Model predicts workplace vigilance from single sleep using under-mattress sensor

1. A machine learning model was developed for estimating vigilance impairment for high-risk shift work occupations using only data from a single prior sleep period.

2. The model was found to be comparable to current fatigue prediction models that relied upon self-reported data and multi-night measurements.

Evidence Rating Level: 1 (Excellent)

The ability to predict vigilance impairment in shift workers is critical to reducing workplace errors and accidents. Current methods for evaluating vigilance involve infeasible self-reported data and laborious multi-night measurements. This study aimed to evaluate a new model for predicting shift work vigilance based on sleep data collected only one day prior using an under-mattress sleep sensor.

Twenty-four participants were recruited for the study using local advertisements. Individuals were included if they were between 18-65 years old, had no symptoms of a sleep disorder, had a typical bedtime of 22:00-00:30, a typical wake time of 07:00-09:00, had a sleep duration between 6 and 9 hours, consumed ≤3 caffeinated beverages per day, consumed ≤10 standard drinks per week, were non-smokers, had no current use of medications or drugs that affected sleep, had no shift work or international travel in the previous 6 months, and had no history of psychiatric, neurologic, cardiac, or respiratory disorders. Study participants came to the laboratory for two 8-day visits, 1 month apart. Each visit began with a flip to a night-shift schedule, involving simulated tasks from 12 am to 8 am. A psychomotor vigilance task (PVT) was administered 6 times throughout the night to evaluate reaction times, reaction speed, and number of lapses in attention. The Withings sleep analyser (WSA) was used underneath the mattress to estimate total sleep time, sleep efficiency, and wake after onset. This data was then utilized in machine learning models to predict PVT performance based on preceding sleep data. The primary outcome measure was reaction time and error on a psychomotor vigilance task.

The final model demonstrated moderate accuracy for predicting PVT performance and was found to be comparable with current models that typically require at least 5 days of consecutive data.
Although the sample size was relatively small, the model was still able to predict fatigue from a single preceding sleep recording, highlighting its efficiency and feasibility. The ability to easily and accurately estimate the vigilance of shift workers will be a valuable tool to manage work schedules to reduce workplace errors.

Click here to read this study in Journal of Sleep Research

Image: PD

©2024 2 Minute Medicine, Inc. All rights reserved. No works may be reproduced without expressed written consent from 2 Minute Medicine, Inc. Inquire about licensing here. No article should be construed as medical advice and is not intended as such by the authors or by 2 Minute Medicine, Inc.