Using time series analysis to forecast the health-related quality of life of post-menopausal women with non-metastatic ER+ breast cancer: A tutorial and case study.

Abstract

Time series models are widely used forecasting techniques in health care for long time series and are typically built in commercial statistical packages. However, for short time series data, such as health-related quality of life (HRQoL), guidance on how to select and use appropriate time series models is lacking. This tutorial provides a step-by-step guide adopting a time series analysis framework for HRQoL forecasting. We walk through a case study examining the forecasting of the effects of adjuvant endocrine therapy on the HRQoL of post-menopausal women with non-metastatic ER+breast cancer using data from the HRQoL sub-protocol of the Tamoxifen arm of the Arimidex, tamoxifen, alone or in combination (ATAC) trial. The forecasting of HRQoL consists of four steps: 1) data extraction and accuracy check, 2) forecasting horizon definition and identification of data pattern, 3) forecasting model identification and fitting using five forecasting approaches appropriate for short time series ((i) double exponential smoothing, (ii) double moving average, (iii) fuzzy forecasting, (iv) grey forecasting, and (v) Volterra series), 4) forecasting model selection. A user-friendly visual basic for applications (VBA) Excel add-in is made available to interested users to facilitate the application of the tutorial. The Grey method and Volterra series appeared to be good candidates to forecast the effects of adjuvant endocrine therapy on the HRQoL of post-menopausal women with non-metastatic ER+breast cancer enrolled in the ATAC trial. It is feasible to forecast the effects of treatments on HRQOL even when the time series is short.