age, serum levels of bilirubin, albumin, and aspartate aminotransferase

Transcription

1 The Relative Role of the Child-Pugh Classification and the Mayo Natural History Model in the Assessment of Survival in Patients With Primary Sclerosing Cholangitis W. RAY KIM, JOHN J. POTERUCHA, RUSSELL H. WIESNER, NICHOLAS F. LARUSSO, KEITH D. LINDOR, JAN PETZ, TERRY M. THERNEAU, MICHAEL MALINCHOC, AND E. ROLLAND DICKSON The Child-Pugh classification is a simple, convenient prognostic measure in patients with liver cirrhosis. We investigated the relative role of the Child-Pugh classification and the Mayo model in the assessment of survival in patients with primary sclerosing cholangitis (PSC). Of the 173 patients described in the original Mayo PSC natural history model, 147 patients had sufficient information in the medical record to allow computation of the Child-Pugh score. We used our most recent modification of the Mayo model to compute the risk score, based on patient s age, serum levels of bilirubin, albumin, and aspartate aminotransferase and history of variceal bleeding. Using the risk score (R), patients were divided into the low- (R F 0), intermediate- (0 I R F 2), and high-risk (R H 2) groups. Kaplan- Meier estimates and proportional hazards analysis were used to evaluate the two prognostic models. Although there was a statistically significant correlation between the Child- Pugh and Mayo risk scores, two-thirds of the patients had a Child-Pugh score of 5 or 6 and a relatively wide range of risk scores ( ). The probability of survival for 7 years in patients in the low-, intermediate-, and high-risk groups was 92%, 74%, and 40% for Child-Pugh class A (n 96) and 100%, 62%, and 28% for Child-Pugh class B patients (n 44), respectively. There were only a small number (n 7) of Child-Pugh class C patients. In our age-adjusted multivariate analysis, each unit increase in the Mayo risk score was associated with a 2.5-fold increase in the risk of death (95% confidence interval: , P F.01), whereas Child-Pugh classification had no significant impact on survival (Child-Pugh B vs. A: risk ratio 1.1 [95% confidence interval: ]; Child-Pugh C versus A: risk ratio 0.6 [95% confidence interval: ]). In contrast to the Child-Pugh classification, which was developed for advanced liver cirrhosis, the Mayo model provides valid survival information, particularly in patients early in the course of PSC. (HEPATOLOGY 1999;29; ) Abbreviation: PSC, primary sclerosing cholangitis. From the Division of Gastroenterology and Hepatology, Section of Biostatistics, Mayo Clinic and Foundation, Rochester, MN. Received December 16, 1998; accepted February 1, Address reprint requests to: W. Ray Kim, MD, Division of Gastroenterology and Hepatology (W19), Mayo Clinic and Foundation, 200 First Street, SW Rochester, MN kim.woong@mayo.edu; fax: Copyright 1999 by the American Association for the Study of Liver Diseases /99/ $3.00/0 The natural history of primary sclerosing cholangitis (PSC) has been the subject of a number of studies. 1-5 Most of the investigators employed the Cox proportional hazards analysis in assessing prognostic variables of patient survival. Certain variables have repeatedly been shown to have prognostic importance in these studies, which include patient age, bilirubin level, hepatosplenomegaly, and histological stage. These models have been used as a clinical tool in measuring the overall disease severity and in the treatment decision making, such as liver transplantation, as well as a research tool for risk stratification or adjustment. 6-8 We published our first model in 1989 to estimate the survival of patients with PSC based on five variables. 1 This study was updated to include expanded data from a number of other institutions. 3 Most recently, we reported a revised natural history model for PSC. This revision was undertaken to enhance the clinical utility of the model by excluding the histological stage as a variable and, thereby, obviating the need for a liver biopsy. 