Flight Attendant Work/Rest Patterns, Alertness, and Performance Assessment: Field Validation of Biomathematical Fatigue Modeling

Fatigue-induced impairments in neurobehavioral performance capacity may compromise safety in 24-hr operational environments, and developing reliable and valid methods of identifying work/rest patterns that produce fatigue and undermine performance is important. One approach is the use of biomathematical modeling as a means of predicting, preventing, and mitigating fatigue-induced safety risks. The Sleep, Activity, Fatigue, and Task Effectiveness model (SAFTE; Hursh et al., 2004) is among the more mature fatigue models currently used in military, shift-work, and various transportation operations. The SAFTE model was constructed empirically, integrating classical physiological and circadian processes with task effectiveness predictions based on the scientific literature of standardized laboratory tests. SAFTE has been validated against accident risk in railroad operations; however, as with virtually all fatigue models, the extent to which variations in model predictions correspond to variations in actual performance capacity in the aviation environment is largely unknown. The present report offers a field validation of the SAFTE model using data from a broad sample of 178 aviation cabin crew from the 2009-2010 Federal Aviation Administration Civil Aerospace Medical Institute (CAMI)-sponsored Flight Attendant Field Study (Roma et al., 2010). Data were collected daily throughout each individual’s continuous 3 to 4-week study period. Objective sleep/wake patterns were determined via actigraphy. In addition, a personal digital assistant device was used to maintain an activity log documenting work schedules and locations, and neurobehavioral performance capacity was assessed via standardized 5-min Psychomotor Vigilance Tests (PVT) taken before and after each work day and sleep episode. Individual sleep, wake, and work patterns were entered into the Fatigue Avoidance Scheduling Tool (FAST) software for continuous records of Predicted Effectiveness (PVT Speed [1/Reaction Time] expressed as a % of individual optimum baseline). SAFTE-FAST performance predictions were then temporally aligned with the 10,659 valid PVT test sessions from the field study, and performance data from each session were expressed as Actual Effectiveness (same as Predicted Effectiveness), Reaction Time (RT), Speed (1/RT), Lapses (RTs>500 msec), and False Starts (FS; premature responses). Linear regression of mean PVT performances across 5% SAFTE prediction bins revealed significant correlations between SAFTE Predicted Effectiveness and PVT Actual Effectiveness (R²=0.884, p<.001), RT (R²=0.745, p<.01), and Lapses (R²=0.486, p<.05). Identical analyses of the 7,533 valid PVT sessions completed while away on a multi-day work “trip” (i.e., excluding sessions while off-duty at home) revealed significant correlations between SAFTE Predicted Effectiveness and mean PVT Actual Effectiveness (R²=0.889, p<.001), RT (R2=0.819, p<.001), Speed (R²=0.808, p<.001), and Lapses (R²=0.484, p<.05). Finally, separate regression analyses of all valid Pre-Work (n=1,712) and Post-Work (n=1,934) PVT sessions revealed a significant Pre-Work correlation between SAFTE Predicted Effectiveness and mean PVT Actual Effectiveness (R²=0.530, p<.05), and significant Post-Work correlations between SAFTE Predicted Effectiveness and mean PVT Actual Effectiveness (R²=0.600, p<.05), RT (R²=0.887, p<.001), Speed (R²=0.539, p<.05), and Lapses (R²=0.901, p<.001). Despite inherent technical limitations and issues of inter-individual variability, these results clearly support the validity of the SAFTE model for population-level prediction of fatigue-induced impairments in objective neurobehavioral performance capacity in extremely dynamic 24-hr field operations such as commercial aviation.