Uncertainties in model-based outcome predictions for treatment planning

Purpose: Model-based treatment-plan-specific outcome predictions (such as n ormal tissue complication probability [NTCP] or the relative reduction in s alivary function) are typically presented without reference to underlying u ncertainties. We provide a method to assess the reliability of treatment-pl an-specific dose-volume outcome model predictions. Methods and Materials: A practical method is proposed for evaluating model prediction based on the original input data together with bootstrap-based e stimates of parameter uncertainties. The general framework is applicable to continuous variable predictions (e.g., prediction of long-term salivary fu nction) and dichotomous variable predictions (e.g., tumor control probabili ty [TCP] or NTCP). Using bootstrap resampling, a histogram of the likelihoo d of alternative parameter values is generated. For a given patient and tre atment plan we generate a histogram of alternative model results by computi ng the model predicted outcome for each parameter set in the bootstrap list . Residual uncertainty ("noise") is accounted for by adding a random compon ent to the computed outcome values. The residual noise distribution is esti mated from the original fit between model predictions and patient data. Results: The method is demonstrated using a continuous-endpoint model to pr edict long-term salivary function for head-and-neck cancer patients. Histog rams represent the probabilities for the level of posttreatment salivary fu nction based on the input clinical data, the salivary function model, and t he three-dimensional dose distribution. For some patients there is signific ant uncertainty in the prediction of xerostomia, whereas for other patients the predictions are expected to be more reliable. In contrast, TCP and NTC P endpoints are dichotomous, and parameter uncertainties should be folded d irectly into the estimated probabilities, thereby improving the accuracy of the estimates. Using bootstrap parameter estimates, competing treatment pl ans can be ranked based on the probability that one plan is superior to ano ther. Thus, reliability of plan ranking could also be assessed. Conclusions: A comprehensive framework for incorporating uncertainties into treatment-plan-specific outcome predictions is described. Uncertainty hist ograms for continuous variable endpoint models provide a straightforward me thod for visual review of the reliability of outcome predictions for each t reatment plan. (C) 2001 Elsevier Science Inc.