Nomogram for Predicting Postoperative Pulmonary Metastasis in Hepatocellular Carcinoma Based on Inflammatory Markers

Abstract

Background

Uncertainty surrounds the usefulness of inflammatory markers in hepatocellular carcinoma (HCC) patients for predicting postoperative pulmonary metastasis (PM). The purpose of this study was to assess the predictive value of inflammatory markers as well as to create a new nomogram model for predicting PM.

Methods

Cox regression was utilized to identify independent prognostic variables and to create a nomogram that predicted PM for comparison with a validation cohort and other prediction systems. We retrospectively analyzed a total of 1109 cases with HCC were included.

Results

The systemic inflammatory response index (SIRI) and aspartate aminotransferase-to-platelet ratio index (APRI) were independent risk factors for PM, with a concordance index of .78 (95% CI: .74-.81) for the nomogram. The areas under the curve of the nomograms for PM predicted at 1-, 3-, and 5-year were .82 (95% CI: .77-.87), .82 (95% CI: .78-.87) and .81 (95% CI: .75-.86), respectively, which were better than those of Barcelona Clinic Liver Cancer and China liver cancer stage. Decision curve analyses demonstrated a broader range of nomogram threshold probabilities.

Conclusion

A nomogram based on SIRI and APRI can accurately predict postoperative PM in HCC.

Introduction

Hepatocellular carcinoma (HCC) is a widely known type of liver cancer in many countries, with an increasing incidence and cancer-specific mortality; ¹ moreover, it is the second leading cause of tumor death in China. ² The current treatment of HCC is recommended using multi-modal and high-intensity strategies, such as local therapy with the addition of immunotherapy-based systemic therapy, including excision, ablation, and intra-arterial therapy. ³ The most effective and comprehensive treatment for HCC is hepatectomy; however, surgical resection has a high recurrence rate and metastasis. ⁴ Pulmonary metastasis (PM) represents the most common type of extrahepatic metastasis, and the possible reason is that the circulation pressure of the lungs is lower than that of other body areas, which is more likely to cause the accumulation of liver cancer cells and lead to the occurrence of metastasis. The treatment of PM remains controversial and poor, ⁵ with a 5-year OS rate of only 2.5% for synchronoust HCC with PM, ⁶ and difficult to obtain a good prognosis. The development of targeted drugs may be a potential therapy for the treatment of PM, Zhang et al demonstrated that targeting the IL-1β/SAA3 axis can inhibit PM, ⁷ but more research investment is still needed. The development of assessment tools to predict PM in HCC patients can help physicians in making prognostic judgments and for intervening in advance for PM conditions.

Current research suggests that inflammation is intimately linked to tumorigenesis, progression and metastasis.^8,9 Inflammatory markers are used to predict the prognosis of other cancers,^10–12 and are also closely associated with the prognosis of HCC. ¹³ Nomograms that have been developed based on inflammatory markers are even more accurate than the commonly used cancer staging systems in forecasting patients’ prognoses for HCC. ¹⁴ Various inflammatory markers, including the prognostic nutritional index (PNI), systemic inflammation response index (SIRI), aspartate aminotransferase-to-platelet ratio (APRI), platelet-to-lymphocyte ratio (PLR) and neutrophil-to-lymphocyte ratio (NLR), have been designed and generally accepted by clinicians as being predictors of overall survival and recurrence-free survival in HCC resection.^15–17 However, few studies have further explored the correlation of inflammatory markers with postoperative PM for prognostic risk assessment. Thus, this study’s goal was to investigate the predictive ability of laboratory inflammatory markers for PM after R0 resection and to create a nomogram in this manner.

