A Model to Predict Poor Survival in Patients Undergoing Transjugular Intrahepatic Portosystemic Shunts

Transcription

1 A Model to Predict Poor Survival in Patients Undergoing Transjugular Intrahepatic Portosystemic Shunts MICHAEL MALINCHOC, 1 PATRICK S. KAMATH, 1 FREDRIC D. GORDON, 2 CRAIG J. PEINE, 3 JEFFREY RANK, 4 AND PIETER C. J. TER BORG 5 Transjugular intrahepatic portosystemic shunts (TIPS) may worsen liver function and decrease survival in some patients. The Child-Pugh classification has several drawbacks when used to determine survival in such patients. The survival of 231 patients at 4 medical centers within the United States who underwent elective TIPS was studied to develop statistical models to (1) predict patient survival and (2) identify those patients whose liver-related mortality post-tips would be 3 months or less. Among these elective TIPS patients, 173 had the procedure for prevention of variceal rebleeding and 58 for treatment of refractory ascites. Death related to liver disease occurred in 110 patients, 70 within 3 months. Cox proportional-hazards regression identified serum concentrations of bilirubin and creatinine, international normalized ratio for prothrombin time (INR), and the cause of the underlying liver disease as predictors of survival in patients undergoing elective TIPS, either for prevention of variceal rebleeding or for treatment of refractory ascites. These variables can be used to calculate a risk score (R) for patients undergoing elective TIPS. Patients with R G 1.8 had a median survival of 3 months or less. This model was superior to both the Child-Pugh classification, as well as the Child-Pugh score, in predicting survival. Using logistic regression and the same variables, we also developed a nomogram that indicates which patients survive less than 3 months. Finally, the model was validated among an independent set of 71 patients from the Netherlands. This Mayo TIPS model may predict early death following elective TIPS for either prevention of variceal rebleeding or for treatment of refractory ascites. (HEPATOLOGY 2000;31: ) The transjugular intrahepatic portosystemic shunt (TIPS) procedure has recently been introduced for the management Abbreviations: TIPS, transjugular intrahepatic portosystemic shunt; OLT, orthotopic liver transplantation; INR, international normalized ratio; ISI, international sensitivity index. From the Division of Gastroenterology, Hepatology, and Internal Medicine (P.S.K.) and the Section of Biostatistics (M.M.), 1 Mayo Clinic and Mayo Foundation, Rochester, MN; 2 Beth Israel Deaconess Medical Center, Boston, MA; 3 Hennepin County Medical Center, Minneapolis, MN; 4 the University of Minnesota, Minneapolis, MN; and 5 the University Hospital, Rotterdam, The Netherlands. Received September 22, 1999; accepted January 18, Supported in part by research grant DK from the National Institutes of Health, Public Health Service, Bethesda, MD. Address reprint requests to: Patrick S. Kamath, M.D., Mayo Clinic, 200 First Street SW, Rochester, MN kamath.patrick@mayo.edu; fax: Copyright 2000 by the American Association for the Study of Liver Diseases /00/ $3.00/0 doi: /he of complications of portal hypertension, 1 namely, control of refractory acute variceal hemorrhage, 2 prevention of recurrent variceal bleeding, 3 and treatment of intractable ascites. 4 Currently, it is not known whether TIPS has any effect on survival in these patients, nor are the factors associated with poor survival known. It is well recognized that the need for an emergency TIPS, that is, TIPS being performed for the control of acute variceal hemorrhage, is associated with a high mortality. 5,6 Some factors predictive of poor survival after elective TIPS, such as advanced liver disease 3,4,7,8 and hyperbilirubinemia, 9 have been recognized. However, there has been no model developed that can accurately predict survival after elective TIPS. Traditionally, Pugh s modification of the Child-Turcotte classification (referred to as the Child-Pugh classification) has been used to assess risk in patients undergoing portosystemic shunt surgery. 