Noninvasive Markers Associated With Portal Hypertension and Varices in Patients With Noncirrhotic Portal Hypertension

Abstract

INTRODUCTION:

Noncirrhotic portal hypertension (NCPH) refers to a diverse group of disorders that affects the hepatic portosinusoidal vascular system resulting in portal hypertension (PH). Unlike cirrhosis, there are no noninvasive criteria to diagnose PH and varices in NCPH.

METHODS:

A prospective cohort of patients with NCPH who had transjugular liver biopsy, liver stiffness (LSM) measured using transient elastography, and laboratory and imaging data were included. PH was defined by the presence of one among the following—varices on endoscopy, portosystemic collaterals, or ascites on imaging. Logistic regression was used to identify predictors. The classification tree approach was used to identify cutoff values for the stepwise decision model.

RESULTS:

Of the 59 patients, 41 (69%) had PH and 36 (62%) had varices. LSM was higher in patients with PH (11.5 kPa vs 5.7 kPa, P < 0.01) and varices (12 kPa vs 5.9 kPa, P < 0.01). Platelet count was lower in patients with PH (79 vs 218 × 10⁹/L, P < 0.01) and varices (75 vs 209 × 10⁹/L, P < 0.01). Multivariate analysis combining LSM and platelet count predicts PH (area under receiver operating characteristic 96% [91%–99%]) and varices (area under receiver operating characteristic 92% [85%–99%]). A stepwise model combining platelet 140 × 10⁹/L and LSM 7 kPa performed with a sensitivity of 100%, negative predictive value of 100%, and accuracy of 90% to detect PH. The same model performed with sensitivity of 100%, negative predictive value of 100%, and accuracy of 81% to detect varices.

DISCUSSION:

Noninvasive model combining LSM with platelet count can aid in identifying NCPH patients with PH and varices. However, this model requires validation in an independent cohort.

INTRODUCTION

Liver cirrhosis is the most common cause of portal hypertension (PH) and varices (1). However, there are rare conditions that primarily affect hepatic vasculature without causing hepatic dysfunction or cirrhosis, but can result in PH, referred to as noncirrhotic PH (NCPH) (2). Various etiologies including immunological, hematological, infections, exposure to toxin or drugs, or genetic predisposition can lead to NCPH. Although the exact pathophysiology of NCPH is unknown, individuals with NCPH often maintain normal liver function and lack or have minimal hepatic fibrosis, yet they remain susceptible to complications associated with PH (3). Despite its diverse origins, NCPH shares clinical similarities with cirrhosis, notably presenting with variceal bleeding as a common initial manifestation of PH (4). Traditionally, hepatic venous pressure gradient (HVPG), measured through a transjugular approach as a surrogate of portal pressure, is considered the gold standard in diagnosing PH. However, in NCPH, HVPG can be unreliable in determining the severity because of the presinusoidal nature of PH and the potential presence of intrahepatic vein to vein communications (5). In addition, due to the invasive nature of HVPG measurement, there has been a push toward the noninvasive diagnostic method to diagnose PH. Similarly, esophagogastroduodenoscopy (EGD) to detect varices also requires sedation, further emphasizing the need for noninvasive approaches to predict varices as well (6).

Both the American Association for the Study of Liver Diseases and Baveno VII consensus recommend combining liver stiffness measurement (LSM) with platelet count to detect clinically significant PH in cirrhotic patients (6,7). However, the cutoff for LSM and platelet count recommended by Baveno VII guidelines are not valid for NCPH. According to Baveno VII guidelines, patients with PH secondary to NCPH should be managed similarly to those with cirrhotic patients (7). Currently, there are no noninvasive markers which are used to identify PH and varices in patients with NCPH. We aimed to identify noninvasive factors associated with PH and varices in NCPH. We hypothesize that markers such as platelet count and liver stiffness measured using vibration-controlled transient elastography (VCTE) are associated with PH and varices in NCPH, similar to findings observed in cirrhosis. In addition, we aim to develop a stepwise prediction model that can aid clinicians in daily practice to decide further evaluation, prioritizing high sensitivity and a strong negative predictive value (NPV) to ensure reliable exclusion of further evaluation in low-risk individuals.

