The ALLEGRO project: what more do we need to know about normal tissue complication probability?

One of the objectives of the ALLEGRO project has been to build and test the validity and accuracy of NTCP models in order to provide either advice concerning their clinical application, or recommendations for further research and development. The investigations have included testing the accuracy with which the dose data (DVH) can be recovered from clinical databases used for model fitting, the validation of NTCP models in three major tumour sites (i.e., head and neck, lung and prostate cancer), and an exploration of ways to improve NTCP models through an understanding of the underlying radiobiological mechanisms leading to specific endpoints.... The model validation has been done in a 4-step approach: (1) to build and validate NTCP-models using existing databases from patients treated with 3D-CRT; (2) to validate these NTCP-models among patients treated with IMRT; (3) to assess if IMRT could be further improved by combining the most relevant DVH-parameters derived from step 1 and 2 and the results of in silico planning comparative studies, and: (4) to estimate the potential benefit of protons compared to photons by using the same methodology. A number of issues have become apparent during the model validation: (1) discrepancies with regard to delineation of OARs ; (2) the need for studies on the clinical relevance of side effects in specific patients populations, in particular in case of multiple endpoints and OARs; (3) the problem of how to deal with the so-called composite endpoints including multiple endpoints related to different anatomical structures; (4) integration of determinants other than DVH-parameters into NTCP-models (e.g., concomitant chemotherapy); (5) how to deal with toxicity endpoints from which the anatomic substrate is still unknown and/or may depend on dose distributions from different OAR (e.g., swallowing dysfunction and patient-rated xerostomia) and methodological problems from a statistical points of view (e.g., how to select the most important prognostic factors and how to test model performance). It has also become clear that there is considerable potential to shed light on the determinants critical to particular endpoints through an improved understanding of the underlying biological mechanisms. Because of the complexity of the biological processes the analysis should be symptom-specific rather than organ specific. The presentation will discuss the ALLEGRO results, and how (and how much) we can hope to improve the usefulness and accuracy of NTCP models so that they can gain wider acceptance in routine clinical practice.