pandemic: prevalence rates from semi-structured interviews, and associated vulnerability and protective factors. Front Psychol 2023;14. 10.3389/fpsyg.2023.1073857 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 34.Minoretti P. Barriers and Facilitators to Mental Health Support Among Airline Pilots: A Narrative Review. Cureus 2025;17. 10.7759/cureus.91340 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 35.Buckley JP, Keil AP, McGrath LJ, et al. Evolving methods for inference in the presence of healthy worker survivor bias. Epidemiology 2015;26:204–12. 10.1097/EDE.0000000000000217 [DOI] [PubMed] [Google Scholar]
- 36.Garcia E, Picciotto S, Costello S, et al. Assessment of the healthy worker survivor effect in cancer studies of the United Autoworkers-General Motors cohort. Occup Environ Med 2017;74:294–300. 10.1136/oemed-2016-104038 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 37.Brown DM, Picciotto S, Costello S, et al. The Healthy Worker Survivor Effect: Target Parameters and Target Populations. Curr Environ Health Rep 2017;4:364–72. 10.1007/s40572-017-0156-x [DOI] [PMC free article] [PubMed] [Google Scholar]
- 38.Cairns MK. Unfit Assessments of Class 1 Medical Certificate Holders. Aerosp Med Hum Perform 2021;92:945–9. 10.3357/AMHP.5881.2021 [DOI] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Data Availability Statement
Data may be obtained from a third party and are not publicly available. The data were obtained from the aviation medical examination database maintained by the Korean Society of Aerospace and Environmental Medicine (KSAEM) under a data use agreement and are not publicly available. Researchers wishing to access the data may apply to KSAEM with a data use plan, subject to approval by the KSAEM Executive Committee. The authors are not permitted to transfer or disclose the data to third parties.