Pilot Fatigue and Sleepiness: Model Predictions with and Without Actigraphy Data Versus Self-Rated Samn-Perelli and Karolinska Sleepiness Scale Scores

Introduction: Fatigue and sleepiness play a crucial role in aviation as they can affect several pilot skills, such as cognitive performance, situation awareness, communication, and decision making. Fatigue risk management systems are usually based upon several components, including aircrew fatigue reactive reports and proactive and predictive approaches based mostly on bio-mathematical models. In this study, the predictions of fatigue and sleepiness from the SAFTE-FAST model [1] were compared with self-rated Samn-Perelli (SP) and Karolinska Sleepiness Scale (KSS) scores from Brazilian airline pilots [2].

Methods: The sample included 51 airline pilots that rated their perceived fatigue and sleepiness three times a day (at the beginning, middle, and end of the work shift), using SP and KSS scales, respectively, for 15 consecutive days. The pilots also wore actigraphs and rated their perceived sleep quality in sleep diaries throughout data collection [2]. SAFTE-FAST calculations were performed either using: (a) a standard set of inputs, which included all the auto-sleep functions of the model with all sleep events considered to be of excellent quality, and (b) a tailored set of inputs that considered more realistic parameters, such as actual sleep from actigraphy and sleep quality from the pilot (diary) reports. In both calculations, the exact date and time of pilot responses were considered, yielding a total of 1,583 validated assessments.

Results: The preliminary results indicated statistically relevant group effects (p < 0.001) comparing the SAFTE-FAST Effectiveness (ESF) for the standard (90.07 ± 0.21%) and tailored (82.82 ± 0.34%) calculations using the Wilcoxon signed-rank test for paired samples. Strong Pearson´s correlations (p < 0.001) were also observed between ESF and SP (KSS) either for the standard rho=-0.204 (rho=-0.292) or for the tailored rho=-0.297 (rho=-0.281) simulations. The risk ratio for low Effectiveness scores (ESF ≤ 77%) was 2.604 (95% CI: 2.177–3.115) comparing the calculations for tailored (actigraphy data) versus the model standard sleep inputs, which presented 23.7 and 9.1% of low scores, respectively.

Conclusion: These findings highlight the importance of adopting model inputs as closely as possible to specific operational scenarios and real worker conditions to obtain more accurate and realistic results. Moreover, current bio-mathematical models do not fully consider the effects of workload on fatigue and sleepiness [3], which may cause further systematic inaccuracies and sub estimations of fatigue and sleepiness scores in operational contexts. References: [1] Hursh, S. R., et al. (2004). Fatigue models for applied research in warfighting. Aviat. Space Environ. Med 75(3), A44-A53. [2] Sampaio, I. T. A. (2023). Regulação e trabalho em jornadas irregulares: o caso de pilotos brasileiros. Implicações para o trabalho e para a saúde. PhD thesis, USP. [3] Rodrigues, T. E., et al (2024). Aircrew rostering workload patterns and associated fatigue and sleepiness scores in short/medium haul flights under RBAC 117 rules in Brazil. arXiv preprint arXiv:2408.08889. Support: The licenses for the SAFTE-FAST software are equally supported by the Brazilian Association of Civil Aviation Pilots (ABRAPAC), Gol Aircrew Association (ASAGOL) and LATAM Aircrew Association (ATL); F.M. Fischer receives a productivity grant from CNPq (306963/2021–3).