A multi-modal artificial intelligence (MMAI) model can accurately predict long-term outcomes in patients with high-risk prostate cancer, according to research presented at the 2023 ASCO Genitourinary Cancers Symposium.

The MMAI model proved more accurate than standard clinicopathological markers for predicting distant metastasis and prostate cancer-specific mortality, said study presenter Daniel E. Spratt, MD, of Case Western Reserve University in Cleveland, Ohio.

The MMAI model, ArteraAI Prostate, was created using pathologic, clinical, and imaging data from multiple phase 3, randomized trials. In the current analysis, Dr Spratt and colleagues tested the model in 1088 patients from 6 trials who had at least 1 high- or very high-risk feature according to National Comprehensive Cancer Network (NCCN) guidelines.

Continue Reading

At baseline, the patients’ median age was 68 (range, 62-73) years, and 76.6% were White. The median prostate-specific antigen (PSA) level was 21.4 ng/mL, 31.5% of patients had a Gleason score of 9-10, 21.2% had T3b-T4 disease, and 49.4% had 1 NCCN high- or very high-risk feature.

The researchers evaluated the cumulative incidence of distant metastasis and prostate cancer-specific mortality using traditional markers and the MMAI model. The median follow-up was 10.4 years.

The MMAI model outperformed traditional markers for predicting distant metastasis at 5 years and prostate-cancer specific mortality at 15 years.

The area under the curve (AUC) for predicting distant metastasis at 5 years was:

  • 0.50 for age
  • 0.56 for PSA as a continuous variable
  • 0.50 for PSA less than 10 ng/mL vs 10-20 ng/mL vs more than 20 ng/mL
  • 0.61 for Gleason score of 7 or lower vs 8 vs 9-10
  • 0.60 for T1-2 disease vs T3-4 disease
  • 0.64 for number of high-risk features
  • 0.71 for the MMAI model.

The AUC for predicting prostate cancer-specific mortality at 15 years was:

  • 0.45 for age
  • 0.51 for PSA as a continuous variable
  • 0.50 for PSA less than 10 ng/mL vs 10-20 ng/mL vs more than 20 ng/mL
  • 0.66 for Gleason score of 7 or lower vs 8 vs 9-10
  • 0.48 for T1-2 disease vs T3-4 disease
  • 0.60 for number of high-risk features
  • 0.73 for the MMAI model.

In a multivariate analysis, the MMAI model was an independent predictor of distant metastasis (subdistribution hazard ratio [sHR], 1.90; 95% CI, 1.57-2.31; P <.001) and prostate cancer-specific mortality (sHR, 2.12; 95% CI, 1.72-2.62; P <.001).

“[T]his MMAI model, which now you can find in NCCN guidelines, is successfully validated,” Dr Spratt said. “The MMAI model has improved discrimination and was independently prognostic compared to our standard care clinical and pathologic variables. But really, ultimately, this biomarker enables improved risk stratification to estimate absolute risk, which can be used to help personalize shared decision-making with our patients.”

Disclosures: This research was partly supported by Artera, Inc., Pfizer, Bristol Myers Squibb, Sanofi Inc., and Takeda Pharmaceuticals. Some study authors declared affiliations with biotech, pharmaceutical, and/or device companies. Please see the original reference for a full list of disclosures.

Reference Spratt DE, Liu VYT, Yamashita R, et al. Patient-level data meta-analysis of a multi-modal artificial intelligence (MMAI) prognostic biomarker in high-risk prostate cancer: Results from six NRG/RTOG phase III randomized trials. ASCO GU 2023. February 16-18, 2023. Abstract 299.

This article originally appeared on Cancer Therapy Advisor

Topics:

Genitourinary (GU) Cancers Symposium Prostate Cancer Urology Conferences