Development and Internal Validation of Machine Learning Survival Models for Overall Survival After Repeat Hepatectomy for Hepatocellular Carcinoma: A Retrospective Cohort Study

Background Repeat hepatectomy is a potentially curative option for intrahepatic recurrence after hepatocellular carcinoma (HCC) resection, but outcomes are heterogeneous and validated risk tools are lacking. We developed and internally validated machine learning models for overall survival (OS) after repeat hepatectomy using routinely available variables. Methods This single-centre retrospective cohort included 267 patients with pure HCC who underwent a second curative-intent hepatectomy (2016–2025). Patients were divided into a training set (n = 186) and a held-out test set (n = 81) at 7:3. A pre-specified panel of 24 features was screened entirely within the training set. Eight survival models were trained, including a prespecified weighted rank ensemble as the delivery model. Discrimination was assessed by Harrell's C-index with bootstrap 95% CIs, and calibration by logistic recalibration at 12, 24 and 36 months. Results The selected eight-feature subset comprised AFP, ACE, ADA, RDW-CV, cholinesterase, D-dimer, anti-HBs and tumour size. The weighted rank ensemble achieved a test C-index of 0.7534 (95% CI 0.6321–0.8518), and the risk score remained independently associated with OS after adjustment for prespecified clinicopathological covariates (HR 10.01 per SD, 95% CI 5.39–18.57; P < 0.001). In the held-out test set, high-risk patients had worse survival (HR 3.78, 95% CI 1.09–13.06; log-rank P = 0.024). Calibration slopes were below 1 at all time points (0.613, 0.327 and 0.411), indicating that absolute risks require recalibration. Conclusions A prespecified weighted rank ensemble showed competitive discrimination and independent prognostic value after repeat hepatectomy for HCC. Recalibration and external validation in larger cohorts are required before clinical use.

Authors

Institutions

Publication Details

Journal
Research Square
Published
2026-11-19
DOI
https://doi.org/10.21203/rs.3.rs-11587697/v1
Primary Topic
Hepatocellular Carcinoma Treatment and Prognosis
Type
preprint
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
preprint

Development and Internal Validation of Machine Learning Survival Models for Overall Survival After Repeat Hepatectomy for Hepatocellular Carcinoma: A Retrospective Cohort Study

Li Li, Zhitao Lin, Shixiong Liang, Dennis Wainaina
Research Square
Hepatocellular Carcinoma Treatment and Prognosis
preprint

Development and Internal Validation of Machine Learning Survival Models for Overall Survival After Repeat Hepatectomy for Hepatocellular Carcinoma: A Retrospective Cohort Study

Li Li, Zhitao Lin, Shixiong Liang, Dennis Wainaina
preprint en

Abstract

Background Repeat hepatectomy is a potentially curative option for intrahepatic recurrence after hepatocellular carcinoma (HCC) resection, but outcomes are heterogeneous and validated risk tools are lacking. We developed and internally validated machine learning models for overall survival (OS) after repeat hepatectomy using routinely available variables. Methods This single-centre retrospective cohort included 267 patients with pure HCC who underwent a second curative-intent hepatectomy (2016–2025). Patients were divided into a training set (n = 186) and a held-out test set (n = 81) at 7:3. A pre-specified panel of 24 features was screened entirely within the training set. Eight survival models were trained, including a prespecified weighted rank ensemble as the delivery model. Discrimination was assessed by Harrell's C-index with bootstrap 95% CIs, and calibration by logistic recalibration at 12, 24 and 36 months. Results The selected eight-feature subset comprised AFP, ACE, ADA, RDW-CV, cholinesterase, D-dimer, anti-HBs and tumour size. The weighted rank ensemble achieved a test C-index of 0.7534 (95% CI 0.6321–0.8518), and the risk score remained independently associated with OS after adjustment for prespecified clinicopathological covariates (HR 10.01 per SD, 95% CI 5.39–18.57; P < 0.001). In the held-out test set, high-risk patients had worse survival (HR 3.78, 95% CI 1.09–13.06; log-rank P = 0.024). Calibration slopes were below 1 at all time points (0.613, 0.327 and 0.411), indicating that absolute risks require recalibration. Conclusions A prespecified weighted rank ensemble showed competitive discrimination and independent prognostic value after repeat hepatectomy for HCC. Recalibration and external validation in larger cohorts are required before clinical use.

Research Square
Guangxi Medical University (CN), Tumor Hospital of Guangxi Medical University (CN), Lanzhou University (CN)
Hepatocellular Carcinoma Treatment and Prognosis
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon