Objectives: We aimed to determine outcomes and predictors of intraoperative-detected portal vein thrombosis in liver transplant recipients.
Materials and Methods: We retrospectively analyzed 806 adult liver transplant recipients from Shiraz, Iran, to determine those with intraoperative-detected portal vein thrombosis. Patients with this complication were compared with age- and sex-matched patients without this complication. Background diseases, surgery parameters, hospital admission, reoperation, rethrombosis, acute rejection, and use of antico-agulants were assessed. Cox proportional hazards, logistic regression, and random classification forest and random survival forest plots were used for data analyses.
Results: Mean age of patients was 44.7 ± 13.2 years. Patients with intraoperative-detected portal vein thrombosis (n = 91; 11.3%) had mortality ratio of 2.9 (range, 1.0-8.6) and 2-year survival of 78% versus 2-year survival rate of 92% in patients without this disease. Median time of survival in patients with this complication who died was 2 weeks versus 10 months in patients who died and did not have this com-plication. Random classification forest plots showed that high fasting blood sugar, autoimmune hepatitis, low prothrombin time, and cryptogenic cirrhosis were (in order) the main predictors of this complication. Random survival forest plots revealed that low prothrombin time, having intraoperative-detected portal vein thrombosis, Model for End Stage Liver Disease score, primary sclerosing cholangitis, diabetes mellitus, and hepatocellular carcinoma were (in order) the main predictors of death in liver transplant recipients. Low body mass index was associated with mortality in patients with intraoperative-detected portal vein thrombosis (by Cox proportional hazards).
Conclusions: One of every 9 liver transplant patients had intraoperative-detected portal vein thrombosis. Hazard of death was 2.9, and death occurred far earlier in patients with this complication. Improvements in diabetes mellitus care, prothrombin time, Model for End-Stage Liver Disease score, and body mass index may improve outcomes of these patients.
Key words : Hazards model, Model for End-Stage Liver Disease score, Mortality, Survival
Liver transplant (LT) is an effective treatment for most patients with end-stage liver disease.1 Portal vein thrombosis (PVT) is usually considered to be a complication of a splanchnic hypercoagulable state in patients with end-stage liver disease and can be an important complication in LT candidates with acute or chronic liver diseases. However, it is no longer considered to be a contraindication for LT.2-4
Preoperative diagnosis of PVT is not an easy task despite use of various imaging techniques; as many as 50% of patients may have PVT detected only at the time of transplant.3 There are contradictory reports on the effects of PVT on outcomes after LT, including survival. Some studies have reported increased mortality and morbidity in LT patients with PVT,2,3,5-8 whereas other reports did not find such effects.9-16 This study aimed to determine the outcome and its predictors in LT patients with intraoperative-detected portal vein thrombosis (IOPVT).
Materials and Methods
In this retrospective cohort study, we assessed the medical records of 806 adult patients who received LT between December 2013 and August 2015 at the Shiraz Organ Transplantation Center, the main referral and training center for LT in Iran.1 All patients with confirmed PVT at time of the LT procedure were compared with an age- and sex-matched control group who received LT and did not have PVT. The control group was also caliper-matched on the time of operation (± 14 days). The thrombosis was classified as partial PVT (PPVT) or complete PVT (CPVT) according to the operation note. Both groups of study patients were followed from the date of LT until death (in patients who died) or until end of study (May 2016) in patients who were alive.
Among patients who died, we included those whose death was caused by LT-related complications. Background medical information, surgical technique (piggy back vs standard), duration of operation, anesthesia duration, graft weight, number of bags of transfused blood, warm ischemia time, cold ischemia time, anhepatic phase duration, type of PVT (PPVT vs CPVT), type of transplanted liver (split vs complete), length of hospital admission, immediate post-LT events (need for reoperation, rethrombosis, acute rejection of graft), and dose and duration of treatment with anticoagulants (such as aspirin, warfarin, and heparin) were also assessed.
