Abstract :
[en] Introduction
Performance of people undergoing critical tasks (like driving) may be impaired completely by the lowering of their vigilance level, due to sleep deprivation for instance. This reduction of performance may be measured by metrics computed from the reaction times (RT) recorded during a 10min Psychomotor Vigilance Test (PVT).
Here, we analyze and compare the sensitivity to sleep deprivation and the subject dependent variability of the PVT metrics performance, with a special emphasis on the time interval sizes.
Materials and Methods
Individuals (22 volunteers; 11 males, 11 females, mean 22.2y., range 19-34 years) follow an uninterrupted 28h sleep deprivation standard PVT protocol during which they achieved two groups of three PVT sessions (in different conditions). The first PVT of each group is in Non-SDP condition (9h30 and 10h30 Day 1) and the second and third PVT of each group are in SDP condition (2h30, 3h30, 10h30 and 11h30 day 2). The subjects fill a sleep journal during the week before the experiment. We checked that they had a normal sleep-wake cycle with no sleep deprivation, jet-lag or shift work and no medication. During the PVT of the first group, the subjects were wearing the glasses of the Phasya’s Drowsimeter.
We compute the usual PVT metrics; meanRT, meanRS (Reaction Speed) and LN500 (500ms lapses number). We also compute LNQ25 (adaptive lapses number computed with a subject dependent threshold) and ELN (Expected Lapse Number, computed from a subject-dependent estimation of the lapse probability).
Results
We use the “Effect Size” (ES, ratio of the mean by the standard deviation of the difference of metrics in the SDP and Non-SDP conditions) to assess the sensitivity to sleep deprivation. For the 10min (resp. 1min, 3min) interval, the ES of LNQ25 and ELN are respectively 1.38 (resp. 0.95, 1.22) and 1.35 (resp. 0.85, 1.14), the ES of meanRS, meanRT and LN500 are 1.23 (resp. 0.91, 1.09), 0.81 (resp. 0.54, 0.68) and 0.85 (resp. 0.63, 0.77).
We classify the intervals on which metrics are computed as SDP or non-SDP. We use a fixed threshold for the metrics, independent of the subject. In the ROC curves, the TPR (for a FPR of 10%) assesses the quality of the classification, and then also the subject independence. For the 10min (resp. 1min, 3min) interval, the TPR of LNQ25 and ELN are respectively 0.86 (resp. 0.56, 0.75) and 0.83 (resp. 0.58, 0.75), the TPR of meanRS, meanRT and LN500 are 0.42 (resp. 0.38, 0.41), 0.40 (resp. 0.39, 0.40) and 0.42 (resp. 0.24, 0.30).
Conclusions
We demonstrate that LNQ25 and ELN enable a quite good classification of the SDP condition for time intervals greater than or equal to 3min, independently of the subject. On the other hand, these metrics provide also a good sensitivity to sleep deprivation. They outperform the usual metrics for both criteria. For time intervals below 3min, the performances degrade first progressively and then much more quickly below 1min.
