Noninvasive prediction model for predicting spontaneous tumor necrosis in hepatocellular carcinoma and prognostic study.

Introduction
Primary liver cancer is the sixth most prevalent malignant tumor worldwide and the third leading cause of cancer-related mortality, of which 75–85% are hepatocellular carcinoma (HCC) [1]. Liver transplantation and hepatectomy are effective modalities for the treatment of HCC. However, postoperative mortality remains high because of the risk of cancer recurrence and metastasis, including factors like microvascular invasion and tumor necrosis [2,3].
Tumor necrosis is a common pathological phenomenon in solid tumors and a prevalent feature of many aggressive, fast-growing tumors, associated with poor prognosis, and a significantly increased risk of metastasis [4]. The accelerated growth of malignant cells may result in inadequate perfusion of the central region of the tumor, leading to deficiencies in nutrients and oxygenation, and ultimately, the tumor tissues undergo necrosis [5]. In addition, tumor necrosis can disrupt the tissue structure and influence the process of tumor neovascularization, resulting in abnormal vascular leakage that facilitates tumor metastasis [6]. According to the literature published in recent years, approximately half of the patients with HCC who underwent hepatectomy developed tumor necrosis. Moreover, tumor necrosis is associated with poorer survival and it also increases the risk of cancer recurrence and extrahepatic metastasis [7,8]. Similarly, studies of tumor necrosis in other malignancies, such as colorectal and kidney cancer, have both shown that spontaneous tumor necrosis is relevant to a reduction in survival rates [9,10]. Therefore, early recognition of the occurrence of tumor necrosis in patients with HCC is of great significance.
Nomograms are widely used in various disease prediction models and have good applicability to help clinicians make decisions [11]. A recent study utilized preoperative imaging features of tumors to forecast the likelihood of tumor necrosis in patients with renal cell carcinoma, incorporating parameters such as tumor size and intratumoral vessels [12]. However, this study only recognized a limited number of predictors. Moreover, a comprehensive predictive model for HCC with tumor necrosis has not yet been developed.
Therefore, a detailed analysis of patients’ preoperative biochemical indicators and imaging data were conducted to determine whether these variables could be used to independently forecast tumor necrosis in patients with HCC. Subsequently, an original nomogram was developed to predict spontaneous tumor necrosis. In addition, we evaluated the prognostic value of tumor necrosis between two groups according to patients’ survival outcomes.

Materials and methods

Study design and population
Preoperative clinical data were collected from patients with HCC who underwent hepatectomy at the Department of Hepatobiliary and Pancreatic Surgery, Zhongnan Hospital of Wuhan University, between 1 January 2015 and 31 May 2024. Inclusion criteria were as follows: (a) patients who received a standard liver resection; (b) postoperative pathologic diagnosis of HCC; (c) aged between 18 and 80 years; (d) Child–Pugh classification as A or B; (e) MRI-enhanced or enhanced computed tomography (CT) liver scan was performed 2 weeks before operation. Exclusion criteria were as follows: (a) received any form of anticancer therapy before surgery; (b) mixed liver cancer; (c) previous history of malignant tumors; (d) presence of extrahepatic lymph node and distant organ metastasis; (e) pathologic diagnosis of tumor necrosis is unclear; (f) missing information. Finally, 495 eligible patients with HCC were enrolled in the study (Fig. 1). A 7 : 3 randomization ratio was employed for the patients in the training cohort (TC, n = 348) and the validation cohort (VC, n = 147). The recurrence of cancer was diagnosed through the detection of new local or distant metastases on imaging. The survival endings included overall survival (OS) and recurrence-free survival (RFS). The OS was calculated according to the following method: the date of hepatectomy was considered as the starting point, and the endpoint was either the date of death from any cause or the last follow-up. The RFS was calculated by the following means: the date of operation to the date of initial cancer recurrence, the date of death from any cause, or the date of the last follow-up.

Histopathology
All patients with HCC received a radical hepatectomy. The presence of tumor necrosis was diagnosed by the histopathological examination of postoperative tumor samples, the results of which were independently evaluated by two senior pathologists.

