Assessment of liver stiffness measurement-related markers in predicting liver-related events in viral cirrhosis with clinically significant portal hypertension

Abstract

BACKGROUND

Clinically significant portal hypertension (CSPH) is a crucial prognostic determinant for liver-related events (LREs) in patients with compensated viral cirrhosis. Liver stiffness measurement (LSM)-related markers may help to predict the risk of LREs.

AIM

To evaluate the value of LSM and its composite biomarkers [LSM-platelet ratio (LPR), LSM-albumin ratio (LAR)] in predicting LREs.

METHODS

This study retrospectively enrolled compensated viral cirrhosis patients with CSPH. The Cox regression model was employed to examine the prediction of LSM, LPR, and LAR for LREs. The model performance was assessed through receiver operating characteristic, decision curve, and time-dependent area under the curve analysis. The Kaplan-Meier curve was used to evaluate the cumulative incidence of LREs, and further stratified analysis of different LREs was performed.

RESULTS

A total of 598 patients were included, and 319 patients (53.3%) developed LREs during follow-up. Multivariate proportional hazards modeling demonstrated that LSM, LPR, and LAR were independent predictors of LREs. LPR had better performance in predicting LREs than LAR and LSM (area under the curve = 0.780, 0.727, 0.683, respectively, all P < 0.05). The cumulative incidence of LREs in the high-risk group were significantly higher than that in the low-risk group (P < 0.001). Among the different LREs, LPR was superior to LSM and LAR in predicting liver decompensation, while the difference in predicting hepatocellular carcinoma and liver-related death was relatively small.

CONCLUSION

LPR is superior to LSM and LAR in predicting LREs in compensated viral cirrhosis patients with CSPH, especially in predicting liver decompensation.

Key Words: Liver stiffness measurement; Liver stiffness measurement-platelet ratio; Liver stiffness measurement-albumin ratio; Liver-related events; Clinically significant portal hypertension; Viral cirrhosis

Core Tip: This study systematically evaluated the predictive value of liver stiffness measurement (LSM)-related markers for the liver related events (LREs) in compensated viral cirrhosis patients with clinically significant portal hypertension. The results showed that LSM-platelet ratio (LPR) is superior to LSM and LSM-albumin ratio in predicting LREs, especially in predicting liver decompensation. LPR can be served as an important tool for identification and individualized management in clinics.

INTRODUCTION

Liver cirrhosis represents the terminal pathophysiological progression of chronic hepatic diseases[1]. In Asia, viral hepatitis, especially hepatitis B virus (HBV) infection, constitutes the predominant etiology of advanced liver disease[2]. HBV-related cirrhosis accounts for more than 50% of cirrhosis-related deaths in Asian countries[3]. Clinically significant portal hypertension (CSPH) is a key pathophysiological feature of cirrhosis and serves as the primary driver of severe complications and mortality[4,5]. Liver-related events (LREs) are key clinical endpoints for assessing the progression of cirrhosis and patient prognosis. Once liver decompensation occurs, the 5-year mortality rate increased 70%-88%[6,7]. Hepatocellular carcinoma (HCC) represents another severe complication affecting cirrhotic individuals, with annual risk of 1%-6%[8,9]. These LREs not only seriously affect the patients’ quality of life, but also significantly increase the health burden. Therefore, early identification of high-risk patients and prediction of the occurrence of LREs are significant for cirrhotic patients with CSPH. Reliable risk stratification tools are also urgently needed to identify patients at highest risk for disease progression.

Hepatic venous pressure gradient (HVPG) evaluation is recognized as the gold standard methodology for portal hypertension[10]. However, HVPG measurement is invasive, technically demanding, and not yet widely available, so reliable non-invasive alternatives are urgently needed to identify high-risk individuals in clinical practice. Non-invasive technologies, such as liver stiffness measurement (LSM) and spleen stiffness, have shown good application prospects in the diagnosis of CSPH. Among them, LSM measured by transient elastography was significantly correlated with HVPG[11]. However, current noninvasive methods for predicting LREs in patients with CSPH remain suboptimal.

