Format

Send to

Choose Destination
J Sleep Res. 2017 Dec;26(6):820-831. doi: 10.1111/jsr.12535. Epub 2017 Apr 24.

Real-time individualization of the unified model of performance.

Author information

1
Department of Defense Biotechnology High Performance Computing Software Applications Institute, Telemedicine and Advanced Technology Research Center, US Army Medical Research and Materiel Command, Fort Detrick, MD, USA.
2
Behavioral Biology Branch, Walter Reed Army Institute of Research, Silver Spring, MD, USA.

Abstract

Existing mathematical models for predicting neurobehavioural performance are not suited for mobile computing platforms because they cannot adapt model parameters automatically in real time to reflect individual differences in the effects of sleep loss. We used an extended Kalman filter to develop a computationally efficient algorithm that continually adapts the parameters of the recently developed Unified Model of Performance (UMP) to an individual. The algorithm accomplishes this in real time as new performance data for the individual become available. We assessed the algorithm's performance by simulating real-time model individualization for 18 subjects subjected to 64 h of total sleep deprivation (TSD) and 7 days of chronic sleep restriction (CSR) with 3 h of time in bed per night, using psychomotor vigilance task (PVT) data collected every 2 h during wakefulness. This UMP individualization process produced parameter estimates that progressively approached the solution produced by a post-hoc fitting of model parameters using all data. The minimum number of PVT measurements needed to individualize the model parameters depended upon the type of sleep-loss challenge, with ~30 required for TSD and ~70 for CSR. However, model individualization depended upon the overall duration of data collection, yielding increasingly accurate model parameters with greater number of days. Interestingly, reducing the PVT sampling frequency by a factor of two did not notably hamper model individualization. The proposed algorithm facilitates real-time learning of an individual's trait-like responses to sleep loss and enables the development of individualized performance prediction models for use in a mobile computing platform.

KEYWORDS:

Bayesian learning; individualized modelling; performance prediction; recursive parameter estimation

PMID:
28436072
DOI:
10.1111/jsr.12535
[Indexed for MEDLINE]

Supplemental Content

Full text links

Icon for Wiley
Loading ...
Support Center