Introduction

The endogenous coagulopathy in hospitalized cirrhosis patients is often considered to be protective against pulmonary thromboembolism (PE) and deep vein thrombosis (DVT) despite the lack of empirical data to validate this presumed low level of risk. A recent retrospective study examined 113 hospitalized cirrhosis patients with newly diagnosed PE/DVT and compared the results with those from an equal number of matched cirrhotic controls [1]. The findings of this study included a 0.5% incidence of newly diagnosed PE/DVT in cirrhotic patients and low serum albumin as an independent predictor of PE/DVT risk. Orthopedic or major surgery patients and those with underlying malignancies are examples of patient subsets with a known increased risk of DVT/PE, and the incidence rate of PE/DVT in these groups of hospitalized patients has been reported to be between 4 and 12% [2]. This reported incidence was significantly higher than that reported by Northup et al. for their cirrhotic patients [1]. The pathogenesis of DVT/PE in cirrhotics is complex and can involve several of the following: concurrent co-morbidities which may precipitate a hypercoagulable state, limited physical activity due to the underlying chronic liver disease symptoms of massive ascites and/or lower extremity swelling [3], and acquired deficiencies in anti-coagulation factors [4]. These factors should be accounted for to fully evaluate the true risk of DVT/PE in these patients.

Given the lack of data in this area, we conducted the retrospective study reported here with the aim of (1) evaluating the incidence of DVT/PE in cirrhotic patients compared to well-matched controls without selected comorbidities who were hospitalized for DVT/PE and (2) assessing the incidence of DVT/PE in cirrhotic patients compared to matched controls with other co-morbidities that were known to increase the risk of DVT/PE (congestive heart failure (CHF), chronic kidney disease (CKD), and malignancies).

Methods

The study was reviewed and approved by the Institutional Review Board at the Indiana University School of Medicine and its affiliated hospitals. Data were collected and analyzed from the Regenstrief Medical Record system (RMRS). The system links five healthcare systems, including three major teaching hospitals on the Indiana University Medical Center campus (Indiana University Hospital, Wishard Memorial Hospital, and Riley Hospital for Children), 11 acute care hospitals, 13 homeless care sites, nearly 100 clinics/offices, and the county health departments in the state of Indiana. Data for more than 2 million patients, consisting of 300 million coded observations, 5.8 million text reports (including medical history, admission, and discharge summary, pathology reports, subspecialty clinic visit, and laboratory test results), 300,000 electrocardiogram tracings, in- and out-patient pharmacy prescription data, and 700,000 images, have been captured using a unique patient identifier [5]. A detailed description of this unique medical record system has been published elsewhere [6].

Patient Selections

This case–control study evaluated patients hospitalized in general medicine from January 1, 1995 to December 31, 2005 using the RMRS. Cases were identified by the appropriate international classification of diseases-9 (ICD-9) codes for cirrhosis (571.2, 571.5, 571.6, 571.8, 275.0, 275.1, 777.8), which had been diagnosed prior to or during the hospitalization. For confirmation of the cirrhosis diagnosis, cases were also required to have previous histories or clinical presentations of at least one of the followings: gastric/esophageal variceal bleed, presence of gastric/esophageal varices from the endoscopy report, hepatic encephalopathy, spontaneous bacterial peritonitis, liver biopsy proven cirrhosis, or imaging consistent with cirrhotic liver changes. The specific program within the RMRS system was used to link the subjects with the ICD-9 code for cirrhosis to the respective text reports of each particular individual during the patient data capturing process. To ensure that we included all potential subjects into the analyses, we also cross-referenced each patient to the ICD-9 and current procedural terminology codes (CPT) of the following diagnoses and/or procedures: variceal bleeding (571.5, 572.3), gastric varices (456.8), esophageal varices (456.1, 456.0, 571.5, 572.3), hepatic encephalopathy (572.2), spontaneous bacterial peritonitis (567.23), and liver biopsy (50.19, 50.11, 50.12) positive for cirrhosis. Another uniqueness of this software is that each subject is not counted twice. We subsequently linked the potential subjects to the in- and out-patient pharmacy prescription data. Those taking warfarin or enoxaparin at the time of admission were excluded. We next identified cases among cirrhotic patients who developed DVT/PE. The ICD-9 codes for PE (415.9, 415.11) and DVT (453.8, 451.11, 453.41, 453.0, 451.81, 453.40, 453.42, 453.9) were used to identify those cirrhosis patients with new onset PE/DVT during the hospital admission. Subjects with a history of PE/DVT prior to January 1, 1995 were excluded. Those patients who developed recurrent DVT/PE during the time of the study period were not re-counted. To make certain that we accurately identified subjects with DVT/PE, such diagnoses were confirmed by positive computerized tomography (CT) using the PE protocol, positive Doppler ultrasound of the extremities, and/or high probability V/Q scan.

