Potential conflict of interest: Nothing to report.
Liver Failure/Cirrhosis/Portal Hypertension
Platelet count is not a predictor of the presence or development of gastroesophageal varices in cirrhosis†
Article first published online: 27 DEC 2007
Copyright © 2007 American Association for the Study of Liver Diseases
Volume 47, Issue 1, pages 153–159, January 2008
How to Cite
Qamar, A. A., Grace, N. D., Groszmann, R. J., Garcia-Tsao, G., Bosch, J., Burroughs, A. K., Maurer, R., Planas, R., Escorsell, A., Garcia-Pagan, J. C., Patch, D., Matloff, D. S. and Makuch, R. (2008), Platelet count is not a predictor of the presence or development of gastroesophageal varices in cirrhosis. Hepatology, 47: 153–159. doi: 10.1002/hep.21941
- Issue published online: 27 DEC 2007
- Article first published online: 27 DEC 2007
- Manuscript Accepted: 23 JUL 2007
- Manuscript Received: 4 APR 2007
- National Institute of Diabetes and Digestive and Kidney Diseases. Grant Number: RO1 46580
Current guidelines recommend esophagogastroduodenoscopy (EGD) in patients with cirrhosis to screen for gastroesophageal varices (GEV). Thrombocytopenia has been proposed as a noninvasive test to predict the presence of GEV. There is no agreement regarding a specific platelet count (PLT) that can reliably predict GEV. The present longitudinal study aims to (1) further investigate the relationship between varices and PLT at the time of endoscopy, (2) investigate whether changes in PLT from the baseline over time can predict the development of GEV, and (3) investigate whether changes in PLT correlate with the hepatic venous pressure gradient (HVPG). A secondary analysis was conducted for 213 subjects with compensated cirrhosis with portal hypertension but without GEV enrolled in a randomized, placebo-controlled, double-blind trial of a nonselective beta-blocker used to prevent GEV. PLTs were obtained every 3 months, and HVPG measurements and EGD were done annually. The PLTs were compared between subjects who did and did not develop GEV. In a median follow-up of 54.9 months, 84 patients developed GEV. PLT was greater than 150,000 in 15% of patients at the development of GEV. A receiver operating curve did not show any PLT with high sensitivity or specificity for the presence of GEV. Subjects with clinically insignificant portal hypertension (HVPG < 10 mm Hg) whose PLT remained greater than 100,000 had a 2-fold reduction in the occurrence of GEV (P = 0.0374). A significant correlation was found between HVPG and PLT at the baseline, year 1, and year 5 (P < 0.0001). Conclusion: Cross-sectional or longitudinal evaluations of PLTs are inadequate noninvasive markers for GEV. Patients with mild portal hypertension whose PLT remains greater than 100,000 have significantly less risk of GEV. Although HVPG correlates somewhat with PLT, changes in PLT cannot be used as a surrogate for HVPG changes. (HEPATOLOGY 2008;47:153–159.)
Variceal hemorrhage is a leading cause of morbidity and mortality in cirrhosis.1–3 Primary prophylaxis with nonselective beta-blockers and endoscopic therapy with band ligation are effective in preventing variceal hemorrhage in patients with large varices.1 Current guidelines recommend screening for gastroesophageal varices (GEV) with esophagogastroduodenoscopy (EGD) in all patients with cirrhosis and starting prophylactic therapy in those with medium to large varices.4–7 In patients without varices, EGD is repeated in 2–3 years, whereas in patients with small varices, the recommendation is to repeat EGD in 1–2 years.2 Because of the cost and invasive nature of endoscopic screening, there is interest in developing a noninvasive predictor of the presence and development of GEV that would decrease the number of EGDs performed. The ideal noninvasive marker should be widely available at a reasonable cost, require minimal expertise, be reproducible, and be relatively unaffected by any other factors. A number of clinical, laboratory, and ultrasonographic variables have been considered.8 However, the accuracy of these variables in predicting varices is currently inadequate to be recommended for clinical practice.2
As varices are a direct consequence of portal hypertension, it is not surprising that the degree of portal hypertension determined by the hepatic venous pressure gradient (HVPG) has been found to be predictive of the development of varices. Specifically, an HVPG greater than or equal to 10 mm Hg and the 1-year change in HVPG have been shown to be strong predictors for the development of GEV.9 This cutoff HVPG has been identified as clinically significant portal hypertension.10
Thrombocytopenia is another complication that is the result of portal hypertension. It is present in up to 64% of patients,11 and it is mainly caused by splenic sequestration as a complication of portal hypertension–induced splenomegaly (that is, hypersplenism), although other factors present in cirrhosis such as decreased thrombopoeitin and interleukin-11 and the suppressive effects of alcohol and viruses may also play a role.12, 13 Numerous cross-sectional studies have shown a wide range of thrombocytopenias to be associated with the presence of GEV. Platelet counts (PLT) from 68,000 to 160,000 have shown sensitivities ranging from 71% to 88% for small varices.14–18 The sensitivities for large varices are better but with unsatisfactory specificities. However, many of these studies were cross-sectional and were limited by the variability and severity of the liver disease.
