Perception of Excessive Sleepiness and Severe Fatigue in Civil Aviation Pilots linked to Irregular Duty Periods

This study assess the relationship between irregular duty periods and labour risk using the self reported perception of sleepiness and fatigue as proxies, through the Karolinska Sleepiness Scale, and the Sam-Perelli Fatigue Scale, respectively at the beginning, middle and end of duty periods or workloads. The study was carried out longitudinally over two semesters and it was evaluated using a cluster-specific binomial model with logistic link.

This repository includes:

• Data: Raw data collected from pilot surveys, actimetry, and other sources (upon request).
• Code: Scripts for data cleaning, analysis, and visualization using relevant Python libraries.
• Analysis: Reports and documentation detailing the methodology, results, and conclusions.
• Jupyter Notebooks: Interactive python Markdown documents for exploring the data and results.

The Jupyter Notebooks follows the order:

• Data preparation, feature extratction and data cleaning feature_engineering.ipynb
• Exploratory Data Analysis EDA.ipynb
• Results related to perception of sleepiness regression_sleepiness.ipynb
• Results related to perception of fatigue regression_fatigue.ipynb

This project aims to contribute to the understanding of work and personal related factors leading to perception of excessive/severe sleepiness/fatigue in civil aviation pilots.

---

Why this project is worth?

Most psychometric studies focusing on the perceptions of sleepiness and/or fatigue using the aforementioned scales treat the response variables as continuous, overlooking their ordinal nature. Due to the scarcity of observations and inspired by the versatility and easy interpretability of binomial models, we have transformed these scales into binary outcomes that denote extreme conditions of the variables. This approach leads to more quantitative and representative results.

---

Detailing

Civil aviation pilots are one of professionals with the most irregular work scale, in order to quantify their labour risk, the present study was conducted longitudinally on 48 participants between the austral summer and autumn of 2022, wherein each participant self-reported their perceived sleepiness and fatigue using the Karolinska and Sam-Perelli scales, respectively at start, middle and end of duty periods. Both the perceived excessive sleepiness and perceived severe fatigue are used as a proxy for pilots' labour risk. Time varying metrics were obtained from punch clock, actimetry and sleep diaries and time invariant metrics such as sociodemographic data and chronotype were obtained through application of questionnaries. The perception of sleepiness and fatigue were collected together using google questionnaires. More than 30 features were analysed and engineered. By employing cluster-specific binomial models with logistic link, we found that duties comprehending periods outside the circadian time (hours outside business hours, Early-Start and Night periods), longer workloads, many hours awake before the start of duty and sleep deprivation are the main enhancement factors for percecption of excessive/severe sleepiness/fatigue. Pilots that have better general work performance at mornings (chronotypes matutines) and pilots in stable union are found to be protection factors. Tailored interventions and potential regulatory adjustments are crucial for pilots’ well-being in order to avoid labour risks associated to sleepiness and fatigue.

---

Main Results

The figures above show the coefficients obtained from regression analysis represented as odds ratio, each effect may be interpreted as the effect of covariate keeping all other covariates constant and the orange bars represent 95% confidence bounds. The baselines for "duty period" is the start of duty, for "hours" is business hours defined as times between 08:00 - 23:59, for "matutine chronotype" is intermediaries and vespertine chronotypes and for "matutines and intermediaries chronotypes" the baseline is vespertine chronotype.

• The more prone individual in the study has 2.29 times more odds to perceive severe fatigue than the less prone.

• The more prone individual in the study has 2.09 times more odds to perceive excessive sleepiness than the less prone.

---

Models Assessment

Perception of Excessive Sleepiness

Above is shown the performance analysis for the model used to assess the perception of severe sleepiness. Once the objective is only to study the effects of covariates in the perception of excessive sleepiness and fatigue, the Area Under the Curve (AUC) is enough for this purpose. As we can see the AUC value is 0.85, following Hosmer & Lemeshow (2013), this can be taken as a good discrimator and a satisfactory model.

Perception of Severe Fatigue

For the model to assess the perception of severe sleepiness, it is found the AUC with value of 0.93, following Hosmer & Lemeshow (2013) it can be taken as an outstanding discriminator and a satisfactory model.

---

Model Description

The binomial mixed model with logistic link can be described as:

$$ g(x_{ij},\beta_{0i},\beta_{s}) = \beta_{0i}+\boldsymbol{x_{ij}^{\prime}}\boldsymbol{\beta_s} $$

$$ \beta_{0i} = \beta_0 + \alpha_i $$

$$ \alpha_i \sim N(0,\sigma_{\alpha}^{2}) $$

where $\beta_0$ is the intercept and $\alpha_i$ is the random intercept due to participant clusters.

The logistic link function is given by:

$$ g(x_{ij},\beta_{0i},\beta_{s}) = ln\left[\frac{\pi(x_{ij},\cdot)}{1-\pi(x_{ij},\cdot)}\right] $$

with,

$$ \pi(x_{ij},\cdot) = \frac{e^{\beta_{0i}+\boldsymbol{x_{ij}^{\prime}}\boldsymbol{\beta_s}}}{1+e^{\beta_{0i}+\boldsymbol{x_{ij}^{\prime}}\boldsymbol{\beta_s}}} $$

---

References

This case study is a reformulation of a consultancy provided by the owner of the repository to the Public Health Faculty of the University of São Paulo through the Centre for Applied Statistics of the University of São Paulo and can be accessed at Repositorio USP