METHODS

Study population

This was a prospective observational study of consecutive patients with biopsy-proven NCPH who were followed in a single center from July 2015 through July 2024. Both patients with and without PH were included in the natural history study (8). Patients without PH were included only if they had liver biopsy showing features of early nodular regenerative hyperplasia or obliterative portal venopathy and an underlying disease known to cause NCPH (8). Exclusion criteria included all other causes of chronic liver disease and cirrhosis. The study was approved by the Institutional Review Board of the National Institutes of Health (NCT02417740), and written informed consent was obtained from all participants.

Data collected

We collected demographic data, clinical history, laboratory testing, liver magnetic resonance imaging, ultrasound of abdomen, LSM, EGD, and liver biopsy. Transjugular liver biopsy with valid portal pressure measurements was available in 49 (83%) participants. HVPG expressed in millimeters of mercury (mm Hg) was calculated by subtracting the free hepatic venous pressure from the wedged hepatic venous pressure. At least 2 readings, from middle or left and right hepatic veins, were obtained to ensure reproducibility and accuracy. All diagnostic assessments were completed within a week of each other to maintain consistency and accuracy. Liver biopsy was considered adequate if it was ≥2 cm length or was otherwise considered adequate for interpretation by the expert pathologist (D.E.K.).

LSM and controlled attenuation parameter measurements were performed using VCTE (FibroScan-Echosens, Paris, France) by trained operators after a minimum 4-hour fasting period. VCTE was equipped with an automatic probe that recommends the appropriate probe for each patient based on real-time assessment. Controlled attenuation parameter was measured in decibels per meter (dB/m) and LSM in kilopascals (kPa). For LSM, at least 10 valid individual measurements with interquartile range/median ≤30% was necessary to consider the test valid and reliable. The Model For End-Stage Liver Disease-Sodium, aspartate aminotransferase to platelet ratio index (APRI), and fibrosis-4 (FIB-4) index were calculated based on previously validated formulas (9,10).

Study definitions

1. PH—defined by the presence of 1 or more of the following: namely, gastric, esophageal, or ectopic varices; portal hypertensive gastropathy on EGD; or ascites or portosystemic shunting in imaging.

2. Varices—defined by the presence of esophageal or gastric varices on EGD.

Outcome

The primary outcomes of interest are the area under receiver operating characteristic (AUROC) values for the model's performance in predicting (i) PH and (ii) varices.

The secondary outcome is overall model performance, including sensitivity, specificity, positive predictive value (PPV), NPV, and accuracy in predicting PH and varices.

Statistical analysis

Statistical analysis was performed using R statistical software (version 4.1.2). Continuous data are presented as median (interquartile range), and categorical variables are presented as frequencies (percentages). Univariate analysis was conducted to assess the association between each marker and the outcomes separately (presence of varices and PH). For continuous variables, the Student t test or Mann-Whitney U test was used to test the difference between the presence and absence of the outcome for normally or nonnormally distributed data, respectively. For categorical variables, the χ² test (or Fisher exact test if frequencies are small) were used. Based on variables selected with statistical significance in the univariate analyses, a multivariable model was constructed using logistic regression for the binary outcome of varices and PH. Multicollinearity testing was performed using a Spearman correlation matrix to identify variables with strong correlation (below −0.7 or above +0.7). From these, clinically meaningful and simple variables were retained in the multivariable model to avoid redundancy. Subsequently, AUROC curve analysis was used to evaluate the discriminatory ability of the individual components and combined model quantified by the area under the ROC curve. In addition, the Hosmer-Lemeshow goodness of fit test was conducted to evaluate adequacy of the logistic regression model in fitting the observed data. The correlation between variables used in final model and HVPG was evaluated using the Spearman rank correlation coefficient. All tests were 2-sided, and P values <0.05 were considered to be significant.

Model construction

The classification tree method was used to identify optimal cutoff for each selected variable to create a simple algorithm with easy applicability for clinicians (11). For example, if both platelet counts and LSM are selected in the multivariate modeling, in the first step, platelet count, which is easily available routine blood test, was used for the initial screening with a cutoff. In the second step, LSM was incorporated to the model with goal to identify patients who will be missed by the platelet count. After the model construction with the selected cutoff, the percentage of outcomes at each node of the stepwise algorithm was calculated, and the Fischer exact test was performed to evaluate difference in proportions between the nodes. A P value less than 0.05 was considered statistically significant, indicating that the proportions of outcomes differed significantly across the nodes. Test characteristics—sensitivity, specificity, NPV, PPV, and accuracy—were calculated to evaluate the performance of the stepwise algorithm for clinicians.