The aim of this nested cohort study was to evaluate, in a homogeneous population of patients with cirrhosis and portal hypertension, the usefulness of PLT and changes in PLT over time in predicting not only the presence of varices but also the development of varices and in correlating them to changes in HVPG over time. The possibility of creating a predictive model for GEV using PLT with other potentially predictive variables associated with cirrhosis was also evaluated.
Patients and Methods
The study was a nested cohort study in the setting of an investigator-initiated, prospective, randomized, double-blind, placebo-controlled, clinical trial designed to evaluate the efficacy of nonselective beta-blockers in preventing GEV and the usefulness of measuring HVPG sequentially. The complete description of the methodology has been published elsewhere.9 The protocol for conducting secondary analysis was approved by the institutional review board.
The patients were enrolled between August 1993 and March 1999 and followed until September 2002. Eligible patients had cirrhosis and portal hypertension as defined by an HVPG of at least 6 mm Hg, did not have GEV, and were older than 18 years of age and less than 75 years of age. Exclusion criteria included ascites requiring diuretics, hepatocellular carcinoma, splenic or portal vein thrombosis, concurrent illness expected to decrease life expectancy to less than 1 year, the use of any drug or procedure affecting the splanchnic hemodynamic or portal pressure, primary biliary cirrhosis or primary sclerosing cholangitis, contraindications to beta blocker therapy, pregnancy, and alcohol intake during the dose titration phase.
A total of 213 patients were enrolled into the study. Full details have been previously published.9
The patients were assessed at the baseline, 1 month after randomization, 3 months after randomization, and every 3 months thereafter. At each visit, the heart rate and alcohol consumption were determined, and blood was obtained for hematological and biochemical measurements. At the baseline and every year thereafter, EGD and HVPG were done as described elsewhere.19 Abdominal ultrasonography was used to evaluate the spleen and portal vein. The primary endpoint was the development of varices (any size) or variceal hemorrhage. The follow-up ended at the time of development of a primary endpoint; otherwise, the patients were followed to the termination of the study in September 2002.
PLTs and GEV.
The PLTs among the subjects that did and did not develop GEV were compared. A receiver operating curve (ROC) was used to determine the sensitivity and specificity of PLT at the time of development of GEV. A longitudinal analysis of PLT was then performed to determine if a specific rate of change could predict GEV. The baseline PLT was compared to the annual PLT at year 1 and onward into the study between the subjects that did or did not develop GEV. PLT was then analyzed to determine if a particular threshold PLT could be identified during the longitudinal follow-up that could serve as a predictor of patients with a higher risk of developing GEV. This was also compared among the subjects with clinically insignificant portal hypertension (HVPG < 10 mm Hg) and clinically significant portal hypertension (HVPG ≥ 10 mm Hg) at the baseline. The relationship between PLT and GEV was analyzed according to the treatment status (timolol or placebo). PLT was also compared to HVPG to determine if a correlation existed at the baseline and then annually. Changes in HVPG were analyzed to determine if it was associated with a change in PLT. Conversely, changes in PLT were analyzed to determine if it was associated with a change in HVPG.
Variables potentially predictive of cirrhosis, such as HVPG, ascites, albumin, aspartate aminotransferase, alanine aminotransferase, the international normalized ratio, and the Child-Pugh score, were evaluated to determine if they could be combined with PLT to create a predictive model for varices.