Preoperative data
We collected the following laboratory variables: aspartate aminotransferase (AST), alanine aminotransferase, total bilirubin, albumin (ALB), γ-glutamyl transferase (GGT), alkaline phosphatase (ALP), hemoglobin (HB), platelets (PLTs), white blood cells, red blood cells, monocytes, lymphocytes, neutrophils, neutrophil-to-lymphocyte ratio (NLR), platelet-to-lymphocyte ratio (PLR), lymphocyte-to-monocyte ratio (LMR), aggregate index of systemic inflammation (AISI), systemic immune inflammation (SII), systemic inflammation response index (SIRI), prognostic nutritional index (PNI), aspartate transaminase-to-platelet ratio index (APRI), aspartate transaminase-to-neutrophil ratio index (ANRI), alpha-fetoprotein (AFP), carcinoembryonic antigen, carbohydrate antigen 199, and hepatitis B surface antigen. AISI, SII, SIRI, and PNI were calculated using the following formulas: AISI = (neutrophils × PLTs × monocytes)/lymphocytes, SII = (neutrophils × PLTs)/lymphocytes, SIRI = (neutrophils × monocytes)/lymphocytes, PNI = [10 × serum ALB (g/dl)] + (0.005 × lymphocytes). We also collected the preoperative enhanced MRI and CT indicators, including tumor size, tumor number, liver cirrhosis status, and vascular invasion. Clinical tumor–node–metastasis (TNM) staging was assessed using the above information. These tests were carried out 2 weeks before the surgical procedure. In addition, we also collected basic demographic data. These potential predictors of tumor necrosis were selected by considering published studies, pathophysiological knowledge, and clinical practice.

Determination of cut-off values for variables
We used receiver operating characteristics (ROCs) curves to calculate the ideal cut-off values for AFP, NLR, PLR, LMR, SII, SIRI, AISI, PNI, APRI, and ANRI. Furthermore, we determined the cut-off values for other continuous variables using their appropriate medical reference ranges.

Statistical analysis
The categorical variables were expressed as numbers and percentages, and the χ2 or Fisher’s exact test was used to compare differences between groups. The continuous variables were expressed as a median and interquartile range or mean ± SD, and the t test or Kruskal–Wallis tests were used to compare differences between groups. The log-rank test and the Kaplan−Meier (KM) survival analysis were used for comparing and estimating survival results. The univariate logistic regression analysis and multivariate logistic regression analysis were performed to determine the hazard variables predicting tumor necrosis independently. The least absolute shrinkage and selection operator (LASSO) regression analysis on these variables was conducted for further screening predictor variables. Spearman correlation analysis was used to test for multicollinearity between variables. The restricted cubic spline (RCS) nested in logistic regression was conducted to investigate nonlinear relations between independent hazard factors and tumor necrosis. We established the prediction model based on these crucial factors and constructed them as a nomogram. The nomogram’s reliability was assessed through ROC curves. The calibration curves were used to assess model accuracy, and the decision curve analysis (DCA) was used to assess clinical significance. P less than 0.05 was considered statistically significant. The data was analyzed using R version 4.3.2 (R Foundation for Statistical Computing, Vienna, Austria).

Ethical statement
The study protocol complies with the ethical guidelines of the 1975 Helsinki Declaration. The study was approved by the Medical Ethics Committee of Zhongnan Hospital of Wuhan University before commencement (No.2024206K). In consideration of the retrospective nature of the research and other objective factors, informed patient consent was waived.

Results

Baseline characteristics
We recruited 495 patients with HCC who underwent hepatectomy and postoperative pathology, with 348 patients in the TC and 147 in the VC. The clinicopathological characteristics of the patients at baseline are presented in Table 1.

Cut-off values for predictors of necrosis
We utilized ROC curves to determine optimal cut-off values for various indicators, including AFP, APRI, ANRI, PLR, NLR, LMR, SII, SIRI, AISI, and PNI which have predictive values for tumor necrosis. Notably, NLR, PLR, SII, SIRI, AISI, and AFP significantly predicted the risk of necrosis in HCC (P < 0.05). The optimal cut-off values and corresponding areas under the ROC curves for each indicator were as follows: NLR: 2.558 [0.717, 95% confidence interval (CI): 0.672–0.762], PLR: 122.111 (0.677, 95% CI: 0.630–0.724), SII: 355.190 (0.720, 95% CI: 0.676–0.765), SIRI: 1.144 (0.717, 95% CI: 0.672–0.762), AISI: 215.490 (0.714, 95% CI: 0.669–0.759), AFP: 3327.485 (0.613, 95% CI: 0.564–0.662) (Table 2).

One-way logistic analysis of necrosis in hepatocellular carcinoma
Univariate analysis revealed that tumor maximum diameter, clinical TNM stage, lymphocyte, HB, PLT, N, PLR, NLR, SII, SIRI, AISI, AFP, ALB, ALP, AST, and GGT were HCC necrosis risk factors (Table 3).

Least absolute shrinkage and selection operator regression analysis to screen for risk factors
Variables with statistically significant one-way logistic regression results (P < 0.05) were incorporated into the LASSO regression analysis, with the optimal model selected based on the lambda.1se. The results showed that maximum tumor diameter, clinical TNM stage, HB, NLR, PLR, SII, SIRI, AFP, AST, and ALP were closely associated with HCC necrosis (Fig. 2).

Results of logistic multifactorial analysis of necrosis in hepatocellular carcinoma
Correlation analysis showed that NLR (r = 0.759, P < 0.001), PLR (r = 0.792, P < 0.001), and SIRI (r = 0.712, P < 0.001) were highly correlated with SII, and AST was correlated with ALP (r = 0.739, P < 0.001). Considering multicollinearity between variables, we selected the variables of tumor maximum diameter, clinical TNM stage, HB, SII, AFP, and ALP for a multifactorial backward stepwise analysis. The factors potentially associated with tumor necrosis based on RCS models are presented in Fig. 3. These four risk factors are linearly associated with necrosis, including SII (P for nonlinear = 0.122), HB (P for nonlinear = 0.499), AFP (P for nonlinear = 0.276), and ALP levels (P for nonlinear = 0.132). The final model was then chosen according to the Akaike information criterion (AIC) index. Finally, the tumor’s clinical TNM stage, SII, AFP, HB, and ALP were included in the prediction model with a minimum AIC of 382.87 (Fig. 4).

Establishment and validation of a necrosis prediction model for hepatocellular carcinoma
Clinical TNM stage, SII, AFP, HB, and ALP were selected to draw a nomogram, and the performance of the models was assessed in both the VC and the full cohort. The results showed that the area under the ROC curve for the model predicting necrosis in HCC was 0.810 (95% CI: 0.766–0.855), 0.758 (95% CI: 0.680–0.837), and 0.795 (95% CI: 0.756–0.834) for the training, validation, and full cohort, respectively (Fig. 5). The calibration curves demonstrated a significant agreement between the nomogram-predicted necrosis development in HCC and the postoperative pathology-based diagnosis of tumor necrosis (Fig. 6). The Hosmer–Lemeshow test demonstrated that the model had a good fit, (χ2 = 1.707, P = 0.974). DCA revealed that our nomogram exhibited significant net gains over the ‘no treatment’ and ‘all treatment’ strategies within a certain probability range (Fig. 7).

Associations between tumor necrosis and overall survival and recurrence-free survival
The median duration of follow-up was 45 (42, 50) months. At the end of the study, there were 162 (32.73%) deaths, and 196 (39.60%) experienced postoperative recurrence. Patients were divided into two groups according to the occurrence of tumor necrosis. At the end of 1, 3, and 5 years, the cumulative survival rates in the presence of the tumor necrosis group were 88.3, 71.8, and 49.8%, respectively, while the absence of the tumor necrosis group had 93.8, 78.7, and 63.1%. According to KM curves, the presence of tumor necrosis was closely related to lower OS (χ2 = 6.8, P = 0.009) and RFS (χ2 = 20.2, P < 0.001) (Fig. 8).

Discussion
This was the first research to construct and validate a nomogram based on hematologic and imaging data to predict spontaneous tumor necrosis for patients with HCC. Our findings showed that independent risk factors such as SII ≥ 355.190, AFP ≥ 3327.485 ng/ml, HB < 120 g/L, ALP ≥ 125 U/L, and clinical TNM stage were significantly associated with tumor necrosis. Furthermore, patients were categorized according to the presence of tumor necrosis, revealing inferior OS and RFS rates within the tumor necrosis subgroup.
Studies have shown that the incidence of spontaneous necrosis of tumors in patients with HCC is about 36.90–60.34% [5,7,8,13]. Our study found that 52.32% of 495 patients with HCC undergoing radical hepatectomy exhibited tumor necrosis, aligning with previous literature reports. Tumor necrosis serves as a crucial indicator of invasive metastasis in HCC, which has a large threat to patients’ prognosis. Notably, the survival rate of patients with HCC with tumor necrosis varies significantly depending on the postsurgical treatment modalities employed [14]. Physicians should consider a combination of treatment options when a high suspicion of tumor necrosis arises in patients with HCC. Consequently, the development of preoperative models capable of accurately predicting tumor necrosis is paramount. Our constructed nomogram demonstrated favorable predictive efficacy.
The American Joint Committee on Cancer TNM staging system is one of the most common tumor staging systems in the world, which guides clinicians in evaluating patients’ conditions and helping to formulate treatment plans [15,16]. Our findings revealed that an increase in the clinical TNM staging would cause an elevated risk of tumor necrosis. Significantly, patients with extrahepatic lymph node metastasis and distant organ metastasis were excluded in this study. Consequently, the TNM staging of these patients was predominantly determined by unilateral tumor factors (T). Similarly, correlations have been found between tumor necrosis and maximum tumor diameter and vascular invasion [8]. Tumor size is indicative of tumor cell proliferation; thus, sufficient oxygen is required for adequate oxygenation. In the event of insufficient local oxygen supply, the tumor will undergo necrosis [17]. Our study also found that the incidence of tumor necrosis significantly increased several times when the tumor invaded the portal or hepatic veins (T4).
The systemic immunoinflammatory index is a new, stable, and well-characterized inflammatory marker that reflects the body’s local immune response and systemic inflammatory response [18]. Our results demonstrated a high degree of accuracy in predicting tumor necrosis at SII greater than or equal to 355.190. Moreover, there was a positive and linear correlation between the SII value and tumor necrosis rate. SII was defined as (neutrophil count × PLT count)/lymphocyte count, which is a composite parameter that reflects the inflammatory state of the body more comprehensively than a single inflammatory indicator [19]. It can also predict the diagnosis and prognosis of a wide range of diseases [20]. During tumor development, it is often accompanied by elevated SII [21]. When necrosis occurs in tumor tissues, cellular debris, and inflammatory factors increase, promoting an inflammatory response in the body; on the other hand, necrotic tissues are recognized by the body’s immune system as a foreign body, causing a subsequent series of inflammatory responses. Therefore, patients with HCC need to be considered for possible tumor necrosis when SII values are too high in the absence of exogenous infectious factors.
AFP is usually used in the clinic to help diagnose HCC, and to evaluate anticancer efficacy [22]. In our result, the risk of tumor necrosis was 7.82 times higher in patients with a level above 3327.485 ng/ml, in comparison to patients below this threshold. This important result indicated the close relationship between AFP and tumor necrosis. One of the reasons may be because of the necrosis of HCC cells, which leads to the release of AFP into the blood, causing the AFP level to rise [23]. Accordingly, further research is necessary to ascertain the most appropriate cut-off value of AFP for predicting tumor necrosis.
HB is an indicator of anemia, which is often seen in chronic liver disease and advanced HCC [24,25]. The findings of our study suggested that decreased HB is a risk factor for necrosis and has clinical significance at HB less than 120 g/L, the significance of which represents anemia. When HB decreases, the oxygen-carrying capacity of erythrocytes decreases, and at the same time, HCC cells consume a large amount of oxygen when proliferating, resulting in insufficient cellular oxygen supply, and local tumor tissues are very susceptible to necrosis [26]. Tumors were more likely to metastasize under hypoxic conditions and hypoxia-inducible transcription factor isoforms were found to be a key determinant of metastatic success [27]. Subsequently, the necrotic material induces an immune-inflammatory response in the body, which in the long term causes a decline in erythropoietin and eventually chronic anemia [28]. In addition, we observed that hemorrhage was often combined with tumor necrosis, and it was hypothesized that tumor necrosis was combined with chronic blood loss, which subsequently led to a decrease in HB.
ALP is a hydrolytic enzyme found in all tissues and organs but accumulates primarily in the liver and is elevated in the presence of cholestasis or injury to the liver parenchyma [29,30]. Our study found that the relation of ALP and tumor necrosis was positively correlated and linear. The risk of tumor necrosis was 2.09 times higher in patients with ALP greater than or equal to 125 U/L. Previous studies have reported a strong correlation between dynamic changes in ALP and liver regeneration [31,32]. Furthermore, several studies indicate that ALP is significantly associated with a poor prognosis of patients with HCC [33]. This suggests that ALP may reflect the state of tumor metabolism and inflammatory response. As a result, the effect of tumor necrosis on the ALP requires further studies.
Some studies have found that tumor necrosis is connected to worse survival in a wide range of solid malignancies, including breast, pancreatic, lung, thyroid, renal, and gastrointestinal tumors [10,34–38]. Our findings also illustrated that the OR and RFS were indeed lower in the tumor necrosis group in HCC. It was also proved that prognosis was poorer in the tumor necrosis subgroup, regardless of tumor size, number of tumors, vascular invasion, and differentiation [8]. In addition, in patients with micronecrosis, adjuvant transcatheter arterial chemoembolization (TACE) after radical liver resection could improve their prognosis. However, in patients without micronecrosis, the survival benefit of TACE is limited [14]. In conclusion, tumor necrosis can serve as an independent prognostic factor in patients with HCC after hepatectomy. Therefore, more research should be carried out to help us choose the most reasonable treatment when cancer patients are suspected of tumor necrosis.
In this study, we initially devised a valuable prediction model and determined the correlation between serum levels of AFP, ALP, HB, and SII in patients with HCC and tumor necrosis. Indeed, our study still had several limitations. First, our data was provided by only one hospital. Accordingly, the accuracy of the prediction model should be confirmed in other medical institutions as well. Second, as our study was retrospective in nature, further prospective studies should be conducted to enhance our model. Ultimately, the number of patients in the present study was inadequate, necessitating the conduct of larger-scale studies to accurately predict the risk of tumor necrosis.

Conclusions
A combination of hematological indicators (including HB, AFP, ALP, and SII levels) and clinical TNM stage was employed to construct and validate a model for patients with HCC to predict tumor necrosis risk. Spontaneous tumor necrosis was closely associated with a worse prognosis among patients with HCC who underwent hepatectomy. Our model demonstrated satisfactory predictive efficacy in estimating the probability of tumor necrosis, thereby aiding clinicians in the formulation of suitable surgical strategies for these patients.

Acknowledgements
We express our highest appreciation to all those who took part in this project. We would also like to extend our gratitude to Zhongnan Hospital of Wuhan University for its invaluable data and financial support.
This study was supported by the Financial Project of Hubei Province (Grant No. : YYXKNLJS2024018) and the Joint Fund Project of Zhongnan Hospital of Wuhan University (Grant No.: ZNLH202207).
Q. H. and C. L. had the idea and designed the study. Q. H. was in charge of the manuscript draft. Z. L. and Z. G. were responsible for collecting and confirming data. C. L. conducted a statistical analysis. N. A. was in charge of grammar checking. K. L. revised the manuscript. All authors agreed to submit the final manuscript.

Conflicts of interest
There are no conflicts of interest.