Letter to the Editor: Early complications in split liver transplantation: An interpretable machine learning model requires multicenter validation

Abstract

We read with interest the study by Wang et al entitled “Interpretable machine learning model for early complication prediction after split liver transplantation”. Split liver transplantation (SLT) expands the donor pool but is associated with an increased risk of early postoperative complications (EPCs) due to extended resection surfaces and altered partial graft hemodynamics. This study aimed to develop an interpretable machine learning (ML) framework for identifying risk factors of EPCs in adult right trisegment SLT. A retrospective analysis was conducted on 109 recipients, of whom 37 developed EPCs. Four ML algorithms were employed, with random forest demonstrating optimal performance. SHapley Additive exPlanations (SHAP) analysis identified four independent predictors: Log-transformed systemic immune-inflammation index (LnSII), Model for End-Stage Liver Disease (MELD) score, partial lobectomy of segment IV, and intraoperative blood loss. The constructed diagnostic nomogram exhibited excellent discrimination and calibration. Survival analysis further revealed that LnSII and MELD score were significantly associated with five-year overall survival (P < 0.05). SHAP analysis bridges the gap between ML predictions and clinical decision-making, holding promising application prospects. However, the single-center retrospective design of this study imposes limitations, necessitating multicenter validation. Future research should incorporate dynamic postoperative variables and clarify the underlying biological mechanisms.

Key Words: Split liver transplantation; Early postoperative complications; Machine learning; Systemic immune-inflammation index; Partial lobectomy of segment IV

Core Tip: The study by Wang et al developed an interpretable machine learning model for predicting early complications after split liver transplantation. Through novel integration of multiple algorithms with SHapley Additive exPlanations (SHAP) analysis to identify the systemic immune inflammation index, the Model for End-Stage Liver Disease score, intraoperative blood loss, and the removal of the fourth segment of the liver lobe as independent predictive factors. The SHAP analysis also made the decision-making process of the model transparent and visible - not only did it globally display the ranking of the contribution of each factor, but it could also present the specific impact of each feature on the predicted risk of individual patients. Ultimately, a visual nomogram integrating inflammation, disease severity, surgical factors, and blood loss was produced. This is an important practice in advancing liver transplantation towards precision medicine.

TO THE EDITOR

Split liver transplantation (SLT) divides a donor’s liver into two parts for two recipients, effectively expanding the donor pool but carrying higher early postoperative complication (EPC) risks[1,2]. Accurate EPC prediction is critical for optimizing perioperative management, and machine learning (ML) has emerged as a transformative tool in this domain[3,4]. We are particularly interested in a recent study by Wang et al[5] published in the World Journal of Gastroenterology, which innovatively used interpretable ML to develop an EPC prediction model for SLT, providing valuable insights for complication prevention.

This study retrospectively analysed 109 adult patients who underwent right trisegment SLT, with 37 (33.9%) developing EPC[5,6]. Four independent predictors were identified: (1) Log-transformed systemic immune-inflammation index (LnSII), calculated as ln[platelet count × neutrophil count/Lymphocyte count], reflected ischemia-reperfusion injury, infection risk, and impaired tissue repair[7]; (2) Model for End-Stage Liver Disease (MELD) score, determined based on serum bilirubin, creatinine, and internal normalized ratio, the gold standard for assessing liver function reserve and short-term survival risk[8]; (3) Intraoperative blood loss, a surrogate for surgical trauma and technical difficulty[9]; and (4) Partial lobectomy of segment IV (IV PL), a protective factor preventing injury, infection, and bile leakage by removing ischaemic segment IV tissue[5,10]. In the present study, among four ML algorithms—random forest, support vector machine, extreme gradient boosting, and logistic regression—the optimal algorithm was random forest. Subsequently, SHapley Additive exPlanations (SHAP) analysis was integrated to transform the black-box ML model into an interpretable application tool[11]. However, whether this tool can be effectively translated into clinical practice and improve recipient prognosis remains to be verified by prospective cohort studies.

Two findings merit particular emphasis. First, LnSII being identified as the top predictor highlights the crucial role of systemic inflammation in post-transplant outcomes, consistent with accumulating evidence regarding the impact of the immune system on surgical recovery[12]. Moreover, the protective effect of IV PL warrants further discussion. Given that segment IV is perfused by both the right and left hepatic arteries in most cases, certain areas may become ischemic after the operation[13]. However, other factors, such as surgical technique and intraoperative bleeding, also require further elucidation.

Nevertheless, this research has certain limitations. As a single-center retrospective study, the sample size (n = 109) limits statistical power, and the predominantly Chinese cohort (with a high prevalence of hepatitis B) limits generalizability. External multicenter prospective validation is therefore indispensable, supplemented by multicollinearity testing and multiple-comparison corrections to improve the reliability of the results[6]. Additionally, postoperative dynamic variables (e.g., serial SII measurements) and radiological assessments (e.g., B-ultrasound or computed tomography to evaluate segment IV perfusion or atrophy) were not included, which may further clarify the prognostic value of IV PL.

Future research should focus on the following directions. Primarily, multicenter collaborative studies are crucial to validate the model’s generalizability. Poeteriorly, the range of variables should be expanded to include preoperative patient body composition parameters, more detailed immunological indicators, and radiomics features, thereby constructing a more robust predictive model. Incorporating emerging biomarkers and imaging features into the model is also anticipated to further enhance predictive accuracy and facilitate the establishment of a dynamic monitoring mechanism, enabling the model to adapt to evolving clinical practices and significantly improve its predictive capacity. Additionally, the interactive mechanisms between different predictive factors warrant in-depth exploration and practical application, particularly the potential correlation between systemic immune inflammation and hepatic segmentectomy. For instance, the inflammatory state of the liver can significantly influence the recipient’s adaptive immune response through multiple pathways, such as enhancing antigen presentation and providing co-stimulatory signals. Furthermore, by exploring various surgical techniques and strategies and integrating preoperative assessment indicators (e.g., graft-to-recipient weight ratio), surgical protocols can be refined to facilitate precise liver transplantation management. From a broader perspective, this research provides valuable methodological foundations and guidance for developing digital twin technologies and intelligent medical solutions in liver transplantation.

This study developed an interpretable ML model to predict EPC in patients who underwent SLT, integrating clinical and surgical variables and accounting for the protective effect of segment IV resection. This enables a robust individualised risk assessment and offers novel insights into the pathogenesis of EPC, thereby supporting surgical optimisation. This study advances personalised and precise postoperative management strategies for liver transplantation.