A Liver Stiffness Measurement-Based, Noninvasive ... - Nature

2 downloads 0 Views 433KB Size Report
Jan 19, 2010 - Prediction Model for High-Risk Esophageal Varices in B-Viral Liver Cirrhosis. Beom Kyung Kim , MD 1, Kwang-Hyub Han , MD 1– 4, Jun Yong ...
1382

ORIGINAL CONTRIBUTIONS

LIVER

A Liver Stiffness Measurement-Based, Noninvasive Prediction Model for High-Risk Esophageal Varices in B-Viral Liver Cirrhosis

nature publishing group

Beom Kyung Kim, MD1, Kwang-Hyub Han, MD1–4, Jun Yong Park, MD1–3, Sang Hoon Ahn, MD, PhD1–3, Ja Kyung Kim, MD1–3, Yong Han Paik, MD, PhD1–4, Kwan Sik Lee, MD1–3, Chae Yoon Chon, MD1–3 and Do Young Kim, MD1–3 OBJECTIVES:

Periodic endoscopic screening for esophageal varices (EVs) and prophylactic treatment for high-risk EVs ((HEVs); (1) medium/large EVs and (2) small EVs with red sign or decompensated cirrhosis) are currently recommended for all cirrhotic patients. However, if a simple, noninvasive test is available, many low-risk patients may safely avoid endoscopy. We developed and validated a new liver stiffness measurement (LSM)-based prediction model for HEVs.

METHODS:

We prospectively enrolled 280 consecutive B-viral cirrhosis patients from 2005 to 2007 (training set) and 121 from 2007 to 2008 (validation set). All underwent laboratory workups, endoscopy, LSM, and ultrasonography. For detection of HEVs, univariate and multivariate analysis were performed, using χ2-test/t-test and logistic regression, respectively. A prediction model was derived from multivariate predictors.

RESULTS:

In the training set, 90 had HEVs, and multivariate analysis showed significant differences in LSM, spleen diameter, and platelet count between patients with and without HEVs. We developed LSMspleen diameter to platelet ratio score (LSPS): LSM × spleen diameter/platelet count. The area under the receiver-operating characteristic curve (AUROC) in the training set was 0.954. At LSPS < 3.5, 94.0% negative predictive value (NPV) was provided (184 patients), whereas 94.2% positive predictive value (PPV) was achieved (69 patients) at LSPS > 5.5. Overall, the likelihood of HEVs was correctly diagnosed in 253 patients (90.3%). Its predictive values were maintained at similar accuracy in subsequent validation set (AUROC = 0.953; 94.7% NPV/93.3% PPV at cutoff 3.5/5.5, respectively).

CONCLUSIONS: LSPS is a reliable, noninvasive method for detection of HEVs. Patients with LSPS < 3.5 may avoid

endoscopy safely, whereas those with LSPS > 5.5 should be considered for appropriate prophylactic treatments. Am J Gastroenterol 2010; 105:1382–1390; doi:10.1038/ajg.2009.750; published online 19 January 2010

INTRODUCTION Portal hypertension is a frequent complication of liver cirrhosis, a contributing factor for ascites and hepatic encephalopathy, and a direct cause of variceal hemorrhage (1,2). An increase in portal pressure leads to the development of portosystemic collateral circulation, of which gastroesophageal varices are the most important clinical feature. Gastroesophageal varices exist in 30–60% of patients with liver cirrhosis, depending on the severity of portal hypertension. The yearly rate of development of new cases is

5–10%, whereas the growth rate from small to large varices ranges between 5 and 30% (3,4). As variceal bleeding is the most important complication of cirrhosis, accelerating the progression of decompensation to a stage at which the patient is at an extremely high risk of death, current guidelines recommend screening all cirrhotic patients by endoscopy to identify those who should undergo prophylactic treatment. Nonselective β-blockers significantly reduce the bleeding rate in more than half of patients with medium/large esophageal

1 Department of Internal Medicine, Yonsei University College of Medicine, Seoul, Korea; 2Yonsei Institute of Gastroenterology, Yonsei University College of Medicine, Seoul, Korea; 3Liver Cirrhosis Clinical Research Center, Seoul, Korea; 4Brain Korea 21 Project for Medical Science, Seoul, Korea. Correspondence: Do Young Kim, MD, Department of Internal Medicine, Yonsei University College of Medicine, 250 Seongsanno, Seodaemun-gu, Seoul 120-752, Korea. E-mail: [email protected] Received 3 October 2009; accepted 14 December 2009

The American Journal of GASTROENTEROLOGY

VOLUME 105 | JUNE 2010 www.amjgastro.com

A New LSM-Based Prediction Model for Esophageal Varix

METHODS This study was conducted in two parts. In the first place, we attempted to identify variables associated with the presence of HEVs and construct the LSM-based formula in the training set. Second, we evaluated the reproducibility of that formula in the © 2010 by the American College of Gastroenterology

subsequent validation set, which included subjects with the same etiology, but enrolled in the different period. Patients

In the first part of the study, between July 2005 and June 2007, we prospectively enrolled 325 consecutive patients with hepatitis B virus-related liver cirrhosis at Severance Hospital, Yonsei University College of Medicine, Seoul, Korea. Cirrhosis was diagnosed using standard laboratory, radiological, and physical examination findings, or by liver histology in equivocal cases. Exclusion criteria were as follows: any etiologies for chronic liver disease other than hepatitis B virus (n = 16); alcohol ingestion in excess of 30 g/day for more than 5 years (n = 6); a body mass index > 35 (n = 1); previous variceal bleeding, β-blocker therapy, or endoscopic treatments (band ligation or sclerotherapy) (n = 9); previous surgery for portal hypertension or transjugular intrahepatic portosystemic stent shunt placement (n = 5); portal vein or splenic vein thrombosis (n = 2); hepatocelluar carcinoma (n = 2); presence of severe ascites that might significantly hamper the accurate assessment of LSM (n = 3); unreliable LSM with fewer than eight successful acquisitions or a success rate of < 60% (n = 1). Finally, 280 patients were included in the study sample (training set). Thereafter, from July 2007 and December 2008, we enrolled 145 B-viral cirrhosis patients in the same manner and 121 were ultimately studied, to validate the reproducibility of those results (validation set). All systematically underwent complete biochemical workups, LSM, ultrasonography, and endoscopy within 2 days. The study protocol was in accordance with the ethical guidelines of the 1975 Declaration of Helsinki. Written, informed consent was obtained from each participant or responsible family member after the possible complications of diagnostic procedures had been fully explained. This study was approved by the Institutional Review Board of Severance Hospital. Endoscopic and ultrasonographic evaluation and assessment of LSM