In addition to Wilcoxon rank sum test, t tests and chi-square tests were applied as statistical tests. Random classification forest (RCF) plots and multiple logistic regression (backward method) were used to identify factors affecting the occurrence of IOPVT. The importance of each factor was compared by odds ratio (OR) in logistic regression and by Gini index for RCF analyses. The prediction value of these 2 classification models were compared in a 10-fold cross-validation, which included the following criteria: sensitivity for identifying IOPVT patients, specificity for correct prediction of non-IOPVT, accuracy, area under the receiver operating curve, and Brier score, which is a common criterion to assess prediction error. Survival rates of IOPVT and non-IOPVT patients were compared by nonparametric Kaplan-Meier (KM) analyses. Furthermore, Cox proportional hazards (CPH) model and random survival forest (RSF) plots were applied as semi-parametric and nonparametric survival models, respectively. Harrell concordance index and Brier score were considered as indicators for evaluating the goodness of fit and prediction of values of survival models. The satisfaction of CPH assumption was evaluated by rho coefficient, which is the correlation between survival time and scaled Schoenfeld residuals. We used R software (version 3.1.0) and SPSS software (SPSS: An IBM Company, IBM Corporation, Armonk, NY, USA) for statistical analyses.
The protocol of this study was approved by the ethics board for the research faculty of the Shiraz University of Medical Sciences (registration number 95-01-62-13170). The study conformed to the ethical guidelines of the 1975 Helsinki Declaration. All patients had provided written informed consent regarding the use of their personal information, and they were identified by codes rather than by initials or identification numbers.
Mean age of patients was 44.7 ± 13.2 years, and the male-to-female ratio was 2:1. Of 806 LT recipients, 91 patients (11.3%) had IOPVT. Average duration of hospitalization after LT was 11.4 days. The longest follow-up period was 2.4 years, with median time of 1.5 years for IOPVT and 1.6 years for non-IOPVT patients.
Of the 91 patients with IOPVT, 13 (14.2%) had died compared with 5 of 91 (5.4%) of patients without IOPVT (chi-square = 3.9; P = .04).All deaths had occurred at the hospital. Two of 91 IOPVT patients (2.1%) had developed rethrombosis, which was similar to that shown in the non-IOPVT group (2 of 91; 2.1%) (Fisher exact test = 0.1). Acute graft rejection was reported in 11 of 91 IOPVT patients (12%) and in 9 of 91 patients without IOPVT (9.8%) (chi-square = 0.2; P = .6).
Our univariate analyses showed that cryptogenic cirrhosis was more common among patients with IOPVT (OR = 2.5; 95% confidence interval, 1.2-5.3). The Model for End-Stage Liver Disease (MELD) score (just before LT), the aspartate aminotransferase-to-alanine aminotransferase ratio, sodium level, albumin level, and duration of LT surgery were significantly higher in IOPVT patients than in patients without IOPVT. Moreover, patients with IOPVT needed more blood transfusion during LT surgery than patients without IOPVT.
Red blood cell count, hematocrit level, white blood cell count, platelet count, alanine amino-transferase, total cholesterol, low-density lipoprotein, and prothrombin time were significantly lower in patients with IOPVT than in patients without IOPVT (Table 1). Other studied variables did not show any significant differences between the 2 groups. According to the logistic regression model, cryptogenic cirrhosis as the cause of end-stage liver disease increased the odds of IOPVT by 2.3 times. Furthermore, an increase for every one MELD score resulted in an increment of 6.2% in odds of IOPVT, with a decrease for each unit in alanine amino-transferase increasing the odds of this disease by 1.8% (Table 2).
In agreement with logistic regression results, RCF analyses recognized cryptogenic cirrhosis as an important predictor of IOPVT. However, RCF showed that fasting blood sugar, autoimmune hepatitis, and prothrombin time were significantly associated with IOPVT (Figure 1). The comparison of RCF and logistic regression revealed that Brier score, which indicates prediction error, was lower with RCF (0.22 vs 0.26), whereas sensitivity (0.64 vs 0.49), specificity (0.63 vs 0.58), accuracy (0.64 vs 0.53), and area under the curve (66.7 vs 61.7) were higher with RCF analyses. Therefore, RCF provided a better prediction of IOPVT.
