부산대학교 도서관

Background Survival prediction is an important aspect of oncology and palliative care. Quantification of activity and sleep-related parameters through actigraphy could result in improved prognostication in patients with advanced cancer, since an activity-based measure, the dichotomy index (I < O), which is a measure of nighttime activity relative to daytime activity, has shown promise in patients receiving chemotherapy. Our aim was to assess the feasibility of acquiring actigraphy and sleep diary data in patients with advanced cancer, and to explore the prognostic value of certain actigraphy-derived activity and sleep-related parameters. Methods Fifty adult outpatients with advanced cancer and an estimated prognosis of < 1 year were recruited. Patients were required to wear an Actiwatch® (wrist actigraph) for 8 days, and complete a sleep diary. Other data included demographic information, performance status, physical and psychological symptoms, routine blood parameters, and date of death (survival). Standard statistical analyses were used to determine correlations between individual factors and survival, and machine learning was used to develop a multifactor prognostic model. Results Forty-nine patients completed the study, and 34 patients died within 1 year. Adequate actigraphy data (≥72 hours) were obtained in 44 patients, and sleep diary data in 46 patients. Forty-two patients had markedly disrupted rest-activity rhythms, however I < O and other activity and sleep-related parameters were not predictive in the univariate analyses, unlike established prognostic indicators (e.g. performance status). A machine learning-derived multivariate model consisting of activity, sleep and routine clinical information identified novel predictors for survival, with the potential of a clinically relevant prognostic algorithm. Conclusion The results demonstrate the feasibility of utilising wrist actigraphy and sleep diaries in studies of survival in advanced cancer, and suggest that machine learning approaches could help to develop more sensitive prognostic algorithms. A larger study is required to further develop/validate the identified multivariate prognostic model.