Methods

Patients

In this research, HCC patients who underwent hepatic R0 resection just at Guangxi Medical University Cancer Hospital between October 2013 and July 2021 were retrospectively analyzed. The following criteria were utilized for inclusion: (1) no preoperative antitumor treatment, such as radiotherapy, surgical treatment and transcatheter arterial chemoembolization procedures, among other procedures; (2) no distant metastases being found at the diagnosis of HCC; (3) liver tumor surgery meeting the criteria of R0 resection; ¹⁸ (4) postoperative pathology confirming HCC; and (5) no history of other malignant tumors. Clinical diagnostic criteria for PM included the presence of new spherical or ovoid nodules in the periphery of the lung on postoperative CT scans, ¹⁹ and elevated serum alpha-fetoprotein (AFP), especially the re-elevation of these levels after the postoperative decline.^20–22 The time of PM ranged from the date of hepatectomy to the date of diagnosis of PM. The following exclusion criteria were used: if patients only underwent palliative resection, died within 30 days after surgery or if case information was missing. A number of 1109 patients were included inside the analysis, and cases were proportionally distributed (7:3) between the training cohort (n = 777) and the validation cohort (n = 332). The study was approved by the Ethics Committee of the Guangxi Medical University Cancer Hospital (No. LW2023032) and followed the Helsinki Declaration 1964 and its subsequent amendments, as well as comparable ethical standards. The study followed relevant Equator guidelines. The reporting of this study conformed to TRIPOD guidelines. ²³ Individual consent for this retrospective analysis was waived, all patient data was anonymized or de-identified.

Examination and Follow-Up

Laboratory tests included preoperative complete blood count and blood biochemistry, as well as postoperative pathological examination. Inflammatory indicators were calculated by using the following formulas: NLR = neutrophils (10⁹/L)/lymphocytes (10⁹/L), LMR = lymphocytes (10⁹/L)/monocytes (10⁹/L), PLR = platelets (10⁹/L)/lymphocytes (10⁹/L), PNI = serum albumin (g/L) + 5 × lymphocytes (10⁹/L), SIRI = monocytes (10⁹/L) × neutrophils (10⁹/L)/lymphocytes (10⁹/L) and APRI = aspartate aminotransferase (U/L)/platelets (10⁹/L). All the resected specimens were cut into slices of approximately 3-5 mm thick and fixed in 1% formalin. The liver slices, which contained tumor tissues and noncancerous adjacent nontumorous tissues, were embedded in paraffin, cut into 4 μm sections, and stained with hematoxylin and eosin. At least one slice of nontumorous liver parenchyma 1 cm away from the tumor edge was examined. ²⁴ Histopathological measurements of excised specimens were independently performed by two pathologists, and consensus was reached through discussion (when necessary).

Patients should be checked in every two months for the first two years after discharge, and then every 3 months after that. The patient follow-up deadline is November 15, 2022.

Statistics

The majority of the statistical studies were performed using R version 4.2.2 software, and packages included “caret”, “autoReg”, “rrtable”, “MASS”, “survminer”, “survival”, “rms”, “ggDCA”, “ggplot2” and “nomogramFormula”. When comparing continuous variables with normal distribution, the mean (standard deviation) was used, while when comparing continuous variables with nonnormal distribution, the median (interquartile range). All of the factors with a P value <.05 in the single-factor Cox regression were entered into the multifactor analysis, and the “backward” method of multifactor Cox regression was used to identify independent risk factors. To assess the nomogram’s capacity for prediction, the receiver operating characteristic (ROC) curve, concordance index (C-index), calibration curves, and decision curve analysis (DCA) were combined. Furthermore, Kaplan‒Meier cumulative risk curves were compared for the subgroups.

Results

Baseline Characteristics of Patients

In our study, the observation window length of patients was from 1 month to 109.6 months, 563 (50.77%) patients were followed up for more than 3 years, and 323 (29.13%) patients were followed up for more than 5 years. The median follow-up time was 35.4 months (range: 14.6-65.8 months) for the training cohort and 44.4 months (range: 19.2-64.9 months) for the validation cohort. Patient clinical and pathological data from the training and validation cohorts were compiled in Table 1. The baseline characteristics of the two randomly allocated cohorts did not differ (P > .05).

Table 1.

Patient Characteristics.

