Main Session
Sep 29
QP 28 - Actionable Biomarkers: Guiding Therapy Selection and De-escalation

1163 - Beyond NCCN: How a Prognostic and Predictive Multimodal AI Biomarker Drives ADT De-Escalation In Prostate Cancer

05:20pm - 05:25pm ET
Room 205

Presenter(s)

Eric Wegener, MBBS Headshot
Eric Wegener, MBBS - Genesis Cancer Care, Newcastle, NSW

E. Wegener1,2, M. Ng3,4, M. Guerrieri3, J. de Leon5, S. Ramani6, M. V. Dreosti7, T. S. Lim8,9, M. Chao10,11, A. C. Ho12, S. Sia13, S. McClintock14, D. Foreman15, K. McMillan16,17, M. Handmer18, M. Frydenberg19,20, L. M. Wong21,22, D. Hayne23, J. Yaxley24, P. Stricker25, and J. M. Martin1,2; 1University of Newcastle, Newcastle, Australia, 2GenesisCare Gateshead, Newcastle, NSW, Australia, 3GenesisCare St Vincent's Hospital, Melbourne, VIC, Australia, 4Department of Medicine, University of Melbourne, Melbourne, Australia, 5GenesisCare St. Vincent's Clinic, Sydney, NSW, Australia, 6GenesisCare John Flynn Private Hospital, Gold Coast, Australia, 7Genesis Cancer Care SA, Adelaide, SA, Australia, 8University of Western Australia, Department of Surgery, Perth, Australia, 9GenesisCare, Fiona Stanley Hospital, Perth, Australia, 10Genesis Care, Ringwood, Melbourne, VIC, Australia, 11Monash University, Melbourne, VIC, Australia, 12Genesis Care St Vincent's Hospital, Sydney, Australia, 13GenesisCare Murdoch, Perth, Western Australia, Australia, 14Gold Coast Private Hospital, Gold Coast, Australia, 15South Terrace Urology, Adelaide, Australia, 16Monash University- Eastern Health Clinical School, Clayton, Australia, 17Knox Private Hospital, Wantirna, Australia, 18John Hunter Hospital, Newcastle, Australia, 19Australian Urology Associates, Malvern, Australia, 20Monash University Faculty of Medicine, Melbourne, Australia, 21Department of Urology, St Vincent’s health, Melbourne, VIC, Australia, 22Department of Surgery, University of Melbourne, Melbourne, Australia, 23The University of Western Australia, Perth, Australia, 24Wesley Urology Clinic, Auchenflower, Australia, 25St Vincent's Hospital Sydney, Sydney, Australia

Purpose/Objective(s): While adding short-term androgen deprivation therapy (ADT) to definitive radiotherapy for intermediate-risk (IR) prostate cancer yields a historic 5% absolute overall survival benefit, universal application causes unnecessary toxicity. We evaluated the real-world impact of a validated multimodal artificial intelligence (MMAI) predictive biomarker on shared decision-making (SDM) regarding ADT in IR prostate cancer.

Materials/Methods: ASTuTE (ACTRN12623000713695p) is a prospective, multicenter study. SDM-ADT recommendations pre- and post-MMAI results were recorded. The primary endpoint was decision discordance (McNemar’s test). Statistical analyses included multivariate logistic regression to identify decision drivers (covariates: MMAI predictive result, ISUP Grade Group, PSA, T-stage, and age), evaluation of discordance based on absolute 10-year distant metastasis (DM) risk, and a practice analysis comparing early (Cohort A) versus late (Cohort B) enrollees separated by interim analysis.

Results: Of 786 analyzed patients (74.1% NCCN Unfavorable IR [UIR]; 25.9% Favorable IR [FIR]) MMAI integration drove significant ADT de-escalation. Notably, 76.5% (176/230) of cases changed from "Use ADT" to "Omit ADT". Conversely, only 4.0% (21/519) escalated to ADT. This yielded a 20.7% net absolute reduction in ADT use (p<0.001).

Pre-test, ISUP Grade Group 3 (OR 3.28, p<0.001) and PSA (Log-OR 2.08, p<0.001) drove SDM-ADT choices. Post-test, the MMAI result (OR 9.25, p<0.001) dominated traditional clinical factors (Table 1).

Adherence to ADT omission was 95.4% versus 34.6% for addition (p<0.001). Discordance was prognostic-driven. ADT was routinely omitted despite the predictive benefit if the absolute 10-year DM risk was low (median 2.9% "Omit ADT" vs. 4.2% "Use ADT", p<0.001).

Post-interim analysis (Nov 2024, n=200), later enrollees (Cohort B) had less FIR disease than early enrollees (Cohort A) (24.3% vs 28.1%). Yet, overall pre-test SDM-ADT recommendations dropped from 32.5% to 29.4%. Specifically for UIR patients, baseline SDM-ADT recommendations fell from 42.5% to 33.9%.

Conclusion: The MMAI predictive biomarker significantly reduces real-world ADT use in IR prostate cancer. Beyond widespread ADT de-escalation, baseline ADT recommendations declined over time, MMAI outweighed clinical factors in SDM, integration of prognostic with predictive biomarker data increased and the use of the tool became more risk-adapted. Follow-up is ongoing regarding clinical outcomes.

Table 1: Multivariate Analysis – Predictors of Post-Test ADT Recommendation

Predictor

Odds Ratio (OR)

95% Confidence Interval

P-Value

MMAI ADT Predictive Result (More vs Less Benefit)

9.25

5.37 – 15.93

< 0.001

ISUP Grade Group 3 vs 1-2

3.73

2.07 – 6.74

< 0.001

PSA (Log Scale)

2.12

1.18 – 3.82

0.012

High T-Stage (T2c vs T1-T2b)

1.03

0.56 – 1.90

0.921

Age (Continuous)

0.95

0.91 – 0.99

0.019