Investigation on factors related to poor CPAP adherence using machine learning: a pilot study.

To improve patients' adherence to continuous positive airway pressure (CPAP) therapy, this study aimed to clarify whether machine learning-based data analysis can identify the factors related to poor CPAP adherence (i.e., CPAP usage that does not reach four hours per day for five days a week). We developed a CPAP adherence prediction model using logistic regression and learn-to-rank machine learning with a pairwise approach. We then investigated adherence prediction performance targeting a 12-week period and the top ten factors correlating to poor CPAP adherence. The CPAP logs of 219 patients with obstructive sleep apnea (OSA) obtained from clinical treatment at Kyoto University Hospital were used. The highest adherence prediction accuracy obtained was an F1 score of 0.864. Out of the top ten factors obtained with the highest prediction accuracy, four were consistent with already-known clinical knowledge. The factors for better CPAP adherence indicate that air leakage should be avoided, mask pressure should be kept constant, and CPAP usage duration should be longer and kept constant. The results indicate that machine learning is an adequate method for investigating factors related to poor CPAP adherence.

Core body temperature estimation by eyeglass-type device: thermal analysis of radiation heat measured from caruncle.

This paper describes a method for estimating core body temperature from radiation heat of the caruncle and an eyeglass-type device for measuring the temperature of the caruncle to prescreen for infectious diseases such as COVID-19. As a precise prescreening method, monitoring a person's continuous core body temperature is desired. By monitoring the continuous core body temperature, including circadian rhythm, in our daily life, infections can potentially be discovered when body temperature is higher than normal. Although monitoring the core body temperature is effective, continuous and precise monitoring requires the use of an invasive instrument. To overcome this, we (1) design an eyeglass-type device for measuring the caruncle temperature and (2) model the correlation between the caruncle temperature and the core body temperature. Experimental results revealed that hypothalamic temperature could be estimated within ± 0.3 °C between 20 and 30 °C by using the eyeglass-type device.