There were two control groups in this study. The first control group consisted of patients with no known history of cirrhosis and selected comorbidities (CHF, CKD, or cancers) who were hospitalized during the study period. These patients were systematically matched to cirrhotic cases by age, race, and gender. We used the same method as described above to capture the diagnoses of DVT/PE in this control group. In order to compare the incidence of DVT/PE in cirrhotic patients to those with other co-morbidities, a second control group was identified. The subjects in this group were those with the following underlying diagnoses: (1) CHF (diagnosed by ICD-9 codes; 428, 428.9, 428.1, and/or by the echocardiogram report), (2) CKD (ICD-9 codes 586, 585.9, 639.3 2), or (3) malignancies (the four most common solid organ cancers in the USA: lung, breast, prostate, and colon cancers [7]). The subjects in each sub-group were age-, gender-, race-, and Charlson Index (see below)-matched to those with cirrhosis.

Given the potential influence of co-morbidities on the occurrence of DVT and PE, we assessed the Charlson Index [8, 9] as a surrogate measurement for the burden of co-morbidities. The Charlson Index contains 19 categories of co-morbidity that are primarily defined using ICD-9-CM diagnoses codes (a few procedure codes are also employed). Each category has an associated weight based on the adjusted risk of 1-year mortality. A higher score indicates a more severe burden of co-morbidity. This index provides an objective measurement of co-morbid disease based on inpatient hospitalizations. A detailed description of this index and how to calculate it has been described elsewhere [8, 9].

Analysis

Statistical analysis was performed at the Regenstrief Institute for Health Care using SAS software (SAS Institute, Cary, N.C.). The descriptive statistics for the cases and controls, including means, standard deviations (SD), ranges, and percentages, were used to better characterize the two patient populations. The incidence of DVT/PE in each particular group was calculated by the number of new cases of DVT/PE diagnosed at admission or during hospitalization divided by the total number of subjects in each group who were hospitalized during the study period. Continuous and categorical variables were compared between groups using the Student t test and chi-square (χ2) test, respectively. Univariate and multi-variate logistic regression analyses were performed to identify the predictors of DVT/PE in the study cohorts. In this study, those variables with the P value < 0.10 from the univariate analysis were then included as the independent variables in the multi-variate models. The strength of association was reported as the odds ratio (OR) and its corresponding 95% confident interval (95% CI). A P value < 0.05 was considered to be statistically significant.

Results

Comparison of the Incidence of DVT/PE in Cirrhotics Versus Non-cirrhotic Controls