5 Instead, more emphasis was placed on reproducibility and generalizability of the variables. The revised model is based on five variables including the patient s age, serum levels of bilirubin, albumin, and aspartate aminotransferase and a history of variceal bleeding (Table 1). This model has been validated using at least one independent data set. In 1997, Shetty et al. 9 made a provocative observation that the Child-Pugh score is able to predict survival in patients with PSC. They also suggested that once the Child-Pugh score is taken into consideration, the Mayo natural history model may not provide additional prognostic information. They advocated the Child-Pugh score to be used for patient counseling, clinical decision-making, and stratifying risks in therapeutic trials as an alternative to the Mayo model. In this work, we re-examined our data at Mayo Clinic to determine what may be the relative role of the Child-Pugh classification and the Mayo model in the assessment of survival in patients with primary sclerosing cholangitis. To the extent that the Child-Pugh classification was designed to determine prognosis in patients with significant portal hypertension, our hypothesis was that the Child-Pugh classification is an insensitive measure for patients in the early phase of disease progression. Based on the data from patients who were described in the original Mayo natural history study, we examined (1) the degree of correlation between the Child- Pugh score and the Mayo risk score and (2) the extent to which the Mayo model provides additional prognostic information, particularly in patients in the early stages of PSC. 1643

2 1644 KIM ET AL. HEPATOLOGY June 1999 TABLE 1. The Current Mayo Natural History Model for Primary Sclerosing Cholangitis Mayo Risk Score R 0.03 Age (years) 0.54 Log e (total bilirubin [mg/dl]) 0.84 Albumin (g/dl) 0.54 Log e (aspartate transaminase [IU/L]) 1.24 Variceal bleeding (yes 1; no 0) NOTE. The Mayo risk score calculated by this formula may be used to estimate survival up to 4 years of follow-up. 5 PATIENTS AND METHODS Patients and Data Extraction. The original Mayo PSC natural history model was based on data from 173 patients in whom the diagnosis of PSC was carefully confirmed based on established criteria. 10 Although a comprehensive array of patient data was prospectively collected and applied in the development of the survival model, there was insufficient information to determine the Child-Pugh score. Specifically, the degree of ascites and encephalopathy was not recorded. Therefore, we retrospectively reviewed the medical records of the study participants and obtained these variables at the time of enrollment into the study. In patients in whom we had sufficient data, we computed the Child-Pugh score using the modification of the score as proposed by Pugh et al. 11 for primary biliary cirrhosis (Table 2). Patients were categorized into three groups (Child-Pugh class A, B, and C) according to Pugh s criteria. We used our most recent version of the Mayo natural history model for PSC. 5 The formula shown in Table 1 was used to compute the Mayo risk score. Statistical Analysis. The relative role of the Child-Pugh score and the Mayo model as a tool for survival estimation was assessed (1) by the correlation between the two scores, (2) by the ability to distinguish individuals by their risk of death, and (3) by examining whether the Mayo risk score provides additional prognostic information once the Child-Pugh score is taken into account. Because a significant number of patients underwent liver transplantation, survival analysis was undertaken twice: first, censoring patients on the day of liver transplantation (equivalent to the patient last seen on that day and lost to follow-up) and second, treating liver transplantation as failure (equivalent to the patient dying on the day of liver transplantation). The correlation between the Child-Pugh and Mayo risk scores was examined using the Pearson correlation. We used the Kaplan- Meier method in assessing the survival of patients. In describing the survival by the Mayo risk score, we divided patients into three groups. The low-risk group had a risk score of 0 or less, the intermediate-risk group between 0 and 2, and the high-risk group 2 or greater. To examine whether the Mayo risk score provides additional prognostic information in addition to the Child-Pugh classification, we divided Child-Pugh class A and B patients into three groups using the Mayo risk score, using the cut-off criteria described earlier. This analysis was not possible for Child-Pugh class C patients because of the small number (n 7). Finally, we used the Cox proportional hazards analysis to examine TABLE 2. The Pugh Modification of the Child-Turcotte Classification 11 Variable Encephalopathy grade none Ascites absent slight moderate Albumin (g/dl) Prothrombin time [seconds prolonged] Bilirubin (mg/dl) NOTE. The criteria for the bilirubin level adapted for primary biliary cirrhosis are shown. Child-Pugh class A 5or6;B 7to9;C 10 to 