10 This system has three inherent problems when applied to patients undergoing TIPS. First, most TIPS candidates are of class C. Thus, the system cannot discriminate among the majority of patients undergoing TIPS. Second, it only divides patients into poor, intermediate, and low risk without quantifying an expected period of survival; for example, it cannot be used to predict which patient will die within a certain period of time and which patient will survive that period. Finally, it uses determinants such as ascites and encephalopathy, which are based on subjective assessment and which can be altered by therapy. The several difficulties and inaccuracies in applying the Child s classification have been detailed in an editorial by Conn. 11 Thus, a score more accurate than the Child-Pugh classification and based on objective data is necessary to determine survival in patients undergoing TIPS. Patients with refractory variceal hemorrhage or refractory ascites are often candidates for liver transplantation. TIPS might best serve as a bridge to liver transplantation; but, timely liver transplantation is not often available. Because the cost of TIPS is not insignificant, a system is needed that can accurately predict survival following the procedure. Patients in whom survival is expected to be in days or weeks (rather than in months or years) should arguably have TIPS only if they are candidates for liver transplantation. Using the statistical methods that have been successful in predicting the survival of patients with primary biliary cirrhosis 12 and primary sclerosing cholangitis, 13 we developed two methods for predicting survival after TIPS. The first method requires at least a programmable calculator, whereas the second method, which uses a nomogram and requires only a ruler and a pencil, predicts which patients are likely to die within 3 months after elective TIPS. We chose a period of 3 months because beyond that period factors, such as stenosis of the shunt and possible variceal rebleeding as a consequence, would compound the analysis. Further, average waiting times 864

2 HEPATOLOGY Vol. 31, No. 4, 2000 MALINCHOC ET AL. 865 TABLE 1. Follow-up Data for Patients Who Underwent Elective TIPS All Patients (n 231) Deaconess (n 90) HCMC* (n 21) Mayo Clinic (n 49) University of Minnesota (n 71) Median followup (yr) Range of followup (yr) Deaths (n) Liver transplantation (n) TIPS dates First 3/20/92 9/21/91 1/6/92 9/28/91 Last 3/16/95 11/16/94 2/16/95 3/21/95 *Hennepin County Medical Center. Among the 121 patients known to be alive at last follow-up. on a list for liver transplantation as a UNOS Status 2 are approximately 3 months. 14 Our model was developed to help physicians counsel patients and their families before deciding to proceed with elective TIPS. PATIENTS AND METHODS Patient Population, Follow-Up, and Data Collection. We followed 231 consecutive patients with cirrhosis but without hepatocellular carcinoma who underwent elective TIPS from September, 1991 through March, 1995 at Deaconess Hospital (Boston, MA), Hennepin County Medical Center (Minneapolis, MN), Mayo Clinic (Rochester, MN), and the University of Minnesota Hospital and Clinic (Minneapolis, MN). Patients with significant cardiopulmonary comorbidity or intrinsic renal disease were excluded from the analysis. Elective was defined as TIPS being performed for the prevention of variceal rebleeding or for treatment of refractory ascites with the patient being hemodynamically stable, without active infection, adequately hydrated, and with the hemoglobin being stable at least 48 hours. Patients in whom TIPS was performed for the control of active variceal bleeding after failure of two sessions of endoscopic therapy within a 24-hour period were designated emergency TIPS and excluded from the analysis. Refractory ascites was defined as ascites that required for control paracentesis more frequently than every 2 weeks despite a sodium-restricted diet and intensive diuretic therapy, consistent with the consensus statement. 15 Of the 231 patients undergoing elective TIPS, 58 (25%) had the procedure for treatment of refractory ascites, and 173 (75%) had TIPS for prevention of recurrent variceal bleeding. Among the 58 TABLE 2. Demographic, Clinical, Biochemical and Hepatic Hemodynamic Features in Patients Who Underwent Elective TIPS All Patients (n 231) Deaconess (n 90) HCMC* (n 21) Mayo Clinic (n 49) University of Minnesota (n 71) Demographic Age (yr) (mean SD) Cause of cirrhosis (%) Alcoholic Cholestatic Viral hepatitis Other Clinical Ascites (%) 0 None Detected only on ultrasound Shifting dullness Tense ascites Hepatic encephalopathy (%) Stage Stage I Stage II Stage III Stage IV Biochemical (mean SD) Albumin (g/dl) Serum bilirubin (mg/dl) Serum creatinine (mg/dl) INR for prothrombin time Childs-Pugh Classification (%) A B C Score (mean SD) Hepatic hemodynamics (mean SD) Pre-TIPS gradient, mm Hg Post-TIPS gradient, mm Hg Antegrade portal flow on doppler ultrasound (%) *Hennepin County Medical Center. Includes Wilson s disease, methotrexate, non alcohol-related steatohepatitis, 1 antitrypsin deficiency, autoimmune, and cryptogenic cirrhosis. Based on hepatic encephalopathy, ascites, bilirubin, albumin, and prothrombin time (seconds). Data not recorded. Post-TIPS portacaval pressure gradient 12 mm Hg.