RESULTS

Baseline characteristics of the cohort

Among 63 patients enrolled in the study, 4 were excluded (1 undergoing evaluation and 3 had cirrhosis on biopsy, Figure 1). Baseline characteristics of the overall cohort is summarized in Table 1. The median age of the cohort was 49 years, with a slight male (56%) predominance. Majority (88%) were of White race. Ten patients (17%) had body mass index (BMI) > 30 kg/m². Underlying etiology was immunodeficiency in 27 (46%) patients, followed by idiopathic in 13 (22%) and autoimmune disorders in 6 (10%) patients. Six patients (10%) had splenectomy. Median HVPG was 9 mm Hg (range: 1–32 mm Hg), and median LSM was 9.4 kPa (range: 3.2–59.6 kPa).

Figure 1.

Consort diagram of the cohort. NCPH, noncirrhotic portal hypertension.

Table 1.

Baseline characteristics of the cohort

Baseline characteristics	Overall cohort (N = 59)
Age (yr)	49 [39–60]
Sex, female	26 [44%]
Race
White	52 [88%]
Black	2 [3%]
Other	5 [9%]
Ethnicity
Non-Hispanic or Latino	53 (90%)
Hispanic or Latino	6 (10%)
Body mass index (kg/m2)	25.1 [22.1–27.5]
Underlying disorder
Immunodeficiency, n (%)	27 [46%]
Autoimmune disorders, n (%)	6 [10%]
Hematological disorders, n (%)	5 [8%]
Exposure to medication, n (%)	4 [7%]
Genetic disorders, n (%)	4 [7%]
Undiagnosed, n (%)	13 [22%]
AST (U/L)	32 [25–63]
ALT (U/L)	40 [28–64]
Alkaline phosphatase (U/L)	106 [81–184]
Total bilirubin (mg/dL)	0.9 [0.6–1.3]
Albumin (gm/dL)	4 [3.7–4.3]
Lactate dehydrogenase (U/L)	187 [163–221]
Gamma glutamyl transferase (IU/L)	62 [30–144]
Platelet count (109/L)	124 [70–186]
FIB-4 score	2.9 [1.5–5.2]
AST to platelet ratio index	1.1 [0.6–1.7]
AST to ALT ratio	1.08 [0.891–1.29]
MELD-NA score	9 [7–10]
Controlled attenuation parameter (dB/m)	189 [172–243]
Liver stiffness measurement (kPa)	9.4 [5.6–13.5]
Hepatic venous pressure gradient (mm Hg)a	9 [4–12]

Data presented as median (IQR) or n (%).

ALT, aspartate aminotransferase; AST, alanine aminotransferase; FIB-4, fibrosis-4; IQR, interquartile range; MELD-Na, Model for End-Stage Liver Disease-Sodium.

^a49 patients had valid HVPG measurement available.

Factors associated with PH

Among 59 patients, 41 (69%) had features of PH. There was no difference in age, sex, race, BMI, and liver-associated enzymes among those with and without PH (Table 2). However, platelet count was significantly lower in those with PH compared with those without (79 vs 218 × 10⁹/L, P < 0.001). On the other hand, LSM was significantly higher in those with PH compared with those without (11.5 vs 5.7 kPa, P < 0.001). In addition, FIB-4 index (4.2 vs 1.4, P < 0.001) and APRI (1.4 vs 0.7, P < 0.001) were significantly higher in those with PH. Median HVPG measurement was significantly higher in those with PH compared with those without (11 vs 4 mm Hg, P = 0.001). However, 6 (18%) patients with PH had HVPG ≤ 5 mm Hg.

Table 2.