While LSM has certain value in evaluating CSPH, LSM alone has limitations due to confounding factors such as liver inflammation and cholestasis. This has prompted the development of composite biomarkers to provide more accurate risk stratification. In recent years, LSM-based composite indices have emerged as potential solutions. LSM-platelet (PLT) ratio (LPR) has shown good diagnostic value in CSPH[12-14]. LSM-spleen-PLT score (LSPS) has shown excellent performance in predicting esophageal varices and bleeding risk[13,15,16]. In addition, the combination of spleen stiffness measurement and von Willebrand factor as non-invasive indicators in patients with cirrhosis also has a certain diagnostic value for CSPH[17]. Although there are few studies on LSM-albumin (ALB) ratio (LAR) in patients with cirrhosis, spleen thickness-age-LSM-ALB and spleen thickness-PLT-ALB have certain predictive value for esophageal varices[18]. However, comparative data on these composite biomarkers for predicting LREs in viral cirrhosis with CSPH remain limited, and their differential performance in different LREs subtypes remains unclear. This requires the development of clinical prediction models to determine risk stratification and intensity of follow-up.

Therefore, this study aimed to systematically compare the predictive performance of LSM, LPR, and LAR for LREs in patients with viral cirrhosis and CSPH, with particular attention to their differential utility among different LREs subtypes, in order to establish a more accurate risk stratification tool for clinical application.

MATERIALS AND METHODS

Study design and subjects

This study retrospectively enrolled compensated viral cirrhosis patients with CSPH treated at Beijing Ditan Hospital between January 2016 and May 2024. Eligible participants were randomly allocated into training and validation datasets using a 7:3 distribution ratio. This study was approved by the Ethics Committee of Beijing Ditan Hospital (approval number: DTEC-KY2024-069-01). All procedures adhered to Declaration of Helsinki.

Inclusion criteria: (1) Age between 18-75 years; (2) Compensated viral cirrhosis without prior ascites, variceal bleeding or hepatic encephalopathy; (3) Take antiviral medicine regularly; (4) Confirmed CSPH diagnosis; and (5) Complete follow-up data. The diagnosis of cirrhosis is based on one of the following criteria[19-21]: (1) Pathological evidence on liver biopsy; and (2) Typical clinical, laboratory, and imaging features, including liver morphological changes, splenomegaly, thrombocytopenia, and portal vein dilation. The definition of CSPH is based on the Baveno VII consensus[10,21-24], including LSM ≥ 25 kPa, or 20 ≤ LSM < 25 kPa and PLT < 150 × 10⁹/L, or 15 ≤ LSM < 20 kPa and PLT < 110 × 10⁹/L. Patients with LSM < 15 kPa and PLT > 150 × 10⁹/L were considered CSPH-negative.

Exclusion criteria: (1) Prior episodes of liver decompensation events; (2) Prior diagnosis of HCC; (3) Severe cardiopulmonary and renal insufficiency; (4) Concurrent with other liver diseases other than viral hepatitis, such as autoimmune or alcoholic liver disease, and non-alcoholic fatty liver disease; (5) Current or previous alcohol consumption exceeding 20 g/day; (6) Pregnant or lactating women; (7) Those who had undergone liver transplantation, splenectomy or transjugular intrahepatic portosystemic shunt surgery; and (8) Incomplete clinical data or follow-up of less than 6 months.

Clinical and laboratory evaluation

Baseline clinical data of all patients were collected, including demographic characteristics (age, gender), comorbidities, and previous treatment history. Laboratory assessments encompassed hematological parameters [neutrophil-to-lymphocyte ratio (NLR); hemoglobin (Hb); PLT], liver function biochemical indicators [total bilirubin (TB); ALB; alanine aminotransferase; aspartate aminotransferase], and coagulation function [international normalized ratio (INR)].

Imaging evaluation

All patients underwent abdominal ultrasound examination, and the spleen diameter and portal vein diameter were recorded. LSM assessment utilized the FibroScan^® device (Echosens, Paris, France) by trained operators. During the measurement, the patient fasted for more than 4 hours, took the supine position and the right upper limb was maximally abducted. The probe was placed in the right intercostal space, perpendicular to the liver surface. Each patient was measured at least 10 valid measurements, and the median was taken as the final LSM value in kilopascals (kPa).

Based on baseline LSM and laboratory parameters, the following two composite indexes were calculated. LPR = LSM (kPa)/PLT count (10⁹/L); LAR = LSM (kPa)/ALB (g/L).

