Seminar: Learning from Every Patient Treated

Time: 14:00 - 15:00

Venue: Room 3.009 Alliance Manchester Business School, Booth Street West, Manchester M5 6PB

Speaker: Marcel van Herk (University of Manchester)

Title: Learning from every Patient Treated

Abstract:

The radiotherapy chain has many uncertainties. With the introduction of image guided radiotherapy, setup error and organ motion are no longer limiting treatment accuracy, while biological uncertainties are. For instance, target volume delineation between doctors is variable. In addition, dose constraints for organs at risk are poorly known. Using a concept of image based data mining, our group is analysing large amounts of radiotherapy data, correlating outcome directly with dose per anatomatical location, using deformable registration to normalise patients into a standard template. Such analysis has allowed detection of subclinical tumour spread in hundreds of prostate cancer patients, and over-sensitivity for radiation of the base of the heart in thousands of lung cancer patients. In order to take this metholodogy forwards, several technical challenges must be addressed. First, multiple testing must be addressed, since thousands of voxels are tested when analysing the data. Typically this is done by testing a single summary statistic, using permutation testing to create the statistics of that variable under the null hypothesis. Then, the impact of confounding variables must be properly taken into account. For instance tumour size influences dose distribution as well as survival. We now utilize a full Cox regresion per voxel during anaysis such that the entire analysis is multivariate. Another way to solve this issue is to use variables that are random during patient treatment, such as minimal variations in image guidance that affect survival in lung cancer and strenghten the evidence. Finally, patient outcomes are always multifactiorial, and it is important to take additional factors into account, preferably analysed quantitatively such as sarcopenia, a form a frailty that can be detected in routine CT. Concluding, the base methodology for learning from every patient is in place, but now the step to clinical implementation of the findings must be taken.