Figure 2 illustrates the survival probabilities estimated by Kaplan-Meier analyses for patients with and without IOPVT. The heavy tail shown in the Kaplan-Meier curves indicates heavy censoring and low number of deaths in the total population of patients. Figure 2 also shows that, after approximately 9 months of LT, the survival rate of IOPVT patients was lower than in those without IOPVT. Moreover, 1-, 1.5-, 2-, and 2.4-year survival rates were 98%, 91%, 78%, and 66% in patients with IOPVT compared with 100%, 96%, 92%, and 85% in those without IOPVT. The median time to death in IOPVT patients and non-IOPVT patients was about 2 weeks and 10 months, respectively. In agreement with Kaplan-Meier analyses, CPH confirmed a 2.9 hazard rate of mortality in patients with IOPVT. Cox proportional hazards analyses also identified diabetes mellitus and low body mass index as factors significantly associated with mortality of IOPVT patients (Table 2).
As shown in Figure 3, in the presence of IOPVT, prothrombin time, MELD score, primary sclerosing cholangitis, diabetes mellitus, and hepatocellular carcinoma (HCC) were (in order) the most important variables for predicting the survival of patients after LT. Concordance index (Figure 4A) and Brier score (Figure 4B) confirmed that RSF showed a greater concordance but a smaller prediction error than CPH. Therefore, RSF provides a more suitable fitness regarding prediction of survival of IOPVT patients. Moreover, attributed R2 values, at the maximum event time, were 71.1% and 50.4%, respectively, for CPH and RSF, which testified the better goodness of fit of the RSF model.
This study revealed that 1 of 9 LT patients had IOPVT and that IOPVT patients had about a 3-time greater hazard of death, with 50% of deaths occurring in the first 2 weeks posttransplant. We also found that cryptogenic cirrhosis, fasting blood sugar, autoimmune hepatitis, prothrombin time, ALT (inversely), and MELD score were significant determinants of IOPVT. Moreover, diabetes mellitus, body mass index (inversely), prothrombin time, MELD score, primary sclerosing cholangitis, and HCC were associated with mortality of IOPVT patients. Finally, we also found that learning algorithms, including RCF and RSF, were more predictive of IOPVT and their survival, respectively, than linear regression and CPH models.
The results of this study were more or less similar to other studies regarding PVT and LT. We found that the prevalence of IOPVT was 11.3%, which is higher than the 7.9%4 and lower than the 15.9%2 reported by other studies regarding patients on wait lists for LT. In a study of 69 LT patients, 53 patients (77%) had cirrhosis, with 16% to 19% (in advanced cirrhosis) having PVT. That study also concluded that PVT often presents in the partial form and is difficult to diagnose preoperatively and that PVT, including CPVT, did not affect survival rate.16
In 2009, Bagheri Lankarani and associates showed that PVT patients had a MELD score of 19.9 ± 5.4, which was lower than in our study (22.1 ± 4.1).4 That study also found that the mean operation time for LT in patients with PVT was 7.2 ± 1.5 hours, the mean transfusion requirement was 5.4 ± 2.8 bags of packed red blood cells, and the duration of hospital stay was 17.7 ± 10.9 days; all of these indicators were about 1.5 times more than their corresponding items in our study.4 Rethrombosis was also reported in up to 28.5% of LT patients with PVT4; this result was far greater than the 2.1% in our study. Moreover, during a similar period of follow-up (16.6 ± 7.9 months), the overall mortality rate in PVT patients was 28%, which was approximately 2 times that of our finding (14.2%).4 These results indicate an improved quality of outcomes after LT from 2009 to 2016 at the Shiraz Transplant Center; however, LT surgery was more complex and mortality was higher in patients with IOPVT than in non-IOPVT patients.4
Regarding the risk factors of IOPVT and its association with posttransplant outcomes, different studies pointed to different findings. These variations were a result of different methods of studies. A retrospective study of 48 570 patients showed that PVT was associated with increased 3-month mortality (OR = 1.7) and graft failure (OR = 1.7).17 In only 31% of patients with PVT was PVT diagnosed and reported before LT, that is, during time on wait list for LT.17 That study also revealed that predictors of PVT in LT patients included fatty or cryptogenic liver disease, ascites, diabetes mellitus, and obesity.17 On the other hand, we found that the hazard of mortality in IOPVT patients and during a 29-month follow-up was higher (2.9 times), although similarly cryptogenic cirrhosis and fasting blood sugar (but not fatty liver disease or body mass index) were predictors of IOPVT.
In a study from Spain, Hernandez Conde and colleagues studied 191 patients with liver cirrhosis who underwent LT. The group reported a prevalence of IOPVT of 9.4% (compared with 11.3% in our study), with most not diagnosed before LT.18 Moreover and in agreement to what was found in our study, they concluded that IOPVT patients had higher MELD scores, although, in contrast to our findings, these patients did not have longer operation time or more need for blood transfusion compared with non-PVT patients.18 Hernandez Conde and colleagues also found that PVT was associated with a higher mortality in month 1 after LT (16.7%) compared with patients without PVT (8.8%).18
In another study, 2% to 26% of patients waiting for LT had PVT, with 50% of PVT cases incidentally found during operation. That study concluded that PVT was a risk factor for early post-LT mortality.3 Bagheri Lankarani and associates found that male sex, previous surgery or endoscopic treatment for portal hypertension, history of variceal bleeding, low platelet count, HCC, and advanced liver failure were risk factors for PVT in cirrhotic patients waiting for LT.4 In our study and among mentioned factors, we found that advanced liver failure (cryptogenic cirrhosis, high-er MELD, and low ALT) was associated with IOPVT.
In a systematic review of studies between 1991 and 2011 and contrary to our study, LT in PVT had a similar outcome compared with that shown in non-PVT patients, despite more blood transfusion and a longer intensive care unit/hospital stay in the first group. This review had a debated view toward the role of anticoagulant treatment in outcomes of transplant patients with PVT.9 Another systematic review of 25 753 LT patients showed that CPVT patients had greater 1-year mortality rate (18.8%) than patients without PVT (15.4%).5 These figures were higher than our findings of 14.2% versus 5.4% mortality in both groups, respectively. Furthermore, the previous review found that rethrombosis occurred in up to 13% of patients with CPVT,5 which is far greater than the 2.1% found in our study and the 1.8% reported in another study.19
In the absence of HCC, previously reported survival rates for patients with PVT at 3, 6, 9, and 12 months and at 3 years after LT were 68%, 64%, 61%, 61%, and 61%, respectively.20 These results are much lower than the 1-year survival rate (98%) and somewhat lower than the 2.4-year survival rate (64%) in our study.
An assessment of 617 LT cases between 1991 and 2008 showed a prevalence of 7.8% of PVT in these patients, including 58.3% with PPVT and 41.7% with CPVT. That study, like ours, concluded that patients with PVT needed more complex surgical techniques, longer operation time, and more blood transfusion, with those with CPVT having a higher mortality rate.6 In another series that included studies between 1991 to 2009, patients who underwent LT with IOPVT had a mean MELD score of 18.3 (< 20 in our research), with survival rates of 85%, 74%, and 63% at 1, 3, and 5 years, respectively, which were similar to our findings.21 Another study conducted from 1998 to 2007 on 2508 adult patients with LT showed a 10% prevalence of PVT. That study found sex as a predictor of PVT, with greater need for blood transfusion in these patients. However, the 1-year survival rate in PVT patients (86.5%) was similar to those without PVT (89.4%).10 These results were close to our findings except for association of sex with PVT and a lower survival rate in PVT patients in that study.
In an older study, adult patients with LT from 1987 through 1996 were assessed, with sex, portal hypertension, Child-Pugh class C, and alcoholic liver disease associated with PVT. That study concluded a 2-time transfusion need (10 vs 5 U), a 2.5-time increased in-hospital mortality rate (30% vs 12.4%), and more postoperative complications (such as renal failure and rethrombosis) in PVT patients than in those without PVT. Moreover, the 5-year survival rate in patients with grade > 1 of PVT (65.6%) was lower than in the control group (76.3%). However, in patients with grade 1 PVT, 5-year survival rate (86%) was identical to the non-PVT group.22
Another previous series found that the main risk factors of PVT after LT were pathologic changes in portal vein, abnormal bloodstream dynamics, hyper-coagulable status, and improper surgical techniques.23 A significant increase in blood trans-fusion (15.2 vs 8.6 U), longer operation time (492 vs 403 minutes), longer hospital stay (32.4 vs 22.1 days), and higher incidence of rethrombosis (8.3% vs 1.2%) in liver transplant patients with PVT (compared with a non-PVT group) was reported in another study, whereas overall morbidity (58.3% vs 50.6%), hospital mortality (8.3% vs 6.5%), and 1-year survival rate (87.5% vs 89.4%) were not different between these 2 groups.11
In an assessment of 133 cases of LT adults, it was concluded that past history of variceal bleeding (OR = 10.6) and surgical shunt surgery (OR = 28.1) were independent risk factors for PVT, with rate of postoperative PVT being significantly higher in patients with PVT than in those without PVT (18.2% vs. 2.7%). However, the 3-year survival rate in PVT patients was not different from those without PVT (73.6% vs 85.3%) (P = .3).12
In another study, incidence of rethrombosis was higher in LT patients with CPVT (compared with those without PVT). However, it was not different between those with PPVT and those without PVT.24 This result was in contrast to our findings, which may be due to a better postoperative management in our study center.
Our study’s limitation was the incompleteness of patient medical records, such as lack of PVT grading, which could have enabled us to detect the association between LT outcomes and different grades of PVT. Moreover, there were some difficulties in conducting verbal autopsy with 1st-degree relatives of dead patients. A larger and multicenter prospective cohort study of patients from before to after LT could lead to a better understanding of all modifiable predictors of morbidity and mortality in those who develop PVT.
Intraoperative-detected portal vein thrombosis was seen in a minority of LT patients. However, it caused poor outcomes in terms of more complicated surgery and lower survival rate compared with patients without IOPVT. There are some predictors of IOPVT such as MELD score, fasting blood sugar, and prothrombin time that are modifiable, and their improvement may decrease the possibility of IOPVT in LT candidates. Moreover, by earlier and better management of diabetes mellitus, body mass index, prothrombin time, and MELD score, it may be possible to reduce the mortality rate among LT recipients with IOPVT, especially in the first weeks after LT.
Volume : 19
Issue : 4
Pages : 324 - 330
DOI : 10.6002/ect.2018.0295
From the 1Health Policy Research Center, Institute of Health, the
Epidemiology, School of Health, the 3Nemazee Hospital, Department of Internal
Medicine, the 4Shiraz Organ Transplant Center, and the 5Shiraz Transplant
Research Center, Shiraz University of Medical Sciences, Shiraz, Iran
Acknowledgements: This research was conducted by Health Policy Research Center without any grant or financial support by others. The authors have no conflicts of interest to declare. We thank all patients and their relatives who helped us conduct this study. We also appreciate the unwavering affection of all members of the Shiraz Liver Transplant Center who provided us the opportunity to conduct this research.
Corresponding author: Behnam Honarvar, Health Policy Research Center, Institute of Health, Shiraz University of Medical Sciences, PO Box 7134853185, Shiraz, Iran
Phone: +98 71 32309615
Figure 1. Associated Factors of Intraoperative Detected Portal Vein Thrombosis in Liver Transplant Patients Based on Random Forests for Classification
Figure 2. Comparison of Survival Between Liver Transplant Patients With Intraoperative Detected Portal Vein Thrombosis and Those Without This Condition Based on Kaplan-Meier Method
Figure 3. Factors Associated With Survival of Patients After Liver Transplantation Based on Random Survival Forests Model
Figure 4. Harrell’s Concordance Index (a) and Brier Score (b) to Compare Cox Regression and Random Survival Forests for Modeling Survival of Patients After Liver Transplant
Table 1. Univariate Variable Analyses of Associated Factors of Intraoperative-Detected Portal Vein Thrombosis in Liver Transplant Patients
Table 2. Associated Factors of Intraoperative-Detected Portal Vein Thrombosis (Logistic Regression: Backward Method) and Survival (Cox Proportional Hazards) in Liver Transplant Patients