Follow-up and endpoint events

Patient surveillance continued from baseline assessment until the occurrence of LREs or the last follow-up, whichever occurred first. The primary endpoint was the occurrence of LREs, including: (1) Liver decompensation (defined as the first occurrence of ascites, esophageal variceal hemorrhage, or hepatic encephalopathy); (2) HCC (diagnosed according to the AASLD or EASL guidelines); and (3) Liver-related death (death directly attributed to liver disease progression or complications).

Statistical analysis

Quantitative data were presented as mean ± SD or median with interquartile ranges based on distribution normality, while qualitative variables were reported as frequencies and percentages. Between-group comparisons utilized Student’s t-test or Mann-Whitney U test for continuous data, and χ² or Fisher's exact tests for categorical data. All patients with missing values were excluded from the analysis.

Cox proportional hazard models were used to analyze risk factors associated with LREs, with results presented as hazard ratio (HR) and 95% confidence interval (CI). Model performance evaluation employed receiver operating characteristic (ROC) curve analysis, and decision curve analysis (DCA). Area under the curve (AUC) was calculated and the differences between three models were compared with DeLong’s test. Harrell’s C-index was utilized to assess concordance between predicted and observed outcomes. Optimal thresholds for LSM, LPR, and LAR were established using Youden’s Index. To develop the predictive model in the training cohort, we used bootstrapping internal validation methods. We also used the time-dependent AUC analysis to compare the predictive performance of each model at different follow-up time points. Schoenfeld residual test was introduced to verify the proportional hazard assumption. Based on ROC-derived optimal thresholds, participants were stratified into high- and low-risk categories, with cumulative event rates compared via Kaplan-Meier analysis and log-rank testing. Subgroup analyses were performed for subendpoints of LREs. Statistical computations utilized R software (v4.3.3), with two-tailed P values < 0.05 indicating statistical significance.

RESULTS

Baseline characteristics of patients

Our study recruited 598 individuals with compensated viral cirrhosis and CSPH, with 419 participants allocated to the training dataset and 179 to the validation dataset (Table 1). The participant population had a median age of 54 (51-58) years, with male predominance at 66.9%. Throughout the observation period, a total of 319 patients (53.3%) experienced LREs, including 46.5% decompensation events, 4.3% HCC, and 1.5% liver-related mortality. The observation period had a median duration of 37 (21-55) months across all participants. The median LSM in the overall cohort was 26.6 (21.1-37.4) kPa, the LPR was 0.37 (0.23-0.57), and the LAR was 0.68 (0.50-1.02). There were no significant differences between the training and validation cohort (all P > 0.05).

Table 1 Baseline characteristics of compensated viral cirrhosis patients with clinically significant portal hypertension, n (%).

	Overall cohort (n = 598)	Training cohort (n = 419)	Validation cohort (n = 179)	P value
Age, years	54 (51-58)	54 (51-58)	53 (51-58.5)	0.498
Sex, male/female (%male)	400 (66.9)	275 (65.6)	125 (69.8)	0.318
LREs	319 (53.3)	222 (53.0)	97 (54.2)	0.786
Decompensation	284 (46.5)	199 (47.5)	85 (47.5)
HCC	26 (4.3)	17 (4.1)	9 (5.0)
Liver-related death	9 (1.5)	6 (1.4)	3 (1.7)
Follow-up time, months	37 (21, 55)	37 (20.5, 56)	37 (22, 53.5)	0.913
Spleen diameter, cm	14.4 (12.7-16)	14.4 (12.7-16.1)	14.4 (12.8-16)	0.638
AST, U/L	33.5 (25.8-48.3)	34 (26.1-47.6)	32.5 (25.5-48.5)	0.402
ALT, U/L	27.1 (19.3-40)	27.6 (19.6-40.6)	26.2 (18.5-39.3)	0.434
TB, μmol/L	19.4 (14.4-28.9)	19.4 (14-29)	20.7 (14.9-28.7)	0.645
ALB, g/L	40.4 (35.7-45)	40.8 (36.1-45.3)	39.1 (35.1-44.3)	0.141
INR	1.17 (1.09-1.23)	1.17 (1.08-1.21)	1.18 (1.09-1.26)	0.196
NLR	1.91 (1.39-2.68)	1.89 (1.365-2.72)	1.95 (1.45-2.66)	0.683
Hb, g/L	138 (122-150)	137 (121-149.5)	140 (124-151)	0.101
PLT, 109/L	78 (55-106)	76 (53.5-107.5)	82.4 (59-104)	0.240
LSM, kPa	26.6 (21.1-37.4)	26.3 (21.1-35.8)	27.4 (21.3-40.3)	0.246
LPR	0.37 (0.23-0.57)	0.37 (0.23-0.58)	0.35 (0.24-0.55)	0.830
LAR	0.68 (0.50, 1.02)	0.68 (0.5, 1.00)	0.68 (0.50, 1.05)	0.535

