5613275 USE OF CONSUMER WEARABLES TO PREDICT PAIN IN PATIENTS WITH SICKLE CELL DISEASE

Abstract

Background: Sickle cell disease (SCD) is associated with increased morbidity and mortality. Episodes of acute and severe pain known as vaso-occlusive crises (VOC) are the most common cause for hospital admission. Mobile health (mHealth) has developed promising, patient-friendly, minimally invasive tools to monitor patients remotely. Our previous work has leveraged data from mHealth apps and wearable devices to evaluate several machine learning (ML) models to accurately predict pain scores in patients with SCD currently admitted for VOC to the SCD Day Hospital. Aims: Evaluate the feasibility of extended monitoring for 30 days post-discharge and to refine the development of ML models to predict pain scores in patients with SCD. Methods: Patients with SCD aged 18 and above who were admitted for a VOC to the SCD Day Hospital or to Duke University Hospital were eligible for this study. All patients were followed for 30 days following discharge. Following informed consent, patients were provided: 1) a mobile app (Nanpar); and 2) an Apple Watch. Patients were instructed to report their pain to nurses as per standard of care monitoring and at least once daily within the Nanpar app. Patients were asked to continuously wear the Apple Watch, removing only to charge. Physiological data collected by the Apple Watch included heart rate, heart rate variability, oxygen saturation and step count. These data were associated with self-reported pain scores to fit five different machine learning classification models. The performance of the models is evaluated by the following metrics: accuracy, F1-score, root mean squared error (RMSE) and area under the ROC curve (AUC). Results: Nineteen patients were included in this study from April through June 2022. The median age at time of inclusion was 30 years (IQR:22-34). The majority of the patients had genotype HbSS (68%) and all were Black or African American. Eleven patients (55%) were enrolled from the SCD Day Hospital. This preliminary dataset consisted of 1480 data points. After micro-averaging due to the imbalanced dataset, the performance of all the models were very similar. The metrics of the best performing model, the random forest model, were: micro-averaged accuracy:0.89, micro-averaged F1-score:0.50, RMSE:1.52, AUC:0.83. There was no correlation between any of the data elements recorded by the Apple Watch. Our random forest model was able to accurately predict higher pain scores not only for patients who were admitted to the hospital, but also for patients after discharge from the hospital. Discussion: Our model was able to predict pain using data from the Apple Watch with high accuracy, but with low F1-score. The consistency in the performance of each model, along with the low F1-scores reflects the high degree of class imbalance and lack of data most probably due to inconsistent pain reporting after discharge. To alleviate this issue of class imbalance, different oversampling approaches and generative models need to be examined. Future efforts will focus on larger numbers of patients and monitoring patients for longer periods of time to provide a larger dataset, in an attempt to further improve the accuracy of pain prediction. Conclusion: The consumer wearable Apple Watch was a feasible method to collect physiologic data and provided accuracy in prediction of higher pain scores. We believe mHealth efforts can provide valuable insights for patient monitoring and pain prediction for patients with SCD.