Update src/cox_train_once.py

src/cox_train_once.py

@@ -11,6 +11,7 @@ from lifelines.exceptions import ConvergenceError
 
 from src.survival_utils import prepare_cox_df, fit_cox
 from src.cox_persist import save_cox_artifacts
+from src.calibration_utils import bootstrap_c_index_ci
 
 
 def _default_cox_dir() -> Path:
@@ -155,6 +156,24 @@ def train_and_save_cox(
     except Exception as e:
         raise RuntimeError(f"Cox fit error: {e}")
 
+    # ------------------------------------------------------------------
+    # Bootstrap 95% CI for the training-cohort C-index.
+    # Risk scores come from the fitted Cox model's partial hazard prediction.
+    # ------------------------------------------------------------------
+    try:
+        partial_hazard = cph.predict_partial_hazard(df_cox).values.ravel()
+        c_point, c_lo, c_hi = bootstrap_c_index_ci(
+            durations=df_cox["OS_time_days"].values,
+            events=df_cox["Event_clean"].values,
+            risk_scores=partial_hazard,
+            n_bootstraps=1000,
+            seed=42,
+        )
+    except Exception:
+        c_point, c_lo, c_hi = (float(cph.concordance_index_), float("nan"), float("nan"))
+ 
+
+
     # Resolve output directory
     out_dir = Path(save_dir) if save_dir is not None else _default_cox_dir()
     out_dir.mkdir(parents=True, exist_ok=True)
@@ -178,6 +197,9 @@ def train_and_save_cox(
             "n_events": int(df_cox["Event_clean"].sum()) if "Event_clean" in df_cox.columns else None,
             "duration_col": "OS_time_days",
             "event_col": "Event_clean",
+            "c_index": float(c_point) if c_point is not None else None,
+            "c_index_ci_low": float(c_lo) if c_lo is not None else None,
+            "c_index_ci_high": float(c_hi) if c_hi is not None else None,
         },
     )