Characteristics	Training Cohort (N = 777)	Validation Cohort (N = 332)	Total (N = 1109)	P Value
Status
Without PM	651 (83.8%)	279 (84%)	930 (83.9%)	.988
PM	126 (16.2%)	53 (16%)	179 (16.1%)
Sex
Male	680 (87.5%)	294 (88.6%)	974 (87.8%)	.701
Female	97 (12.5%)	38 (11.4%)	135 (12.2%)
Age (years)
≤50	423 (54.4%)	183 (55.1%)	606 (54.6%)	.887
>50	354 (45.6%)	149 (44.9%)	503 (45.4%)
NLR, median (IQR)	2.1 (1.5-3.0)	2.0 (1.4-2.8)	2.0 (1.5-2.9)	.393
LMR, median (IQR)	3.8 (2.9-5.0)	3.7 (2.8-4.8)	3.8 (2.9-4.9)	.441
PLR, median (IQR)	114.3 (85.8-157.4)	111.1 (81.6-159.4)	113.2 (84.6-158.0)	.433
PNI, mean±SD	48.2 ± 5.5	48.4 ± 5.5	48.3 ± 5.5	.695
SIRI, median (IQR)	.9 (.6-1.6)	.9 (.6-1.5)	.9 (.6-1.5)	.95
APRI, median (IQR)	.2 (.1-.3)	.2 (.1-.3)	.2 (.1-.3)	.241
AFP (ng/ml)
≤400	418 (53.8%)	178 (53.6%)	596 (53.7%)	1
>400	359 (46.2%)	154 (46.4%)	513 (46.3%)
HBsAg
Negative	86 (11.1%)	36 (10.8%)	122 (11%)	.996
Positive	691 (88.9%)	296 (89.2%)	987 (89%)
Liver cirrhosis
No	87 (11.2%)	34 (10.2%)	121 (10.9%)	.717
Yes	690 (88.8%)	298 (89.8%)	988 (89.1%)
Tumor thrombus
Absence	666 (85.7%)	277 (83.4%)	943 (85%)	.377
Presence	111 (14.3%)	55 (16.6%)	166 (15%)
MVI
Absence	443 (57%)	185 (55.7%)	628 (56.6%)	.74
Presence	334 (43%)	147 (44.3%)	481 (43.4%)
Largest tumor diameter (cm)
≤5	291 (37.5%)	122 (36.7%)	413 (37.2%)	.877
>5	486 (62.5%)	210 (63.3%)	696 (62.8%)
Tumor number
Single	612 (78.8%)	252 (75.9%)	864 (77.9%)	.331
Multiple	165 (21.2%)	80 (24.1%)	245 (22.1%)
Recurrent lesions
No	679 (87.4%)	295 (88.9%)	974 (87.8%)	.559
Yes	98 (12.6%)	37 (11.1%)	135 (12.2%)
Child‒Pugh grade
A	768 (98.8%)	330 (99.4%)	1098 (99%)	.521
B	9 (1.2%)	2 (.6%)	11 (1%)
Edmondson stage
I	15 (1.9%)	10 (3%)	25 (2.3%)	.744
II	354 (45.6%)	150 (45.2%)	504 (45.4%)
III	380 (48.9%)	160 (48.2%)	540 (48.7%)
IV	28 (3.6%)	12 (3.6%)	40 (3.6%)
BCLC stage
O	0 (0%)	1 (.3%)	1 (.1%)	.219
A	415 (53.4%)	162 (48.8%)	577 (52%)
B	174 (22.4%)	84 (25.3%)	258 (23.3%)
C	188 (24.2%)	85 (25.6%)	273 (24.6%)
CNLC stage
I a	211 (27.2%)	85 (25.6%)	296 (26.7%)	.647
I b	329 (42.3%)	130 (39.2%)	459 (41.4%)
II a	112 (14.4%)	56 (16.9%)	168 (15.1%)
II b	11 (1.4%)	6 (1.8%)	17 (1.5%)
III a	114 (14.7%)	55 (16.6%)	169 (15.2%)

Abbreviations: PM: pulmonary metastasis; NLR: neutrophil-to-lymphocyte ratio; LMR: lymphocyte-to-monocyte ratio; PLR: platelet-to-lymphocyte ratio; PNI: prognostic nutritional index; SIRI: systemic inflammation response index; APRI: aspartate aminotransferase-to-platelet ratio index; AFP: alpha-fetoprotein; HBsAg: hepatitis B virus surface antigen; MVI: microvascular invasion; BCLC: Barcelona Clinic Liver Cancer; CNLC: China liver cancer stage.