15. the relative importance of the Child-Pugh and Mayo risk scores in the assessment of the survival of patients. Three variables were considered, namely patient s age, Child-Pugh classification, and Mayo risk score. The age and Mayo risk score were used as continuous variables, whereas dummy variables were created to code Child-Pugh classification as a categorical variable. Statistical significance was determined using the likelihood 2 test for model improvement. Risk ratios with 95% confidence intervals and twotailed P values are presented. RESULTS Of the 173 patients in our natural history model cohort, 147 (85%) had sufficient data in the medical record to compute the Child-Pugh score. Table 3 summarizes the clinical characteristics of these 147 patients. The demographic characteristics of our patients appear representative of PSC patients (median age, 39 years; 65% male). The majority of patients belonged in the Child-Pugh class A or B and rarely had complications of advanced liver disease and portal hypertension. The median length of follow-up for the entire group was just over 5 years. During the follow-up, 26 patients underwent liver transplantation, and 56 patients died. Figure 1 shows the correlation between the Child-Pugh and Mayo risk scores. There was a statistically significant correlation between the two scores (r.67, P.01). However, the range in the Mayo risk score for a given Child-Pugh score was quite wide. In addition, there were a large number of subjects at the lower end of the Child-Pugh score, namely 5 and 6. Figure 2A shows survival of patients in the low-, intermediate-, and high-risk groups by the Mayo risk score, with liver transplantation censored. Similarly, Fig. 2B shows the survival of patients in Child-Pugh class A, B, and C. Whereas the survival of patients in the three risk groups by the Mayo risk score was widely separated, the survival of Child s B and C overlapped with each other. Results were the same when liver transplantation was treated as failure (data not shown). Figure 3A and B shows the survival of patients by the Mayo risk score in the Child-Pugh class A and B, respectively. The Mayo risk score of patients in the Child-Pugh class A ranged Variable TABLE 3. Patient Characteristics Median (Range) or Proportion (%) n 147 Clinical variables Age (years) 39 (21-76) Sex (% male) 65% Ascites 3.4% Encephalopathy 0% Variceal bleeding 4.1% Laboratory variables Bilirubin (mg/dl) 1.8 ( ) Albumin (g/dl) 3.4 ( ) Prothrombin time (seconds) 10.6 ( ) Aspartate transaminase (IU/l) 82 (14-343) Child-Pugh classification A 65.3% B 29.9% C 4.8% Mayo risk score 0.97 ( ) Deaths 38.1% (n 56) Liver transplantation 17.6% (n 26) Follow-up (months) 62.8 ( )

3 HEPATOLOGY Vol. 29, No. 6, 1999 KIM ET AL FIG. 1. Correlation between Mayo risk score and Child-Pugh score. Although there was a statistically significant correlation between the two scores, the range in the Mayo risk score for a given Child-Pugh score was quite wide. There were a large number of subjects at the lower end of the Child-Pugh score, namely 5 and 6. from 1.1 to 3.0 with a median of 0.6. Those in the Child-Pugh class B had Mayo risk scores ranging from 0.6 to 4.3 with a median of 1.9. Within the same Child-Pugh classification, the survival of the three risk groups were clearly distinguished from one another. Moreover, the configuration of survival curves for the low-, intermediate-, and high-risk groups was essentially identical for Child-Pugh class A and B patients, indicating that to which Child-Pugh class a given patient belonged did not influence the risk of death as determined by the Mayo risk score. Finally, Table 4 summarizes the proportional hazards models. Three variables (age, Child-Pugh classification, and Mayo risk score) were considered. In the univariate analysis, all three variables were statistically significant. However, in the multivariate model, the Child-Pugh Class B or C was no longer associated with an increased risk of death compared with Child-Pugh Class A, whereas age and Mayo risk score remained important factors. Specifically, each unit increase in the risk score was associated with a 2.1-fold increase in the risk of death. DISCUSSION The Child-Pugh classification has been used widely as a disease severity index in patients with liver cirrhosis. For example, it is being used in the minimal listing criteria for liver transplant candidates. 