One of the big challenges with prostate cancer is it’s a very common disease, particularly the localised prostate tumours, yet there’s a wide spectrum in risk. So some tumours behave in a really indolent, slow fashion and others, unfortunately, behave aggressively and can metastasise quickly. We have some guide on pulling apart these different scenarios but it’s really quite limited and it’s really important for men that we have a better guide because the treatments come with important side effects, some of which can be quite life changing. There is an option to survey and, equally, there is a small proportion of men who develop metastases and we need to treat them more aggressively.

How was the multimodal AI biomarker validated using the CHHiP trial data?

So the multimodal AI test consists of input from the diagnostic biopsy H&E stained images which are digitised, so they were digitised in the CHHiP trial cohort, together with age, presenting PSA and tumour stage, or T stage. Those inputs were combined to create the MMAI test result and then we reconciled and matched that with the long and very complete follow-up data we had in terms of recurrence free survival and development of distant metastases in the CHHiP trial.

What were the key findings regarding its prognostic accuracy and potential clinical utility?

We had data from almost 1,800 men, so we had a really large, well-powered cohort. The key findings, and I would emphasise we’ve used performance metrics that are considered gold standard for evaluation of digital pathology biomarkers in the recently published ESMO guidelines, the key findings were that the addition of the MMAI test really significantly improved model performance. So with the MMAI the prognostic models fit the data much better.

So what that means in real terms for men is we pulled out a small but significant high-risk group that had a particularly high risk of developing distant metastases and who recurred quickly. So this group, although it’s a minority, it’s an important minority that needs intensification. On the other end of the spectrum the MMAI test pulled out a large low-risk group which is five times larger than the low-risk group that we can identify with our existing re-stratification systems.  For that group you could argue that less treatment is in their best interests and potentially even no treatment, just active surveillance. So the MMAI provides added information to guide these treatment decisions much more robustly.

How might this AI biomarker influence risk stratification or treatment decision-making in practice?

I would hope that in the near future we could have a clinical care situation where patients H&E biopsy images were all digitised and they had MMAI test readouts and was a test result that we could get back quickly. So as we sat down discussing treatment options with patients we would have this information. Obviously a key part of these treatment decisions is shared decision-making with patients and patients themselves will have preferences. But if we can empower them to make the right decision for them with additional, robust data, it’s a kinder way to treat men with prostate cancer.