Lesion selection and external evaluation of clinical-imaging survival models after transarterial chemoembolization for hepatocellular carcinoma

In 69 HCC-TACE-Seg patients with 15 deaths by day 365, using CT measurements from all retained tumors did not yield more accurate first-year mortality probabilities than clinical-only modeling.

• Why it matters: Pretreatment CT can quantify tumor features for TACE in HCC, but it is unclear whether expanding lesion measurements beyond shared clinical predictors improves mortality prediction in a new cohort.
• What they did: The study compared clinical-only models against models adding CT features from the largest, two largest, or all retained annotated tumors, using a regularized discrete-time survival estimator and repeated nested cross-validation in 228 WAW-TACE patients plus frozen external evaluation in 69 HCC-TACE-Seg patients.
• The result: Externally, the all-lesion model showed the highest discrimination point estimates, but its mortality probabilities were less accurate and showed calibration consistent with overly extreme predicted risks; local recalibration reduced day-365 error yet left median adjusted error higher than the adjusted clinical model.

AI summary of the abstract · may contain errors · not medical advice · read the paper.