Abstract 5261: Development of a computational algorithm to detect fatigue of gynecologic cancer patients using lifelogs

Abstract

Abstract The improvement of the quality of life (QOL) of gynecological cancer patients is a critical purpose in cancer treatment. At present, there is no other methodology to evaluate QOL other than patients reported outcomes using QOL questionnaires. Therefore, management of QOL is not performed clinically because there is no objective and easy method to evaluate QOL. In this study, we extracted the worst symptom of QOL in gynecological cancer patients and examined whether QOL can be evaluated by lifelogs objectively. Lifelogs (diets, sleep duration, the number of steps, pulse rate, voice, and so on) and QOL questionnaires (EORTC-qlq-c30, PHQ9, and so on) were collected from 139 gynecological cancer patients at our hospital using mobile applications. Scales of symptoms and functions were calculated from EORTC-qlq-c30. Logistic regression analyses and calculation of feature impact in the predictive model depicting receiver operating characteristic (ROC) curve were performed to extract symptoms and functions that contributed most to global health status (GHS) indicating a general QOL status. Fatigue-related metabolites in patient serum were measured by ELISA. Correlations between each lifelog, fatigue scale, and fatigue-related metabolites were examined. A computational model to detect high fatigue scale status was developed using lifelogs. The fatigue scale was the highest among all treatment periods. The fatigue scale, appetite loss scale, and physical functioning scale were symptoms that deteriorate QOL frequently in gynecologic cancer patients (odds ratio: 0.9708, 0.9797, and 1.0374, respectively; 95% CI: 0.9493 to 0.9927, 0.9677 to 0.9919, and 1.0155 to 1.0597, respectively). ROC curve indicated that an accurate prediction of GHS using scales of symptoms and functions (area under the curve, AUC: 0.85). From the scales of symptoms and functions, the fatigue scale showed the highest feature impact to predict GSH. Several lifelogs were significantly correlated with fatigue scores (p<0.05, r=-0.25 to -0.15 for each), and with the blood concentration of fatigue-related metabolites (p<0.05, r=-0.44 to -0.37 for each). A computational algorithm using the fatigue-related lifelogs was successful to detect high fatigue score status (AUC=0.79). We found that two cases with persistent high fatigue scores and a low fatigue-related lifelog during chemotherapy led to depression and depressive relapse. In summary, fatigue is a significant problem contributing to the decline in QOL of gynecological cancer patients. Fatigue can be monitored and diagnosed using digital data using lifelogs. These findings lead to the development of future management strategies of QOL for cancer patients. Citation Format: Ken Yamaguchi, Nozomi higashiyama, Akihiko Ueda, Shiro Takamatsu, Masayo Ukita, Mana Taki, Koji Yamanoi, Junzo Hamanishi, Masaki Mandai. Development of a computational algorithm to detect fatigue of gynecologic cancer patients using lifelogs [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2022; 2022 Apr 8-13. Philadelphia (PA): AACR; Cancer Res 2022;82(12_Suppl):Abstract nr 5261.