Artificial Intelligence in Medicine: Predicting Prostate Cancer Progression During Active Surveillance

“Despite increasing adoption of active surveillance (AS) for low-risk prostate cancer, AS discontinuation rates are high, investigators report.” This quote, from a September 2021 Renal & Urology News report, referred to a study of over 16,000 Canadian prostate cancer (PCa) patients who had been diagnosed with low risk disease from 2008-2014. 51% of these patients initially chose to go on Active Surveillance (AS). AS is increasingly recommended—and adopted—by low-risk patients as a way to avoid immediate treatment that could impair their urinary or sexual function.

According to a 2019 JAMA publication, from 2010-2015 use of surveillance among men with low-risk PCa increased from 14.5% to 42.1%^[i]. While this trend has been very encouraging for experts in the field, the Canadian study struck a blow to their optimism. The authors’ analysis found that by 5 years into AS, 52% of patients had discontinued monitoring in favor of active treatment; the average time from initial decision to active treatment was 16 months^[ii]. And, concerningly, about half of those who went on to treatment were subsequently found to have progressed to a higher grade. This is called PCa progression.

Predicting progression

This raises the question, if a patient could look into a crystal ball and see that his cancer had a probability of progression, would he go on AS to begin with? AS entails a psychological problem for many patients, who are uncomfortable living with uncertainty. Patients who had advance warning of probable progression would need to be extra diligent about monitoring, perhaps with more frequent PSA/biomarker tests as well as multiparametric MRI. Perhaps knowing their situation from the start would mentally prepare them for the task…or perhaps they would simply skip AS and opt for immediate treatment.

Now it appears that something more effective than a crystal ball is at hand. A collaborative team whose members include personnel from Harvard, MIT, Massachusetts General Hospital and McMaster University (Hamilton, ONT) have applied a form of Artificial Intelligence called Machine Learning (ML) to more traditional statistical approaches to predicting progression on AS. If successful, both efficiency and accuracy would be improved, making it possible to tailor AS protocols to an individual patient’s risk level.

Their training dataset included 790 AS cases with records of all clinical factors, of which 239 demonstrated PCa grade progression over an average of 6.29 years. They tested several ML classifiers against a traditional logistic regression classifier, finding that a support vector machine classifier offered the best performance, while the traditional classifier had the poorest performance. The concluded, “While our specific models require further validation, we anticipate that a ML approach will help produce robust prediction models that will facilitate individualized risk-stratification in prostate cancer AS.”^[iii]

Further development and testing will be needed to validate their ML tool for predicting which low-risk patients are likely to progress if they go on AS. Perhaps fewer men will find AS appealing if they understand they are at some risk, but on the bright side it’s equally possible the more men who opt for AS will do so because they reasonably expect to be on it for many years—perhaps the rest of their lives—without ever needing treatment. This is another example of Artificial Intelligence’s potential to serve the needs of PCa patients.

NOTE: This content is solely for purposes of information and does not substitute for diagnostic or medical advice. Talk to your doctor if you are experiencing pelvic pain, or have any other health concerns or questions of a personal medical nature.

[i] Mahal BA, Butler S, Franco I, et al. Use of Active Surveillance or Watchful Waiting for Low-Risk Prostate Cancer and Management Trends Across Risk Groups in the United States, 2010-2015. JAMA. 2019;321(7):704–706.
[ii] Timilshina N, Komisarenko M, Martin LJ, Cheung DC et al. Factors Associated with Discontinuation of Active Surveillance among Men with Low-Risk Prostate Cancer: A Population-Based Study. J Urol. 2021 Oct;206(4):903-913.
[iii] Nayan M, Salari K, Bozzo A, Ganglberger W et al. A machine learning approach to predict progression on active surveillance for prostate cancer. Urol Oncol. 2021 Aug 28;S1078-1439(21)00366-5.