Longitudinal changes in PLT were analyzed with a repeated-measure analysis of variance (ANOVA). The correlation between PLTs and HVPG at the baseline, year 1, and year 5 was analyzed with Spearman's correlation coefficient. A Student t test and a Wilcoxon test were used for numerical values. The ROC, area under the curve, and its 95% confidence interval were used to examine the relationship between PLT and GEV. Univariate and multivariate analyses for variables associated with cirrhosis were evaluated for the possibility of creating a predictor model for GEV.
PLT as a Predictor of the Presence of GEV.
The median PLT at the time of occurrence of GEV in the current cohort was 91,000 (interquartile range: 65,000–123,000). A proportion of the subjects had a normal PLT at the time of development of GEV (Fig. 1). Fifteen percent of the subjects who developed GEV had a PLT greater than 150,000 at the time of diagnosis (150,000–200,000 = 9%, >200,000 = 6%). Similarly, 14% of the subjects that developed large varices or variceal hemorrhage had a PLT greater than 150,000 (150,000–200,000 = 7%, >200,000 = 7%). With ROC analysis, no PLT cutoff could be identified that accurately predicted the presence of GEV (area under the curve = 0.630, 95% confidence interval = 0.554–0.706; Fig. 2).
Change in PLT as a Predictor of GEV.
A repeated-measure ANOVA showed that PLT significantly decreased over time in all study subjects (P = 0.0029). ANOVA showed that PLT significantly decreased over time in all subjects despite a treatment with a placebo or timolol (P = 0.0033). As shown in Fig. 3, PLT was significantly lower in the GEV group than that of the non-GEV group throughout the study (P = 0.0023). PLT also remained significantly lower in the GEV group when adjusted for the treatment with a placebo or timolol (P = 0.0024). At the end of the first year of the study, the subjects who developed GEV had a lower mean PLT (PLT in the GEV patients was 104,000 versus 120,000 in the non-GEV patients; Fig. 3). This difference remained significant during follow-up (at year 4, PLT in the GEV patients was 97,000 versus 120,000 in the non-GEV patients; at year 7, PLT in the GEV patients was 84,000 versus 117,000 in the non-GEV patients). The rate of change in PLT between the GEV group and non-GEV group was not significantly different for the full duration of the study (P = 0.3675). However, there was a significant difference in the rate of change in PLT between the GEV and non-GEV groups between the baseline and years 2 and 3. The rate of change in PLT from the baseline to year 2 in the patients who developed varices was −6.3% per year; it was −2.8% in the patients who did not develop varices (P = 0.0149). Although the change in PLT from the baseline to year 3 in the patients who developed varices was −5.1% per year, it was −3.2% in the patients who did not develop varices (P = 0.0315).
Longitudinal Follow-Up of PLT as a Predictor of GEV.
Serial PLTs were analyzed to determine if the longitudinal follow-up increased the clinical accuracy for predicting the development of GEV. In the current cohort, the median PLT at the time of occurrence of GEV was 91,000. A PLT of 100,000 was used as a cutoff, and the subjects in whom PLT remained greater than 100,000 throughout the study were compared to those subjects in whom this did not occur. There was less occurrence of GEV among the subjects in whom PLT remained greater than 100,000 (Table 1) compared with those with any PLT less than 100,000, but this did not achieve statistical significance (P = 0.3903). An adjustment for the treatment with timolol or a placebo did not affect this conclusion (P = 0.3919). Interestingly, the subjects with clinically insignificant portal hypertension at the baseline (HVPG < 10 mm Hg) for whom PLT remained greater than 100,000 had significantly less occurrence of GEV than the subjects for whom any PLT was less than 100,000 [19.4% (7/36) for PLT > 100,000 versus 41.5% (17/41) for PLT < 100,000, P = 0.0374]. None of the patients with mild portal hypertension with a PLT greater than 100,000 developed large varices/variceal hemorrhage (LV/VH) versus 29% of the patients with a PLT less than 100,000. There was no difference in the occurrence of GEV between the subjects with clinically significant portal hypertension (HVPG ≥ 10 mm Hg) for whom PLT did or did not remain greater than 100,000. There was no difference in the occurrence of LV/VH between the two groups with severe portal hypertension.
