drankitjoshi/NAHT-Model-Prostate_Cancer_Outcome_Predictor

1---2title: NAHT Outcome Predictor3emoji: 🩺4colorFrom: blue5colorTo: indigo6sdk: gradio7sdk_version: 6.20.08python_version: "3.12"9app_file: app.py10pinned: false11license: cc-by-nc-4.012---13 14# NAHT Outcome Predictor — High-Risk Prostate Cancer15 16Machine-learning models predicting outcomes after **6 months of Neoadjuvant Hormonal Therapy (NAHT)17followed by Robot-assisted Radical Prostatectomy** with bilateral pelvic lymph node dissection.18 19## Models20 21| Model | Timing | Cross-validated AUC (95% CI) |22|---|---|---|23| Positive surgical margin | Pre-operative | 0.759 (0.685– 0.833) |24| Lymph node positivity | Pre-operative | 0.842 (0.781– 0.894) |25| Biochemical recurrence | Post-operative | 0.782 (0.710– 0.852) |26| BCR-free survival (Cox) | Post-operative | C-index 0.755 (0.702–0.807) |27 28Performance was estimated by **nested cross-validation** (n=200, single centre): the algorithm and its29hyperparameters were selected only inside training folds, so no test observation influenced model30selection. These figures are therefore lower — but more trustworthy — than those from a single31favourable train/test split.32 33## Key finding34 35NAHT response metrics (PSA reduction %, volume reduction %, T-stage downstaging) add **statistically36significant** predictive value beyond baseline characteristics:37 38- Biochemical recurrence: AUC 0.648 → 0.758 (ΔAUC +0.110, DeLong p = 0.0049)39- Positive margin: AUC 0.612 → 0.759 (ΔAUC +0.147, DeLong p = 0.0005)40- Node positivity: AUC 0.821 → 0.842 (p = 0.39, not significant)41 42## Limitations43 44- **Research tool only.** Not a medical device; not validated for clinical decision-making.45- **No external validation.** Single-centre, retrospective, n=200.46- Pathological complete response (n=6, 3.0%) was too rare to model reliably and is not predicted.47 48## Contact49 50[**drankitjoshiurologist@gmail.com**]51