EVs were classified as small (minimally elevated veins above the esophageal mucosal surface), medium (tortuous veins occupying less than one-third of the esophageal lumen), or large (those occupying more than one-third of the esophageal lumen). In this study, HEVs were defined as patients with medium/large EVs and patients with small EVs with red signs or decompensated cirrhosis (3,5). Two independent experienced endoscopists performed examinations with good agreement in evaluation of the size of EVs (the κ value 0.96). In the cases of discrepancy between two operators, the final reports based on their consensus opinion were adopted. Within 1 day after or before endoscopy, all the patients underwent ultrasonographic examination of the upper abdomen, performed by two independent experienced operators with experience of more than 10 years (K.H. Han and D.Y. Kim). A spleen bipolar diameter was defined as the greatest longitudinal dimension at the level of splenic hilum on the image monitor using electronic calipers (18). Its measurement was performed in all patients by two operators, which was technically feasible in all patients. The interobserver coefficient of variation was 1.1% and according to The American Journal of GASTROENTEROLOGY

LIVER

varices (EVs) and in patients with small EVs with red signs or decompensated cirrhosis (2,3,5). However, as the prevalence of high-risk EVs (HEVs) at any given point in time is ~15–25%, the majority of subjects undergoing screening endoscopy either do not have varices, or have varices that do not require prophylactic therapy. Therefore, periodic endoscopic screening in all cirrhotic patients, especially low-risk groups, might unnecessarily induce an invasive and expensive procedure, ultimately increasing not only the medical workload of endoscopy units, but also the financial burden to patients, as the number of patients with chronic liver disease increases and their survival improves. Furthermore, compliance may be limited, because even asymptomatic patients may be required to repeatedly undergo an unpleasant endoscopic procedure, which usually requires conscious sedation decreasing work productivity, and has a small but not insignificant risk of complications. Accordingly, there is considerable interest in developing noninvasive models to predict the presence of varices by nonendoscopic methods, using biochemical, clinical, and ultrasonographic parameters (6–12). Among these, liver stiffness measurement (LSM), relying on the calculation of liver elasticity (or stiffness) from the velocity of a low-frequency elastic wave inside the liver, has recently been evaluated for the prediction of clinically significant portal hypertension. Liver stiffness shows an excellent correlation with fibrosis and with the hepatic venous pressure gradient, because portal hypertension is a direct consequence of the fibrotic transformation of liver tissue (7,9,13–15). Several investigators have evaluated the utility of LSM for prediction of EVs and found variable efficacies among studies (7,9,16). Therefore, although LSM has a potential theoretical role in assessment of portal hypertension, complementary processes using other parameters, rather than using LSM alone, are still required for reliable noninvasive screening. So far, few studies have tried to generate and validate a prediction model for EVs, which uses LSM in combination with other parameters reflecting portal hypertension. Furthermore, etiologically, this is the first study to be carried out primarily in a population with hepatitis B virus-related cirrhosis (7,9,16). Although LSM has been shown to predict progression of fibrosis and cirrhosis with good accuracy, some differences in cutoffs and efficacies that depend on the underlying etiology still exist (3,17). Physicians may find it useful to identify patients with HEVs who require further treatments and more vigorous surveillance at the time of diagnosis and to conduct periodic follow-ups by means of a simple, LSM-based noninvasive model. The aims of this study are to assess the diagnostic efficacy of LSM for HEVs, to generate a simple, LSM-based noninvasive prediction model, and to validate the usefulness of this model.

1383

LIVER

1384

Kim et al.

the above definition, we selected the longer value of two ones by two operators, as the final spleen diameter. After a complete upper abdomen ultrasound examination, transient elastography was performed using a FibroScan (Echosens, Paris, France) by a single, well-trained technician. Details of the technical description and examination procedure have been reported earlier (19). For accurate assessment of LSM, before transient elastography, we tried to control ascites through diurecits/ albumin, or to drain out ascites, if necessary. Patients with severe ascites enough to hamper accurate assessment of LSM significantly were excluded from the enrollment. The results were expressed in kilopascals (kPa). Fewer than eight successful acquisitions or a success rate of < 60% was considered unreliable. All operators of endoscopy, abdomen ultrasonography, and transient elastography were independent from one another and blinded to the others’ instrumental results and the patients’ clinical and laboratory data. Statistical analysis

The major goals of this study were to detect HEVs using an LSMbased prediction model consisting of clinically relevant variables, identify whether and when patients should undergo prophylactic medical or endoscopic treatment, and to validate the model’s usefulness. In the first place, univariate analysis was performed, to detect candidate variables among various clinical factors for generation of new formula. Differences between continuous and categorical variables were examined statistically using Student’s t-test (or Mann–Whitney test, if appropriate) and χ2-test, respectively, in the training set. Thereafter, variables with P < 0.05 in the univariate analysis were included in subsequent multivariate analysis, where multiple logistic regression analysis was used to select variables to be maintained in the final model. Factors with P < 0.05 were finally selected as components of the new formula. On the basis of these multivariate predictors, we derived a multiple fractional equation for prediction of HEVs. To assess the diagnostic accuracy of the models, the receiver operating characteristic (ROC) curves were constructed and each area under ROC curve (AUROC) was computed. The sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV) were calculated using ROC curves. For the comparison of AUROC values among various tests, we used the method suggested by Hanley and McNeil (20). Thereafter, in subsequent validation set, we tested the diagnostic value of the formula that was derived from the training set. In addition, we tried to determine another optimal cutoff values to adjust alanine aminotransferase (ALT) level, as LSM was reported to increase with respect to necroinflammatory activity and variations in LSM value and its diagnostic performance was observed during ALT flares in chronic viral hepatitis in several studies (21,22). So, according to ALT level, we performed subgroup analyses, where patients were divided into those with ALT < upper limit of normal (ULN: defined as 45 IU/L) (Nomral-ALT group; 177 patients) and those with ALT ≥ULN (high-ALT group; 103 patients), in order to suggest the optimal cutoffs fitted in each subgroup. The American Journal of GASTROENTEROLOGY

A probability level (P) of 0.05 was chosen for statistical significance. Statistical analyses were performed using SAS software version 9.1.3 (SAS, Cary, NC).