12 The original description of the Child s classification is found in a monograph by Drs. Child and Turcotte, titled, The Liver and Portal Hypertension, published in The purpose of this system was to assess the operative risk in patients undergoing surgical portosystemic shunt. It appears that this score represents collective wisdom accumulated over time by surgeons performing portosystemic shunts. 14,15 In 1973, Pugh et al. 11 used a modified version of the Child-Turcotte classification in describing the outcome of patients undergoing surgical ligation of esophageal varices, by assigning a score ranging from 1 to 3 to each of the variables in the classification. Classes A, B, and C were designated by criteria applied to the sum of the individual scores. While the original Child-Turcotte score was based on five variables including hepatic encephalopathy, ascites, bilirubin, albumin, and nutritional status, Pugh replaced nutritional status with prothrombin time. 11,13 Although the Child-Pugh classification was not developed based on a strong statistical foundation, it has been repeatedly shown to be useful in the assessment of prognosis in patients with liver cirrhosis. Christensen et al. 16 applied the Child- Pugh score in a heterogeneous group of patients with liver cirrhosis. They found that each of the five individual variables as well as the Child class had prognostic significance. In another retrospective study in 620 patients with chronic liver disease of a variety of causes, the Child-Pugh classification was able to distinguish patients by their survival likelihood. 17 A number of investigators tried to improve the classification, by either adding new variables or employing more sophisticated measures to produce a prognostic index In general, these efforts resulted in only a marginal improvement while lacking practicality or generalizability of the existing score. These strengths not withstanding, the Child-Pugh score has two major limitations relevant to our discussion. First, to the extent that the score was designed as a tool to assess the operative risk for shunt operation in patients who had already experienced variceal hemorrhage, the score focuses on patients with decompensated liver cirrhosis and, thus, may not be as useful in patients with less advanced liver disease. This is apparent in our data in which the majority of patients were aggregated at the lower end of the Child-Pugh score. This phenomenon may be characterized as a floor effect, in which the measurement tool (Child-Pugh score, in this case) is not sensitive to detect differences below a certain level (floor) of the variable assessed. Second, the ability of the Child-Pugh classification to distinguish subjects by their mortality risk is rather crude. As shown in our study, there was a wide range in the risk of death among patients within a given Child-Pugh classification. One of the reasons for the lack of precision may be that the Child-Pugh classification has no interval scale properties. In other words, there is no assurance that a one unit increase in one variable (for example, ascites) has the same prognostic implication as one unit increase in another variable (such as the bilirubin level). Thus, patients with the same Child-Pugh score may have a different mortality risk depending on what combination of scores in variable(s) led to that score. In addition to these main caveats, some of the variables in the Child-Pugh score are subject to change with treatment. For example, the degree of ascites or encephalopathy may increase or decrease enough to result in a shift from one Child-Pugh class to another. Finally, the lack of standardization across different laboratories may add to the variability of the Child-Pugh classification. For example, prothrombin time may be measured using many different thromboplastin reagents, which may result in a significant variability in the results. It has been shown that replacing the prothrombin time with the international normalized ratio (INR) may not circumvent this issue. 21 In contrast to these limitations to the Child-Pugh classification, there are a number of advantages to the Mayo model for PSC. First, because it was developed and validated based on patients with wide range of disease severity, the Mayo model