3 866 MALINCHOC ET AL. HEPATOLOGY April 2000 patients with refractory ascites, 24 fulfilled criteria for diuretic resistant ascites and 34 for diuretic intractable ascites. 15 The 231 patients were followed from their date of TIPS until death, liver transplantation, or study closure (August, 1995). During follow-up, 110 died and 28 underwent liver transplantation, 7 of these liver transplants taking place within 3 months of TIPS. Deaths in the 110 were related to complications of liver disease. The follow-up of the 28 who underwent TIPS and later orthotopic liver transplantation (OLT) was censored at the day of transplantation; that is, it was assumed that the OLT date was their last day of follow-up. This censoring was done to remove the effect of OLT when modeling the survival of patients who undergo TIPS. If these patients were not censored, deaths due to surgical mortality related to OLT might have influenced the selection of variables that are prognostic for the TIPS procedure. On the other hand, OLT may prolong survival compared with patients who do not undergo TIPS. We found predictably that survival in patients who underwent OLT was significantly improved compared with those who did not undergo OLT. Thus, patients were censored at the day of OLT. Of the remainder, 72 were still alive and being followed and 21 were lost to follow-up. Eighteen of these 21 patients had less than 3 months follow-up. The average duration of follow-up was 1.1 years (range, 0.1 to 3.8 years) (Table 1). We studied factors predicting survival after all elective TIPS and separately for prevention of variceal rebleeding and for treatment of refractory ascites. Table 2 presents descriptive statistics on patient features that were recorded within 24 hours of the TIPS procedure. Ascites was present in 80% of the patients and graded as indicated in Table 2. Large volume paracentesis was performed and spontaneous bacterial peritonitis was excluded before TIPS. Some degree of hepatic encephalopathy was observed in 60% and staged using standard criteria. 16 Prothrombin time in seconds was converted to international normalized ratio (INR) using the international sensitivity index (ISI) for thromboplastin. 17 The Child-Pugh mean score was with about 56% of the patients being Child-Pugh class C. Portacaval pressure gradient, measured both before and after the placement of the shunt, was reduced by an average of mm Hg to a mean of mm Hg post-tips. The model was validated by applying it to an independent set of 71 patients who underwent TIPS at Rotterdam, the Netherlands from January 1, 1992 to December 31, These patients were followed, on average, for 1.4 years (range, 0.1 to 5.8 years) during which 33 died, 14 within 3 months. No patient underwent OLT or was lost to follow-up within 3 months of TIPS. The cause of liver disease was alcohol related or cholestatic (50%) or viral hepatitis (20%) with the remaining 30% having other types of liver disease. The mean SD levels of bilirubin, creatinine, and INR were mg/dl, mg/dl, and , respectively. Survival Modeling. The starting time for all survival analyses was the date of the TIPS procedure. Patients lost to follow-up were censored at the date they were last known to be alive. The term censored indicates that the patient was alive at that date and that was the extent of follow-up. Cox proportional-hazards regression 18 was the main statistical tool for survival modeling. Model estimation was done by computer with the PHREG procedure of SAS 19 and the coxph function of S-plus. 20 To pick a small set of variables that adequately predicted survival, we used the backwards elimination variable selection method with criteria for retaining variables in the model being P.01. The candidate variables were those factors shown in Table 2 that were significant (P.05) in a univariate Cox model. To lessen the influence of extreme laboratory values we transformed the quantitative variables to their natural logarithms. Based on the data shown in Fig. 1, the cause of cirrhosis was coded: 0 for alcohol related; 0 for cholestatic liver disease; 1 for viral hepatitis; and 1 for other liver diseases. The diseases grouped together as other liver disease have been so grouped because their survival was similar to patients with viral hepatitis who are coded as 1. The basis for this grouping is more apparent from the univariate analysis in the Results section. Because of variation in timing of the TIPS procedure, 50 patients did not have one of these parameters albumin, prothrombin time, level of hepatic encephalopathy, or ascites recorded within 24 hours of placement of the shunt. These missing values were imputed for survival modeling. We used bilirubin and creatinine, which were recorded for all patients, in a linear regression model to predict the missing values. Creation of a Nomogram to Predict Mortality Within Three Months. Seventy of the 231 patients (46 of 173 with variceal bleeding and 24 of 58 with intractable ascites) died of liver failure within 3 months. Using the independent predictors of survival that were identified in the multivariate survival modeling, we used logistic regression to predict the probability of death within 3 months of TIPS. For this analysis, deaths from liver disease within 3 months of TIPS were recorded as events. Patients who survived longer than 3 months were recorded as nonevents. Patients who had less than 3 months of follow-up (n 18) or who underwent OLT within 3 months of TIPS (n 7) were dropped from the analysis. Based on the logistic regression model, we developed a simple pocket chart FIG. 1. Comparison of actual (Kaplan-Meier) survival after elective TIPS for patients with cholestatic, alcoholic, viral hepatitis, or other types of liver disease. The survival of patients with alcoholic or cholestatic liver disease was similar (P.92). The survival of patients with viral hepatitis or other types of liver disease was similar (P.78). The difference in survival between alcohol-related and cholestatic liver disease on one hand, and viral and other liver disease on the other was statistically significant (P.02).

