Task Progress:
|
This proposal addresses the Human Research Program (HRP) Risk of Adverse Cognitive or Behavioral Conditions and Psychiatric Disorders and several other critical HRP risks and gaps. This study utilizes Reaction Self-Test (RST) data collected by the Principal Investigator (PI) and his team in N=24 astronauts on 6-month International Space Station (ISS) missions, arguably the largest cognitive dataset ever collected in spaceflight. RST consists of a survey module and a 3-minute version of the Psychomotor Vigilance Test (PVT). Our main objective is to additionally obtain data on key environmental stressors (i.e., CO2 and O2 levels, temperature, noise, and radiation) and combine them with RST data and other operational data collected by the PI and his team (Specific Aim 1). All data will be integrated in one carefully annotated database, which will be delivered to NASA at the end of the project and could be later amended and mined by other researchers (Deliverable 1). We will then develop an individualized dynamic prediction model that informs future PVT performance based on environmental data, survey data, prior PVT administrations, and person-specific characteristics using state-of-the-art machine learning techniques such as functional concurrent regressions and neural networks for time series forecasting (Specific Aim 2). We will perform model selection and identify those variables that have the highest predictive value for PVT performance (Deliverable 2) and could preferentially be collected on future spaceflight missions to inform relevant changes in cognition and behavioral health. At the end of the study, we will deliver an algorithm to NASA that, for the first time, can predict adverse cognitive conditions in astronauts early and with an unprecedented precision (Deliverable 3). The predictive algorithms can be translated to several settings on Earth where high performing individuals have to sustain high levels of cognitive performance while facing several environmental or other challenges (e.g., US Navy personnel on submarines).
We obtained approval from the Lifetime Surveillance of Astronaut Health Board, the Institutional Review Board (IRB) of Johnson Space Center, European Space Agency (ESA), and Japan Aerospace Exploration Agency (JAXA), and the IRB of the University of Pennsylvania to use already collected RST data for the purposes of this project. N=7 non-US astronauts who had not provided broad consent for the re-use of their RST data were re-consented. These included 6 non-US astronauts, for which NASA previously had no process in place for re-consenting them. We also obtained key environmental variables from the ISS for the duration of the RST project (2009-2014): O2 and CO2 levels, noise levels, radiation levels, temperature levels. The temporal resolution of these data varies. We also extracted key demographic (prior days in space, prior missions, DOB, sex, nationality, space agency, military background, degree, marital status, number of children), and operational (number of ISS occupants, date of extravehicular activities (EVAs) or spacecraft dockings/undockings, sleep schedule)) information for the relevant time period. The four data streams (RST data, environmental data, demographic data, operational data) were combined in a data matrix. We have identified those variables that have the highest predictive value for PVT performance combining a number of variable selection methods: Past PVT performance Self-reported RST PC scores (Workload, Sleep Quality, Sleepiness, Physical Exhaustion, Mental Fatigue, Stress), Radiation, Temperature, O2 and CO2 levels, Sleep duration, predicted lapses based on sleep history, and Age. We investigated predictive properties of three classes of prediction models (Linear Mixed Effect Model, Functional Concurrent Regression Model, Random Forest Model) and an Ensemble Model that combines predictions from the previously mentioned three models. The ensemble model was found to provide the highest prediction accuracy. Finally, we built a Shiny App for visualizing the effects of several stressors on PVT performance, which can be a powerful tool for investigating the effects of individual stressors, their combination, and to generate hypothesis. This tool could be amended to include additional predictors and/or outcomes.
|