Medians (IQR) or counts (proportions). LREs: Liver related events; HCC: Hepatocellular carcinoma; AST: Aspartate aminotransferase; ALT: Alanine aminotransferase; TB: Total bilirubin; ALB: Albumin; INR: International normalized ratio; NLR: Neutrophil-to-lymphocyte ratio; Hb: Hemoglobin; PLT: Platelets; LSM: Liver stiffness measurement; LPR: Liver stiffness measurement-platelet ratio; LAR: Liver stiffness measurement-albumin ratio.

Analysis of risk factors for LREs

In the training cohort, univariate Cox regression analysis revealed that LSM (HR = 1.03, 95%CI: 1.02-1.04, P < 0.001), LPR (HR = 2.79, 95%CI: 2.10-3.69, P < 0.001), and LAR (HR = 2.32, 95%CI: 1.79-3.00, P < 0.001) were markedly related to LREs (Table 2). In addition, spleen diameter, TB, ALB, INR, NLR, HB, and PLT were also obviously associated with LREs. Following adjustment for potential confounders, multivariable analysis showed that LSM (aHR = 1.03, 95%CI: 1.02-1.04, P < 0.001), LPR (aHR = 2.30, 95%CI: 1.55-3.42, P < 0.001) and LAR (aHR = 2.37, 95%CI: 1.66-3.38, P < 0.001) were independent predictors. In addition, NLR were also independent predictors in the three models.

Table 2 Risk factors for liver related events in compensated viral cirrhosis patients with clinically significant portal hypertension.

	Univariate analysis		Multivariate analysis of LSM model		Multivariate analysis of LPR model		Multivariate analysis of LAR model
	HR (95%CI)	P value	aHR (95%CI)	P value	aHR (95%CI)	P value	aHR (95%CI)	P value
LSM	1.03 (1.02-1.04)	< 0.001	1.03 (1.02-1.04)	< 0.001
LPR	2.79 (2.10-3.69)	< 0.001			2.30 (1.55-3.42)	< 0.001
LAR	2.32 (1.79-3.00)	< 0.001					2.37 (1.66-3.38)	< 0.001
Age	1.01 (1.00-1.03)	0.146
Gender	0.88 (0.67-1.15)	0.337
Spleen diameter	1.06 (1.01-1.11)	0.020	0.99 (0.93-1.04)	0.600	0.99 (0.93-1.03)	0.436	0.99 (0.94-1.05)	0.677
AST	1.00 (1.00-1.01)	0.330
ALT	1.00 (1.00-1.00)	0.388
TB	1.01 (1.01-1.02)	0.002	1.00 (1.00-1.01)	0.821	1.00 (0.99-1.01)	0.813	1.00 (0.99-1.01)	0.501
ALB	0.96 (0.94-0.97)	< 0.001	0.99 (0.96-1.02)	0.552	0.99 (0.96-1.01)	0.337	1.01 (0.98-1.04)	0.490
INR	5.18 (1.98-13.54)	0.004	1.96 (0.73-5.28)	0.183	2.48 (0.94-6.54)	0.066	2.43 (0.91-6.46)	0.076
NLR	1.12 (1.04-1.20)	0.002	1.09 (1.01-1.18)	0.029	1.09 (1.01-1.18)	0.022	1.10 (1.02-1.18)	0.018
Hb	0.99 (0.98-0.99)	< 0.001	0.99 (0.99-1.01)	0.083	1.00 (0.99-1.00)	0.167	0.99 (0.99-1.00)	0.143
PLT	0.99 (0.99-1.00)	< 0.001	0.99 (0.99-1.00)	< 0.001	1.00 (0.99-1.00)	0.556	0.99 (0.99-1.00)	0.001

HR: Hazard ratios; aHR: Adjusted hazard ratios; LSM: Liver stiffness measurement; LPR: LSM-platelet ratio; LAR: LSM-albumin ratio; AST: Aspartate aminotransferase; ALT: Alanine aminotransferase; TB: Total bilirubin; ALB: Albumin; INR: International normalized ratio; NLR: Neutrophil-to-lymphocyte ratio; Hb: Hemoglobin; PLT: Platelets.