|Number||Occurrence of GEV||SV||LV||VH|
|All subjects (P = 0.3903)|
|PLT > 100,000||66||20 (30.3%)||18 (90%)||1 (5%)||1 (5%)|
|PLT < 100,000||147||64 (43.5%)||52 (81%)||7 (11%)||5 (8%)|
|Clinically significant portal hypertension (P = 0.998)|
|PLT > 100,000||30||13 (43.3%)||11 (85%)||1 (8%)||1 (8%)|
|PLT < 100,000||106||47 (44.3%)||40 (85%)||3 (6%)||4 (9%)|
|Clinically insignificant portal hypertension (P = 0.0374)|
|PLT > 100,000||36||7 (19.4%)||7 (100%)||0 (0%)||0 (0%)|
|PLT < 100,000||41||17 (41.5%)||12 (71%)||4 (24%)||1 (6%)|
Consideration of PLT with Other Potentially Predictive Variables of Cirrhosis To Create a Predictor Model for GEV.
The creation of a predictor model combining PLT with potentially predictive variables of cirrhosis for the occurrence of GEV was investigated. In a univariate analysis, six variables were significant: ascites (P = 0.0008), albumin (P = 0.0482), aspartate aminotransferase (P = 0.0599), alanine aminotransferase (P = 0.0921), the international normalized ratio (P = 0.0181), and the Child-Pugh score (P = 0.0689). These were included in a model-building process as potential predictors. A multivariate analysis found that only HVPG status (≥10 mm Hg versus <10 mm Hg) and ascites were predictors of varices (P = 0.0005).
Correlation Between HVPG and PLT.
There was a significant correlation between HVPG and PLT at the baseline with a Spearman correlation curve (r = − 0.44, n = 213, P < 0.0001; Fig. 4). This correlation remained at year 1 and year 5 in the study (year 1, r = −0.53, n = 154, P < 0.0001; year, 5 r = −0.57, n = 44, P < 0.0001). The correlation between HVPG and PLT was present in the patients with both alcoholic and nonalcoholic cirrhosis.
Association of Changes in HVPG and PLT.
A ≥10% increase in PLT between the baseline and year 1 was not associated with any change in HVPG (−3.3%) in comparison with the subjects with a <10% increase in PLT (−3.6%; P = not significant). A Wilcoxon test showed no difference in the change in PLT among the subjects with a 10% decrease in HVPG (n = 69) between the baseline and first year in comparison with all others (n = 85; −0.8% versus −3.6%, P = not significant; Tables 2 and 3). Even in the subjects with severe portal hypertension (HVPG ≥ 10 mm Hg), no difference was observed in the change in PLT among the subjects with a 10% decrease in HVPG (n = 44) between the baseline and first year in comparison with all others (n = 47; −3.5% versus −8.6%, P = not significant). There continued to be no effect on PLT when there was a 10% decrease in HVPG at year 1 among the subjects with alcoholic or nonalcoholic cirrhosis.
|PLT||Number||Percentage Change in HVPG||Range|
|≥10% increase||32||−3.3 ± 31.5||−71.1 to 81.3|
|<10% increase||122||−3.6 ± 27.6||−57.1 to 133.3|
|HVPG||Number||Percentage Change in PLT||Range|
|≥10% decrease||69||−0.8 ± 23.7||−55.8 to 116.8|
|<10% decrease||85||−3.6 ± 29.6||−73 to 142.6|
Despite recent advances in prevention and management, variceal hemorrhage remains a serious complication of cirrhosis.2 Because there is an effective treatment to prevent variceal hemorrhage, practice guidelines have recommended screening patients with cirrhosis for varices with EGD.4–7 However, the adherence of practicing gastroenterologists to these guidelines is unsatisfactory, as shown by two recent studies.20, 21 There has been considerable interest in developing noninvasive markers of GEV to decrease the burden of EGD and provide a diagnostic indicator for clinicians with respect to the selection and timing for screening EGD. A number of different variables have been investigated, including the PLT, spleen size, portal vein diameter, prothrombin time, and ascites. D'Amico and Morabito8 summarized these studies, and an ROC showed that none of these markers were reliable for the presence of GEV. In a recent multicenter, cross-sectional study conducted by Giannini et al.22 using a novel PLT to spleen diameter ratio cutoff of 909, the diagnostic accuracy for GEV was 86% with a negative predictive value of 87%. The accuracy was maintained when the severity and etiology of the disease subgroups were considered.
