Model-based selection of patients for advanced cancer therapies

Aim

To establish and continuously improve models for unbiased selection of patients for proton therapy and other novel radiation therapy modalities.

Background

Many technological developments, including proton therapy, are problematic to implement based on randomized trials primarily because of the equipoise but also because the expected improvements are often related to very late toxic endpoints and prohibitive study sizes. To address this issue, a model-based selection of patients has been proposed, possibly in combination with a randomized subset. The method is critically dependent on the generalizability of the underlying models of toxicity and/or tumor control and FA 8 will focus on supporting such model reliability. By model-based selection, moderate to severe morbidity for the individual patient is predicted by comparing dose-volume relationships of relevant organs at risk on treatment plans for protons and photons. Possible model misspecifications or unexpected effects on, e.g. tumor control, must then be monitored in non-randomized fashion by following all patients. FA 8 will use data of relevant and specific toxicity and tumor control data for the development and verification of Normal Tissue Complication Probability (NTCP) and tumor control probability models (TCP).

Methods

FA8 will establish a biological modeling competence center (BMCC). This will together with the leaders of FA8 include senior statistician and international experts in biological modeling. BMCC will provide general guidelines for biological modeling and provide support to interventional protocol workgroups (IP) that have or are planning collection of clinical data for NTCP or TCP modeling. The BMCC will provide individual support in the planning phase, during the conduction of the study in pilot testing and in the final analysis of the study.

Examples of IPs with data for NTCP modeling: 1) DNOG has neurocognitive data on patients treated for brain tumors, 2) DAHANCA has data on swallowing dysfunction after head and neck cancer radiotherapy, 3) DBCG has data on patients with coronary artery disease after left-sided breast cancer, 4) DAPROCA has data on gastrointestinal (GI) morbidity for prostate cancer radiotherapy, and finally 5) LET- and BED modelling based on the observed toxicities. An example of TCP modeling: Rigshospitalet has data on local, regional and distant control of a cohort of NSCLC patients treated with radiotherapy. These data will be modeled and validated in an independent dataset.

Data for analysis will be provided from the interventional protocols and existing DMCG outcome databases or existing institutional series.

An initial round of workshops for IPs and professionals with interests in biological modeling is planed in spring 2018 to set the scene and to present potential materials for modeling. There will be a period to analyze data exposed on the workshop and to describe avenues for future research. This will be followed by a second workshop to generate a tangible research program and writing of a major grant application.

Expected results

The initiative will establish models for prediction of normal tissue reaction and tumor control after radiation therapy and it will provide a quantitative framework for comparing competing treatment strategies including conventional and advanced radiotherapy technologies (proton therapy and MR guided radiotherapy) at the individual patient level.

Impact/relevance/ethics

The proposed initiative will provide important analytical support to IPs and improve the quality of their studies. Biological modeling of outcomes of large patient cohorts will improve the precision in the selection process for cancer therapy modalities.

The IP is responsible for the authorization of data collection. There are no ethical concerns related to the initiative.