No abstract

We have developed a system with multimodality that monitors objective biomarkers for screening the mental distress in the office. A field study using a prototype of the system was performed over four months with 39 volunteers. We obtained PC operation patterns using a PC logger, sleeping time and activity levels using a wrist-band-type activity tracker, and brain activity and behavior data during a working memory task using optical topography. We also administered two standard questionnaires: the Brief Job Stress Questionnaire (BJS) and the Kessler 6 scale (K6). Supervised machine learning and cross validation were performed. The objective variables were mental scores obtained from the questionnaires and the explanatory variables were the biomarkers obtained from the modalities. Multiple linear regression models for mental scores were comprehensively searched and the optimum models were selected from 2,619,785 candidates. Each mental score estimated with each optimum model was well correlated with each mental score obtained with the questionnaire (correlation coefficient = 0.6-0.8) within a 24% of estimation error. Mental scores obtained by means of questionnaires have been in general use in mental health care for a while, so our multimodality system is potentially useful for mental healthcare due to the quantitative agreement on the mental scores estimated with biomarkers and the mental scores obtained with questionnaires.

A new concept, 'Layered mental healthcare' for keeping employees mental well-being in the workplace to avoid losses caused by both absenteeism and presenteeism is proposed.A key factor forming the basis of the concept is the biometric measurements over three layers, i.e., behaviour, physiology, and brain layers, for monitoring mental/distress conditions of employees. Here, the necessity of measurements in three layers was validated by the data-driven approach using the preliminary dataset measured in the office environment. Biometric measurements were supported by an activity tracker, a PC logger, and the optical topography; mental/distress conditions were quantified by the brief job stress questionnaire. The biometric features obtained 1 week before the measurement of mental/distress scores were selected for the best regression model. The feature importance of each layer was obtained in the learning process of the best model using the light graded boosting machine and was compared between layers. The ratio of feature importance of behaviour:physiology:brain layers was found to be 4:3:3. The study results suggest the contribution and necessity of the three-layer features in predicting mental/distress scores.