4 HEPATOLOGY Vol. 31, No. 4, 2000 MALINCHOC ET AL. 867 that can be used bedside to predict a patient s chance of dying within 3 months of placement of TIPS as shown in Fig. 2A and B. Model Validation. The 71 independent patients from the Netherlands were stratified into 2 risk (R) groups: R 1.8 and R 1.8. These were the risk scores that stratified patients in the 231 model development group into those with a median survival less than 3 months (R 1.8) and those with a median survival greater than 3 months (R 1.8). Within each risk group the actual survival was calculated using the Kaplan-Meier procedure 21 and the predicted survivals were compared using the one sample long rank test. 22 P.05 was used to indicate significant differences between observed and predicted survival. To validate the nomograms, we compared the actual death rates with the predicted death rates. Further, we calculated standard indices of validity such as sensitivity, specificity, positive predictive value, negative predictive value, and area under the receiver operating curve. RESULTS Univariate Analyses. The 231 elective TIPS patients had a median survival time of 1.4 years. The 6-month, 1-year, and 2-year survival rates were 62%, 56%, and 46%, respectively. Figure 1 indicates that patients with alcohol-related or cholestatic liver disease have similar survival rates (P.92). Also, patients with viral hepatitis and patients with nonviral, non alcohol-related, noncholestatic liver disease (termed other liver disease ) have similar survival rates (P.78). Patients with non alcohol-related or noncholestatic types of liver disease (such as viral hepatitis) or other liver disease have a 64% (P.02) greater risk of dying after TIPS compared with alcoholic or cholestatic patients. Table 3 reports that patients who undergo elective TIPS for the treatment of refractory ascites have a 66% (P.01) greater chance of dying after TIPS compared with patients who undergo TIPS for preventing variceal bleeding. Similarly, increasing levels of ascites, hepatic encephalopathy, Child- Pugh class and Child-Pugh score, bilirubin, creatinine, and INR significantly increased the risk of death, whereas increasing levels of serum albumin indicated significantly better survival. Neither the portacaval pressure gradient nor direction of portal blood flow pre-tips were associated with survival after TIPS. Multivariate Analyses. Of the candidate variables derived from univariate analyses only log e creatinine, log e bilirubin, log e INR, and cause of cirrhosis were independently predictive of survival after TIPS (Table 4). These variables were predictive of survival whether the patient underwent TIPS for prevention of recurrent bleeding or for treatment of refractory ascites; that is, refractory ascites was not an independent variable in the multivariate analysis. A 100% increase in creatinine multiplies the risk of death by 1.94 times; a 100% increase in bilirubin multiplies risk of death by 1.30 times; and a 100% increase in INR more than doubles (2.17) the risk of death. The presence of viral hepatitis or other liver disease increases the risk of death by 90%. Table 5 presents the underlying survival function, that is, the survival after TIPS of a patient with a risk score of (mean risk score of the 231 patients). Calculating the TIPS Risk Score and Predicted Survival Probabilities. Risk scores (R) for individual patients can be calculated by combining their 4 prognostic values with the regression coefficients reported in Table 4. That is, R log e (creatinine mg/dl) log e (bilirubin mg/dl) log e (INR) (cause of cirrhosis). For example, for a hypothetical patient with cirrhosis caused by hepatitis C virus who has a serum creatinine concentration of 1.9 mg/dl, a serum bilirubin concentration of 4.2 mg/dl, and an INR value of 1.2, the risk score would be as follows: R (0.957 log e 1.9) (0.378 log e 4.2) (1.120 log e 1.2) ( ) To obtain the probability of survival for at least t years after TIPS, one calculates the risk score R, reads S 0 (t) from Table 5, and computes S(t) using the equation S(t) S 0 (t) exp(r R 0 ).R 0 is the risk score of the average patient in the series; namely For example, for the hypothetical hepatitis C patient discussed above, the probability of surviving at least 3 months after TIPS is exp( ) FIG. 2. Nomogram for predicting the probability of death within 3 months of placement of TIPS stent for patients with cirrhosis due to alcoholic or cholestatic liver disease (A) and for patients with cirrhosis due to non alcohol-related or noncholestatic liver disease, such as viral hepatitis (B). To use the nomogram the actual values of bilirubin (mg/dl), INR for prothrombin time, and creatinine (mg/dl) are plotted. Connect the bilirubin to the INR and extend the line to the middle scale. The point on the middle scale is connected to creatinine and the line extended to the predicted probability scale. The predicted probability of death within 3 months of placement of TIPS is read off this scale. *Internal normalized ratio for prothrombin time.

