Deep-Learning Approach to Predict Survival Outcomes Using Wearable Actigraphy Device Among End-Stage Cancer Patients

Abstract

Survival prediction is highly valued in end-of-life care clinical practice, and patient performance status evaluation stands as a predominant component in survival prognostication. While current performance status evaluation tools are limited to their subjective nature, the advent of wearable technology enables continual recordings of patients' activity and has the potential to measure performance status objectively. We hypothesize that wristband actigraphy monitoring devices can predict in-hospital death of end-stage cancer patients during the time of their hospital admissions. The objective of this study was to train and validate a long short-term memory (LSTM) deep-learning prediction model based on activity data of wearable actigraphy devices. The study recruited 60 end-stage cancer patients in a hospice care unit, with 28 deaths and 32 discharged in stable condition at the end of their hospital stay. The standard Karnofsky Performance Status score had an overall prognostic accuracy of 0.83. The LSTM prediction model based on patients' continual actigraphy monitoring had an overall prognostic accuracy of 0.83. Furthermore, the model performance improved with longer input data length up to 48 h. In conclusion, our research suggests the potential feasibility of wristband actigraphy to predict end-of-life admission outcomes in palliative care for end-stage cancer patients.Clinical Trial Registration: The study protocol was registered on ClinicalTrials.gov (ID: NCT04883879).

Deep Learning Prediction Model for Patient Survival Outcomes in Palliative Care Using Actigraphy Data and Clinical Information

“…The results of our current study are in line with our previous related study results, including two prospective observational studies [ 28 , 29 ] and a scoping review [ 44 ]. One of our previous findings showed that the majority of the included studies in the scoping review, which utilized wrist-worn wearable devices in cancer populations, focused on physical activity, sleep analysis, and heart vital signs and showed a positive correlation between patient-reported and wearable outcomes [ 44 ], while in the other study, automatic survival prediction using an LSTM DL model showed feasibility in clinical settings and possible benefits in end-of-life care settings without healthcare professionals [ 29 ]. Additionally, in the third study, wearable devices reported greater angle and spin movements as early as within the first 48 h of observation in the cancer patients who were still alive after discharge from the hospice inpatient unit [ 28 ].…”

“…The validated cutoff values for the KPS and PPI were 50% and 6.0, respectively [ 29 ]. Sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), overall accuracy, and the area under the receiver operating characteristic (AUROC) curve were used to evaluate the predictive accuracy of the KPS and PPI.…”

“…Recently, in our previous study, we evaluated the feasibility of using actigraphy-based patient monitoring to predict survival outcomes and found that wearable devices (WDs) can be useful prognostic tools for palliative care patients nearing the end of their lives. These devices reported greater angle and spin movements as early as in the first 48 h of observation in cancer patients who were still alive following discharge from a hospice inpatient unit [ 28 , 29 ]. Applied research has explored the applications of ML techniques for health monitoring, elderly care, and fitness tracking in the last decade and is growing over time [ 30 ].…”

Recent advances in artificial intelligence applications for supportive and palliative care in cancer patients

“…Many studies in Table 1 and approximately half of all studies included in total focused on predicting mortality as a clinical outcome, which includes predicting short-term mortality risk 14▪,15,18▪,24▪ and survival over a longer horizon 19–21,22▪▪,23▪▪,25,29. Mortality risk and survival time were both usually predicted using machine learning (ML) models that analyze various patient factors such as clinical parameters, changes during treatment, and symptoms 14▪,15,18▪,20,21,22▪▪,23▪▪,24▪,25. The accuracy of ML and deep learning (DL) models is typically evaluated by their area under the curve (AUC) value, which measures the accuracy of predictions and a model’s discriminative ability where 1.0 represents the highest possible AUC score indicating perfect discrimination 11,39.…”

“…The accuracy of ML and deep learning (DL) models is typically evaluated by their area under the curve (AUC) value, which measures the accuracy of predictions and a model’s discriminative ability where 1.0 represents the highest possible AUC score indicating perfect discrimination 11,39. The models in Table 1 had AUC values between 0.70 and 0.92 indicating generally good model performance 14▪,15,18▪,19–21,23▪▪,24▪,25,26,40. Interestingly, models predicting 60-day and especially 180-day mortality had lower AUC values compared to 30-day mortality models 19,20,24▪,40, suggesting that predicting acute mortality risk could be more reliable than prediction over a longer term.…”