Factors associated with portal hypertension

Characteristics	Portal hypertension presentc (N = 41)	Portal hypertension absentc (N = 18)	P valuea
Age (yr)	49 [40–59]	50 [35–61]	0.9
Sex, female	17 (41%)	9 (50%)	0.6
Race			0.4
White	37 (90%)	15 (83%)
Black	1 (2%)	1 (6%)
Other	3 (7%)	2 (11%)
Body mass index (kg/m2)	24.1 [22.5–28.2]	25.1 [21.8–26.9]	0.7
AST (U/L)	32 [24–59]	35 [25–81]	0.8
ALT (U/L)	41 [29–57]	36 [26–69]	0.8
Alkaline phosphatase (U/L)	115 [81–178]	90 [81–175]	0.7
Total bilirubin (mg/dL)	1.0 [0.7–1.5]	0.6 [0.4–1.0]	0.1
Albumin (g/dL)	3.9 [3.5–4.2]	4.2 [3.8–4.3]	0.3
Lactate dehydrogenase (U/L)	187 [164–211]	189 [162–222]	0.7
Gamma glutamyl transferase (IU/L)	69 [37–134]	35 [22–144]	0.3
Platelet count (109/L)	79 [64–124]	218 [150–262]	<0.001
Fibrosis 4 index	4.2 [2.0–6.1]	1.4 [1.1–2.1]	<0.001
AST to platelet ratio index	1.4 [1.0–2.6]	0.7 [0.4–0.9]	<0.001
AST to ALT ratio	1.08 [0.90–1.29]	1.11 [0.79–1.34]	0.7
MELD-NA score	9 [8–11]	8 [7–10]	0.1
Controlled attenuation parameter (dB/m)	189 [151–245]	189 [175–233]	0.9
Liver stiffness measurement (kPa)	11.5 [8.3–20.3]	5.7 [4.3–6.2]	<0.001
Hepatic venous pressure gradient (mm Hg)b	11 [6–15]	4 [2–6]	0.001

Data presented as median (IQR), or n (%).

ALT, aspartate aminotransferase; AST, alanine aminotransferase; IQR, interquartile range; MELD-Na, Model for End-Stage Liver Disease-Sodium.

^aP values show a comparison among the groups using either Student t test or the Mann-Whitney U test for continuous variables and corrected χ² test or a 2-sided Fisher exact test for categorical variables.

^b49 patients (33 with and 16 without portal hypertension) had valid hepatic venous pressure gradient.

^cFeatures of portal hypertension were defined by presence of gastric, esophageal, or ectopic varices, portal hypertensive gastropathy, or the presence of ascites or portosystemic shunting in imaging.

Factors associated with varices

Among 41 patients with PH, varices were the most common feature of PH seen in 36 (88%) patients on EGD. Similar to findings with PH, there was no difference in age, sex, race, BMI, and liver-associated enzymes among those with and without varices. Median platelet count was significantly lower in those with varices compared with those without (75 vs 209 × 10⁹/L, P < 0.001). Median LSM was significantly higher in those with varices compared with those without (12 vs 5.9 kPa, P < 0.001). The FIB-4 index (4.6 vs 1.4, P < 0.001) and APRI (1.4 vs 0.6) were significantly higher in those with varices (Table 3). Median HVPG measurement was significantly higher in those with varices compared with those without (11 vs 5 mm Hg, P = 0.001). Five (18%) patients with varices had HVPG ≤ 5 mm Hg, while 3 (14%) patients without varices had HVPG ≥10 mm Hg. Three patients (7%) with varices presented with moderate to severe ascites that was evident on clinical examination. These patients underwent therapeutic paracentesis before undergoing LSM evaluation.

Table 3.

Factors associated with varices

Characteristics	Varices present (N = 36)	Varices absent (N = 23)	P valuea
Age (yr)	50 [42–59]	47 [32–61]	0.4
Sex, female	16 (44%)	10 (43%)	0.9
Race			0.4
White	32 (89%)	20 (87%)
Black	1 (3%)	1 (5%)
Other	3 (8%)	2 (9%)
Body mass index (kg/m2)	24.1 [22.8–27.6]	25.1 [21.2–27.2]	0.6
AST (U/L)	35 [27–66]	30 [23–56]	0.4
ALT (U/L)	45 [34–62]	30 [26–65]	0.1
Alkaline phosphatase (U/L)	121 [85–212]	88 [78–132]	0.1
Total bilirubin (mg/dL)	1.0 [0.7–1.5]	0.6 [0.5–1.1]	0.06
Albumin (g/dL)	3.9 [3.5–4.3]	4.1 [3.8–4.3]	0.08
Lactate dehydrogenase (U/L)	187 [163–228]	187 [162–214]	0.9
Gamma glutamyl transferase (IU/L)	79 [41–189]	35 [21–84]	0.06
Platelet count (109/L)	75 [63–129]	209 [128–254]	<0.001
Fibrosis 4 index	4.6 [2.5–6.6]	1.4 [1.2–2.4]	<0.001
AST to platelet ratio index	1.39 [0.93–2.44]	0.62 [035–0.84]	<0.001
AST to ALT ratio	1.09 [0.89–1.29]	1.07 [0.82–1.29]	0.6
MELD-NA score	9 [8–11]	8 [7–10]	0.1
Controlled attenuation parameter (dB/m)	189 [147–243]	184 [174–240]	0.7
Liver stiffness measurement (kPa)	12 [8.3–26.3]	5.9 [4.5–8.4]	<0.001
Hepatic venous pressure gradient (mm Hg)b	11 [6–15]	5 [3–7]	0.001

Data presented as median (IQR), or n (%).

ALT, aspartate aminotransferase; AST, alanine aminotransferase; IQR, interquartile range; MELD-Na, Model for End-Stage Liver Disease-Sodium.

^aP values show a comparison among the groups using either Student t test or the Mann-Whitney U test for continuous variables and corrected χ² test or a 2-sided Fisher exact test for categorical variables.

^b49 patients (28 with and 21 without varices) had valid hepatic venous pressure gradient.

Noninvasive predictors of PH and varices

In the next step, we performed a multivariable analysis using statistically significant variables in univariable analysis to identify predictors of PH and varices. As shown in univariable analysis, platelet count, FIB-4 index, APRI, and LSM were significantly different between those with and without outcomes. However, the correlation matrix showed that platelet count was strongly correlated with FIB-4 index (−0.76) and APRI (−0.74), but only weakly correlated with LSM (−0.29). Since platelet count was a simple and clinically meaningful variable compared with FIB-4 and APRI, we decided to include platelet count in the multivariable analysis. FIB-4 demonstrated comparable performance with platelet count alone in predicting PH (AUROC = 0.84) and varices (AUROC = 0.81), supporting our decision to prioritize platelet count for the model.

Each unit increase in LSM was associated with 1.84 times (95% CI 1.33–2.99; P = 0.002) higher odds of having PH and 1.42 times (95% CI 1.15–1.85; P = 0.003) higher odds of having varices. However, every 10 unit increase in platelet count was associated with decreased odds of both PH (odds ratio = 0.85, 95% CI 0.73–0.95; P = 0.01) and varices (odds ratio = 0.88, 95% CI 0.78–0.96; P = 0.01) (Table 4). Hosmer and Lemeshow goodness of fit test for the PH model was χ² = 4.56, df = 8, P = 0.80 and for the varices model was χ² = 7.93, df = 8, P = 0.44, showing that both models had good fit.

Table 4.

Multivariable model to predict outcome

Predictive model	Odds ratio (95% CI)	P value
Model for portal hypertension
Liver stiffness measurement (kPa)	1.85 (1.33–2.99)	0.002
Platelet count (×109/L)a	0.85 (0.73–0.95)	0.01
Model for varices
Liver stiffness measurement (kPa)	1.42 (1.15–1.85)	0.003
Platelet count (×109/L)a	0.88 (0.78–0.96)	0.01

^aOdds ratio calculated for every 10 unit change in platelet count.

Correlation between HVPG and factors associated with PH

In cirrhosis, HVPG is considered the gold standard in assessing PH. We assessed whether HVPG correlates with platelet count and LSM. The correlation between LSM and HVPG was found to be moderately positively correlated (rho = 0.53; 95% CI 0.29–0.71, P < 0.001). The correlation between HVPG and platelet count was found to be moderately negatively correlated (rho = −0.43; 95% CI −0.64 to −0.17, P = 0.002) (Supplementary Figure 1, http://links.lww.com/CTG/B421).

Model development

A model combining platelet count and LSM performed with AUROC of 96% (95% CI 91–99%) in predicting PH and 92% (95% CI 85–99%) in predicting varices (Figure 2). We decided to create a stepwise algorithm for easy applicability for clinicians. Using classification tree, we initially identified individual cutoff values for both the platelet count and LSM. For both PH and varices, LSM cutoff was found to 6.9 kPa. For PH, platelet cutoff was found to be 140 × 10⁹/L, and for varices, the cutoff was found to be 202 × 10⁹/L (Supplementary Table 1, http://links.lww.com/CTG/B421). For a simpler model applicable to clinical use, we decided to use platelet cutoff of 140 × 10⁹/L for both PH and varices prediction. Patients with platelet count < 140 × 10⁹/L should undergo assessment for PH and varices. However, those with platelet count ≥140 × 10⁹/L, should get LSM and based on which can decide whether further evaluation is required or not (Figure 3).

Figure 2.

Performance of the combined model including platelet count and liver stiffness measurement in comparison with the individual model for predicting (a) portal hypertension and (b) varices. AUROC, area under receiver operating characteristic; LSM, liver stiffness measurement; ROC, receiver operating characteristic.

Figure 3.

Noninvasive decision model. Clinical decision model for evaluating disease severity in a patient suspected of having noncirrhotic portal hypertension: Patients with platelet count less than 140 × 10⁹/L should undergo assessment for portal hypertension and varices. If ≥140 × 10⁹/L, liver stiffness measurement (LSM) using vibration-controlled transient elastography should be performed. If LSM ≥7 kPa patients should undergo further evaluation, if <7 kPa portal hypertension and varices can be safely ruled out. LSM, liver stiffness measurement.

Model performance

The model performed with 100% sensitivity and NPV for predicting PH and varices (Table 5). However, specificity was modest at 67% for predicting PH and 52% in predicting varices. The accuracy in predicting PH and varices was 90% and 81%, respectively. No patient with platelet count > 140 × 10⁹/L and LSM < 7 kPa had PH or varices. The Youden index for identifying PH was 0.67. A proportion of patients identified at each node is shown in Supplementary Figure 2 (http://links.lww.com/CTG/B421). There were significant differences in the proportions across the 3 nodes for both PH and varices (P < 0.001 for both), suggesting that the distribution of these outcomes varied significantly between the nodes.

Table 5.

Performance of decision tool

Test characteristics	Predict portal hypertension (95% CI)	Predict varices (95% CI)
Sensitivity	100% (91–100)	100% (90–100)
Specificity	67% (41–87)	52% (31–73)
Negative predictive values	100% (74–100)	100% (74–100)
Positive predictive value	87% (74–95)	77% (62–88)
Accuracy	90% (79–96)	81% (70–90)

DISCUSSION

This study shows that LSM, measured using a VCTE, along with platelet count, can help identify NCPH patients with PH and varices. The platelet threshold identified is comparable with what is recommended for cirrhotic patients, but the LSM threshold for detecting PH was much lower that what is typically used in cirrhosis (12). The decision model was designed with a focus on high sensitivity and NPV to serve as a screening tool for determining whether further evaluation is needed. Notably, the model successfully identified all patients with PH or varices.

Among blood-based biomarkers, platelet count is an important factor which aids in identifying PH or varices. This is well established in cirrhosis because PH results in splenomegaly, which in turn results in platelet sequestration. In this study, the platelet cutoff identified for NCPH (140 × 10⁹/L) was similar to the threshold used in cirrhosis (150 × 10⁹/L) (12,13), suggesting a shared underlying pathophysiology driving thrombocytopenia. Since platelet count is readily available, we chose it as the initial screening step. However, in some patients with NCPH, underlying conditions such as bone marrow disease or splenectomy can make platelet count unreliable (14). In these cases, if thrombocytopenia is due to an alternative explanation such as a myeloproliferative disorder, LSM could be considered for further risk assessment. Unlike cirrhosis, NCPH is not associated with alteration in liver-associated enzymes and liver synthetic function, so most common blood-based biomarkers including aspartate aminotransferase/alanine aminotransferase ratio and Model For End-Stage Liver Disease-Sodium score used to distinguish severity of liver disease in cirrhosis does not apply here (3,15). This could be because in NCPH, the hepatic parenchyma is usually preserved or minimally fibrotic. Although FIB4 and APRI were significantly different between patients with or without PH or varices, this difference was primarily driven by platelet count.

VCTE is widely available but primarily limited to hepatology clinics, making platelet count a useful initial screening tool for risk stratification before further evaluation. To overcome the drawback of platelet count, LSM was incorporated into the model. The LSM threshold of 7 kPa was comparable with the median LSM reported in patients with NCPH in previous studies (16). Previous studies have shown LSM in NCPH is significantly higher than in healthy controls with no liver disease, but lower than those with cirrhosis (16,17). Although LSM > 15 kPa is typically seen in patients with cirrhosis, some patients with NCPH in this study exceeded this threshold without histological evidence of advanced fibrosis, and similar findings were noted in previous studies (18,19). The exact cause for the elevation in LSM in NCPH is unclear, but vascular and sinusoidal pressure changes could be contributing factors. Importantly, LSM and platelet count have prognostic significance in patients with NCPH because our previous studies have shown they can predict the risk of death or the need of a liver transplantation (8).

Recently, there has been increased awareness NCPH, and the list of associated conditions continues to grow (20). In this NCPH cohort, various etiologies were included, although there was a higher prevalence of immunological condition. This 2-step model successfully identified all patients with PH and varices, but its PPV and specificity were suboptimal. As a result, 6 patients without PH and 11 patients without varices would undergo further evaluation. However, we could confidently avoid unnecessary testing in 12 patients, improving efficiency. An important advantage of this model is its ease of use. If platelet count is low, patients should undergo abdominal imaging to assess for signs of PH. Those with unreliable platelet counts like those who underwent splenectomy or have conditions associated with thrombocythemia, should undergo LSM to guide the next steps in management. This model is clinically meaningful in identifying patients with varices, especially since most with varices remain asymptomatic in earlier stages and (21) these can progress over time. Intervention with agents such as beta blockers therapy could help prevent progression and reduce risk of complications (22).

As our group has previously shown that HVPG measurement is often unreliable in patients with NCPH, unlike in cirrhosis (5). Although LSM and platelet count correlated with HVPG, the correlation was not as strong as seen in patients with cirrhosis (23). These findings suggest that HVPG measurement may not be clinically necessary for assessing patients with NCPH.

This study has some limitations. One of the major limitations is the absence of a validation cohort, which is challenging to obtain due to the rarity of NCPH. The relatively small cohort size restricted our ability to perform multivariable analyses, so these findings should be generalized with caution. Nevertheless, the study was conducted in a single center with comprehensive evaluations, and all other chronic liver diseases were excluded. Another limitation is the potential selection bias because patients referred to our center may have unique characteristics compared with the general population. However, we included all consecutive patients irrespective of their disease severity. Splenic stiffness measurement is a potential method to assess PH in cirrhotic patients (24), but we did not have splenic stiffness assessments available. In addition, some patients had undergone splenectomy, making it impossible to measure splenic stiffness. Most of our patients were White, raising questions about generalizability of these findings to other races. Further studies are needed to validate these results in diverse population.

In summary, a noninvasive model combining LSM and platelet count can help identify NCPH patients with PH and varices.

CONFLICTS OF INTEREST

Guarantor of the article: Theo Heller, MD.

Specific author contributions: All authors had a substantial contribution to this work. All authors provided approval for the final submitted version of the manuscript. Individual contributions are listed as follows: H.G.: conception and design; data acquisition, analysis, interpretation of data; drafted the initial work and substantively revised it. M.M., N.A.B., D.E.K., C.K.: data acquisition, interpretation, and revision of the article. S.S., J.N.W.: acquisition of data. T.H.: conception and design; analysis, interpretation of data and substantively revised it.

Financial support: This work was supported by the Division of Intramural Research Program at National Institute of Diabetes and Digestive and Kidney Diseases (1ZIADK075008).

Potential competing interests: None to report.

Data availability statement: Data in aggregate available on request from the corresponding author.

Study Highlights.

WHAT IS KNOWN

• ✓ Validated noninvasive criteria can predict portal hypertension and varices in cirrhotic patients.

• ✓ No noninvasive model is available to predict portal hypertension and varices in noncirrhotic portal hypertension (NCPH).

WHAT IS NEW HERE

• ✓ Platelet count and liver stiffness measurement measured using transient elastography can be used to identify portal hypertension and varices in NCPH.

• ✓ A simple stepwise model combining platelet count and liver stiffness measurement can aid in identifying which patients with NCPH need evaluation for esophageal varices.

Supplementary Material

ct9-17-e00956-s001.docx ^{(98.4KB, docx)}

ABBREVIATIONS:

AASLD

American Association for the Study of Liver Diseases

ALT

alanine aminotransferase

APRI

aspartate aminotransferase to platelet ratio index

AST

aspartate aminotransferase

AUROC

area under the receiver operating curve

EGD

esophagogastroduodenoscopy

FIB-4

fibrosis 4

HVPG

hepatic venous pressure gradient

LSM

liver stiffness measurement

MELD Na

Model for End-Stage Liver Disease Sodium score

NCPH

noncirrhotic portal hypertension

NPV

negative predictive value

PH

portal hypertension

PPV

positive predictive value

VCTE

vibration-controlled transient elastography