Cross-sectional studies have shown that PLT cutoffs from 68,000 to 160,000 have sensitivities and specificities ranging from 71% to 90% and from 36% to 73%, respectively.14–18 Three of the four studies with higher sensitivities had poor specificity. In a study by Madhotra et al.,17 in which the sensitivity and specificity for a PLT of 68,000 were 71% and 73%, respectively, a significant proportion of the patients had Child C cirrhosis. In fact, most studies of noninvasive markers are limited by their heterogeneous populations and by their cross-sectional evaluation. A recent study by Sanyal et al.18 addressed this issue in a homogeneous cohort of 1016 patients, of whom 418 had cirrhosis. Among the 418 patients with cirrhosis, a cross-sectional evaluation of PLT using a cutoff of <150,000 had a negative predictive value of 96.2% and a positive predictive value of 15.1% for medium or large varices. The current study evaluates whether the longitudinal follow-up of PLT and cross-sectional data are effective as noninvasive markers of GEV in patients with compensated cirrhosis. A secondary data analysis was conducted for a homogeneous cohort of well-compensated patients, with approximately 90% having Child A cirrhosis and the remaining 10% having Child B disease. Arguably, this group would benefit the most from endoscopic screening. Seventy-seven percent of the subjects in this study had thrombocytopenia at the baseline. At year 5, a similar percentage of the patients had thrombocytopenia (77%). This is slightly higher than the value of 64% reported in a previous study.11 As previously shown, with an ROC, no absolute PLT could reliably predict the occurrence of GEV. Furthermore, 15% of the subjects who developed GEV had a normal PLT. Among this subgroup, 2 (15%) patients presented with variceal hemorrhage. This suggests that recommendations using PLT as a noninvasive marker would miss a proportion of patients with varices with devastating consequences.
In the current study, the longitudinal follow-up showed a lower mean annual PLT among the subjects who developed GEV compared with those that did not throughout the study (P = 0.0023). An adjustment for the treatment with a placebo or timolol did not affect this finding. A significant rate of change between the GEV and non-GEV groups was not statistically observed for the entire study. However, a significant rate of reduction in PLT in the GEV group compared with the non-GEV groups occurred from the baseline to years 2 and 3 of the study (year 2, P = 0.0149; year 3, P = 0.0315) that was consistent with the negative PLT slope. Although it has been known that patients with GEV have lower PLT, this dynamic reduction in PLT has not been reported before. This novel finding suggests that patients with GEV have a greater rate of reduction in PLT compared with patients who do not develop GEV.
As the subjects with GEV had a lower mean annual PLT with a greater rate of reduction, it was hypothesized that a longitudinal evaluation of PLT would be a better predictor of the development of GEV in a homogeneous cohort of Child A and B patients. The median PLT at the time of occurrence of GEV in this study was 91,000. As shown in Fig. 3, the subjects who did not develop GEV had a mean annual PLT greater than 100,000 throughout the study. Because the ROC could identify a particular PLT for the detection of GEV, a PLT of 100,000 was chosen. With a PLT of 100,000 used as a cutoff, each subject's PLT was followed longitudinally throughout the study, during which values were obtained every 3 months. The analysis suggested less development of GEV among subjects whose PLT remained greater than 100,000, but this did not achieve statistical significance (P = 0.3903). The sensitivity and specificity were poor. The findings were not affected by the treatment with a placebo or timolol.