Predictive model performance evaluation

ROC analysis within the training dataset revealed that the LPR model had the best predictive efficacy (AUC = 0.780, 95%CI: 0.736-0.824), which was significantly better than the LAR model (AUC = 0.727, 95%CI: 0.679-0.775, P = 0.015) and the LSM model (AUC = 0.683, 95%CI: 0.632-0.733, P < 0.001) (Figure 1A, Table 3). Harrell's C-index demonstrated similar discriminative performance, with the LPR model achieving the highest C-index of 0.780 (95%CI: 0.736-0.825), followed by the LAR model (C-index = 0.727, 95%CI: 0.678-0.776) and the LSM model (C-index = 0.683, 95%CI: 0.631-0.734) (Table 3). DCA further verified the clinical practicality of the three models. The LPR model provided the highest net benefit when the risk threshold was between 0.2 and 0.8, while the net benefits of the three models were similar when the risk threshold was below 0.2 (Figure 1B). Time-dependent AUC analysis showed that the predictive efficacy of the three models was low in the early follow-up period (1-2 years). As the follow-up time increased, the predictive accuracy increased slightly and remained relatively stable (Figure 1C). The bootstrap validation showed the same trend. The LPR model had the best effect, with a C index of 0.659 (95%CI: 0.619-0.699); followed by the LAR model, with a C index of 0.637 (95%CI: 0.594-0.679); and the LSM model had the worst effect, with a C index of 0.616 (95%CI: 0.573-0.660). Validation dataset analysis confirmed this performance, with LPR achieving the highest discrimination (AUC = 0.753, 95%CI: 0.682-0.824), followed by LAR (AUC = 0.686, 95%CI: 0.608-0.765, P = 0.033) and LSM (AUC = 0.661, 95%CI: 0.581-0.740, P = 0.004) (Figure 1D, Table 3). Harrell's C-index confirmed the superior performance of the LPR model (C-index = 0.753, 95%CI: 0.681-0.825). The DCA curves and time-dependent AUC analysis in the validation cohort showed a similar trend (Figure 1E and F).

Figure 1 Performance evaluation of liver stiffness measurement and its composite indices in predicting liver-related events. A: Receiver operating characteristic (ROC) of liver stiffness measurement (LSM), LSM-platelet ratio (LPR), and LSM-albumin ratio (LAR) models in the training cohort; B: Decision curve analysis of LSM, LPR, and LAR models in the training cohort; C: Time-dependent area under the curve (AUC) analysis of LSM, LPR, and LAR models in the training cohort; D: ROC curve of LSM, LPR, and LAR models in the validation cohort; E: Decision curve analysis of LSM, LPR, and LAR models in the validation cohort; F: Time-dependent AUC analysis of LSM, LPR, and LAR models in the validation cohort.

Table 3 Prognostic performance of liver stiffness measurement and its composite markers for liver related events.

Cohort	AUROC (95%CI)	Sensitivity	Specificity	PPV	NPV	C-index (95%CI)
Training cohort
LSM	0.683 (0.632-0.733)	0.57	0.73	0.65	0.66	0.683 (0.631-0.734)
LPR	0.780 (0.736-0.824)	0.7	0.73	0.69	0.73	0.780 (0.736-0.825)
LAR	0.727 (0.679-0.775)	0.57	0.8	0.71	0.68	0.727 (0.678-0.776)
Validation cohort
LSM	0.661 (0.581-0.740)	0.56	0.7	0.61	0.65	0.661 (0.580-0.742)
LPR	0.753 (0.682-0.824)	0.62	0.8	0.73	0.72	0.753 (0.681-0.825)
LAR	0.686 (0.608-0.765)	0.76	0.6	0.61	0.74	0.686 (0.607-0.766)

LSM: Liver stiffness measurement; LPR: Liver stiffness measurement-platelet ratio; LAR: Liver stiffness measurement-albumin ratio; AUROC: Area under the receiver operating characteristic; PPV: Positive predictive value; NPV: Negative predictive value.

