Abstract
There are strong individual differences in performance during sleep deprivation. We assessed whether baseline features of Psychomotor Vigilance Test (PVT) performance can be used for classifying participants' relative attentional vulnerability to total sleep deprivation. In a laboratory, healthy adults (n = 160, aged 18-30 years) completed a 10-min PVT every 2 h while being kept awake for ≥24 hours. Participants were categorized as vulnerable (n = 40), intermediate (n = 80), or resilient (n = 40) based on their number of PVT lapses during one night of sleep deprivation. For each baseline PVT (taken 4-14 h after wake-up time), a linear discriminant model with wrapper-based feature selection was used to classify participants' vulnerability to subsequent sleep deprivation. Across models, classification accuracy was about 70% (range 65-76%) using stratified 5-fold cross validation. The models provided about 78% sensitivity and 86% specificity for classifying resilient participants, and about 70% sensitivity and 89% specificity for classifying vulnerable participants. These results suggest features derived from a single 10-min PVT at baseline can provide substantial, but incomplete information about a person's relative attentional vulnerability to total sleep deprivation. In the long term, modeling approaches that incorporate baseline performance characteristics can potentially improve personalized predictions of attentional performance when sleep deprivation cannot be avoided.