Independent Predictors of PM

In the training cohort, both univariate and multivariate analyses were carried out. The univariate analysis showed that 12 indicators were risk factors for PM, and these factors were included in the multivariate analysis, which showed that age, SIRI, APRI, AFP, tumor thrombus, MVI, largest tumor diameter and recurrent lesions were independent risk factors for PM (Table 2).

Table 2.

Univariate and Multivariate Analyses of PM in the Training Cohort.

Variable	Univariate Analysis			Multivariate Analysis
	HR	95% CI	P Value	HR	95% CI	P Value
Sex	1.115	.669-1.859	.677
Age	.522	.356-.766	.001	.566	.38-.843	.005
NLR	1.127	1.071-1.185	<.001
LMR	.791	.699-.895	<.001
PLR	1.004	1.002-1.006	<.001
PNI	.945	.915-.976	.001
SIRI	1.114	1.058-1.172	<.001	1.088	1.02-1.16	.011
APRI	2.046	1.394-3.003	<.001	2.32	1.461-3.684	<.001
AFP	2.343	1.624-3.38	<.001	1.576	1.074-2.312	.02
HBsAg	1.263	.68-2.345	.459
Cirrhosis	1.288	.694-2.391	.423
Tumor thrombus	2.329	1.527-3.552	<.001	1.571	1.02-2.418	.04
MVI	2.238	1.564-3.204	<.001	1.574	1.086-2.281	.017
Largest tumor diameter	6.28	3.6-10.955	<.001	4.071	2.294-7.225	<.001
Tumor number	1.145	.76-1.726	.517
Recurrent lesions	3.789	2.504-5.734	<.001	2.585	1.693-3.945	<.001
Child‒Pugh class	1.114	.156-7.971	.914
Edmondson stage	1.309	.985-1.741	.064

Abbreviations: NLR: neutrophil-to-lymphocyte ratio; LMR: lymphocyte-to-monocyte ratio; PLR: platelet-to-lymphocyte ratio; PNI: prognostic nutritional index; SIRI: systemic inflammation response index; APRI: aspartate aminotransferase-to-platelet ratio index; AFP: alpha-fetoprotein; HBsAg: hepatitis B virus surface antigen; MVI: microvascular invasion; BCLC: Barcelona Clinic Liver Cancer; CNLC: China liver cancer stage.

Development of the Nomogram

Based on the multivariate Cox regression analysis of independent risk factors, a nomogram was created to forecast the likelihood of PM in HCC patients at 1-, 3-, and 5-year (Figure 1).

Figure 1.

Nomogram for 1-, 3-, and 5-year PM prediction. Each independent prognostic factor was used to assess the patient and given a score on the nomogram, the probability of PM at 1-, 3-, and 5-year PM could be predicted. Abbreviations: SIRI: systemic inflammation response index; APRI: aspartate aminotransferase-to-platelet ratio index; AFP: alpha-fetoprotein; MVI: microvascular invasion; PM: pulmonary metastasis.

Performance of the Nomogram

Both cohorts had greater area under the curve (AUC) values in a nomogram for predicting 1-year, 3-year and 5-year PM than those of Barcelona Clinic Liver Cancer (BCLC) and China liver cancer stage (CNLC) in the same period (Figure 2). The training cohort’s C-index for the nomogram was .78 (95% CI: .74-.81), which was better than that of .62 (95% CI: .57-.66) for BCLC and .68 (95% CI: .64-.72) for CNLC. The C-index for the nomogram in the validation cohort was .76 (95% CI: .70-.82), which was higher than .61 (95% CI: .54-.68) for BCLC and .64 (95% CI: .58-.71) for CNLC. The results of the C-index and AUC of the nomogram in both cohorts were similar and better than those of the other two staging prediction models, thus indicating the high accuracy of nomogram prediction.

Figure 2.

