Nomograms for Predicting Hepatocellular Carcinoma Recurrence and Overall Postoperative Patient Survival

Lidi Ma, Kan Deng, Cheng Zhang, Haixia Li, Yingwei Luo, Yingsi Yang, Congrui Li, Xinming Li, Zhijun Geng, Chuanmiao XieFrontiers in Oncology 2022 · DOI

TL;DR

This paper introduces novel nomograms that combine clinical, pathological, and radiological data to accurately predict the 5-year overall survival (OS) and early recurrence of hepatocellular carcinoma (HCC) patients after surgical resection. These nomograms offer improved predictive power over the existing BCLC staging system, providing clinicians with a valuable tool for personalized follow-up and treatment strategies.

What Problem Does This Paper Address?

Hepatocellular carcinoma (HCC) is a deadly cancer with a high recurrence rate after surgical resection, often leading to poor patient outcomes. While the Barcelona Clinic Liver Cancer (BCLC) staging system is widely used, it has limitations in precisely predicting individual patient prognosis, especially for early recurrence and long-term survival. Existing models often focus on either imaging or pathological factors, but rarely integrate both comprehensively. This study aims to bridge this gap by developing more accurate predictive models that incorporate a broader range of data, thereby enabling better clinical decision-making and potentially improving patient survival.

Methods

The researchers employed a retrospective approach, collecting data from 494 HCC patients who underwent hepatectomy. They utilized a two-stage modeling strategy:

  1. Random Survival Forest (RSF): This machine learning algorithm was used to identify the most relevant features from a large set of clinicopathological and radiological data. RSF is adept at handling non-linear relationships and complex interactions within the data, making it suitable for identifying key prognostic factors.
  2. Multivariate Cox Regression: Following feature selection by RSF, multivariate Cox regression was applied to identify independent prognostic factors and construct the nomograms. This statistical method is a standard for survival analysis and allows for the quantification of the risk associated with each identified factor.

The performance of the developed nomograms was rigorously evaluated using concordance indices (C-index) for discrimination and calibration curves for accuracy. Decision curve analysis (DCA) was also used to assess the clinical utility of the nomograms. The models were internally validated on a separate testing cohort and externally validated using data from two independent cohorts.

Datasets

The study utilized data from a total of 494 HCC patients who underwent hepatectomy.

  • Training Cohort: 346 patients who underwent surgery prior to April 2015.
  • Testing Cohort: 148 patients who underwent surgery after April 2015.
  • External Cohort 1: 86 HCC patients from Southern Medical University Zhujiang Hospital.
  • External Cohort 2: 52 HCC patients from Hunan Cancer Hospital.

The data included a comprehensive set of clinicopathological information (age, sex, HBsAg status, liver cirrhosis, Child-Pugh class, BCLC stage, tumor differentiation, microvascular invasion, etc.) and radiological features derived from multiparametric MRI scans obtained within one month before surgery.

Key Insights and Ideas

  • Integration is Key: The study strongly suggests that combining clinical, pathological, and radiological data is superior to relying on any single category for predicting HCC prognosis. This holistic approach captures a more complete picture of the disease's behavior.
  • RSF as a Powerful Feature Selector: The use of Random Survival Forest proved effective in sifting through a large number of potential predictors to identify the most impactful ones, reducing dimensionality and improving model robustness.
  • Nomograms Offer Actionable Insights: The developed nomograms provide a visual and quantitative tool that clinicians can use to estimate a patient's specific risk of recurrence and survival, facilitating personalized treatment planning and follow-up schedules.
  • Reoperation as a Viable Option: For patients who experience recurrence and are in good condition, reoperation (repeat hepatectomy) appears to be a more beneficial treatment strategy compared to palliative care or comprehensive treatments, leading to better overall survival after recurrence.

Key Results

The study successfully developed two nomograms: one for predicting 5-year overall survival (OS) and another for predicting 2-year recurrence-free survival (RFS).

  • OS Nomogram: Based on seven factors (BCLC stage, non-smooth margin, maximum tumor diameter, microvascular invasion, age, AST, and differentiation), it achieved a C-index of 0.787 in the training cohort and 0.711 in the testing cohort.
  • RFS Nomogram: Based on four factors (BCLC stage, non-smooth margin, maximum tumor diameter, and microvascular invasion), it achieved a C-index of 0.717 in the training cohort and 0.701 in the testing cohort.
  • Superiority over BCLC: Both nomograms demonstrated significantly better predictive performance (higher C-index) and calibration compared to the BCLC staging system across training, testing, and external validation cohorts. For instance, the OS nomogram's C-index was 0.787 vs. 0.678 for BCLC in the training cohort.
  • External Validation: The nomograms showed good performance in external validation cohorts, with C-indices for RFS ranging from 0.716 to 0.729 and for OS from 0.748.
  • Treatment Strategy: Patients who underwent reoperation after recurrence had a higher survival rate compared to those receiving comprehensive treatments or supportive care.

Limitations and Future Work

The authors acknowledge several limitations:

  • HBsAg Dominance: A high proportion of patients (85.6%) had a history of positive HBsAg, which is a major cause of HCC in China. This limits the generalizability of the nomograms to Western populations without further validation.
  • External Cohort Size: While external validation was performed, the sample sizes of the external cohorts were relatively small.
  • Limited Treatment Comparison: The study primarily compared surgical vs. non-surgical treatments for recurrent HCC. Future work should involve comparing various specific treatments for recurrent HCC.
  • Model Scope: The current nomograms combine clinicopathological and radiological data. Future research could explore incorporating deep learning, artificial intelligence, or genomic data to build even more robust models.

The authors suggest the need for large-scale studies and multi-center clinical trials, particularly in diverse geographical regions, to further validate and refine these prognostic models.

Summary updated

Abstract

The nomogram, based on clinical, pathological, and radiological factors, demonstrated good accuracy in estimating OS and recurrence, which can guide follow-up and treatment of individual patients. Reoperation may be the best option for patients with recurrence in good condition.
BibTeX
@inproceedings{ma2022nomograms,
  author = {Lidi Ma and Kan Deng and Cheng Zhang and Haixia Li and Yingwei Luo and Yingsi Yang and Congrui Li and Xinming Li and Zhijun Geng and Chuanmiao Xie},
  title = {{Nomograms for Predicting Hepatocellular Carcinoma Recurrence and Overall Postoperative Patient Survival}},
  booktitle = {Frontiers in Oncology},
  year = {2022},
  doi = {10.3389/fonc.2022.843589},
}
Figure