A Statistical Framework for Forecasting Cumulative Dose Metrics in Adaptive Radiotherapy

Adaptive radiation therapy (ART) seeks to maintain accurate dose delivery by monitoring anatomical changes during treatment and modifying plans accordingly, yet commonly used approaches for estimating cumulative dose rely on heuristic, deterministic assumptions and do not quantify uncertainty in future dose delivery. We present a statistical framework based on simple exponential smoothing (SES) to forecast cumulative dose metrics probabilistically throughout the course of ART. Adapting lead time SES methods from inventory control, we model cumulative dose trajectories and derive closed form prediction intervals and threshold exceedance probabilities for clinically relevant dose metrics. The framework was evaluated using per fraction dose volume histogram data recomputed on daily volumetric imaging for 32 prostate and 19 head and neck patients, allowing assessment of empirical interval coverage, interval width, and adaptive decision support performance. SES based forecasts demonstrated robust coverage probabilities across a range of smoothing parameters and provided appropriately scaled uncertainty estimates for remaining course cumulative dose. In addition, a difference based formulation enabled in course estimation of the probability that an adaptive plan would improve delivered dose metrics relative to the original plan. This SES based forecasting approach offers a computationally efficient and statistically grounded method for anticipating cumulative dose delivery, supporting quantitative, individualized adaptive monitoring and evidence based replanning decisions within routine ART workflows.