Comparison of the 1-, 3-, and 5-year PM predictions made by time-dependent ROC curves. The training cohort’s PM 1-, 3-, and 5-year ROC values were shown in (A), (B), and (C), respectively, as were the validation cohort’s PM 1-, 3-, and 5-year ROC values (D–F). Abbreviations: PM: pulmonary metastasis; ROC: receiver operating characteristic curve; BCLC: Barcelona Clinic Liver Cancer; CNLC: China liver cancer stage.

In addition, the predictions of the nomogram and actual observations were in good agreement, as shown by the calibration curves for the training and validation cohorts predicting 1-, 3-, and 5-year PM (Figure 3).

Figure 3.

Calibration curves for the nomogram. (A) The nomogram predicted the probabilities of postoperative PM at 1-, 3-, and 5-year in the training cohort; (B) the nomogram predicted the probabilities of postoperative PM at 1-, 3-, and 5-year in the validation cohort. Abbreviations: PM: pulmonary metastasis.

When observing the clinical decision curve of the nomogram (Figure 4), the nomogram outperformed the BCLC and the CNLC in predicting PM risk within a certain risk threshold, thus yielding more net benefits and greater clinical application value.

Figure 4.

Comparison of DCA in two cohorts. (A–C) DCA for 1-year (A), 3-year (B) and 5-year (C) PM prediction in the training cohort. (D–F) DCA for 1-year (D), 3-year (E) and 5-year (F) PM prediction in the validation cohort. Abbreviations: DCA: decision curve analysis; BCLC: Barcelona Clinic Liver Cancer; CNLC: China liver cancer stage.

The predicted scores were calculated using the nomogram, and the patients were split into the following two groups using the median as the cutoff value: low-risk group and high-risk group. Both two cohorts were plotted using Kaplan-Meier cumulative risk curves, and the outcomes demonstrated that the predicted probability of PM in HCC patients could be significantly stratified (Figure 5).

Figure 5.

Kaplan‒Meier cumulative risk curves in the low-risk and high-risk groups as defined by the nomogram. (A) Training cohort for PM; (B) validation cohort for PM. Abbreviations: PM: pulmonary metastasis.

Discussion

In our study, a nomogram was created to accurately predict the likelihood of PM in HCC patients after R0 resection, and the nomogram outperformed BCLC and CNLC when it came to clinical utility, calibration, and discrimination.

The findings of this study demonstrated that the preoperative inflammatory markers SIRI and APRI are independent prognostic factors for PM, thus indicating a potential association between inflammatory markers and PM. It has been shown that platelets promote the growth and development, invasion and metastasis of HCC, ²⁵ and preoperative platelet counts can predict early extrahepatic metastasis after radical treatment of HCC, ²⁶ possibly because platelets express P-selectin, which regulates the adhesion of tumor cells to platelets and promotes PM. ⁴ Tumor-infiltrating lymphocytes have many specific antigens on the surface, including Foxp3 + T cells, which are associated with HCC migration. ²⁷ Monocytes are believed to play a surveillance role in the immune system, and programmed death ligand 1 and protein tyrosine phosphatase receptor type O of monocytes are valuable prognostic indicators in patients with HCC after surgery. ²⁸ Monocytes in the tumor infiltrates of HCC patients can regulate the PFKFB3-NF-κB signaling pathway through a glycolytic switch, thus causing the accumulation of neutrophils in the tumor; ²⁹ in addition, neutrophils are associated with HCC migration by the formation of extracellular traps, thus causing a tumor inflammatory response that promotes HCC metastasis. ³⁰ Future research must go into greater detail to clarify the processes by which inflammatory markers control the occurrence of PM because they are not currently clear.

