In this cross-sectional study, we enrolled consecutive patients with a diagnosis of liver cirrhosis referred to the Department of Health, Medical, and Surgical Sciences of the University of Trieste (Italy) and evaluated at the Liver Clinic between January 2016 and December 2018. From an original cohort of 650 individuals with liver cirrhosis, only 220 (Fig. 1) patients with compensated advanced chronic liver disease (cACLD) met the enrollment criteria. Both liver and spleen elastography measures were reliable in 210 patients, 84 of whom (40%) had EVs. We had previously measured SS in a control group of 100 healthy individuals, where we also assessed SS inter-operator agreement [21,22].

Fig. 1.

From the original cohort of 650 patients diagnosed with liver cirrhosis, 150 had ongoing liver injury. From the remaining 500, we selected the ones (363) without previous history of decompensating events. Then, we chose patients without therapeutic confounding factors (such as current use of non-selective beta-blockers and previous esophageal band ligation), that could alter spleen stiffness values. From this group of 220 patients, 210 had reliable liver and spleen stiffness measurements.

(0.19MB).

We selected adult patients with liver cirrhosis, whose diagnosis was confirmed by imaging, elastography and/or histological evaluation and blood tests. As per the standard diagnostic procedure, the patients underwent an EGD with an evaluation of the presence/absence of esophageal and/or gastric varices and congestive gastropathy. Patients also underwent an elastographic examination of the liver and spleen with a maximum interval of 6 months from the endoscopy. Only patients with liver stiffness>11.34kPa [25], a value compatible with severe fibrosis (F4 according to METAVIR), were included.

Patients with no informed consent were categorically excluded. We also excluded pregnant women, patients with current alcohol abuse, patients with ongoing HCV infection (or SVR<12 months), patients with presence of decompensating events (such as hepatic encephalopathy, variceal hemorrhage, ascites, and spontaneous bacterial peritonitis), previous endoscopic EVs banding ligation, ongoing intake of non-selective beta-blockers (NSBB), history of portal vein thrombosis, placement of transjugular intrahepatic portosystemic shunt (TIPS), non-cirrhotic causes of PH, and current/recent diagnosis of hepatocellular carcinoma [26].

A complete evaluation of the liver and splenic-portal axis was performed using a Philips Affiniti 70 ultrasonography system with a 1–5MHz convex probe. The sampling of the portal flow was performed at the hepatic hilum level with a probe positioned in the intercostal window, and with the acquisition of at least three values. The portal caliber was evaluated at the level between the portal vein and the hepatic artery with the probe positioned at the epigastric-sub-focal level. The portal vein diameter was expressed in cm, while the portal vein flow velocity was expressed in cm/s. The spleen was evaluated in the supine position patient via the intercostal window and trying to acquire the broadest possible scan that included the splenic hilum. Both the bipolar spleen diameter (expressed in cm) and the splenic area (expressed in cm2) were measured at the organ hilum.

In order to avoid confounding factors in stiffness measurement, the patients had to arrive while fasting for at least 3h and with no caffeine intake during the previous hour [27,28].

Liver and spleen elastography were performed with the same instrument used for ultrasonography, operating the ElastPQ evaluation protocol. The machine converted the measure from m/s to kPa, with the formula: E=ρcs2 – where E is the Young Modulus (kPa), ρ is the density of the tissue (kg/m3), and cs is the shear wave speed (m/s). All measures were acquired by an experienced operator with four years of experience in ultrasound and elastography.

Patients were positioned in supine decubitus with the right arm (liver) or left arm (spleen) in maximal abduction in order to increase the intercostal acoustic window. The region of interest (ROI) was placed between the VII and VIII segments at least 1.5cm from the hepatic capsule (LS) and at the splenic hilum or lower pole at least 1cm from the splenic capsule (SS). The ROI was accurately located in an area without large liver vessels, bile ducts, and rib shadows. During the acquisition, the patient was requested to hold his/her breath for 5s. All measures obtained after a deep inspiration; maximal expirations and Valsalva maneuver were discarded [29,30]. In 10% of cases, breath-hold was practiced with the patient prior to initiating elastography.

Ten different valid elastographic measurements were obtained in all subjects both in the liver and in the spleen, and the median value was used. The measure obtained was acquired only if its standard deviation was <30%. According to Boursier et al. stiffness measurements were considered poorly reliable when they showed an interquartile (IQR)/median (M) ratio≥0.35; reliable when they showed 0.15≤IQR/M<0.35 and very reliable if IQR/M<0.15 [31]. We defined “technical failure” the impossibility to obtain any value or an IQR/M≥0.35 and selected values with an IQR/M<0.30.

All patients underwent a complete endoscopic examination conducted up to the second duodenal portion. The esophagus was evaluated in search of esophageal varices, which, if present, were classified with the Beppu classification (1981) [32]. In detail, the localization (L), the shape, and size (F), the color (C), the presence or absence of red color signs (RS), and the presence or absence of esophagitis (E) were evaluated. The gastric chamber, also evaluated in retroversion with adequate visualization of the fundus and cardiac region, was inspected for signs of congestive gastropathy and/or gastric varices.