During the study period, 963 cirrhotic cases (age 50.5 ± 11 years, females 32%, Caucasians 60%) and 12,405 controls without selected co-morbidities (age 50.5 ± 12 years, females 36%, Caucasians 60%) met criteria for inclusion in the study. Details on the demographics, baseline laboratory tests, Charlson Index, and the total events of DVT/PE in each group are shown in Table 1. The admission diagnoses for the control group were as follows: trauma, motor vehicle accidents, and skeletal fractures, 30%; burn, 20%; community-acquired pneumonia, 15%; psychiatric illnesses, 12%; gastrointestinal (GI)-related illnesses (acute appendicitis, acute pancreatitis, acute cholecystitis), 9%; obstetrics/gynecology-related issues, 7%; ophthalmologic problems, 1%; others, 6%. The incidence of DVT/PE in patients with cirrhosis and those without cirrhosis was 1.87 and 0.98%, respectively (OR 1.78, 95% CI 1.1–2.2, P = 0.007). As expected, the Charlson Index in the cirrhotic group was significantly higher than that in the non-cirrhotic patients (3.2 ± 1.9 vs. 0.96 ± 1.5, P < 0.001). To determine whether the presence of cirrhosis is an independent predictor for DVT/PE, multi-variate analysis was performed to control for other potential confounders, such as Charlson Index and International Normalized Ratio (INR). The Charlson Index score (OR 0.93, 95% CI 0.74–1.16) and presence of cirrhosis (OR 0.87, 95% CI 0.2–2.6) were not independently associated with DVT/PE. Partial thromboplastin time (PTT) (OR 0.88: 95% CI 0.84–0.94) and serum albumin (OR 0.47, 95% CI 0.23–0.93) were the independent predictors of DVT/PE in this study cohort (Table 2). We also compared the risk of DVT/PE between compensated (Child A) and decompensated (Child B/C) cirrhotic patients and found that the incidence of DVT/PE in Child B/C and in Child A was 2.6 and 0.8%, respectively. Although, there was the trend towards the higher incidence of DVT/PE in decompensated cirrhotic patients (P = 0.10), the difference was not statistically significant.

Table 1 Selected demographics and characteristics of individuals with cirrhosis and controls without selected co-morbidities
Full size table
Table 2 Factors associated with DVT/PE in the entire study cohort (n = 13,368)
Full size table

Comparison of the Incidence of DVT/PE in Cirrhotics Versus Other Co-Morbid Illnesses

We next compared the incidence of DVT/PE in cirrhotic patients to age-, gender-, race-, and Charlson index-matched controls with underlying co-morbid illnesses. The detailed demographic data and the incidence of DVT/PE in each sub-group are shown in Table 3. The incidence of DVT/PE in subjects with cirrhosis (1.87%) was significantly lower than that in patients with CKD (7%; OR 0.25, 95% CI 0.15–0.41), CHF (7.75%; OR 0.23, 95% CI 0.14–0.37), and cancer (6.1%; OR 0.29, 95% CI 0.17–0.52). Figure 1 summarizes the odds ratio for the risk of developing DVT/PE when we used the incidence of DVT/PE in non-cirrhotic controls as a baseline reference.

Table 3 Selected demographics and characteristics of individuals with cirrhosis and controls with selected co-morbid illnesses
Full size table
Fig. 1
figure 1

The incidence of deep vein thrombosis/pulmonary embolism (DVT/PE) [reported as odds ratio (OR) and 95% confidence intervals] was significantly lower in cirrhotic patients than in patients with other medical illnesses. Although the incidence of DVT/PE was higher in cirrhotics when compared to non-cirrhotic controls, the difference was not significant in the multivariate analysis. CRF Chronic renal failure, CHF congestive heart failure. *P = 0.06

Full size image

Discussion

The results of our study demonstrate that hospitalized cirrhotic patients did not have a lower risk of DVT/PE than the appropriately matched non-cirrhotic controls without selected co-morbidities. However, the risk of developing DVT/PE was low when compared to age-, gender-, race-, and Charlson Morbidity index-matched subjects with other co-morbid illnesses that are well-known risk factors for DVT/PE [10].

The underlying coagulopathy in patients with cirrhosis has led to the assumption that these patients may be at a lower risk for developing DVT/PE. However, this perception is contrary to the results of several studies that have shown an increase in thromboembolic risk in cirrhotic patients [11]. Although the levels of several coagulation factors are low in cirrhosis patients [12], they also develop acquired deficiencies in the anti-coagulation factors (including protein C, protein S, and anti-thrombin III) [13]. Additionally, these patients often suffer from physical conditions which prohibit active lifestyles, thus perpetuating the risk of thromboembolism. Taken together, the pathogenesis of DVT/PE in cirrhotic patients is complex, with several factors involved, leading to the inconsistency in the reported incidence of DVT/PE in these patients [1, 10].

