Developing a predictive model for baseline detection and follow-up of risk of prostate cancer progression on active surveillance

Prostate cancer is the most common cancer in men in 112 countries, and its incidence and mortality will have doubled by 2040. Nearly 50% of newly presenting patients in the UK harbour low and intermediate-favourable-risk localised disease, for which active surveillance (AS) offers non-inferior survival outcomes compared to radical treatment. AS involves repeat testing to defer or avoid treatment and its side effects until progression to more aggressive disease (which can then be treated with a curative intent) or a graduation to watchful waiting (when life expectancy is too short for a curative treatment to yield a significant benefit). Although the practice of AS varies significantly across countries, guidelines and individual centres, the key tests used both at baseline and during follow-up include serum prostate-specific antigen (PSA), MRI and biopsy.

Despite the growing global uptake of AS, the afore- mentioned significant variation in its protocol hinders its optimal implementation. Specifically, up to 38% of patients are deterred from AS due to the need for repeat prostate biopsies, which are not well tolerated by most men but are still performed without evidence of disease progression (eg rising PSA or lesion growth on MRI) according to rigid protocols. This highlights the discrepancy between the standardised approach recommended by guidelines and the risk-adapted approach favoured by both patients and clinicians. In addition, up to 30% of men switch to treatment over the first five years due to true disease progression, highlighting the failure of current baseline risk stratification approaches to offer AS to the right patients.

It is, therefore, clear that developing an objective, risk-adapted approach to AS is key to delivering its benefits to more eligible patients. One way to address this is to make the most of routinely available PSA, MRI and biopsy data by combining them into a model that would predict the risk of prostate cancer progression at all stages of AS. With MRI being a crucial component of AS, the subjectivity of PI-RADS3 and PRECISE4 scores used for baseline and follow-up tumour assessment can be overcome by extracting quantitative, or radiomic, features from the visible lesions or the whole prostate. By doing so, the model trained on expert-level data would no longer rely on the experience of a radiologist, but rather use objective image-derived data to quantify the risk of disease progression. Our research has, therefore, aimed to answer two questions both in the baseline and follow-up AS settings:
• Can AI-driven radiomic analysis of prostate MRI achieve similar performance to expert radiologists for predicting tumour progression on AS?
• Can a joint predictive model combining PSA, MRI and biopsy data outperform each of these parameters that are currently used individually in clinical practice?

To access the full article as a pdf please sign up to our newsletter.

The content on this page is provided by the individuals concerned and does not represent the views or opinions of RAD Magazine.