Proportional hazards assumption verification and risk stratification analysis

To verify the clinical application efficacy of the three prediction models, we performed risk stratification and proportional hazard hypothesis tests (Figure 2). Training dataset analysis revealed markedly elevated LREs rates among high-risk vs low-risk participants, with events predominantly occurring during early observation phases (Figure 2A). A similar trend was also verified in the validation cohort (Figure 2C), confirming the model’s effective identification ability for high-risk populations. In the training cohort, the Schoenfeld individual test P value of the three models (LSM, LPR, and LAR) were 0.269, 0.906, and 0.210, respectively (Figure 2B), indicating that the three models met the proportional hazard assumption during the follow-up period. The results of the validation cohort further supported this conclusion (Figure 2D). The predictive effects of the three models on LREs remained relatively constant at different follow-up time points, further supporting their stability and reliability in long-term follow-up management. Among them, the LPR model had the highest P-value in the Schoenfeld test (0.906 for the training cohort and 0.576 for the validation cohort), indicating that it has the best stability as a proportional hazard model (Figure 2B and D).

Figure 2 Proportional hazard hypothesis test and risk stratification analysis. A: The risk stratification diagram in the training cohort; B: The Schoenfeld residual diagram of liver stiffness measurement (LSM), LSM-albumin ratio (LAR), and LSM-platelet ratio (LPR) in the training cohort; C: The risk stratification diagram in the validation cohort; D: The Schoenfeld residual diagram of LSM, LAR, and LPR in the validation cohort. The horizontal axis represents the follow-up time (days), the vertical axis represents the cox regression coefficient of the models, and the solid line and the dotted line represent the fitted spline smooth curve and the standard deviation (confidence interval) of 2 units above and below the fitted curve, respectively.

Cumulative incidence analysis of LREs

Using ROC-derived optimal thresholds, participants were stratified into high-risk and low-risk categories (Figure 3, Table 4). In the training cohort (Figure 3A), patients with LSM ≥ 25.25 kPa, LPR ≥ 0.35, and LAR ≥ 0.57 all showed a significantly elevated cumulative incidence of LREs (all Log-rank P < 0.001). The validation cohort (Figure 3B) further confirmed the above findings.

Figure 3 Cumulative incidence of liver-related events in compensated viral cirrhosis patients with clinically significant portal hypertension. A: Cumulative incidence of decompensation events in patients stratified by liver stiffness measurement (LSM) risk groups (≥ 25.25 kPa vs < 25.25 kPa), LSM-platelet ratio (LPR) risk groups (≥ 0.35 vs < 0.35), and LSM-albumin ratio (LAR) risk groups (≥ 0.57 vs < 0.57) in the training cohort; B: Cumulative incidence of decompensation events in patients stratified by LSM risk groups (≥ 25.25 kPa vs < 25.25 kPa) , LPR risk groups (≥ 0.35 vs < 0.35), and LAR risk groups (≥ 0.57 vs < 0.57) in the validation cohort.

Table 4 Cumulative incidence of liver-related events according to liver stiffness measurement and its composite indicators.

Cohort	1-year cumulative incidence (%)	3-year cumulative incidence (%)	5-year cumulative incidence (%)
Training cohort
LSM ≥ 25.25	3.3 (1.0-5.5)	38.9 (32.4-45.5)	71.7 (64.8-78.6)
< 25.25	2.4 (0.1-4.6)	13.2 (7.6-18.8)	44.6 (34.4-54.8)
LPR ≥ 0.35	3.2 (0.9-5.5)	39.7 (32.9-46.5)	77.3 (70.8-83.8)
< 0.35	1.5 (0.2-3.3)	14.8 (9.2-20.3)	38.4 (28.7-48.1)
LAR ≥ 0.80	3.1 (1.0-5.2)	37.5 (31.3-43.8)	73.1 (66.7-79.6)
< 0.80	2.6 (0.1-5.1)	12.4 (6.6-18.1)	37.4 (26.5-48.2)
Validation cohort
LSM ≥ 24.15	1.8 (0.7-4.4)	31.9 (22.7-41.1)	77.2 (67.4-86.9)
< 24.15	2.9 (1.1-7.0)	23.5 (12.0-34.9)	54.1 (37.3-71.0)
LPR ≥ 0.35	3.1 (0.0-6.6)	34.1 (24.1-44.1)	78.2 (68.4-87.9)
< 0.35	1.2 (0.0-3.6)	24.1 (13.6-34.7)	51.5 (36.7-66.2)
LAR ≥ 0.80	1.7 (0.0-4.1)	31.1 (22.2-40.0)	74.4 (64.8-84.1)
< 0.80	3.3 (0.0-7.7)	24.3 (12.0-36.6)	59.6 (40.0-79.1)

