Disease trajectory prediction among cancer patients

Abstract

Cancer is the leading cause of morbidity and mortality in Lithuania and around the world. In order to monitor patients between clinical encounters, passively collected data taken from smartphones could revolutionise the way in which cancer patients are cared for by enabling real-time data analysis and individual patient monitoring. This information can also be used to predict how patients will behave in the future. Using ARIMA (autoregressive integrated moving average), Holt’s Trend, TBATS, Simple Exponential Smoothing, and Naive models, this study aims to forecast the activity period per day estimated by passive data. The most suitable model is selected after evaluating the training error (MAPE), and forecasts are provided for the next 30 days. Root mean square error (RMSE), mean absolute percentage error (MAPE) and mean absolute error (MAE) are used to evaluate the accuracy of the models. According to the results, there is no single most accurate model that can be used for all 30 patients. There are, however, two models that are most appropriate: ARIMA and Holt