RESULTS Baseline patient characteristics of the training and validation set

Baseline patient characteristics of the training and validation set are summarized in Table 1. Among 280 patients in the training set (mean age 52.5 years, 196 male), 124 patients (44.3%) had EVs (39 classified as small, 71 as medium, and 14 as large) and 90 had high-risk cases (32.1%). Portal hypertensive gastropathy was observed in 34 patients (12.1%) and 221 were classified as Child– Pugh class A (78.9%), 54 were class B (19.3%), and 5 were class C (1.8 %). Mean LSM value was 25.37±12.90 kPa, whereas spleen diameter was 12.03±2.62 cm. The LSM showed the positive correlation, as ALT level increases (r = 0.401, P = 0.001). One hundred and fifteen patients had a history of antiviral therapy. Those given antiviral therapy had a tendency of lower LSM (22.93±11.6) than those without antiviral therapy (27.08±13.1 P = 0.075), whereas the proportion of EVs were also lower in those given antiviral therapy (37.4%) than in those without antiviral therapy (49.1%, P = 0.053). There were no significant differences of baseline clinical characteristics between the training and validation set. Comparisons of variables according to the presence of HEVs in the training set

Table 2 shows the various parameters of patients according to the presence of HEVs in the training set. Significant differences in sex (P = 0.001), LSM (P < 0.001), white blood cell count (P = 0.017), hemoglobin (P < 0.001), platelet count (P < 0.001), prothrombin time (P < 0.001), total bilirubin (P = 0.001), albumin (P < 0.001), aspartate aminotransferase (P < 0.001), and spleen diameter (P < 0.001) were observed between the two groups, whereas no differences in age, body mass index, ALT, blood urea nitrogen (BUN), and creatinine were seen (Table 2). These univariate predictors were entered into a stepwise logistic regression model. Ultimately, LSM (P < 0.001; odds ratio 1.107; 95% confidence interval (CI) 1.074–1.142), spleen diameter (P < 0.001; odds ratio 1.844; 95% CI 1.477–2.302), and platelet count (P < 0.001; odds ratio 0.976; 95% CI 0.964–0.989) were confirmed as independent predictors of HEVs. The AUROC were 0.886 (95% CI 0.853–0.919) for LSM, 0.885 (95% CI 0.854–0.916) for spleen diameter, and 0.809 (95% CI 0.760–0.858) for platelet count (Figure 1). Formula for prediction of HEVs and comparisons to other tests for diagnostic accuracy in the training set

On the basis of the above multivariate analysis, we derived a multiple fractional equation for prediction of HEVs that included LSM (odds ratio > 1.0) and spleen diameter (odds ratio > 1) as the numerator, and platelet count (odds ratio < 1.0) as the denominator, to amplify the effect of each factor in the progression of portal hypertension and development of HEVs. We propose a new model called LSM-spleen diameter to platelet ratio score (LSPS): VOLUME 105 | JUNE 2010 www.amjgastro.com

Variable

Training set (n =280)

Validation set (n =121)

P value

Age (years)

52.52 ± 9.12

53.88 ± 8.32

0.165

196:84

81:40

0.543

Sex (male:female) 2

BMI (kg/m )

23.39 ± 2.85

24.08 ± 2.83

0.055

LSM (kPa)

25.37 ± 12.90

25.37 ± 13.40

0.839

EVs, n (%)

124 (44.3%)

60 (49.6%)

0.781

Small

39 (13.9%)

18 (14.9%)

Medium

71 (25.4%)

35 (28.9%)

14 (5%)

7 (5.8%)

High-risk EVs

Large

90 (32.1%)

43 (35.5%)

0.508

Portal hypertensive gastropathy

34 (12.1%)

12 (9.9%)

0.521

White blood cell count (/μl)

4,753 ± 1,899

4,415 ± 1,653

0.091

Hemoglobin (g/dl)

13.28 ± 2.09

13.08 ± 1.98

0.392

Platelet count (10 /l)

117.7 ± 54.3

110.0 ± 48.8

0.181

Prothrombin time (INR)

1.15 ± 0.28

1.13 ± 0.14

0.462

9

Total bilirubin (mg/dl)

1.94 ± 0.47

1.31 ± 0.54

0.13

Albumin (g/dl)

4.05 ± 0.64

4.01 ± 0.63

0.506

AST (UI/l)

47.52 ± 21.19

45.40 ± 29.98

0.673

ALT (UI/l)

43.27 ± 25.78

41.23 ± 21.37

0.752

BUN (mg/dl)

13.43 ± 5.56

13.09 ± 4.11

0.551

Creatinine (mg/dl)

1.10 ± 0.53

0.91 ± 0.37

0.290

Spleen diameter (cm)

12.03 ± 2.62

12.33 ± 2.48

0.290

A

221 (78.9%)

103 (85.1%)

B

54 (19.3%)

16 (13.2%)

C

5 (1.8%)

2 (1.7%)

Child–Pugh classification

0.335

ALT, alanine aminotransferase; AST, aspartate aminotransferase; BMI, body mass index; BUN, blood urea nitrogen; EVs, esophageal varices; INR, international normalized ratio; LSM, liver stiffness measurement.