4 1646 KIM ET AL. HEPATOLOGY June 1999 FIG. 2. Survival (Kaplan-Meier estimates) of patients by the Mayo model (A) and Child-Pugh classification (B). Patients undergoing liver transplantation were censored on the day of transplantation. Whereas the Mayo model was able to distinguish patients by survival, the separation among the Child-Pugh classes was not as clear-cut. The cut-off criteria of the Mayo risk score to divide patients into low-, intermediate-, and high-risk groups were 0.0 and 2.0. is applicable to patients with a wide spectrum of disease severity. Thus, in comparison to the Child-Pugh score, it is particularly useful in patients with less advanced disease in which the Child-Pugh classification is insensitive. Second, the Mayo model has an interval scale: a one-unit increment in the risk score has a uniform impact on the risk of death, regardless of how the risk score was reached. Third, unlike the Child-Pugh score, the variables in the Mayo model were selected to minimize variability from medical interventions. Finally, the Mayo model is able to provide survival estimates with relatively narrow confidence intervals. The results of our analysis are in disagreement with those of Shetty et al. 9 who suggested that the Child-Pugh score is more useful in predicting survival than the Mayo model for patients with primary sclerosing cholangitis. There may be several reasons for the discrepancy. First, our patients had less advanced liver disease, as indicated by significantly lower frequencies of patients with ascites (P.01) and encephalopathy (P.04), making the data less conducive to the application of the Child-Pugh score due to the floor effect. Second, Shetty et al. 9 used the previous version of the Mayo model, in which there may be lack of reproducibility in some

5 HEPATOLOGY Vol. 29, No. 6, 1999 KIM ET AL FIG. 3. Survival (Kaplan-Meier estimates) by the Mayo model in patients with Child-Pugh class A (A) and B (B). Within the same Child- Pugh class, the survival of the three risk groups were clearly distinguished from one another. The shape of the three survival curves was very similar for Child-Pugh class A and B patients, suggesting that Child-Pugh classification did not influence the survival determined by the Mayo risk score. Variable TABLE 4. Cox Proportional Hazards Model Univariate Risk Ratio (95% CI) P Multivariate Risk Ratio (95% CI) P Age (per 10 years) 1.8 ( ) ( ).01 Child-Pugh class B versus A 2.5 ( ) ( ).75 C versus A 2.9 ( ) ( ).36 Mayo risk score 2.4 ( ) ( ).01 Abbreviation: CI, confidence interval. of the variables. For example, one may define splenomegaly using different methods (by physical examination or by imaging studies), leading to a reduced reliability of the measurement. In fact, in the univariate analysis by Shetty et al., splenomegaly and histological stage were found to have no significant impact on patient survival. In our current model, we purposefully selected variables that are reproducible and generalizable. Finally, because a sizable proportion of patients underwent liver transplantation at both centers, institutional policy regarding the timing decision for transplantation and the survival results following liver transplantation may account for some of the differences between the two

6 1648 KIM ET AL. HEPATOLOGY June 1999 series. 5,22 For example, there were more liver transplants and less deaths in Shetty s data, reflective of the more recent accrual of patients. We must point out some caveats to our analysis. First, to the extent that we used in this study the data which was used in the creation of the Mayo model, there is likely to be an inherent tendency for the Mayo model to work well in this group of patients. Nonetheless, our data is illustrative of the fact that the Child-Pugh score has a floor effect and that there is a wide range of survival even within one Child-Pugh class. Second, the computation of the Child-Pugh score is much easier than that of the Mayo model, although the formula for the Mayo risk score may be conveniently stored in a scientific calculator or a personal computer. Perhaps, this is why the Child-Pugh score has been incorporated in the minimal listing criteria for PSC patients. We have developed a user-friendly software that can be used to compute the Mayo risk score and estimate survival of PSC patients using our current model. This software has been included in our on-line worksheet, available via the Internet at the worldwide web site of Mayo Clinic (address: int-med/gi/model/mayomodl.htm). In summary, despite a number of limitations, the Child- Pugh score has proved to be useful in risk assessment in patients with chronic liver disease. The utility of the Child- Pugh score is greatest in patients with decompensated liver cirrhosis. For example, the Child-Pugh score is probably a reasonable measure as minimal listing criteria for patients considered for liver transplantation. 