5 868 MALINCHOC ET AL. HEPATOLOGY April 2000 TABLE 3. Univariate Assessment of Risk Factors for Death Caused by any Cause Among Elective TIPS Patients Variable Patients Regression Coefficient Regression Coefficient Standard Error P TABLE 4. Survival Model for Patients Undergoing Elective TIPS (n 231, death 110) Variable Regression Coefficient Regression Coefficient Standard Effort P Demographic Age Medical center Cause of cirrhosis* TIPS for refractory ascites Clinical Ascites Hepatic encephalopathy Childs-Pugh Classification Score Biochemical Albumin (log e value) Serum bilirubin (log e value) Serum creatinine (log e value) INR (log e value) Hepatic hemodynamics Pre-TIPS gradient, mm Hg Post-TIPS gradient, mm Hg Gradient reduction, mm Hg Antegrade portal flow on ultrasound NOTE. All analyses were Cox Proportional Hazard Regressions except for Medical Center and Child-Pugh Classification which were log rank tests. Entries are the regression coefficients and their standard errors from a univariate Cox proportional-hazards regression model. Positive coefficients imply that the risk of death increases with increasing values of a risk factor. The P values reflect the role of chance in these findings (that is, the test that the regression coefficient 0). The relative risk attributable to a risk factor can be estimated from the regression coefficients. For example, in this study, every unit increase in the Child-Pugh score increases the risk of death after TIPS by 32% (e 0.28 ). *Cause of cirrhosis coded 0 for alcohol-related liver disease; 0 for cholestatic liver disease; 1 for viral hepatitis; and 1 for Wilson s disease, Budd-Chiari, methotrexate, non alcohol-related steatohepatitis, 1 -antitrypsin efficiency, autoimmune, or cryptogenic. Coded: 0 none to 3 tense ascites. Coded: 0 stage 0 to 4 stage IV. Predicting Three-Month Survival Using the Nomogram. Figure 2 can be used at the bedside to predict the likelihood of dying within 3 months after placement of TIPS. There is a separate nomogram for alcohol-related liver disease and cholestatic liver disease, and for viral hepatitis and other liver diseases. To use the nomograms, the actual values of bilirubin (mg/ dl), INR for prothrombin time, and creatinine (mg/dl) are first plotted. Logarithmic transformation of the variables should not be performed. Second, bilirubin should be connected to INR and the line extended to the middle scale. Third, the middle scale point is connected to creatinine, and the line extended to the predicted probability scale. The predicted probability of death within 3 months of placement of TIPS is thus read off the scale. Serum creatinine (log e value) Serum bilirubin (log e value) INR (log e value) Cause of cirrhosis* *For cause of cirrhosis, use 0 for alcohol-related liver disease or for cholestatic liver disease; 1 for all other causes. Model Validation and Stability. Figure 3 depicts the Kaplan Meier survival and the expected survival for the high risk (R 1.8) and low risk (R 1.8) group from the Netherlands. Median survival was 2.8 months and 1.3 years, respectively. The P values from the one sample log rank test were not significant, indicating that deaths expected from our Mayo model and deaths observed in the Netherlands group were similar in each group (P.41 and P.88, respectively). Thus, the Mayo model very accurately predicted survival among the group from the Netherlands. The nomograms represent a continuum of risk for TIPS rather than a specific cutoff point. In the US cohort, the area under the curve for the receiver operating characteristics was 85%, 95% confidence interval 78% to 91%. An area under the curve greater than 50% suggests the model has utility. Table 6 shows the utility of the model. For example, in the US group where we predicted mortality of 0% to 20%, 9% of patients died, whereas in the group with a 3-month predicted mortality of 80% to 100%, 93% died. Using a cutoff of a predicted probability of death of 0.25 within 3 months, the sensitivity of the model was 77%, specificity 79%, positive predictive value 63%, and negative predictive value 88%. In the cohort from The Netherlands, using a cutoff of 0.25, the sensitivity was 43%, specificity 82%, positive predictive value 38%, and negative predictive value 85%. Comparison Between the Mayo Model and the Child-Pugh Classification/Score. Increasing levels of Child-Pugh classification and Child-Pugh score were associated with an increased risk of death by univariate analysis. However, on multivariate analysis, neither the Child-Pugh classification nor Child-Pugh score could accurately predict survival. This was especially so in patients with Child-Pugh class B with impaired renal function. In support of this, there were 12 patients with a Child-Pugh score of 8 or 9 with creatinine greater than 1.6 in whom the Mayo model predicted a survival of less than 3 months 9 of these 12 patients died within that period, 3 of procedure-related complications leading to liver failure. Procedure-related intra-abdominal bleeding occurred in 2 patients, whereas the third in whom TIPS was performed for refractory ascites had pulmonary edema and subsequently TABLE 5. Underlying Survival Function for the Survival Model in Table 4 for Patients Undergoing Elective TIPS t (days) S o (t) S o (t) gives the estimated survival probability for an average patient undergoing elective TIPS. The average patient has a risk score of To calculate the survival of a given patient, use the equation S (t) S o (t) exp(r 1.127). R is calculated from the model given in Table 4.