Each patient underwent blood tests and a complete physical examination at the time of the visit. We evaluated: weight, height, body mass index (BMI), etiology of cirrhosis, clinical signs of liver disease decompensation (presence/absence of ascites, and hepatic encephalopathy). Subsequently, the following blood laboratory values were collected: AST, ALT, GGT, creatinine, total bilirubin, albumin, INR, and platelet count. The following scores were then calculated: Model for End-stage Liver Disease (MELD), Child–Pugh (CP), AST to Platelet Ratio Index (APRI), PSR, LSPS.

Most continuous variables were not Gaussian-distributed, and all are reported as median (50th percentile) and interquartile range (IQR, 25th and 75th percentiles). Discrete variables are reported as the number and proportion of subjects with the characteristic of interest.

Between-group comparisons of discrete variables were performed using Pearson's Chi-square test and those of continuous variables using median regression [33]. The association between varices (discrete; 0=no, 1=yes) and the four continuous predictors of interest (liver stiffness, spleen stiffness, spleen diameter and platelets) was evaluated using univariable logistic regression [34].

We used univariable fractional polynomials to test whether the logits of the predictors were linear and transformed both liver stiffness and spleen stiffness using an inverse square root transformation to make their logits linear [35]. To evaluate the collinearity of SS, LS, spleen size, and platelet count, correlations were assessed using the Spearman rank-order correlation coefficient [36].

We compared models using Akaike information criterion (AIC) and the Bayesian information criterion (BIC) and calculated Nagelkerke pseudo-R2 and the area under the receiver-operating characteristic curve (AUROC). The diagnostic accuracy was calculated using sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), accuracy, positive likelihood ratio (+LR), and negative likelihood ratio (−LR). Optimal cut-off values were chosen to safely rule-out patients with EVs and HRVs for screening purposes. Accordingly, we selected SS and LS cut-off threshold with maximal sensitivity and NPV and a minimal −LR.

The four logistic regression models (M1 to M4) used to evaluate the association between varices (0=no; 1=yes) and transformed liver stiffness (M1), transformed spleen stiffness (M2), spleen diameter (M3) and platelets (M4) are given in Table 2. M2, based on transformed spleen stiffness, was the best model according to all discriminative and calibration metrics (lowest AIC, lowest BIC, highest AUROC, and higher Pseudo-R2).

The plots in Fig. 2 give the observed vs. the expected probability of varices as estimated from models M1 to M4. Calibration is plotted across ten percentiles (deciles of risk). The spike plot at the bottom of the graph plots the distribution of events (1=presence of varices) and non-events (0=absence of varices) and a lowess smoother is superimposed on the graph to allow a rough assessment of the calibration at the individual level.

Fig. 2.

Calibration plots corresponding to models M1–M4. The spike plot at the bottom of the graph plots the distribution of events (1=presence of varices) and non-events (0=absence of varices).

(0.21MB).

The cut-off value that allowed us to rule out the presence of EVs of any grade was 31kPa. This value showed a sensitivity of 100%, specificity of 60%, NPV of 100%, PPV of 62%, accuracy of 76%, +LR of 2.47. On the contrary, the cut-off value that allowed us to rule in the presence of EVs of any grade was 69kPa. This cut-off showed sensitivity of 14%, specificity of 100%, NPV of 64%, PPV of 100%, accuracy of 65%, +LR of −LR of 0.86. Regarding high-risk varices (HRVs), the cut-off of 46kPa showed sensitivity of 100%, specificity of 84%, NPV of 100%, PPV 27%, +LR of 6.19 and accuracy of 85%. Fig. 3 represents the probability of varices according to SS (A) as calculated from model 2 (M2), the grey area defines the upper and lower 95% C.I. of the probability defined by the black line. The probability formula has been derived in three steps from values reported in Table 2:

• 1.

  Calculation of Transformed Spleen Stiffness (TSS):

  

• 2.

  Calculation of the Linear Predictor (LP):

  

• 3.

  Calculation of the Spleen Stiffness Probability Index (SSPI) of varices from LP:

  

Fig. 3.

Plots the probability of varices according to liver stiffness (as calculated from model 1) and spleen stiffness (as calculated from model 2). Note that the units of the x-axis are retransformed to the original scale, i.e. kPa).

(0.11MB).

According to Fig. 3, the probability of EVs is <7% with SS values <30kPa (this value is the first inflection point of the curve). The slope of the curve rapidly increases between 30 and 50kPa. In particular, at 40kPa the probability of EVs has seen a six-fold (around 60%) gain, and at 50kPa the probability is nine times greater (around 90%). After 50kPa the probability steadily increases and reaches a plateau at 70kPa, where the probability is >97%.

LS showed an AUROC of 0.79 (95% C.I. 0.72–0.85, p<0.001) to discriminate the presence of EVs of any grade. SS and LS AUROCs were compared, and they were found to be statistically different (DeLong p<0.001). The cut-off of 20kPa showed a sensitivity of 73%, specificity of 78%, NPV of 81%, PPV of 69%, accuracy of 76%, +LR of 3.27, and −LR of 0.35. Using 20kPa as a cut-off, we would have missed 5 (out of 12) HRVs. Fig. 3 represents the probability of varices according to LS (B) as calculated from model 1 (M1), the grey area defines the upper and lower 95% C.I. of the probability defined by the black line. The probability formula has been derived in three steps from values reported in Table 2:

• 1.

  Calculation of Transformed Spleen Stiffness (TLS):

  

• 2.

  Calculation of the Linear Predictor (LP):

  

• 3.

  Calculation of the Liver Stiffness Probability Index of varices from LP:

  

According to Fig. 3, a patient with a LS equal to 20kPa has a probability of having EVs equal to 50%. Moreover, the probability of having EVs is between 0 and 10% with values of LS=11–12kPa. The slope of the curve rapidly increases between 12 and 26kPa (where the probability is equal to 60%). Then the curve reaches a plateau at 50kPa, and the probability stabilizes between 75 and 80%.

PSR showed an AUROC of 0.83 (95% C.I. 0.77–0.88, p<0.001) to discriminate the presence of EVs of any grade. The cut-off of 909 showed a sensitivity of 80%, specificity of 70%, NPV of 83%, PPV of 64%, accuracy of 74%, +LR of 2.46, and −LR of 0.29. Using this cut-off value, we would not have missed any HRVs. LSPS showed an AUROC of 0.91 (95% C.I. 0.88–0.95, p<0.001) to discriminate the presence of EVs of any grade. The cut-off of 1.72 showed a sensitivity of a sensitivity of 93%, specificity of 62%, NPV of 93%, PPV of 62%, accuracy of 74%, +LR of 2.44, and −LR of 0.12. Using this cut-off value, we would have missed 1 (out of 12) HRVs. SS and PSR/LSPS AUROCs were compared, and they were found to be statistically different (DeLong p<0.001).

Using Baveno VI criteria, we would have missed 5 (out of 12) patients with HRVs. Those patients had LS<20kPa, but all showed SS>46kPa. Forty-five patients (21.4%) matched the criteria to rule-out HRVs, and none of them presented EVs during the EGD. At the same time, we would have uselessly enrolled for EGD 76 patients (36.1%) who matched the criteria mentioned above to rule-in HRVs. Twenty-three had no endoscopic sign of esophageal varices, whereas 53 had low-risk varices.

A correlation between platelet count and stiffness measurements was found. This correlation was stronger with SS (r=0.67) than LS (r=0.4), p<0.001. In patients with SS<31kPa (i.e., all without EVs) the median (IQR) value for platelet count was 145 (IQR 120.5–173.5) g/L, whereas in patients with SS>69kPa (i.e., all with EVs) the median (IQR) value for platelet was 75 (IQR 60.5–86.7) g/L. In patients with SS>46kPa, the median (IQR) value for platelet count was 86 (IQR 66–111) g/L. In patients with EVs and LS > 20kPa, the median (IQR) value for platelet was 87 (IQR 65–111) g/L, whereas patients without EVs and LS<20kPa the median (IQR) was significantly higher (p<0.001) and equal to 170 (IQR 120–176) g/L.

SS showed a higher correlation with both splenic diameter and splenic area (r=0.52 and r=0.55 respectively) if compared to LS (r=0.30 and r=0.25 respectively). In patients with SS<31kPa (i.e. all without EVs) the median (IQR) value for spleen bipolar diameter was 12 (IQR 10–13) cm, whereas the median (IQR) spleen area measured at hilum was 51 (IQR 43–65) cm2. In patients with SS>69kPa (i.e. all with EVs) the median (IQR) value for spleen bipolar diameter was 15 (IQR 14–16) cm, whereas the median (IQR) spleen area measured at hilum was 92 (IQR 79–94) cm2.

One hundred controls consisted of 51 females (51%) and 49 males (49%). Their median (IQR) age was 52 (IQR 28–56). The median (IQR) LS and SS values were 4.86 (IQR 4.17;5.59) kPa and 17.50 (IQR 15.63;20.52) respectively. LS and SS distribution between healthy subjects and patients with liver cirrhosis are reported in Fig. 4.

Fig. 4.

Spleen Stiffness (left) and Liver Stiffness (right) distribution in Healthy Subjects, Cirrhotics without EVs and Cirrhotics with EVs. Values are reported in kPa (y-axis). Spleen Stiffness values have been found to overlap between healthy subjects and cirrhotic patients who have not developed EVs yet. No overlap was found between liver stiffness values in the three sub-groups.

(0.12MB).