We found that underlying cirrhosis did not reduce risk of DVT/PE when cirrhotic patients were compared to non-cirrhotic patients without other co-morbidities. Our results are in accordance with that reported by Northup et al. [1]; however, several issues deserved further discussion. First, the risk of DVT/PE reported by Northup et al. was 0.5% compared to the 1.8% obtained in this study. The difference in the incidence rate relative to controls is likely due to the use of different control groups; cirrhotic patients without DVT/PE in the former and non-cirrhotic patients in the latter. Another possibility could be the manner in which the incidence rate was calculated. The denominator of our incidence rate was the number of cirrhotic subjects hospitalized during the study period; in the report by Northup et al. [1], the authors used the total inpatient cases who were hospitalized during the study period, using those without cirrhosis as their denominator. Second, we also found that serum albumin was an independent predictor for DVT/PE in cirrhotic patients. Because albumin is exclusively synthesized in the liver, low levels of serum albumin may serve as a marker for poor hepatic reserve. As such, these patients are likely to have deficiencies of anti-coagulation factors, thus placing them at risk for DVT/PE [1].

The findings that the incidence of DVT/PE in cirrhotic patients is much lower than that in subjects with other medical illnesses deserve further consideration. The reported incidence of DVT/PE in patients with CHF in our study is similar to that found in the recently published report based on the use of General Practice Research Database (GPRD), the computerized patient care system in the UK. However, the incidence of DVT/PE in chronic liver disease and cancer patients was reported to be 0.6 and 16%, respectively, in that study. Such differences may be attributed to the following reasons. First, we only studied patients with the five most common cancer diagnoses in the USA. Therefore, we did not include any hematological malignancies or other gastrointestinal cancers in the analysis. Second, the study based on the GPRD simply looked at the risk of thromboembolism in patients with chronic liver disease. In comparison, our inclusion criteria for cirrhosis may lead to the selection of more advanced liver disease cases. The risk of DVT/PE in cirrhotics in our study is higher. This correlates with studies showing that the thrombotic condition worsens as the liver disease progresses [11, 14]. Regardless of the reported incidence, the results from these two studies are in agreement that the risk of DVT/PE in cirrhosis is significantly lower than that in patients with CHF or cancers.

This study suffers from multiple inherent weaknesses. First, this is a retrospective study utilizing an electronic medical record system. Thus, it has the shortcoming of any retrospective study design. To overcome this issue, we have carefully selected the subjects with cirrhosis by cross-referencing with sign/symptoms of portal hypertension and/or decompensated liver diseases. Additionally, we have aggressively verified cases with DVT/PE, not only from the ICD-9 diagnosis, but also from the text reports and ‘accepted standard diagnostic procedures (V/Q scan, CT scan of the chest with specific PE protocol)’ in order to ensure that we did not over-estimate and recruit non-DVT/PE cases in the analyses. Second, as previously mentioned, we have developed rigorous criteria to identify cirrhotic cases. The result of utilizing these criteria could lead to a selection bias for more advanced cirrhotic patients; this would result in the inclusion of cases having a higher prevalence of confounding features, such as refractory ascites and limited physical activity. With this in mind, we carefully controlled for co-morbidity using the Charlson Index in the analyses. Finally, when a study is conducted to examine a disease/condition with a low incidence rate (such as the incidence of DVT/PE in cirrhosis), an essential requirement is that the sample size is large enough to avoid type II error (β error). In other words, this is the error of failing to observe a difference when in truth there is one. However, given the strength of the RMRS database, we believe that we were able to capture a large number of subjects and eliminate this type of statistical error from our analyses.

In conclusion, we did not find a reduced risk of DVT/PE in cirrhotic patients when these were compared to non-cirrhotic controls, although this risk is much lower in the former than in patients with other co-morbid illnesses, such as CHF and malignancy. We also found that serum albumin and PTT were independent predictors of DVT/PE in the cirrhotic cohorts of our study. Our findings should alert healthcare providers to consider DVT/PE in the differential diagnosis when cirrhotic patients present with signs and symptoms compatible with venous thromboembolism.