6 HEPATOLOGY Vol. 31, No. 4, 2000 MALINCHOC ET AL. 869 FIG. 3. Survival of 71 independent TIPS patients from the Netherlands who were stratified according to their risk score into two risk groups, namely a high risk group with a median survival less than 3 months (R 1.8) and a low risk group with a median predicted survival more than 3 months (R 1.8). Actual (Kaplan-Meier) and expected survival using the Mayo model were compared using the one sample log rank test. For the low- and high-risk patients, the observed and expected survival were similar (P.88 and P.41, respectively). liver failure. On the other hand, there were 63 patients with a Child-Pugh score greater than 10 with normal renal function. The high Child-Pugh score would suggest a poor outcome; however, our model correctly predicted survival longer than 3 months in 52 of these 63 patients. DISCUSSION TIPS functions in effect like a side-to-side portocaval shunt and promptly reduces portal pressure. The decrease in hepatic sinusoidal pressure decreases the risk of variceal bleeding as well as the formation of ascites. However, by diverting portal blood flow, TIPS may worsen liver function as well as increase the risk of hepatic encephalopathy in these patients. Some patients do well after TIPS whereas others fair poorly and their survival might, in fact, be reduced. Thus, it is important to be able to predict the patient who is most likely to benefit from TIPS. In those patients in whom predicted survival can be counted in days (or weeks), the procedure might be best performed only as a bridge to liver transplantation. In this study, we have shown that a model using creatinine, bilirubin, INR for prothrombin time, and the cause of liver disease may accurately predict patients who have a median survival of 3 months after the elective procedure (R 1.8). This model was developed using a heterogeneous group of patients, from geographically diverse areas of the United States and accurately predicted survival in an independent group of patients from The Netherlands. The Child-Pugh classification, 10 used to stratify patients, is based on arbitrary criteria and does not quantify a patient s TABLE 6. Predicting Three-Month Mortality Post-TIPS Predicted 3-Month Mortality* 0%-20% 20%-40% 40%-60% 60%-80% 80%-100% Number of patients Number of deaths Observed mortality 9% 31% 69% 61% 93% *Predicted mortality based on logistic regression model as presented in Fig. 3. expected survival. The Child-Pugh classification is easy to calculate and can differentiate between patients with poor liver function and preserved liver function. It is not as useful when applied to patients undergoing TIPS as the majority of patients are of Child-Pugh class C, and the classification does not discriminate within class C. Also, the Child-Pugh classification uses serum bilirubin, albumin, prothrombin time, presence of ascites, and encephalopathy. Bilirubin and prothrombin time measured within 24 hours of TIPS may accurately reflect underlying liver function, but albumin, ascites, and encephalopathy are subject to medical intervention. Moreover, the Child-Pugh classification treats all patients with a bilirubin greater than 3 mg/dl (10 mg/dl if they are cholestatic), or prothrombin time greater than 6 seconds prolonged, or albumin less than 2.8 g the same. Clearly, the prognosis of a patient of cirrhosis with a bilirubin of 4 mg/dl is better than that of a patient with a bilirubin of 20 mg/dl, but the Child-Pugh classification does not recognize this. Albumin tends to be low immediately after a bleed and after red cell transfusions. Further, there are difficulties as well as considerable observer variation in scoring ascites and encephalopathy. 11 During vigorous fluid resuscitation, ascites can become more prominent. Encephalopathy is also more common after a bleed but can respond promptly to treatment such as lactulose. 23 Moreover, ascites and encephalopathy rely heavily on physical examination for diagnosis. Consequently, within the span of 24 hours, a patient s Child-Pugh classification can change by one class depending on how carefully the patient is examined. 11 Another difficulty is that the Child- Pugh classification uses prothrombin time in seconds. With the use of the INR for prothrombin time and the reporting of results using an INR, 17 the scoring becomes more difficult. Both the prothrombin time in seconds as well as the prothrombin index expressed as a percentage of control vary depending on the sensitivity of the thromboplastin used. If a hypothetical patient at the Mayo Clinic (where the control for the prothrombin time is 10 seconds and the ISI of thromboplastin 1.0) had a prothrombin time of 15 seconds, that same patient at the Boston center (where the control is 12 seconds