A multivariate analysis created a predictive model for GEV using an HVPG greater than or equal to 10 mm Hg and ascites. This suggests that the occurrence of ascites in patients with mild portal hypertension would indicate the occurrence of GEV. However, the occurrence of ascites is an indicator of decompensated cirrhosis, and GEV screening guidelines differ in this population.4–7 This limits the use of ascites as a predictor of GEV in compensated cirrhosis. Interestingly, the patients with clinically insignificant portal hypertension (HVPG < 10 mm Hg) for whom at the baseline PLT remained greater than 100,000 throughout the study had less development of GEV than the subjects with any PLT less than 100,000, and this achieved statistical significance (P = 0.0374). Although the sensitivity and specificity were not ideal, the negative predictive value was 81%. Interestingly, none of the patients with clinically insignificant portal hypertension and a PLT greater than 100,000 developed LV/VH. In contrast, the subjects with clinically insignificant portal hypertension and any PLT less than 100,000 during the study had an occurrence of all GEV and LV/VH similar to that of the subjects with clinically significant portal hypertension. This shows that patients with clinically insignificant portal hypertension whose PLT remains greater than 100,000 have significantly less risk for GEV and particularly LV/VH. Overall, these data suggest that patients whose PLT remains above 100,000 have less risk of developing GEV, particularly in clinically insignificant portal hypertension. By the use of a threshold PLT greater than 100,000, there could be a reduction in the number of unnecessary EGDs and the associated cost. These data also provide further support for obtaining HVPG measurements in cirrhosis. Recent studies have shown that patients with HVPG greater than or equal to 10 mm Hg have a higher risk of clinical decompensation and hepatocellular carcinoma.23, 24 Obtaining HVPG measurements at the baseline may allow better risk stratification of patients with cirrhosis. It may also allow better utilization of resources for the care of complicated patients with cirrhosis.
Portal hypertension is believed to play a significant role in thrombocytopenia in cirrhosis. However, there is little confirmatory evidence to support this. A cross-sectional study by Dittrich et al.25 showed a weak correlation between PLT and HVPG. The study was limited by the varying severity of the liver disease and the diverse group of etiologies. The current study, looking at well-compensated patients with Child A/B cirrhosis, shows a significant inverse correlation between HVPG and PLT. This correlation is similar from the baseline up to 5 years into the study and among subjects with either alcoholic or nonalcoholic liver disease. However, the correlation also suggests that other factors influence PLT in addition to portal hypertension, such as growth factor effects (thrombopoeitin and interleukin-11) and suppressive effects (alcohol-related and virus-related) on bone marrow platelet production. Changes in PLT were analyzed to determine whether they could be used as surrogate markers for HVPG changes. In the current study, no significant change in HVPG occurred when PLT increased by 10% at year 1. This remained unchanged when the etiology and severity of the baseline portal hypertension were taken into consideration. A 10% decrease in HVPG was not associated with an increase in PLT. This finding is consistent with the concept that thrombocytopenia in cirrhosis is indeed related to portal hypertension. However, PLT cannot be used as a surrogate marker for HVPG or changes in HVPG.
In conclusion, a cross-sectional or longitudinal evaluation of PLT is not an adequate noninvasive marker for GEV, and the current guidelines for endoscopic screening for GEV should remain. Patients with mild portal hypertension whose PLT remains greater than 100,000 have significantly less risk of GEV, particularly large varices and variceal hemorrhage. PLT is not useful as a surrogate for HVPG measurements in the evaluation and treatment of patients with compensated cirrhosis. This study supports a role for obtaining baseline HVPG measurements in the management of cirrhosis.
- 7www.omgc.org/globalguidelines/guide08.html. Accessed on May 2003.. WGO-OMGE practice guideline: treatment of esophageal varices. Available at:
- 10Predictive models in portal hypertension. In: de FranchisR, ed. Portal Hypertension IV: Proceedings of the Fourth Baveno International Consensus Workshop on Methodology of Diagnosis and Treatment. Oxford, UK: Blackwell Publishing; 2006: 47–102., , , , , , et al.
- 11Prevalence of peripheral blood cytopenias (hypersplenism) in patients with nonalcoholic chronic liver disease. Am J Gastroenterol 2000; 95: 2936–2939., , .Direct Link:
- 13Thrombocytopenia in liver disease. Can J Gastroenterol 2000; 14(Suppl D): D60–D66..
- 16Factors predicting the presence of esophageal or gastric varices in patients with advanced liver disease. Am J Gastroenterol 1999; 94: 3292–3296., , , , .Direct Link:
- 20The use of screening and preventive therapies for gastroesophageal varices in patients referred for evaluation of orthotopic liver transplantation. Am J Gastroenterol 2001; 96: 833–837., , , .Direct Link:
- 21Changing compliance to the American College of Gastroenterology guidelines for the management of variceal hemorrhage: a regional survey. Am J Gastroenterol 2004; 99: 645–649., , , , .Direct Link: