Real-Time Caffeine Optimization During Total Sleep Deprivation

Overview

Sleep deprivation (SD) has a powerful degrading effect on cognitive performance, particularly psychomotor vigilance (PV) and reaction time. Caffeine is well known to be an effective countermeasure to the effects of SD. However, individuals differ in both their response to SD and to the administration of caffeine. This has made it difficult to provide individualized recommendations regarding the use of caffeine to sustain alertness when needed. For the past two decades, the Army's Biotechnology HPC Institute (BHSAI), in collaboration with the Walter Reed Army Institute of Research, have been developing statistical models to predict individual performance during prolonged SD. Recently, this resulted in the publication of the 2B-Alert app, a computer algorithm based on large datasets that can learn an individual's response to SD by combining actigraphic sleep data with simultaneously acquired PV performance data. The 2B-Alert algorithm can predict an individual's sleep need and performance after \~2 weeks of training the model. Recently, the model has been extended to incorporate individualized responses to caffeine. This was recently validated in a retrospective study published by BHSAI in 2019. The present study is designed to test the predictive capacity of the 2B-Alert app in real time. During Phase 1 a total of 21 healthy participants will wear an actigraph \& complete multiple daily PV tests on a personal cell phone. After 2 weeks, these individuals will attend Phase 2 involving an in-laboratory stay \& SD. Participants will have an 8-hour period of sleep in the laboratory, followed by 62 hours of continuous wakefulness. During these 62 hours, participants will complete PV and mood testing every 3 hours. The 2B-Alert app will be used to predict individual caffeine need to sustain performance at near-baseline levels based on the statistical model. At 44 hours SD, participants will undergo a 6-hour "alertness window" where they may receive individualized doses of caffeine based on the recommendations of the model. After 62 hours of SD, Phase 3 begins, involving a night of monitored recovery sleep and additional sessions of PV and mood testing until release from the study at 6 pm on the final day. It is hypothesized that the 2B-Alert app will be effective at providing caffeine dosing recommendations that return PV and mood performance to normal levels during the alertness window.

The first objective of the present study is to determine whether the 2B-Alert Caffeine Optimization Model (2BAlert) can create individualized caffeine schedules that effectively recover PVT performance during a specified window of time. The 2BAlert model is the linchpin of the comprehensive fatigue management system being developed by Walter Reed Army Institute of Research (WRAIR) and Biotechnology High Performance Computing Software Applications Institute (BHSAI). It is a software tool that "learns" (quantifies) the relationship between an individual's sleep/wake parameters (as objectively measured via wrist actigraphy) and his/her psychomotor performance (as objectively measured on a PVT). It quantifies the extent to which an individual is sensitive/resilient to the effects of sleep loss, and produces individualized performance predictions that can be used to inform decisions regarding current and future readiness, as well as the application of fatigue countermeasures such as naps or caffeine. The most recent version of 2BAlert can apply fatigue countermeasures directly to an individual based on their individualized performance prediction. This model gives recommendations for when and how much caffeine to use in order to optimize performance during a specified period of time. While this model has been applied post-hoc to previously collected data, it has yet to be tested in real-time. This study will provide the first opportunity to directly test if the model can be effectively used, in real-time, to recover PVT performance to a desired level (i.e., 275ms) when required (i.e., a six hour period following 44 hours of sleep loss). A second objective of the present study is to investigate if 2BAlert can not only recover PVT performance during a specified period, but also recover increases in self-reported stress and anxiety related to sleep-loss. Data from a previous study by WRAIR shows that self-reported stress and anxiety increase after 1 night of sleep loss and continues to increase after a second night of continuous sleep deprivation. These measures return to baseline following 12 hours of recovery sleep and map directly onto PVT performance. Therefore, the investigators hypothesize that if PVT performance can be recovered by caffeine, self-reported stress and anxiety can also return close to baseline levels with caffeine. A tertiary objective of the study is to assess whether images acquired using a smartphone camera are suitable for developing a passive, non-intrusive computer-vision system that could substitute the PVT for assessing alertness. Currently, the algorithm uses PVT data because: (a) compared to other performance measures, the PVT is relatively sensitive to sleep loss and the circadian rhythm of alertness; and (b) there are no learning effects on the PVT. However, the PVT requires an individual to actively engage in a 3- to 10-min long test, which must be performed at least a dozen times during sleep deprivation, making it a sensitive but impractical test to measure a Soldier's alertness in operational settings. All participants will participate in the following continuous study phases. Caffeine gum may be administered during this study. Phase 1: At Home Sleep/Wake Measurement: All participants will be instructed to maintain their normal sleep/wake schedule for the 13 days/12 nights immediately preceding phase 2 (the in-laboratory portion of the study). Compliance will be verified objectively via wrist actigraphy. Participants will be given a smartphone and asked to complete a PVT on it every 3 hours while awake and log their normal caffeine use as well as daily sleep duration. Phase 2: In-Laboratory Sleep/Wake and PVT Performance Assessment: Participants will report to the sleep lab at 1900 hrs on Day 13. While awake in the laboratory, they will complete various cognitive tests, including the PVT, and record a 3-minute video of their face every 3 hours. They will be allowed to sleep from 2300 hrs on Day 13 until 0700 hrs on Day 14 and sleep will be monitored via PSG and actigraphy. This will be followed by 62 hours of continuous wakefulness (i.e., from 0700 hrs on Day 14 until 2100 hrs on Day 16). At 1900 on Day 15, 2BAlert will be run for each individual and will use each individual's previous sleep and performance to create an individualized caffeine dosing schedule to optimize performance from 0300-0900 on Day 16. Phase 3: In-Laboratory Recovery: All participants will undergo a recovery phase consisting of 12 hours time in bed from 2100 hrs on Day 16 until 0900 hrs on Day 17 \& sleep will be monitored via PSG and actigraphy. The cognitive testing and facial video recording schedule will continue from Phase 2 during wake hours. After being evaluated by a study physician they will be released from the study at approximately 1800 hrs on Day 17. Thus, participants will be in the laboratory for a total of 95 hours. The main endpoint for this study is psychomotor vigilance test performance (PVT), measured using Smart-PVT. A secondary endpoint is self-reported stress and anxiety, as measured by the Stress Visual Analog Scale and the Spielberger State-Trait Anxiety Inventory, respectively. A third endpoint of the study is to test if facial images captured with the phone camera could be used to assess the alertness level of subjects under sleep deprivation. Hypotheses to be tested: (a) Psychomotor performance data collected on the Smart-PVT will stay at or below 275ms during the Peak Alertness Window of optimized performance thus indicating that 2BAlert is valid and ready to be utilized in future field studies and operational settings. (b) Self-reported stress and anxiety will return to baseline levels during the Peak Alertness Window and map onto the recovery of PVT performance during this time window. BACKGROUND: Sleep loss-induced neurobehavioral deficits are a recognized threat to safety and productivity in both civilian and military operational settings. Highly-publicized fatigue-related accidents and mishaps (including commercial mishaps such as those occurring at 3-Mile Island, Chernobyl, and Bhopal; and military mishaps such as the ambush of the 507th Maintenance Company) continue to draw attention to the problem of sleep loss/sleepiness in operational environments. Such accidents highlight detrimental effects of sleep loss on decision-making, vigilance, problem-solving and other mental abilities critical to military effectiveness. Publications and manuals from the Army, U.S. Marine Corps, and the U.S. Navy have documented militarily-relevant deleterious effects of sleep loss on alertness and performance. The Army's fifth Mental Health Advisory Team survey of Warfighters serving in Operation Iraqi Freedom and Operation Enduring Freedom revealed that service members who reported less sleep also reported higher rates of accidents and mistakes, and more frequently endorsed negative mental health items. The Department of Defense has funded development of a computational model for quantifying the effects of daily sleep amounts on neurobehavioral performance. The most recent (and most advanced) version of the mathematical sleep/performance prediction model is the 2BAlert model, developed by BHSAI. Like its predecessors, 2BAlert was based primarily on PVT data collected in the WRAIR sleep research laboratory. PVT data was used because, for the purpose of constructing sleep/performance prediction models, it is generally superior to data from other performance measures in several important respects: (a) compared to other performance measures, the PVT is relatively sensitive to sleep loss and the circadian rhythm of alertness; (b) there are no learning effects on the PVT; and (c) it has previously been demonstrated that a PVT-based sleep/performance prediction model has good ecological validity. That is, the PVT-based model predictions of performance effectiveness have been shown to correlate well with 'risk of accidents' in actual railroad operations. The current 2BAlert tool comes in two forms: 1)a web-based tool that takes sleep and caffeine schedules and displays the predicted GROUP AVERAGE performance \& 2)a smartphone app that utilizes individual performance measured with the PVT on a smartphone to individualize the performance predictive model. This version of the tool was successfully validated in a recent study by WRAIR. The most recent version of the smartphone app uses individualized performance predictions to compute individualized caffeine scheduling that optimize performance during specific periods. While this tool was developed using data collected at WRAIR and has been run post-hoc on data previously collected at WRAIR, it has yet to be tested in real-time. Therefore, the purpose of this study is to validate this caffeine optimization algorithm in real-time. The investigators will be utilizing the same study protocol as the recent WRAIR study because it is known there is variability in performance in a healthy population over this time period and that overall performance is slower than 275ms and self-reported stress and anxiety increase significantly without a caffeine intervention. Additionally, by utilizing a previous study design the investigators can minimize the burden on the study team in creating new documentation and schedules. Taking a step back, it is important to address why caffeine optimization is important and relevant for the military. While caffeine is widely accepted and utilized as a stimulant to counter the effects of fatigue associated with shift work and sleep loss, a previous study conducted by WRAIR has shown that caffeine loses its effectiveness and can even slow recovery with repeated use and chronic sleep loss. This surprising result indicates that caffeine has a limit and a cost, and, therefore, the Warfighter would be more effective if he or she utilized caffeine optimally. For previous caffeine studies with similar lengths of continuous sleep deprivation as the current protocol, participants were given 800mg of caffeine during each night. While this research suggests that the investigators could recover performance by giving all participants in this study 800mg of caffeine, the 2BAlert algorithm will allow the investigators to give some participants less caffeine but still reach the same outcome as if everyone had been given 800mg of caffeine. Recent research applied the 2BAlert algorithm post-hoc using parameters laid out in this protocol. The model results show that all participants recovered performance during the Peak Alertness Window, i.e. between 44-50 hours awake, where recovery is defined as 275ms or faster, and over half the subjects required less than half the max total allowable amount of caffeine (800mg), while three subjects required no caffeine at all. The investigators hope to validate 2BAlert by applying it to real-time data utilizing the original study design. In addition to testing the Caffeine Optimization model, the investigators are also interested in testing the hypothesis that if PVT performance can be recovered with caffeine, self-reported stress and anxiety can also be recovered. Recent work found that self-reported stress and anxiety increased with increased sleep loss and recovered to baseline after recovery sleep, following the same trend as PVT performance data. While previous work has shown that self-reported stress and anxiety increase with 1 night of sleep loss, these are the first data to show that self-reported stress and anxiety continue to increase after 2 nights of sleep loss. There are currently no studies reporting if and how self-reported stress and anxiety can be recovered with caffeine during sleep loss.