LSM × spleen diameter/platelet count. The AUROC of LSPS was 0.954 (95% CI 0.934–0.974), showing the superiority of diagnostic accuracy over each single factor: LSM (P = 0.001), spleen diameter (P < 0.001), and platelet count (P < 0.001). LSPS gave better AUROC results when compared with other noninvasive tests such as aspartate aminotransferase/ALT ratio (0.694, 95% CI 0.626– 0.763; P < 0.001), age–platelet index (0.616, 95% CI 0.550–0.683; P < 0.001), aspartate aminotransferase-to-platelet ratio index (0.856, 95% CI 0.829–0.883; P < 0.001), spleen-to-platelet ratio index (0.870, 95% CI 0.835–0.905; P < 0.001), and age–spleen– platelet ratio index (0.864, 95% CI 0.829–0.899; P < 0.001) from earlier studies (23–26). In subjects with Child–Pugh class A (n = 221) among the training set, AUROC of LSPS was 0.941 (95% CI 0.901–0.982), which was similarly high with that of LSPS (0.954, 95% CI 0.934–0.974) from the overall training set. Furthermore, the superior diagnos© 2010 by the American College of Gastroenterology

1385

tic efficacy of LSPS over AST/ALT ratio (0.711, 95% CI 0.644– 0.771, P < 0.001), age–platelet index (0.619, 95% CI 0.550–0.685, P < 0.001), AST-to-platelet ratio index (0.829, 95% CI 0.771–0.878, P = 0.002), spleen-to-platelet ratio index (0.865, 95% CI 0.811– 0.908, P = 0.018), and age–spleen-to-platelet ratio index (0.870, 95% CI 0.816–0.912, P = 0.027) was also maintained in subjects with Child–Pugh class A. In remaining patients with Child–Pugh class B/C (59 patients), AUROC of LSPS was 0.877 (95% CI 0.760– 0.994), which was slightly lower.

LIVER

Table 1. Baseline patients characteristics of the training and validation sets and comparisons between the two groups

A New LSM-Based Prediction Model for Esophageal Varix

Determination of LSPS cutoff value for detection of HEVs in the training set

Table 3 shows sensitivity, specificity, NPV, and PPV of the cutoff values of LSPS obtained from ROC curves, for prediction of HEVs in the training set. A cutoff value of LSPS 3.5 had an NPV of 94.0 (95% CI 89.2–96.8%), a PPV of 82.3% (95% CI 72.8–89.1%), a sensitivity of 87.7% (95% CI 78.7–93.4%), and a specificity of 91.1% (95% CI 85.8–94.5%). In the patients with LSPS < 3.5 (184 patients of 280, 65.7%), absence of HEVs was correctly identified with high accuracy (94.0% NPV). Endoscopy might be safely avoided in these low-risk patients. Similarly, when LSPS was set at 5.5, an NPV of 88.1% (95% CI 82.8–92.0%), a PPV of 94.2% (95% CI 85.0–98.1%), a sensitivity of 72.2% (95% CI 61.6–80.9%), and a specificity of 97.8% (95% CI 94.3–99.3%) were observed in 69 patients (24.6%). Patients with LSPS > 5.5 could be identified as having HEVs with high accuracy (94.2% PPV), and empirical treatments with nonselective β-blocker or endoscopy for prophylactic variceal ligation should be considered for these patients, based on the clinician’s decision. Overall, the likelihood of HEVs in cirrhotic patients was correctly diagnosed with a nonendoscopic method in 253 patients (90.3%). In addition, for NALT group from the training set, we suggested the lower cutoff of 3.3 (an NPV of 94.0% (95% CI 88.1–97.6%), a PPV of 75.0% (95% CI 62.1–85.3%), a sensitivity of 86.5% (95% CI 74.2–94.4%), and a specificity of 88.0% (95% CI 81.0–93.1%)) and the higher cutoff 5.2 (an NPV of 89.8% (95% CI 83.5–94.3%), a PPV of 95.0% (95% CI 83.1–99.2%), a sensitivity of 73.1% (95% CI 59.0–84.2%), and a specificity of 98.4% (95% CI 94.3–99.7%)). In HALT group from the training set, we also suggested the lower cutoff of 3.5 (an NPV of 94.0% (95% CI 85.0–98.3%), a PPV of 89.5% (95% CI 75.2–97.0%), a sensitivity of 89.5% (95% CI 75.2–97.0%), and a specificity of 93.9% (95% CI 85.0–97.0%)) and the higher cutoff of 5.6 (an NPV of 84.0% (95% CI 73.7–91.4%), a PPV of 93.0% (95% CI 76.5–98.9%), a sensitivity of 68.4% (95% CI 51.4–82.5%), and a specificity of 96.9% (95% CI 89.3–99.5%)) (Table 3). Validation of LSPS cutoffs derived from the training set

An AUROC in the validation set was 0.953 (95% CI 0.930–0.977). The cutoff values derived from the training set are validated (Table 4), displaying excellent agreement from the training and validation set. In the validation set, for prediction of HEVs, when applying a cutoff value of LSPS 3.5, an NPV of 94.7% (95% CI 86.3–98.3%) was achieved (76 patients, 62.8%), whereas a PPV of 93.3% (95% CI 76.5–98.8%) was obtained at a cutoff value of LSPS 5.5 (30 patients, 24.8%). Overall, 87.6% of patients were correctly diagnosed, using that formula. Therefore, based on these analyses, The American Journal of GASTROENTEROLOGY

1386

Kim et al.

Patients without high-risk EVs (n =190)

Patient with high-risk EVs (n = 90)

52.48 ± 9.32

52.60 ± 8.71

0.915



121:69

75:!5

0.001

0.072

BMI (kg/m2)

23.47 ± 2.84

23.22 ± 2.87

0.570



Variable Age (years) Sex (male:female)

Univariate analysis (P value)

Multivariate analysis (P value)

LSM (kPa)

16.56 ± 10.71

43.97 ± 21.67

< 0.001

< 0.001

White blood cell count (/μl)

4,941 ± 1867

4,361 ± 1915

0.017

0.608

Hemoglobin (g/dl)

13.82 ± 1.77

12.15 ± 2.26

< 0.001

0.157

Platelet count (109/l)

134 ± 52.9

82.2 ± 37.6

< 0.001

< 0.001

Prothrombin time (INR)

1.08 ± 0.14

1.29 ± 0.43

< 0.001

0.928

Total bilirubin (mg/dl)

1.21 ± 0.49

3.06 ± 2.61

0.001

0.277

4.26 ± 0.51

3.61 ± 0.65

< 0.001

0.402

AST (UI/l)

Albumin (g/dl)

37.26 ± 22.54

69.19 ± 40.22

< 0.001

0.835

ALT (UI/l)

36.07 ± 19.44

58.46 ± 45.11

0.070



BUN (mg/dl)

13.31 ± 4.62

13.69 ± 7.17

0.638



Creatinine (mg/dl)

1.16 ± 0.91

0.98 ± 0.74

0.481



Spleen diameter (cm)

10.93 ± 2.09

14.34 ± 2.08

< 0.001

< 0.001

ALT, alanine aminotransferase; AST, aspartate aminotransferase; BMI, body mass index; BUN, blood urea nitrogen; EVs, esophageal varices; INR, international normalized ratio; LSM, liver stiffness measurement.

1.0

Table 3. Diagnostic accuracy of LSPS for prediction of high-risk EVs in the training set and its subgroups (NALT and HALT groups) Training set

0.8

LSPS cutoff values 3.5

Sensitivity

LIVER

Table 2. Comparisons and uni/multivariate analysis of variables according to presence of high-risk EVs in the training set

0.6

0.4

LSPS

5.5

NPV (%)

94.0 (89.2–96.8)

88.1 (82.8–92.0)

PPV (%)

82.3 (72.8–89.1)

94.2 (85.0–98.1)

Sensitivity (%)

87.7 (78.7–93.4)

72.2 (61.6–80.9)

Specificity (%)

91.1 (85.8–94.5)

97.8 (94.3–99.3)

Interpretation

Absence of HEV

Presence of HEV

NALT group

3.3

5.2

LSM

0.2

Spleen diameter Platelet count

NPV (%)

94.0 (88.1–97.6)

89.8 (83.5–94.3)

PPV (%)

75.0 (62.1–85.3)

95.0 (83.1–99.2)

Sensitivity (%)

86.5 (74.2–94.4)

73.1 (59.0 -84.2)

0.0 0.0

0.2

0.4

0.6

0.8

1.0

1- Specificity Figure 1. Receiver-operating characteristic curve of LSM-spleen diameter to platelet ratio score (LSPS), liver stiffness measurement (LSM), spleen diameter, and platelet count for the diagnosis of high-risk esophageal varices in the training set.

we propose a management strategy algorithm, with reference to current guidelines, as shown in Figure 2 (1,5). In addition, in subgroup analyses for ALT adjustment, the lower cutoff of 3.3 provided an NPV of 94.0% (95% CI 83.4–98.7%), whereas the higher cutoff of 5.2 gave a PPV of 91.0% (95% CI 70.0– 98.6%) in NALT group from the validation set. Similarly, in HALT The American Journal of GASTROENTEROLOGY

Specificity (%)

88.0 (81.0–93.1)

98.4 (94.3–99.7)

Interpretation

Absence of HEV

Presence of HEV

HALT group

3.5

5.6

NPV (%)

94.0 (85.0–98.3)

84.0 (73.7–91.4)

PPV (%)

89.5 (75.2–97.0)

93.0 (76.5–98.9)

Sensitivity (%)

89.5 (75.2–97.0)

68.4 (51.4–82.5)

Specificity (%)

93.9 (85.0–97.0)

96.9 (89.3–99.5)

Interpretation

Absence of HEV

Presence of HEV

ALT, alanine aminotransferase; EVs, esophageal varices; HALT, high-ALT; HEV, high-risk EVs; LSPS, LSM-spleen diameter to platelet ratio score; NALT, normal-ALT; NPV, negative predictive value; PPV, positive predictive value. In parentheses, 95% confidence interval.

VOLUME 105 | JUNE 2010 www.amjgastro.com

Table 4. Diagnostic accuracy of LSPS for prediction of high-risk EVs in the validation set and its subgroups (NALT and HALT groups) Validation set

LSPS cutoff values 3.5

5.5

NPV (%)

94.7 (86.3 – 98.3)

83.5 (73.9 – 90.2)

PPV (%)

86.6 (72.5 – 94.4)

93.3 (76.5 – 98.8)

Sensitivity (%)

90.7 (76.9 – 96.9)

65.1 (49.0 – 78.5)

Specificity (%)

92.3 (83.4 – 96.8)

97.4 (90.2 – 99.5)

Interpretation

Absence of HEV

Presence of HEV

NALT group

3.3

5.2

NPV (%)

94.0 (83.4 – 98.7)

84.5 (72.6 – 92.6)

PPV (%)

86.2 (68.3 – 96.0)

91.0 (70.0 – 98.6)

Sensitivity (%)

89.3 (71.8 – 97.6)

67.9 (47.7 – 84.1)

Specificity (%)

92.2 (81.1 – 97.8)

96.1 (86.5 – 99.4)

Interpretation

Absence of HEV

Presence of HEV

HALT group

3.5

5.6

NPV (%)

96.0 (80.0 – 99.3)

84.4 (67.2 – 94.7)

PPV (%)

82.4 (56.6 – 96.0)

100.0 (69.0 – 100.0)

Sensitivity (%)

93.3 (68.0 – 98.9)

66.7 (38.4 – 88.1)

Specificity (%)

99.9 (70.8 – 97.5)

100.0 (87.1 – 100.0)

Interpretation

Absence of HEV

Presence of HEV

ALT, alanine aminotransferase; EVs, esophageal varices; HALT, high-ALT; HEV, high-risk EVs; LSPS, LSM-spleen diameter to platelet ratio score; NALT, normalALT; NPV, negative predictive value; PPV, positive predictive value In parentheses, 95% confidence interval.

Liver cirrhosis

LSM, spleen diameter, and routine laboratory tests: calculate LSPS

LSPS < 3.5

LSPS 3.5 ~ 5.5

LSPS > 5.5

No endoscopic screening, only periodic follow-up

Endoscopic screening according to guidelines

Perform endoscopy or start empirical β-blocker

Figure 2. Proposed management algorithm for screening and surveillance of esophageal varices in patients with cirrhosis. LSM, liver stiffness measurement; LSPS, LSM-spleen diameter to platelet ratio score.

group from the validation set, the lower cutoff of 3.5 provided an NPV of 96.0% (95% CI 80.0–99.3%), whereas the higher cutoff of 5.6 gave a PPV of 100.0% (95% CI 69.0–100.0%) (Table 4).

DISCUSSION As variceal hemorrhage remains the leading cause of morbidity and mortality for cirrhotic patients, periodic endoscopic screening for EVs and appropriate prophylactic treatments for highrisk cases of medium/large EVs or small EVS with red sign or © 2010 by the American College of Gastroenterology

decompensated cirrhosis are currently recommended for all cirrhosis patients (2,5). However, a screening strategy involving all cirrhotic patients implies a number of unnecessary endoscopies, and may be hampered by a lack of compliance, especially when periodically undertaken, because it is an invasive, expensive, and unpleasant procedure. Therefore, it is clinically relevant for physicians to have a noninvasive screening tool that allows endoscopy to be restricted to selected, high-risk patients, and to relieve the medical and financial burden of unnecessary screens. Several nonendoscopic screening tools have recently been reported, but our study has a number of unique features. First of all, this is the first study to develop a simple, LSM-based prediction model in combination with other parameters reflecting portal hypertension. In this study, we show the improved diagnostic accuracy of an LSM-based model, and assess and validate its usefulness. A few investigators have already evaluated LSM, but found that results from LSM alone were neither consistent nor reliable enough to be universally applied as an alternative (3,7,16). We conducted this study under the assumption that LSM may be theoretically correlated to EVs, because it is a phenomenon caused directly by portal hypertension and fundamentally by hepatic fibrosis, and we also assumed that complementing factors might enhance the diagnostic power (13). Second, in contrast to previous studies, the distribution of our study population was homogeneous and representative of the B-viral cirrhotic patients seen in clinical practice. Earlier studies have focused primarily on populations with chronic hepatitis C or heterogenous causes (7,9,16). Considering differences in cutoff values and efficacies of LSM according to the underlying etiology, this study may provide more generalizable results, as they are for patients with B-viral cirrhosis. Third, we prospectively enrolled a relatively large number of consecutive subjects from a single center. Fourth, we performed subgroup analyses to suggest the optimal cutoffs, adjusting ALT level, as LSM and its diagnostic performance were reported to be affected, with respect to necroinflammatory activity (21,22). Furthermore, in contrast to other studies, our primary goal was the prediction of HEVs, rather than the presence of EVs of any size, because the decision of whether and when prophylactic treatments should be initiated is an important clinical issue (5,7,11). Empirical therapy with β-blockers for all cirrhotic patients without endoscopy is not currently recommended because long-term benefit has not been established in patients with small EVs and no criteria for increased bleeding risk, not to mention, patients without EV and about 15% of subjects experienced severe adverse effects necessitating withdrawal (5). Furthermore, prophylactic endoscopic variceal ligation should now be considered only for selected patients with medium/ large EVs. In the course of this study, we were able to identify not only who should be screened endoscopically for EVs but also who should be eligible for prophylactic treatments. In our training set, 90 cases had HEVs. Significant differences in LSM, spleen diameter, and platelet count were observed among patients with or without HEVs, using multivariate analysis. These three variables are reported to be closely affected by portal hypertension (6,27,28). As cirrhotic transformation increases liver tissue stiffness through structural changes, major angio-architectural The American Journal of GASTROENTEROLOGY

1387

LIVER

A New LSM-Based Prediction Model for Esophageal Varix

LIVER

1388

Kim et al.

modifications as well as the accumulation of fibrillar extracellular matrix, portal hypertension subsequently develops (9). LSM had significant prediction power for portal hypertension and its consequence, EVs (7,9,13). Splenomegaly in cirrhotic patients is most likely the result of vascular disturbances, which is almost always the expression of greater portal pressure, whereas thrombocytopenia might be caused by either portal hypertension-induced splenic sequestration, or other mechanisms such as decreased thrombopoietin synthesis, shortened platelet mean lifetime, or myelotoxic effects of drugs or hepatitis viruses (6,29,30). Consistent with these hypotheses, LSM, spleen diameter, and platelet count had considerable diagnostic power even when applied alone, although platelet count had a relatively lower diagnostic value for prediction of EVs, because of pathogenetic causes other than portal hypertension. By integrating LSM and spleen diameter and platelet count in a single ratio, a new formula, LSPS, was derived from our analysis. LSPS has excellent diagnostic accuracy in terms of AUROC, showing superiority over each factor alone and other noninvasive tests from earlier studies (23–26). The strength of LSPS can be explained in the pathophysiological basis that supports its conception. Giannini et al. (6,11) reported that platelet count/spleen diameter ratios for prediction of EVs had a 100% NPV with AUROC 0.921 at the cutoff point of 909 in one study, and an 87% NPV with AUROC 0.860 in the other. In our study, LSPS also gave a better AUROC than the platelet count/spleen diameter ratio, which was essentially same as the spleen-to-platelet ratio index. LSPS may conceivably give improved outcomes because of the combination of LSM with the previously known two factors, as LSM itself was pathophysiologically correlated to portal hypertension and considerably accurate even when applied alone. The good diagnostic performance of LSPS was also maintained similarly in patients with Child–Pugh class A (0.941 (95% CI 0.901–0.982)), which was still better than that in patients with Child–Pugh class B/C. Hence, authors carefully conjectured that patients with Child–Pugh A class would benefit most in avoiding unnecessary endoscopy through this noninvasive tool, as patients with decompensated cirrhosis most probably have more advanced EVs. Exclusion of HEVs could be identified with a high NPV of 94.7% at a cutoff value of LSPS 3.5 in the validation set, so it can be confidently applied without significant risk of missing important diagnoses. On the basis of this value, 76 patients in the validation set (62.8%) would avoid endoscopy, and only periodic follow-up with this noninvasive tool may be sufficient for those low-risk patients. Likewise, the presence of HEVs could be predicted at a high PPV of 93.3% at a cutoff value of LSPS 5.5. Two further strategies may be suggested in this situation: either endoscopy for both diagnostic and therapeutic aims, or beginning β-blocker therapy without performing endoscopy. Patients in the border zone (3.5 < LSPS < 5.5; 15 patients, 12.4%) should undergo screening endoscopy according to the current guidelines (1,5). These results allowed us to propose a management strategy algorithm with reference to current guidelines that may help physicians make clinical decisions about diagnosis, prophylaxis, and surveillance (1,5). According to our proposal, endoscopy could be safely avoided in 62.8–87.6%, depending on physicians’ decisions. Besides, we also tried to suggest the optimal The American Journal of GASTROENTEROLOGY

