webinar: what is the future of genito-urinary cancers clinical research?

Genito-urinary cancer research looks to the future

The EORTC webinar on the future of genito-urinary cancer research held on 19 November started with a look at the past. Professor Silke Gillessen presented the International Germ Cell Cancer Co-operative Group’s (IGCCCG) classification of metastatic germ cell cancer, published in 1997, and now an accepted standard for trials as well as for clinical use.  The classification was based on treatments delivered between 1975 and 1990, she explained, and not all patients had received cisplatin-based treatments. « We therefore wanted to validate the original criteria and update outcomes, as well as trying to identify new prognostic factors that might explain the variance in outcomes in the original groups, » she said.

The new study included metastatic germ cell cancer patients from trials and registries who had received treatment between 1990 and 2013. Endpoints were progression-free survival (PFS) and overall survival (OS). Over 12000 patients were eligible for the study.

« We established that original classification still categorised patients correctly.  The consortium that we put together to do this is the largest-ever database of metastatic GCT treated with modern chemotherapy, and will be very useful for future studies in the field, » said Prof Gillessen. « Our research has showed that, although PFS had only improved in those patients with a poor prognosis, OS improved across all groups. This is a very encouraging finding. »

The researchers have submitted two papers on their results for publication, and these will be accompanied by a WebApp to facilitate their practical implementation in the clinic.

Professor Bertrand Tombal, EORTC President, raised the question: what kind of trial should we be doing now? Both randomised controlled trials (RCTs) and cohort studies had advantages as well as disadvantages, he said. « GU cancers have been submerged by thousands of drug trials in the past 10 years. But many of these trials had super-selected patients, optimised for the benefit of the drug being tested. However, in cohort studies it is difficult to identify controls, there are problems of hidden confounders, and blinding is difficult. »

Prof Tombal quoted the example of the Latitude trial of abiraterone plus prednisone in metastatic prostate cancer as an example of the problem of drug-centred versus patient-centred trials. « When the trial started, abiraterone was already available to patients progressing to the metastatic state, » he said, « so the question for the clinician appeared to be whether s/he should start abiraterone up front. The results showed a distinct benefit in doing so, but on further analysis, these benefits came from Eastern European countries where the drug was not available to patients outside the trial. »

However, it is not impossible to combine both RCTs and cohort studies to the benefit of patients, he said. Mixed trials with a focus on good research questions should be promoted, and obstacles to the participation of patients in RCTs removed. « The future of academic research is to remove these boundaries and have a flexible mixed system that combines both registry and RCT information, » he concluded.

The third speaker, Professor Luc Bidaut, member of the EORTC Imaging Group, explained how artificial intelligence (AI) applied to imaging may be able to support clinical trials in the future. « Imaging in drug trials makes things faster, more efficient, and thus overall cheaper », he said. It allows accurate and early measurement of response to treatments, and, provided proper quality control, has now risen to the status of biomarker. A lot of information can be extracted from imaging, and such a process is now encompassed under the umbrella term of Radiomics.  However, extracting and combining complex radiomic data from various modalities means that there is far too much information for mere humans to analyse, and this is where machines are most needed to come to the rescue.  Machines can in fact help at all stages of the data extraction and analysis, not just at the final step, he said.

Machines generally learn via neural networks, which, once properly trained, can then rapidly detect trends and identify patterns that may be too complex for humans to notice, he said.  In a likely sign of things to come, some studies have already shown that the interpretation of images by AI plus a radiologist is superior to either a radiologist or AI alone.

Although combining Radiomics and AI shows a lot of potential in imaging, problems remain.  For instance, implementing Radiomics at scale remains difficult, he said. While results in a very controlled environment (e.g. a single site) are promising, they do not necessarily translate well in the broader and more diverse context of multiple sites and clinical settings.

Accessing large amount of suitable data is difficult too, which is a significant problem since AI methods such as deep learning networks require large quantity of realistically diverse items in order to be properly trained. To date, this can be mitigated only in part by using AI to synthesize realistic virtual data as part of the training set.

For the foreseeable future, human participation will still be required, firstly to control the AI and assess its results, and also, perhaps more pragmatically, because of liability concerns – if the AI says a patient has a particular kind of disease and needs a specific therapy, who (or what) is ultimately responsible for the decision?

AI will become more controversial as its use increases, and people will need to be assured that it will be accurate, trustworthy, explainable, and employed ethically. « Our ambition is to have some kind of digital twin of every oncology patient, so that we can devise treatment and predict response in the overarching context of personalised medicine. This is an aspiration, and we are not there yet. We need to think in a multidisciplinary fashion, which may be challenging at times, and so we – the Imaging Group in the EORTC context – are very keen to get ever more closely involved in the design and delivery of relevant oncology trials, » he concluded.

The whole webinar may be seen on Vimeo.

Back to news list