It has been shown that preoperative tumor burden (tumor number and tumor diameter) and AFP levels (>400 ng/ml) synergistically affect the long-term prognosis of HCC resection, ³¹ and that AFP plays a role in encouraging HCC invasion and metastasis.^32,33 Our study suggests that MVI and tumor thrombosis are risk factors for PM after hepatectomy. They are vascular invasions of liver cancer cells. The vascular invasion of cancer cells leads to the activation of the epithelial mesenchymal transition transcriptional program and hepatic venous invasion, which may be the cause of PM. ³⁴ Chen et al ³⁵ showed that linc00261 had a lower expression in HCC patients with tumor diameter ≥5 cm or MVI, and that the downregulation of linc00261 was associated with HCC PM. Yokoo et al ³⁶ similarly suggested that MVI and tumor diameter >5 cm were strongly correlated with extrahepatic metastasis. In comparison with the nomogram of Li et al, ³⁷ the present study also confirmed that AFP, MVI and tumor diameter were strongly correlated with PM. Our model achieved a predictive C-index of .78, thus offering the advantage of being cost-effective and allowing for repeatable measurements.

Patients with HCC have lower long-term survival rates when they undergo R0 resection and have early postoperative recurrence (<1 year), ³⁸ with recurrence occurring in more than 50% of patients who undergo R0 resection. ³⁴ In this study, the prognosis of HCC R0 resection was linked to the recurrence of tumor lesions, and 12.2% of patients were found to have recurrent intrahepatic lesions on the first follow-up imaging within 2 months after surgery, with a recurrence rate much lower than 50%. This result was likely due to the short time to recurrence that was observed in this study. PM occurred in 21.7% of these patients during subsequent follow-up, thus suggesting that the presence of recurrent intrahepatic lesions within 2 months may reflect PM (P < .001). It is suggested that a preliminary prediction of HCC type could be established within two months post-surgery. This prediction could be based on a combination of surgical outcomes, pathology reports, as well as a decline in AFP levels and abnormal prothrombinogen (DCP) levels. ³⁹ These factors collectively serve as an initial reference point for assessing the potential risk of PM. According to some studies, the main cause of early recurrence after hepatectomy comes from occult tumor metastases, ⁴⁰ and Erstad et al ³⁴ suggested that early recurrence is due to the presence of residual microscopic lesions. These metastases and residual tumors cannot be identified prognostically by imaging and microscopy and may be the true culprit for PM outcomes.

There is a correlation between age and HCC, and some studies have suggested that the overall survival of patients going through a radical resection for HCC is worse with each 10-year increase in age, ⁴¹ and that overall survival after radical resection is significantly worse in older HCC patients (≥75 years) than in younger patients. ⁴² However, most of the studies discussed the correlation between age groups and HCC and lacked a uniform cutoff value. In our research, the age cutoff value for the occurrence of PM was 50 years old, which was obtained through ROC analysis, and a larger sample size is still needed to determine a more clinically meaningful cutoff value.

The development of BCLC and CNLC is based on European population and Chinese patients, respectively, ⁴³ and is widely used for tumor staging, treatment recommendations, and prognosis prediction.^44,45 We found that the nomogram, BCLC, and CNLC all predicted PM after hepatectomy, and the nomogram was superior to the other two models. It is worth noting that the CNLC was superior to the BCLC, possibly because our cases were from Chinese patients. Zhong et al ⁴⁵ study also verified that in Chinese HCC patients, CNLC performed better than BCLC in predicting prognosis after initial treatment with transarterial chemoembolization.

This study did have some limitations. First, the nomogram’s foundation was HCC patients with cirrhosis and hepatitis B virus as the main symptoms. Second, it is vital to validate our model in additional centers because it was developed in just one. Third, being a retrospective study, it was inevitable that there would be patient selection bias and insufficient adherence to the postoperative follow-up protocol.

Conclusion

In conclusion, we created and verified a nomogram relying on inflammatory markers to forecast PM in HCC patients undergoing R0 resection of liver cancer. The nomogram based on SIRI and APRI has good predictive performance, which is convenient for clinicians to assess each patient’s prognosis individually, thus avoiding or reducing the occurrence of PM risk factors as well as facilitating the benefits of patient survival after surgery.

Ethical Statement

Ethical Approval

The study was approved by the Ethics Committee of the Guangxi Medical University Cancer Hospital (Approval number: LW2023032) and followed the Helsinki Declaration 1964 and its subsequent amendments, as well as comparable ethical standards.

Patient Consent

The requirement for patient consent was waived due to the retrospective nature of the study.

ORCID iD

Huanjie Zhou https://orcid.org/0009-0004-1174-4993