cutoffs fitted in each subgroup according to ALT level, as the positive correlation between ALT and LSM was observed. LSPS 3.3/5.2 and 3.5/5.6 were suggested and validated in NALT and HALT group, respectively, with a similarly high accuracy. In addition to the excellent diagnostic value of LSPS, it had several advantages in the clinical field. From a practical point of view, it is easy to calculate at bedside or in the outpatient clinic, whereas from a technical point of view, LSM and spleen diameter can easily be assessed, and are highly reproducible with low intra- and interobserver variability (31,32). Therefore, this method had several strengths over other noninvasive tests using ultrasonographic parameters such as portal vein velocity, portal vein diameter, hepatic impedance indexes, and splenic impedance indexes (13,33). Regarding the effect of antiviral therapy in delaying the progression to liver cirrhosis and development of its complications, we observed a tendency of the lower LSM and the less frequency of EVs in patients who underwent antiviral therapy. Theoretically, the appropriate antiviral intervention against viral replications, which could be detected through periodic surveillance, is expected to delay not only the progression from simple carrier to liver cirrhosis, but also the development of the associated complications (e.g., EVs). However, to elucidate the beneficial effect of antiviral therapy to prevent hepatic decompensation, another longitudinal study in the prospective manner may be needed. This study had several limitations. First, it was cross-sectional, and the diagnostic value in prediction of subsequent development of HEVs with sequential LSPS measurement needs to be further examined in a longitudinal study. In addition, to validate the effectiveness of this noninvasive strategy, the long-term follow-up is required to determine not only cumulative bleeding event rates, especially in those at low probability for HEV, but also its cost and patient preferences. Second, if the hepatic venous pressure gradient, which is an invasive but standard method for the assessment of portal pressure and related complications, had been measured in this study, the quantitative analysis among various parameters might have been better established, although LSM has been also reported as having a good correlation with hepatic venous pressure gradient (7,9). Third, our formula cannot be applicable to patients with severe ascites, as LSM is not eligible for them. Finally, even though LSPS has an excellent NPV in excluding HEVs, it had relatively lower sensitivity, still resulting in additional endoscopies for patients who will not have HEV. In summary, a new formula, LSPS, was developed in the course of this prospective, large-scale study. The assessment of LSPS provides excellent diagnostic accuracy in prediction of HEVs. These results allowed us to safely reduce the number of unnecessary endoscopies in our patient sample. On the basis of these results, we hope that, in future studies, other researchers will evaluate the reproducibility of the LSPS for the noninvasive diagnosis of HEVs in independent populations with different clinical backgrounds. ACKNOWLEDGMENTS

We thank Eun Hee Choi, PhD, (Department of Biostatistics, Yonsei University College of Medicine, Seoul, Korea) and Hae Ryoung Song,

VOLUME 105 | JUNE 2010 www.amjgastro.com

PhD (Department of Research Affair, Yonsei University College of Medicine, Seoul, Korea) for critical comments on statistics. CONFLICT OF INTEREST

Guarantor of the article: Do Young Kim, MD. Specific author contributions: Study design, data collection, data analysis, and writing of the manuscript: Beom Kyung Kim; study design, data collection, data analysis, and writing of the manuscript: Kwang-Hyub Han; data collection and data analysis: Jun Yong Park; study design, data collection, and data analysis: Sang Hoon Ahn; data collection and writing of the manuscript: Chae Yoon Chon; data collection, data analysis, and writing of the manuscript: Ja Kyung Kim; study design, data collection, and data analysis: Yong Han Paik; data collection and data analysis: Kwan Sik Lee; study design, data collection, data analysis, and writing of the manuscript: Do Young Kim. Financial support: This study was supported by the Liver Cirrhosis Clinical Research Center, in part by a grant from the Brain Korea 21 Project for Medical Science, and by a grant form Ministry for Health, Welfare and Family Affairs, Republic of Korea (no. A050021). Potential competing interests: None.

Study Highlights WHAT IS CURRENT KNOWLEDGE

3Endoscopic screening for esophageal varices (EVs) and prophylactic treatment for high-risk EVs (HEVs) are currently recommended for all cirrhotic patients; however, the majority of subjects either do not have varices or have varices that do not require prophylactic therapy.

3If a simple, noninvasive test is available, many low-risk patients may safely avoid endoscopy. 3Liver stiffness measurement (LSM), although potentially

useful in noninvasive screening for EVs, still requires complementary methods using other parameters, for more reliablitiy.

WHAT IS NEW HERE

3An LSM-based new formula, LSM-spleen diameter to platelet ratio score (LSPS), consisting of LSM, spleen diameter, and platelet count, showed excellent diagnostic accuracy for prediction of HEVs.

3At LSPS < 3.5, 94.0% negative predictive value was

provided, whereas 94.2% positive predictive value was achieved at LSPS > 5.5 for prediction of HEVs and the overall likelihood of HEVs was correctly diagnosed in 90.3% of population.

3On the basis of these, LSPS can be proposed as a reli-

able noninvaise screening tool for HEVs, and patients with LSPS < 3.5 may avoid endoscopy safely, whereas those with LSPS > 5.5 should be considered for appropriate prophylactic treatments.

REFERENCES 1. de Franchis R. Non-invasive (and minimally invasive) diagnosis of oesophageal varices. J Hepatol 2008;49:520–7. 2. de Franchis R. Evolving consensus in portal hypertension. Report of the Baveno IV consensus workshop on methodology of diagnosis and therapy in portal hypertension. J Hepatol 2005;43:167–76.

© 2010 by the American College of Gastroenterology

3. Bosch J, Berzigotti A, Garcia-Pagan JC et al. The management of portal hypertension: rational basis, available treatments and future options. J Hepatol 2008;48 (Suppl 1): S68–92. 4. Merli M, Nicolini G, Angeloni S et al. Incidence and natural history of small esophageal varices in cirrhotic patients. J Hepatol 2003;38:266–72. 5. Garcia-Tsao G, Sanyal AJ, Grace ND et al. Prevention and management of gastroesophageal varices and variceal hemorrhage in cirrhosis. Hepatology 2007;46:922–38. 6. Giannini E, Botta F, Borro P et al. Platelet count/spleen diameter ratio: proposal and validation of a non-invasive parameter to predict the presence of oesophageal varices in patients with liver cirrhosis. Gut 2003;52:1200–5. 7. Kazemi F, Kettaneh A, N’Kontchou G et al. Liver stiffness measurement selects patients with cirrhosis at risk of bearing large oesophageal varices. J Hepatol 2006;45:230–5. 8. Gentile I, Viola C, Graf M et al. A simple noninvasive score predicts gastroesophageal varices in patients with chronic viral hepatitis. J Clin Gastroenterol 2008; e-pub ahead of print. 9. Vizzutti F, Arena U, Romanelli RG et al. Liver stiffness measurement predicts severe portal hypertension in patients with HCV-related cirrhosis. Hepatology 2007;45:1290–7. 10. Berzigotti A, Gilabert R, Abraldes JG et al. Noninvasive prediction of clinically significant portal hypertension and esophageal varices in patients with compensated liver cirrhosis. Am J Gastroenterol 2008;103:1159–67. 11. Giannini EG, Zaman A, Kreil A et al. Platelet count/spleen diameter ratio for the noninvasive diagnosis of esophageal varices: results of a multicenter, prospective, validation study. Am J Gastroenterol 2006;101:2511–9. 12. Kovacevic N, Alempijevic T. Right liver lobe diameter: albumin ratio: a new non-invasive parameter for prediction of oesophageal varices in patients with liver cirrhosis (preliminary report). Gut 2007;56:1166–7; authro reply 1167. 13. Rockey DC. Noninvasive assessment of liver fibrosis and portal hypertension with transient elastography. Gastroenterology 2008;134:8–14. 14. Castéra L, Vergniol J, Foucher J et al. Prospective comparison of transient elastography, Fibrotest, APRI, and liver biopsy for the assessment of fibrosis in chronic hepatitis C. Gastroenterology 2005;128:343–50. 15. Lucidarme D, Foucher J, Le Bail B et al. Factors of accuracy of transient elastography (fibroscan) for the diagnosis of liver fibrosis in chronic hepatitis C. Hepatology 2009;49:1083–9. 16. Castéra L, Le Bail B, Roudot-Thoraval F et al. Early detection in routine clinical practice of cirrhosis and oesophageal varices in chronic hepatitis C: comparison of transient elastography (FibroScan) with standard laboratory tests and non-invasive scores. J Hepatol 2009;50:59–68. 17. Marcellin P, Ziol M, Bedossa P et al. Non-invasive assessment of liver fibrosis by stiffness measurement in patients with chronic hepatitis B. Liver Int 2009;29:242–7. 18. Dittrich M, Milde S, Dinkel E et al. Sonographic biometry of liver and spleen size in childhood. Pediatr Radiol 1983;13:206–11. 19. Ziol M, Handra-Luca A, Kettaneh A et al. Noninvasive assessment of liver fibrosis by measurement of stiffness in patients with chronic hepatitis C. Hepatology 2005;41:48–54. 20. Hanley JA, McNeil BJ. A method of comparing the areas under receiver operating characteristic curves derived from the same cases. Radiology 1983;148:839–43. 21. Chan HL, Wong GL, Choi PC et al. Alanine aminotransferase-based algorithms of liver stiffness measurement by transient elastography (Fibroscan) for liver fibrosis in chronic hepatitis B. J Viral Hepat 2009;16:36–44. 22. Coco B, Oliveri F, Maina AM et al. Transient elastography: a new surrogate marker of liver fibrosis influenced by major changes of transaminases. J Viral Hepat 2007;14:360–9. 23. Kim BK, Kim SA, Park YN et al. Noninvasive models to predict liver cirrhosis in patients with chronic hepatitis B. Liver Int 2007;27:969–76. 24. Wai CT, Greenson JK, Fontana RJ et al. A simple noninvasive index can predict both significant fibrosis and cirrhosis in patients with chronic hepatitis C. Hepatology 2003;38:518–26. 25. Sheth SG, Flamm SL, Gordon FD et al. AST/ALT ratio predicts cirrhosis in patients with chronic hepatitis C virus infection. Am J Gastroenterol 1998;93:44–8. 26. Poynard T, Bedossa P. Age and platelet count: a simple index for predicting the presence of histological lesions in patients with antibodies to hepatitis C virus. METAVIR and CLINIVIR Cooperative Study Groups. J Viral Hepat 1997;4:199–208. 27. Bosch J, Garcia-Pagán JC, Berzigotti A et al. Measurement of portal pressure and its role in the management of chronic liver disease. Semin Liver Dis 2006;26:348–62.

The American Journal of GASTROENTEROLOGY

1389

LIVER

A New LSM-Based Prediction Model for Esophageal Varix

1390

Kim et al.

LIVER

28. Lim JK, Groszmann RJ. Transient elastography for diagnosis of portal hypertension in liver cirrhosis: is there still a role for hepatic venous pressure gradient measurement? Hepatology 2007;45:1087–90. 29. Giannini EG. Review article: thrombocytopenia in chronic liver disease and pharmacologic treatment options. Aliment Pharmacol Ther 2006;23:1055–65. 30. Poynard T, Trabut JB, Ratziu V et al. Prediction of oesophageal varices with platelet count/spleen diameter ratio or platelets alone. Gut 2004;53:913–4; author reply 914.

The American Journal of GASTROENTEROLOGY

31. O’Donohue J, Ng C, Catnach S et al. Diagnostic value of Doppler assessment of the hepatic and portal vessels and ultrasound of the spleen in liver disease. Eur J Gastroenterol Hepatol 2004;16:147–55. 32. Sandrin L, Fourquet B, Hasquenoph JM et al. Transient elastography: a new noninvasive method for assessment of hepatic fibrosis. Ultrasound Med Biol 2003;29:1705–13. 33. Liu CH, Lin JW, Tsai FC et al. Noninvasive tests for the prediction of significant hepatic fibrosis in hepatitis C virus carriers with persistently normal alanine aminotransferases. Liver Int 2006;26:1087–94.

VOLUME 105 | JUNE 2010 www.amjgastro.com