LSM: Liver stiffness measurement; LPR: Liver stiffness measurement-platelet ratio; LAR: Liver stiffness measurement-albumin ratio.

Risk stratification analysis of subendpoints of LREs

Additional assessment of LSM-based markers’ predictive capacity for specific LREs was conducted through subgroup analysis (Figure 4). In the training cohort, risk stratification based on LSM revealed significantly elevated rates of HCC, decompensation events, and liver-related death in the high-risk vs low-risk participants (2.9% vs 4.9%, 32.0% vs 58.3%, 0% vs 2.4%, respectively, P < 0.001) (Figure 4A). The stratification effect of the LPR indicator was more significant (Figure 4C), with the proportion of decompensation events in the high-risk participants as high as 65.6%, while it was only 13.6% in the low-risk individuals (P < 0.001). Similarly, the LAR indicator also showed a good stratification effect (Figure 4E). Accordingly, the cumulative incidence curve showed that the high-risk individuals had more decompensation events than the low-risk individuals, while the differences of HCC and liver-related death was relatively small (Figure 4B, D, and F). The validation cohort further confirmed the risk stratification efficacy of the three models. Among the three indicators, the stratification effect of LPR was the most obvious, especially for the prediction of decompensation events.

Figure 4 Subgroup analysis of liver-related events stratified by liver stiffness measurement and its composite indices A: Distribution of liver-related events (LREs) in liver stiffness measurement (LSM) risk groups (≥ 25.25 kPa vs < 25.25 kPa) in the training and validation cohort; B: Cumulative incidence curve of LREs in LSM risk groups (≥ 25.25 kPa vs < 25.25 kPa) in the training and validation cohort; C: Distribution of LREs in LSM-platelet ratio (LPR) risk groups (≥ 0.35 vs < 0.35) in the training and validation cohort; D: Cumulative incidence curve of LREs in LPR risk groups (≥ 0.35 vs < 0.35) in the training and validation cohort; E: Distribution of LREs in LSM-albumin ratio (LAR) risk groups (≥ 0.57 vs < 0.57) in the training and validation cohort; F: Cumulative incidence curve of LREs in LAR risk groups (≥ 0.57 vs < 0.57) in the training and validation cohort.

DISCUSSION

This study systematically evaluated the predictive value of LSM and its composite markers (LPR and LAR) in predicting LREs in compensated viral cirrhosis patients with CSPH. The results showed that LPR, as a composite marker, showed better predictive performance than single LSM and LAR. It provided an important basis for the identification and individualized management of high-risk patients in clinical practice.

As an emerging composite index, LPR has attracted widespread attention in recent years. Berzigotti et al[25] suggested that utilizing LPR substantially enhanced the precision of CSPH prediction. Integrating LSM, spleen diameter, and PLT count enables identification of cirrhotic individuals with highest probability of developing CSPH and esophageal varices[25]. However, Ryu et al[26] proposed that the LSM cutoff values for predicting HVPG ≥ 10 and ≥ 12 mmHg in alcoholic cirrhosis were 32.2 and 36.6 kPa, respectively. LPR and LSPS had no significant differences. LPR also demonstrated optimal performance in determining fibrosis severity in nonalcoholic fatty liver disease, serving as an accessible, non-invasive substitute for hepatic biopsy in routine practice[27]. Our study found that LPR had an AUROC of 0.780 for predicting LREs, which was superior to LSM and LAR. This advantage may be due to the ability of LPR to simultaneously capture two key pathological mechanisms of liver fibrosis and portal hypertension, which play a crucial role in the progression of cirrhosis. Particularly, a decrease in PLT count not only reflects hypersplenism and portal hypertension, but may also be associated with decreased liver synthetic function, thus providing comprehensive information on liver disease progression. In contrast, although LAR incorporates ALB, an important synthetic index of liver function, its predictive ability is slightly inferior to that of LPR. Because ALB levels are affected by a variety of non-liver factors, such as nutritional status and renal function. The results of DCA further support the clinical utility of LPR. In addition, the predictive efficacy of the three models was low in the early follow-up period (1-2 years), and the predictive accuracy increased slightly and remained relatively stable as the follow-up time increased. This finding suggests that these indicators may be more suitable for medium- and long-term risk prediction, while they may need to be combined with other clinical parameters to improve accuracy for short-term risk assessment. The Schoenfeld residual test also verified that the LPR model had the highest P-value, indicating that it has the best stability as a proportional hazard model. This feature is particularly important for long-term follow-up management because it ensures the consistency and reliability of the prediction effect at different time points.

The subendpoint analysis of this study provides important insights into the specificity of LSM and its composite index in predicting different LREs. Our study showed that although all three models showed significant value in predicting overall LREs, there were obvious differences in predicting specific types of LREs. Particularly, LPR performed best in predicting liver decompensation events with the proportion of decompensation events in the high-risk group as high as 65.6%, while it was only 13.6% in the low-risk group. By integrating LSM with PLT count, LPR can more comprehensively reflect this pathophysiological process and therefore performs well in predicting decompensation. In contrast, for HCC and liver-related death, although the cumulative incidence of the high-risk individuals was higher, the difference was relatively small. It may be attributed to multiple factors. First, the occurrence of HCC involves complex pathophysiological mechanisms, including viral integration, chronic inflammation, oxidative stress, genetic and epigenetic changes, which may not be completely dependent on portal hypertension or the degree of liver fibrosis[28-31]. Although cirrhosis is the main risk factor for HCC, HBV-related HCC can occur in a non-cirrhotic setting[32], which also explains why markers based on liver fibrosis and portal hypertension may have limited predictive power for HCC. Second, the number of HCC and liver-related deaths in this study was relatively small, which may limit the power and reliability of the statistical analysis. It is worth noting that among all three indicators, LPR has a better predictive ability for liver decompensation than LAR and single LSM, which may reflect the sensitivity of PLT count in the early stages of portal hypertension. Thrombocytopenia is not only the result of hypersplenism, but also related to reduced production of thrombopoietin, increased production of PLT autoantibodies, and bone marrow suppression[33-35]. These factors together make PLT count a sensitive marker for portal hypertension syndrome. Conversely, ALB, as a marker of liver function, may not have decreased significantly in the early stages of portal hypertension, so the predictive ability of LAR is relatively weak.

Our investigation presents several constraints. Primarily, as a single-center retrospective analysis, there may be selection bias, and multicenter prospective validation in different populations and medical institutions is needed in the future. Second, our study subjects were mainly Chinese patients with viral cirrhosis, potentially restricting the generalizability of results to Western demographics or populations with heterogeneous disease etiologies (alcoholic, non-alcoholic steatohepatitis, or other viral hepatitis). The pathophysiology and disease progression patterns of different populations and different etiologies may vary significantly, which may affect the predictive performance of LSM-based indicators. In addition, the definition of CSPH used non-invasive surrogate indicators instead of gold standard HVPG measurements, which may introduce misclassification bias and affect the accuracy of risk stratification. Future research directions include conducting multicenter prospective studies to verify the predictive value of LAR and LPR, exploring combination models that integrate other noninvasive indicators, and incorporating artificial intelligence into predictive models to further improve prognostic accuracy and clinical applicability.

CONCLUSION

In patients with compensated viral cirrhosis with CSPH, LPR, as a simple and noninvasive composite indicator, showed better predictive performance than LSM and LAR in predicting LREs, especially liver decompensation events. It provides an important reference for the screening, risk stratification and individualized management of high-risk patients in clinical practice.

ACKNOWLEDGEMENTS

Authors are grateful to all members of Center for Integrative Medicine of Beijing Ditan Hospital for their contributions to the manuscript preparation.