0084 Differently Worded Questions Do Not Substantially Influence Subject Predictions of Performance

Abstract Introduction Being able to accurately predict when one’s performance is likely to be impaired by sleep loss can help avoid costly errors. However, the literature is mixed on how accurately individuals are able to make such predictions. Subjective predictions of performance made during total sleep deprivation (TSD) are often only as accurate as judgments of subjective sleepiness, which are poorly correlated. We investigated whether the wording of a question about anticipated performance on a Psychomotor Vigilance Task (PVT) influences the ability to predict actual performance during TSD. Methods N=56 healthy adults (ages 25.9±5.3y, 32 females) were randomly assigned to 38h TSD in a 4-day/3-night laboratory study. The PVT and the Karolinska Sleepiness Scale (KSS) were administered approximately every 2-4h during scheduled wakefulness. We created four differently worded questions to elicit ratings of predicted performance on a scale from 1 (worse performance) to 9 (better performance). The questions asked about expected performance at the beginning or end of the PVT, expected duration of good performance, or expected time until impairment. Participants completed the KSS and answered one of the questions (randomly selected), and then performed the PVT. Using separate linear regressions with a fixed effect for rating and a random intercept over participants, we examined whether any of the questions, or the KSS ratings, were better able to predict PVT lapses (RT>500ms). Results were compared using root mean square error (RMSE), a measure of absolute model fit. Results Using the predictive accuracy of the KSS as a reference (RMSE=4.71), the RMSE values for the questions ranged from 10.0% lower (RMSE=4.24), indicating slightly better accuracy for predicting subsequent PVT performance, to 13.0% higher (RMSE=5.32), indicating slightly worse accuracy for predicting subsequent PVT performance. Conclusion Our results suggest that the wording of a question about anticipated performance on the PVT influences its predictive utility, but with no more than 10% improvement over KSS-rated subjective sleepiness as a performance predictor, the predictive gain was not substantial. This puts into question whether self-predictions of performance, regardless of how they are framed, can be relied upon for assessments of readiness in operational settings. Support (if any) CDMRP W81XWH-16-1-0319