AI-Based Predictive Model Could Spare Use of Prostate Cancer ADT

ReachMD Healthcare Image

08/09/2023

A new predictive model based on artificial intelligence (AI) may increase the number of men with intermediate-risk prostate cancer who can avoid androgen deprivation therapy (ADT).

The model, described in NEJM Evidence, found that most men in this patient population do not benefit from the addition of ADT to radiation therapy (RT).

“Currently, outside of NCCN [National Comprehensive Cancer Network] risk groups and other prognostic tools, there are no predictive biomarkers to specifically guide which men need ADT with radiotherapy,” said first author Daniel Spratt, MD, chair of radiation oncology and a professor in the department of radiation oncology at UH Seidman Cancer Center and Case Western Reserve University, Cleveland, Ohio.

The researchers used digital pathology images from pretreatment prostate tissue and clinical data from 5727 patients enrolled in 5 phase 3 randomized trials (ClinicalTrials.gov numbers NCT00767286, NCT00002597, NCT00769548, NCT00005044, and NCT00033631). The patients received RT with or without ADT. These data were used to develop and validate an AI–derived predictive patient-specific model that could identify patients who would develop the primary endpoint of distant metastasis.

Using this model, validation was performed with data from NRG Oncology/Radiation Therapy Oncology Group (RTOG) 9408, which included 1594 patients randomly assigned to RT plus or minus 4 months of ADT. With a median follow-up of 14.9 years, ADT significantly improved time to distant metastasis. Among these patients, 543 (34%) were model positive, and ADT significantly reduced the risk of distant metastasis compared with RT alone. Among the remaining 1051 patients who were model negative, ADT did not provide benefit.

In the overall validation cohort, the 15-year distant metastasis estimate was 5.9% in the ADT group compared 9.8% for recipients of RT alone, result in a significant 36% lower risk for distant metastasis in the ADT group.

Among patients who were predictive model positive, the 15-year distant metastasis estimate was 4.0% for those who had ADT added to RT compared with 14.4% among patients who received RT alone. The ADT group had a significant 66% lower risk for metastasis compared with the RT-only group.

Among patients who were predictive model negative, the 15-year distant metastasis estimates were 6.9% for ADT recipients and 7.4% for the RT-only group, a nonsignificant difference.

The study successfully demonstrated, in the largest completed phase 3 randomized trial of RT with or without ADT, that the AI biomarker could identify men for whom ADT did or did not significantly decrease their risk for distant metastasis, Dr Spratt said. “Approximately two-thirds of the men [for whom] we currently recommend ADT could safely be spared ADT use,” he said.

The investigators also assessed prostate cancer–specific mortality (PCSM) as a secondary endpoint. In the overall validation cohort, the short-term ADT group had a 15-year event estimate of 4.4%, whereas the RT only group had a 15-year event estimate of 8.6%, with ADT group experiencing a significant 48% lower risk for PCSM.

For patients who were predictive model positive, the 15-year PCSM estimates were 2.6% for ADT recipients compared with 12.7% for those receiving RT alone. The ADT group had a significant 72% lower risk for PCSM. In contrast, for patients who were predictive model negative, 15-year PCSM estimates were 5.3% in the ADT group and 6.5% for those receiving RT alone, a nonsignificant difference between treatment arms.

The clinical trials included in the development of the AI model enrolled men at more than 100 centers across the United States and Canada. About 20% of the patients in these trials were Black. In previous US-based clinical trials, Black men have made up approximately 10% of the participants, according to the investigators.

Dr Spratt said this predictive model may help further a personalized medicine approach for prostate cancer treatment. Because the study included a high percentage of Black patients, the new predictive model also may benefit a patient population at much higher risk for prostate cancer and is known to be undertreated.

The predictive model is analogous to the original Oncotype Dx for breast cancer that was similarly validated using completed phase 3 randomized trial data and endorsed by clinical guidelines approximately 20 years ago, according to Dr Spratt. Currently, Oncotype Dx is used to personalize the use of chemotherapy in women with breast cancer. The new AI-based model can assist with shared decision-making between doctors and patients, and potentially help men avoid the detrimental side effects of ADT without compromising oncologic outcomes, Dr Spratt said.

ADT consistently is one of the major drivers of declines in patient reported quality of life. Dr Spratt said having a predictive biomarker may help avoid overtreatment and substantially benefit a significant number of men.

This AI tool and other tests like the genomic tests are being commonly used, especially among patients who are doing active surveillance.

David Lee, MD, professor of urology and acting chair of the department of urology at the University of California Irvine, said he expects wide adoption of biomarkers using multimodal digital pathology AI-derived platforms.

Noting that ADT is associated with adverse side effects as well as a financial burden to the health care system, Dr Lee commented, “If you can safely avoid it, then having these types of tests can be among one of the best tools we have to help make this decision whether to use ADT or not in intermediate-risk prostate cancer. This AI tool and other tests like the genomic tests are being commonly used, especially among patients who are doing active surveillance.”

Michael Whalen, MD, associate professor of urology and director of the urologic oncology division at George Washington University in Washington, DC, said having training models that can recognize defining tissue features and then correlate these features with outcomes after specific treatments may be highly beneficial. The training set of the new AI biomarker comprised many randomized clinical trials that were conducted in the 1990s.

“The treatment paradigms have significantly evolved since this time,” Dr Whalen said.

“Furthermore, the validation cohort from RTOG 9202 used now-antiquated ADT agents and only for 4 months duration. Current NCCN guidelines recommend 6 months of ADT for unfavorable intermediate-risk patients along with RT. Certainly, an advantage of these older trials is long follow-up time, but we would consider these treatment protocols outdated at this point.” That being said, according to Dr Whalen, a promise of AI in general is the possibility of the learning models to evolve along with the treatment landscape.

The current armamentarium for prognostication includes nomograms, which weigh various clinical and pathologic features to predict a likelihood of a given oncologic outcome. Dr Whalen pointed out, however that currently available tissue-based biomarker assays and gene expression profiling tools (such as Decipher, Oncotype DX, and Prolaris) do not specifically predict ADT-responsiveness. “So, this AI model is certainly interesting. The fact that it has analyzed tissue alone without input of clinical features, unlike a nomogram, is certainly an exciting novel approach,” Dr Whalen said. “I applaud the authors for the exciting momentum here and look forward to more work being done.”

This article originally appeared on Renal and Urology News

Facebook Comments

Register

We’re glad to see you’re enjoying Global Women's Health Academy…
but how about a more personalized experience?

Register for free