7 870 MALINCHOC ET AL. HEPATOLOGY April 2000 and the ISI 2.5) would have a prothrombin time of 14 seconds. The same patient under identical circumstances would get a score of 2 for the prothrombin time at Mayo under the Child-Pugh score, but only 1 in Boston. Finally, in our study we have shown that neither the Child-Pugh class nor the Child-Pugh score can predict survival after TIPS as accurately as the risk factors used in the Mayo model, especially in patients with impaired renal function. Our model uses bilirubin, creatinine, and INR for prothrombin time, which are objective and tend to be stable. Bilirubin is the most robust of the variables in the Child-Pugh classification. We also used total bilirubin, which derives from both hemolytic factors as well as hepatocellular function and biliary excretion. It is known that hemolysis, which increases unconjugated bilirubin, is a poor prognostic factor in patients with alcohol-related liver disease. 24 Moreover, bilirubin is probably the closest of the liver function tests to a true test of liver function. 11 Renal dysfunction carries a poor prognosis in patients with chronic liver disease, 15,25,26 and in agreement with our findings, other studies have noted the association of elevated creatinine with poor survival in patients undergoing TIPS. 4 It is notable that for patients with end-stage liver disease not undergoing TIPS, other reports have shown that prolonged prothrombin time and elevated creatinine were associated with a higher risk of dying. 25 The Child-Pugh classification has the drawback that it does not use creatinine, which is an important determinant of survival. The cause of the liver disease is the fourth prognostic variable in our model, a feature that the Child-Pugh classification lacks. From our data, it was not possible to come up with a single model for all liver diseases. Difficulties arise when a patient has a combination of two or more causes for the liver disease, such as alcohol abuse and hepatitis C. For the purposes of our model, the more likely cause of the liver disease was used. For instance, a patient with alcohol abuse and hepatitis C is presumed to have the cause of liver disease to be hepatitis C and a score of 1 should be used. The improved outcome of patients with cholestatic or alcoholrelated liver disease compared with those with other diseases such as viral hepatitis might suggest either that viral hepatitis has a more rapidly progressive course, or that patients with viral hepatitis have more impaired hepatocellular function and thus are less tolerant to portosystemic shunting. An obvious explanation in patients with alcohol-related liver disease might be abstinence from alcohol following the procedure, thus, accounting for improvement in liver function and survival. We could not accurately assess the period of abstinence before TIPS in these patients. Although the 58 patients in the study who had intractable ascites as the indication for placement of TIPS had an increased risk of death, refractory ascites was not an independent risk factor for survival. We noted that our model did not significantly change when the variceal bleed patients and refractory ascites patients were studied separately. That is, the interaction of the risk factors and the group variable did not significantly improve the model (P.199). This suggests that TIPS, whether performed for ascites or for variceal bleeds, has the same determinants of outcome; namely, creatinine, bilirubin, INR, and cause of liver disease. In accordance with this report, a recently published study on the outcome of ascites after TIPS found the survival to be shorter in those patients who had a creatinine level of greater than 1.8 mg/dl and a bilirubin level of greater than 1.3 mg/dl. 4 The hypothesis that patients in whom partial blood flow is maintained after portosystemic shunts have a better outcome has intuitive appeal. 27 However, we found determinants of portal hypertension; namely, direction of blood flow (hepatopetal vs. hepatofugal), initial portocaval pressure gradient, post-tips portacaval pressure gradient, and decrease in the portacaval pressure gradient postshunt had no effect on survival. Although surprising, these data are consistent with findings from other studies. 9 Patients in whom liver transplantation was performed were censored at the date of transplantation; that is, the date of transplantation was considered the last date of follow-up. This is because we believe that survival after transplant is likely to be related to factors other than placement of TIPS. This hypothesis would need to be tested in a large group of patients with TIPS who have undergone OLT. Short-term studies, however, show no benefit of TIPS on operative mortality or blood transfusion requirements over patients who have not previously undergone TIPS. 28 An ideal survival model should use commonly available determinants, be easily calculated, and be widely applicable to a geographically diverse patient. We studied patients from four centers in the United States having a wide range of age (18-81 years), of bilirubin ( mg/dl), and creatinine ( mg/dl). The model was validated in an independent set of patients from a different continent. Moreover, a nomogram that can be used at the bedside and does not require either a calculator or computer skills can predict those patients not likely to survive 3 months post-tips, both in the United States, as well as in the Netherlands. Such patients with poor survival may die before they receive a liver transplant. Our model can be applied to predict longer survival too patients with a risk score R 1.5 have a median survival of 6 months and patients with R 1.3 have a median survival of 12 months. The model is likely to lose its accuracy the longer a patient is followed, because other factors such as development of shunt stenosis, or a hepatocellular carcinoma may negatively impact survival. Our fourvariable model could be used for predicting the survival of TIPS candidates and determining candidacy for liver transplantation. The model uses easily determined and objective variables. The utility of this model in