12 On the other hand, the Mayo model for PSC has a stronger statistical foundation and is applicable over a much wider spectrum of severity of primary sclerosing cholangitis. Thus, the Mayo model has a particular advantage in patients with less advanced disease. For example, to describe the overall severity of liver disease in PSC in clinical trials, the Mayo model is obviously more efficient classifying patients than the Child-Pugh score because most of the patients enrolling in clinical trials most likely have well-compensated liver disease (hence, Child s A). Moreover, at a time when the median waiting time for liver transplantation approaches 2 or more years, the ability to estimate patient survival early in their course is increasingly important. In conclusion, we propose that the usefulness of the Child-Pugh score in PSC patients is limited to patients in late stage diseases and the Mayo model, particularly our most recent revision continues to be valuable in estimating the natural history of patients with PSC. REFERENCES 1. Wiesner RH, Grambsch PM, Dickson ER, Ludwig J, MacCarty RL, Hunter EB, Hunter EB, et al. Primary sclerosing cholangitis: Natural history, prognostic factors and survival analysis. HEPATOLOGY 1989;10: Farrant JM, Hayllar KM, Wilkinson M, Karani J, Portmann B, Westaby D, Williams R. Natural history and prognostic variables in primary sclerosing cholangitis. Gastroenterology 1991;100: 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: Broome U, Olsson R, Loof L, Bodemar G, Hultcrantz R, Danielsson A, Prytz H, et al. Natural history and prognostic factors in 305 Swedish patients with primary sclerosing cholangitis. Gut 1996;38: Kim, WR, Lindor KD, Poterucha JJ, Benson JT, Therneau TM, Dickson ER. The Mayo risk score increases rapidly in the terminal phase of primary sclerosing cholangitis [Abstract]. Gastroenterology 1998;114: A Lindor KD. Urosodiol for primary sclerosing cholangitis. N Engl J Med 1997;336: Farges O, Malassagne B, Sebagh M, Bismuth H. Primary sclerosing cholangitis: liver transplantation or biliary surgery. Surgery 1995;117(2): Ahrendt SA, Pitt HA, Kalloo AN, Venbrux AC, Klein AS, Herlong HF, Coleman J, et al. Primary sclerosing cholangitis: resect, dilate, or transplant? Ann Surg 1998;227(3): Shetty K, Rybicki L, Carey WD. The Child-Pugh classification as a prognostic indicator for survival in primary sclerosing cholangitis. HEPATOLOGY 1997;25: LaRusso NF, Wiesner RH, Ludwig J, MacCarty RL. Current concepts. Primary sclerosing cholangitis. N Engl J Med 1984;310(14): Pugh RN, Murray-Lyon IM, Dawson JL, Pietroni MC, Williams R. Transection of the oesophagus for bleeding oesophageal varices. Br J Surg 1973;60(8): Lucey MR, Brown KA, Everson GT, Fung JJ, Gish R, Keeffe EB, Kneteman NM, et al. Minimal criteria for placement of adults on the liver transplant waiting list: a report of a national conference organized by the American Society of Transplant Physicians and the American Association for the Study of Liver Diseases. Liver Transplant Surg 1997;3: Child CG, Turcotte JG. Surgery and portal hypertension. In: Child CG, ed. The liver and portal hypertension. Philadelphia: Saunders, 1964; Wantz GE, Payne MA. Experience with portacaval shunt for portal hypertension. N Engl J Med. 1961:265; Conn HO. A peek at the Child-Turcotte classification. HEPATOLOGY 1981;1(6): Christensen E, Schlichting P, Fauerholdt L, Gluud C, Andersen PK, Juhl E, Poulsen H, et al. Prognostic value of Child-Turcotte criteria in medically treated cirrhosis. HEPATOLOGY 1984;4(3): Propst A, Propst T, Zangerl G, Ofner D, Judmaier G, Vogel W. Prognosis and life expectancy in chronic liver disease. Dig Dis Sci 1995;40(8): Siegel JH, Williams JB. A computer based index for the prediction of operative survival in patients with cirrhosis and portal hypertension. Ann Surg 1969;169(2): Milani A, Marra L, Siciliano M, Rossi L. Prognostic significance of clinical and laboratory parameters in liver cirrhosis. A multivariate statistical approach. Hepatogastroenterology 1985;32(6): Adler M, Verset D, Bouhdid H, Bourgeois N, Gulbis B, Le Moine O, Van de Stadt J, et al. Prognostic evaluation of patients with parenchymal cirrhosis. Proposal of a new simple score. J Hepatol 1997;26(3): Robert A, Chazouilleres O. Prothrombin time in liver failure: time, ratio, activity percentage, or international normalized ratio? HEPATOLOGY 1996;24(6): Schluchter MD. Methods for the analysis of informatively censored longitudinal data. Statistics in Medicine 1992;11: