1165 - Assessing MMAI Algorithmic Fairness within Racial and Age Subgroups In NRG/RTOG Post-Radical Prostatectomy Phase III Trials

05:30pm - 05:35pm ET

Room 205

Presenter(s)

Shalini Moningi, MD - Cleveland Clinic Taussig Cancer Institute, Cleveland, OH

S. Moningi¹, Y. Ren², A. Pollack³, D. Croucher⁴, K. O'Shaughnessy⁴, E. Stewart⁵, A. Esteva⁴, H. Lukka⁶, A. G. Martin⁷, J. P. Bahary⁸, J. M. Michalski⁹, A. G. Balogh Jr¹⁰, A. D. Currey¹¹, A. P. Dicker¹², M. Duclos¹³, N. S. Kapadia¹⁴, S. Pugh¹⁵, P. L. Nguyen¹⁶, P. Tran¹⁷, and A. C. Olson¹⁸; ¹Cleveland Clinic, Cleveland, OH, ²Artera, Inc., Los Altos, CA, ³University of Miami, Miami, FL, ⁴ArteraAI, Los Altos, CA, ⁵Artera, Los Altos, CA, ⁶Juravinski Cancer Centre, Hamilton, ON, Canada, ⁷CHU de Québec – Université Laval, Québec, QC, Canada, ⁸Department of Radiation Oncology, Centre Hospitalier de l'Université de Montréal, Montreal, Quebec, Montreal, QC, Canada, ⁹Washington University School of Medicine, St. Louis, MO, ¹⁰University of Calgary, Calgary, AB, Canada, ¹¹Department of Radiation Oncology Medical College of Wisconsin, Milwaukee, WI, ¹²Department of Radiation Oncology, Sidney Kimmel Medical College at Thomas Jefferson University, Philadelphia, PA, ¹³Department of Radiation Oncology, McGill University Health Centre, Montreal, QC, Canada, ¹⁴Dartmouth Hitchcock Medical Center, Lebanon, NH, ¹⁵NRG Oncology Statistics and Data Management Center, Philadelphia, PA, ¹⁶Mass General Brigham, Boston, MA, ¹⁷MD Anderson, Houston, TX, ¹⁸UPMC-Shadyside Hospital, Pittsburgh, PA

Purpose/Objective(s): Consistent prognostication across demographic subgroups is critical for clinical adoption of prostate cancer biomarkers. African American (AA) men have more aggressive disease and worse prostate cancer–specific mortality. Older age influences disease management and outcomes. Prognostic models that perform differently across race or age may exacerbate existing disparities. A locked multimodal artificial intelligence (MMAI) digital pathology (DP) model was previously developed and its performance validated for prognostication in post–radical prostatectomy (RP) patients. Prior MMAI fairness was shown in a definitive RT cohort. Here, we evaluate Post-RP MMAI score distributions and prognostic associations across racial and age-based subgroups to assess potential demographic bias.

Materials/Methods:

The locked Post-RP MMAI model (V1.1) was applied to patients from RTOG 0534 and 9601 trials (used for model development and internal validation). Men with available pre-treatment RP histopathology slides and clinical data were included. MMAI scores were generated using RP whole slide images and clinicopathologic features (age, PSA, pathological Gleason grade group, surgical margin status, and pT-stage). Score distributions were compared across race (AA vs non-AA) and age (<70 vs =70 years). Associations between MMAI score and time to DM were evaluated using Fine-Gray competing risk regression (death without DM as competing event). Interaction terms assessed differential prognostic associations by race and age, and subgroup-specific conditional hazard ratios (HRs) were estimated within demographic subgroups, adjusting for trial to account for baseline risk differences.

Results:

The study cohort included 1,855 men, of whom 185 (10%) were AA and 416 (22%) were aged =70 years. DM incidence was similar across racial and age subgroups. MMAI score distributions showed substantial overlap across demographic subgroups, with no significant differences in median scores (Wilcoxon p>0.05). MMAI score was prognostic for DM in the overall cohort (sHR 2.59, 95% confidence intervals (CI): 2.25-2.99, p<0.001), while race and age were not. No significant interaction was observed between MMAI score and race (interaction p=0.35) or age (interaction p=0.38), indicating no evidence of effect modification by these factors. Subgroup-specific conditional HRs showed consistent prognostic associations between MMAI score and DM within AA (HR: 3.17, 95% CI: 2.00-5.02) and non-AA (HR: 2.52, 95% CI: 2.16-2.94) patients, and within younger (HR: 2.68, 95% CI: 2.26-3.17) and older (HR: 2.31, 95% CI: 1.75-3.06) patients.

Conclusion: In this multi-trial post-RP cohort, the MMAI DP biomarker demonstrated similar score distributions and comparable prognostic DM associations across race and age subgroups, supporting generalizability of MMAI prognostic performance in these populations.