predicting survival in patients with chronic liver disease not undergoing a TIPS procedure is presently being studied. However, before the model can be widely accepted it will need to be validated by other investigators in independent groups of patients undergoing TIPS. Although continued refinement and improvement in the model is anticipated by the authors, its present form may be useful in assisting clinicians in the selection and timing of TIPS. We recommend that it be used to counsel patients and their families before carrying out TIPS. Acknowledgment: The authors thank E.R. Dickson, M.D. and T. Therneau, Ph.D. for constructive criticism, and Linda Veer for expert secretarial assistance. REFERENCES 1. Kamath PS, McKusick MA. Transvenous intrahepatic portosystemic shunts. Gastroenterology 1996;111: Burroughs AK, Hamilton G, Philips A, Mezzanotte G, McIntire N, Hobbs KEF. A comparison of sclerotherapy with staple transection of the esophagus for the emergency control of bleeding from esophageal varices. N Engl J Med 1989;321:

8 HEPATOLOGY Vol. 31, No. 4, 2000 MALINCHOC ET AL Rossle M, Haag K, Ochs A, Sellinger M, Noldge G, Berarnaug M, Berger E, et al. The transjugular intrahepatic portosystemic stent shunt procedure for variceal bleeding. N Engl J Med 1994;330: Ochs A, Rossle M, Haag K, Hauenstein KH, Derbert P, Siegersteter V, Huonker M, et al. The transjugular intrahepatic portosystemic stentshunt procedure for refractory ascites. N Engl J Med 1995;332: Sanyal AJ, Freedman AM, Luketic VA, Purdum PP, Shiffman ML, Tisnado J, Cole P. Transjugular intrahepatic portosystemic shunts for patients with active variceal hemorrhage unresponsive to sclerotherapy. Gastroenterology 1996;111: Patch D, Nikolopoulou V, McCormick A, Dick R, Armonis A, Wannamethee G, Burroughs A. Factors related to early mortality after transjugular intrahepatic portosystemic shunt for failed endoscopic therapy in acute variceal bleeding. J Hepatol 1998;28: Sanyal AJ, Freedman AM, Luketic VA, Purdum PP, Shiffman ML, DeMeo J, Cole PE, et al. The natural history of portal hypertension after transjugular intrahepatic portosystemic shunts. Gastroenterology 1997; 112: Jabbour N, Zajko AB, Orons PD, Irish W, Bartoli F, Marsh WJ, Dodd GD 3rd, et al. Transjugular intrahepatic portosystemic shunt in patients with end-stage liver disease: results in 85 patients. Liver Transplant Surg 1996;2: Rouillard SS, Bass NM, Roberts JP, Doherty CA, Geel, Bacchetti P, Somberg KA. Severe hyperbilirubinemia after creation of transjugular intrahepatic portosystemic shunts: natural history and predictors of outcome. Ann Int Med 1998;128: Pugh RNH, Murray-Lyon IM, Dawson JL, Pictioni MC, Williams R. Transection of the esophagus for bleeding oesophageal varices. Br J Surg 1973;60: Conn HO. A peek at the Child-Turcotte classification. HEPATOLOGY 1981;1: Dickson ER, Grambsch PM, Fleming TR, Fisher LD, Langworthy A. Prognosis in primary biliary cirrhosis: model for decision making. HEPATOLOGY 1989;10: Dickson ER, Murtaugh PA, Wiesner RH, Grambsch PM, Fleming TR, Ludwig J, LaRusso NF, et al. Primary sclerosing cholangitis: refinement and validation of survival models. Gastroenterology 1992;103: Everhart JE, Lombardero M, Detre KM, Zetterman RK, Wiesner RH, Lake JR, Hoofnagle JH. Increased waiting time for liver transplantation results in higher mortality. Transplantation 1997;64: Arroyo V, Gines P, Gerbes AL, Dudley FJ, Gentilini P, Laffi G, Reynolds TB, et al. Definition and diagnostic criteria of refractory ascites and hepatorenal syndrome in cirrhosis. HEPATOLOGY 1996;23: Conn HO, Leevy CM, Vlahcevic ZR, Rodgers JB, Maddrey WC, Seeff L, Levy LL. Comparison of lactulose and neomycin in the treatment of chronic portosystemic encephalopathy. A double-blind controlled trial. Gastroenterology 1977;72: Ven der Besselarr AM. Precision and accuracy of the international normalized ratio for prothrombin time in oral anticoagulation control. Hemostasis 1996;26: Cox DR. Regression models and life-tables. J R Stat Soc [B] 1972;34: Downton F. Discussion. J R Stat Soc [B] 1972;34: SAS Institute SAS User s Guide: Volume 1. Cary, North Carolina: SAS Institute, Statistical Sciences, Inc. S-Plus Reference Manual Version 3.2 Seattle: Stat Sci, A Division of MathSoft, Inc., Kaplan EL, Meier P. Non-parametric estimation from incomplete observations. J Am Stat Assoc 1958;53: Harrington DP, Fleming TR. A class of rank test procedures for censored survival data. Biometrika 1982;69: Riordan SM, Williams R. Current concepts. Treatment of hepatic encephalopathy. N Engl J Med 1997;337: Grahn EP, Dietz AA, Stefani SS. Burr cells, hemolytic anemia and cirrhosis. Am J Med 1968;45: Cooper GS, Bellamy P, Dawson NV, Desbiens N. Fulkerson WJ Jr, Goldman L, Quinn LM, et al. A prognostic model for patients with end-stage liver disease. Gastroenterology 1997;113: Gonwa TA, Klintmalm GB, Levy M, Jennings LS, Goldstein RM, Husberg BS. Impact of pretransplant renal function on survival after liver transplantation. Transplantation 1995;59: Sarfeh IJ, Rypins EB, Mason GR. A systematic appraisal of portacaval H-graft diameters. Clinical and Hemodynamic Perspectives. Ann Surg 1986;204: Somberg KA, Lombardero MS, Lawlor SM, Ascher NL, Lake JR, Wiesner RH, Zetterman RK. A controlled analysis of the transjugular intrahepatic portosystemic shunt in liver transplant recipients. The National Institute of Diabetes and Digestive and Kidney Diseases (NIDDK) Liver Transplantation Database. Transplantation 1997;63: