Article Text

Download PDFPDF

Original article
Reliable prediction of clinical outcome in patients with chronic HCV infection and compensated advanced hepatic fibrosis: a validated model using objective and readily available clinical parameters
  1. Adriaan J van der Meer1,
  2. Bettina E Hansen1,
  3. Giovanna Fattovich2,
  4. Jordan J Feld3,
  5. Heiner Wedemeyer4,
  6. Jean-François Dufour5,
  7. Frank Lammert6,
  8. Andres Duarte-Rojo3,
  9. Michael P Manns4,
  10. Donatella Ieluzzi7,
  11. Stefan Zeuzem8,
  12. W Peter Hofmann8,
  13. Robert J de Knegt1,
  14. Bart J Veldt1,
  15. Harry L A Janssen1,3
  1. 1Department of Gastroenterology and Hepatology, Erasmus MC University Medical Center Rotterdam, Rotterdam, The Netherlands
  2. 2Department of Medicine, University of Verona, Verona, Italy
  3. 3Liver Centre, Toronto Western Hospital, University Health Network, Toronto, Ontario, Canada
  4. 4Departments of Gastroenterology, Hepatology, and Endocrinology, Medical School Hannover, Hannover, Germany
  5. 5Hepatology, Department of Clinical Research, University of Bern, Bern, Switzerland
  6. 6Department of Medicine II, Saarland University Medical Center, Homburg, Germany
  7. 7Department of Surgery, University of Verona, Verona, Italy
  8. 8Medizinische Klinik 1, Klinikum der Johann Wolfgang Goethe-Universität, Frankfurt am Main, Germany
  1. Correspondence to Adriaan J van der Meer, Department of Gastroenterology and Hepatology, Erasmus MC, University Medical Center, Gravendijkwal 230, Room Ha 206, 3015 CE Rotterdam, The Netherlands; a.vandermeer{at}erasmusmc.nl

Statistics from Altmetric.com

Significance of this study

What is already known about this subject?

  • Model for End-stage Liver Disease (MELD) and Child–Turcotte–Pugh (CTP) score are used for the assessment of short-term mortality among patients with cirrhosis with end-stage liver disease.

  • Long-term prognosis varies to a great extent among patients with chronic hepatitis C virus (HCV) infection and compensated cirrhosis.

  • Intensive and costly surveillance for cirrhosis-related complications is recommended for all patients with chronic HCV infection and cirrhosis.

What are the new findings?

  • A risk score combining age, sex, platelet count and aspartate aminotransferase/alanine aminotransferase ratio accurately and objectively assesses the long-term mortality risk among patients with chronic HCV infection and advanced hepatic fibrosis.

  • Adding HCV genotype optimised the predictive accuracy of the risk score for clinical disease progression, which could identify patients highly unlikely to develop liver failure or hepatocellular carcinoma (HCC) within 5 years.

  • The risk scores for mortality and clinical disease progression were validated in a large and independent cohort of patients with chronic HCV infection and cirrhosis.

How might it impact on clinical practice in the foreseeable future?

  • The proposed risk scores can be used when counselling the growing population of patients with chronic HCV infection and advanced hepatic fibrosis regarding their prognosis.

  • If further validated, initiation of costly and intensive semi-annual surveillance with ultrasound might be postponed among patients with the lowest risk for cirrhosis-related complications.

Introduction

Chronic hepatitis C virus (HCV) infection is a major cause of liver cirrhosis, liver failure and hepatocellular carcinoma (HCC).1 It is expected that the incidence of HCV-related cirrhosis and its complications will increase substantially during the upcoming years.2 ,3 Among patients with chronic HCV infection and advanced liver disease, antiviral therapy resulting in sustained virological response (SVR) was associated with reduced occurrence of liver failure, HCC and mortality.4–8 Unfortunately, many patients with advanced fibrosis or cirrhosis fail to attain SVR with current interferon-based treatment regimens, even with the addition of direct-acting antiviral drugs for those infected with HCV genotype 1.9–13

Patients with advanced liver disease who do not achieve SVR do not have a good prognosis. A recent meta-analysis indicated these patients have an overall annual risk of 2.9% of developing liver failure, 3.2% of progressing to HCC and 2.7% of dying of liver-related causes.14 Clinical disease progression and survival, however, may vary considerably among patients with compensated advanced liver disease. Thus, reliable risk scores to assess an individual patient's long-term prognosis would be helpful for counselling and clinical decision making. Due to the rising costs of the new antiviral treatment regimens with high virological efficacy, these scores are relevant to assess the clinical efficacy of therapy. Especially with universally high SVR rates, the number needed to treat to prevent hard clinical endpoints is very much dependent on the baseline risk for events.15 Costly antiviral therapy with high cure rates will thus have most clinical effect among patients at highest risk of cirrhosis-related events. Furthermore, although intensive monitoring is currently advised in all patients with HCV-induced cirrhosis, this may not be necessary or cost effective for those with lowest risks.16 ,17 Validated predictive tools for mortality, as a definite and robust clinical endpoint, or cirrhosis-related complications in general are currently lacking.

We previously identified SVR as the strongest factor independently associated with reduced all-cause mortality and liver-related morbidity in our multicentre cohort of interferon-treated patients with chronic HCV infection and advanced hepatic fibrosis who were followed in the long term.4 Patients with HCV-induced cirrhosis and SVR also showed reduction in portal pressure and regression of hepatic fibrosis, which were related to improved clinical outcome.18–21 Together, these data suggest that patients with cirrhosis and SVR should be considered separately from patients with cirrhosis and ongoing HCV infection. Therefore, in this study we focus on patients who did not attain SVR. Among these patients numerous cirrhosis-related events and deaths occurred during follow-up, which enabled identification of prognostic baseline factors. The primary aim of this study was to develop prediction scores based on objective and readily available clinical variables for mortality and clinical disease progression among patients with HCV-induced advanced hepatic fibrosis who failed to attain SVR. Next, we sought to validate the risk scores in an independent cohort of patients with chronic HCV infection and cirrhosis.

Patients and methods

Derivation cohort

The derivation cohort included all consecutive patients with chronic HCV infection who failed to attain SVR on their initial interferon-based treatment between 1990 and 2003 following histological proof of advanced hepatic fibrosis or cirrhosis (Ishak fibrosis score 4–6).22 Patients were treated in five large hepatology units of tertiary care centres in Europe and Canada. Characteristics of this international multicentre cohort and study design have been described in detail previously.4 Briefly, all included patients had compensated liver disease as patients with decompensated liver disease were not eligible for antiviral therapy. Coinfection with HIV or hepatitis B virus (hepatitis B surface antigen and/or HBV DNA positivity) was an exclusion criterion. Survival and occurrence of liver failure, HCC and liver transplantation was retrospectively assessed by reviewing medical charts. In case the follow-up was not complete, the patient was invited to visit the outpatient clinic or, if this was not feasible, the patient or primary care physician was asked to answer a structured questionnaire over the telephone. Baseline laboratory markers of liver disease severity (platelet count, bilirubin, albumin, aspartate aminotransferase (AST), alanine aminotransferase (ALT)) were registered if available 6 months prior to the start of treatment.

The study was conducted in accordance with the guidelines of the Declaration of Helsinki and the principles of Good Clinical Practice. Informed consent was obtained according to the standards of the local ethics committees.

Validation cohort

The validation cohort was derived from a second international multicentre cohort from Europe (EUROHEP cohort), which included all consecutive patients with non-A, non-B chronic hepatitis and compensated biopsy-proven cirrhosis between January 1982 and December 1992 from seven tertiary care centres. The general design of this study has been described previously.23 The EUROHEP cohort included patients with cirrhosis who received interferon treatment during follow-up and patients who remained untreated. Both were included in the current study as these patients had similar clinical outcome.23 A second data collection round updated the follow-up to 1 January 1997, by review of medical charts, consultation of population registries and an invitation for all patients to return to the outpatient clinics for evaluation.24

To be included in the validation cohort, patients from the EUROHEP cohort were required to have confirmed anti-HCV antibodies, which were retrospectively tested. Patients were not included in the validation cohort if they tested negative for HCV RNA or in case the parameters incorporated in the primary risk score for mortality were not available. Furthermore, patients enrolled and followed at the Erasmus Medical Center in Rotterdam were not considered to prevent possible overlap with patients in the derivation cohort.

Study endpoints

The primary outcome of this study was all-cause mortality. Clinical disease progression was analysed as a secondary endpoint, to which liver failure, HCC, liver transplantation or death contributed. In case of multiple events in an individual patient, only the first event was considered for this combined endpoint. Episodes of ascites, bleeding varices, jaundice or overt hepatic encephalopathy were considered as liver failure. The diagnosis of HCC was based on cytohistological confirmation, two coincident imaging techniques (ultrasonography, CT or MRI) showing a focal lesion larger than 2 cm with arterial hypervascularisation or one imaging technique showing a focal lesion larger than 2 cm with arterial hypervascularisation in the presence of an α-fetoprotein level greater than 400 ng/mL.25

Statistical analysis

In the derivation cohort, follow-up started 24 weeks after cessation of antiviral treatment, as at this time the distinction is made between sustained virological responders and non-responders. Since we specifically aimed to assess survival and clinical disease progression among patients without SVR, patients who did attain SVR were not included. In the validation cohort, follow-up started at the time of diagnosis as not all patients received antiviral therapy. In both cohorts patients were censored at the time of SVR due to (re-)treatment, since viral eradication is likely to alter the natural course of advanced hepatic fibrosis.4–7 ,18–21 Otherwise, if the clinical endpoint did not occur, patients were censored at the last follow-up visit.

In the derivation cohort, Cox proportional hazard regression analysis was used to assess the association between baseline factors and time to event. To construct objective risk scores, which can be reliably reproduced, only the objective variables age, sex, body mass index (BMI), HCV genotype, antiHBc status, platelet count, albumin, bilirubin and AST/ALT ratio were considered. These factors have been linked with clinical outcome in the past. To have sufficient power, a cut-off of minimally 10 events per variable was used.26 Linearity of the association of continuous variables was assessed by including polynomial terms of the variables, which remained included in multivariate analyses in case these were statistically significantly associated with the outcome measure. To create parsimonious final models, which are most easily used and reproduced in clinical practice, variables that were no longer statistically significant in multivariate analyses were removed.27 Potential confounding was checked. Bootstrapping with 1000 replications was performed as internal validation analyses to check the stability of variables included in the final models. Linear prediction equations, representing the risk scores, were derived from the final Cox model. For visualisation purposes, three risk groups were created based on the outcome of the risk scores in the derivation cohort. The highest quartile of patients represented the high-risk group, the lowest quartile the low-risk group and the remaining 50% of patients the intermediate-risk group. Cumulative incidence rates of mortality or clinical disease progression among these three groups were determined by Kaplan–Meier analyses and tested with the log-rank test in the derivation cohort. The absolute 2.5-year, 5-year and 7.5-year risks for mortality or clinical disease progression according to the outcome of the risk scores were determined with the baseline hazard as derived from the final Cox models.

In the derivation and the validation cohort the C statistic was used to assess the predictive accuracy of the risk scores.28 ,29 Predicted mortality and clinical disease progression rates were compared with those observed.

To check the stability of the final model, sensitivity analyses using multiple imputation with replacement to impute missing values was performed to construct 10 complete datasets.30 ,31

All statistical tests were two sided and a p<0.05 was considered to be statistically significant. Interaction terms between the variables included in the final models were assessed in the complete case analyses and the multiple imputation analyses. The significance level for interactions was set at p<0.01 to correct for multiple testing. The proportional hazard assumption was checked by assessing the interaction between the variables and time (see online supplementary table S1). SPSS V.17.0.2 (SPSS Inc., Chicago, Illinois, USA) and SAS V.9.2 PROC GENMOD (SAS institute, Cary, North Carolina, USA) were used for all statistical analyses.

Results

Characteristics of the derivation cohort

The derivation cohort consisted of 405 patients out of the 546 patients who started an interferon treatment regimen, as 125 patients attained SVR and 8 patients were lost, 3 were diagnosed with HCC and 5 experienced liver failure before the start of follow-up (see online supplementary figure 1). Baseline characteristics are shown in table 1. Median follow-up duration was 8.1 (IQR 5.7–11.1) years. Follow-up was complete in 359 (89%) patients. Retreatment during follow-up resulted in SVR for 67 additional patients after a median of 5.8 (IQR 3.1–8.5) years and these patients were censored at the time of achieving SVR.

Table 1

Patient characteristics*

In total, 100 patients died: 70 of liver-related causes, 15 of non-liver-related causes and for another 15 the cause of death was not known. The overall 5-year survival probability was 91.4% (95% CI 88.5 to 94.3) (figure 1A). One patient experienced liver failure before initiating retreatment and died during this course of pegylated interferon and ribavirin due to severe infection. Two patients died within 6 months following cessation of a repeated treatment course with pegylated interferon and ribavirin. The first patient, who also had a history of liver failure prior to retreatment, discontinued therapy at week 15 because of a deteriorating liver function and died approximately 2 months later, shortly after being diagnosed with HCC. The other patient received a full 48-week treatment course and died 5 months after therapy following the onset of liver failure.

Figure 1

Overall (A) survival and (B) event-free survival probability in patients with advanced liver disease and chronic hepatitis C virus infection in the derivation cohort (continuous line) and validation cohort (dashed line).

Clinical disease progression was experienced by 169 (42%) patients. The overall 5-year event-free survival was 77.6% (96% CI 73.5 to 81.7) (figure 1B). The first cirrhosis-related complication was liver failure in 87 (51%) patients, HCC in 60 (36%) and death in 19 (11%). Three (2%) patients underwent liver transplantation for progressive liver disease, although no clear event of liver failure or HCC could be registered.

Construction of the mortality risk score

In the derivation cohort, higher age, lower platelet count, higher total bilirubin level, lower albumin level and higher AST/ALT ratio were significantly associated with mortality in univariate Cox regression analyses (table 2). Multivariate analyses showed that bilirubin and albumin were not independently associated with mortality (table 2, model A). Since HCV genotype was no longer significantly associated with mortality after bilirubin and albumin were excluded (HR 1.842, 95% CI 0.922 to 3.682, p=0.084), it was removed to create the final model (table 2, model B). Also after bootstrapping and multiple imputation analyses all variables included in the final model remained statistically significantly associated with mortality. The final analysis included 305 (75%) representative patients of the derivation cohort, for whom platelet count and AST/ALT ratio were available at baseline. The 305 patients included in the analyses did not differ significantly from the 100 patients who were not included in the analyses with respect to other baseline characteristics of liver disease severity, age, and survival (log rank test, p=0.957) (see online supplementary figure S2 and table S2).

Table 2

Cox regression analyses for mortality in the derivation cohort

The linear prediction equation for mortality as derived from the final Cox model was represented by: Rm=(6 * age in years) − (platelet count per 109/L) + (258.8 * log10(AST/ALT)) + (64.5 for male patients).

The predictive accuracy of this risk score for mortality, as measured by the C statistic, was 0.78 (95% CI 0.72 to 0.83). To visualise cumulative mortality rates, patients were categorised into three risk groups according to the IQRs of Rm. After 5 years of follow-up the cumulative mortality rate was 0% in the low-risk group (Rm<87.1), 7.2% (95% CI 2.9 to 11.5) in the intermediate-risk group (87.1≤Rm≤221.2) and 22.2% (95% CI 12.6 to 31.8) in the high-risk group (Rm > 221.2) (p<0.001) (figure 2A).

Figure 2

The mortality risk score is represented by Rm=(6 * age in years) – (platelet count per 109/L)+(258.8 * log10(AST/ALT))+(64.5 for male patients). (A) Cumulative mortality rate according to three risk groups based on the IQRs of Rm in the derivation cohort. The low-risk group included 76 (25%) patients with Rm<87.1, the intermediate-risk group included 153 (50%) patients with Rm≥87.1 and ≤221.2, and the high-risk group included 76 (25%) patients with Rm>221.2. Mortality rates were determined by Kaplan–Meier analysis and compared by the log rank test. (B) Absolute risk for mortality at 2.5, 5 or 7.5 years as a function of the mortality risk score, which was estimated with the baseline survival derived from the final Cox model in the derivation cohort (absolute mortality risk=100* (1 – (baseline survival(t)  ×)); with x=exp(Rm/100)).

To assess a patient's individual likelihood for mortality, figure 2B illustrates the estimated 2.5-year, 5-year and 7.5-year mortality risk as a function of the mortality risk score. The baseline survival, as derived from the final Cox model, was 0.9965, 0.9870 or 0.9752 at 2.5 years, 5 years or 7.5 years, respectively (under the condition that Rm=0). Among the patients with an estimated 5-year mortality risk < 5% (n=130 (43%), mortality risk score < 137), the observed 5-year cumulative mortality rate was 0.9% (95% CI 0.0 to 2.7) (table 3). The observed cumulative 2.5-year and 7.5-year mortality rates were also in line with the predicted estimates (see online supplementary table S3).

Table 3

Predicted against observed 5-year mortality

Validation of the mortality risk score

Of the 319 patients who were eligible to be included in the validation cohort, 23 were not considered because the baseline platelet count and/or AST/ALT ratio were missing. Table 1 summarises the characteristics of the remaining 296 patients with cirrhosis. During a median follow-up of 6.6 (IQR 4.4–9.0) years, 166 (56%) patients underwent at least one interferon-based treatment course. Eight (5%) patients attained SVR. In total, 58 patients died and the overall 5-year survival probability was 88.1% (95% CI 84.2 to 92.0) (figure 1A). In the validation cohort, the C statistic of the mortality risk score was 0.76 (95% CI 0.69 to 0.83). Table 3 and online supplementary table S3 show the predicted against the observed mortality rates, which were comparable to those in the validation cohort.

As a sensitivity analysis, 63 (21%) patients with non-A, non-B chronic hepatitis were excluded in whom HCV RNA was not assessed. In this subgroup the C statistic of the mortality risk score remained similar (0.75, 95% CI 0.65 to 0.84).

Construction of a risk score for clinical disease progression

With a similar approach a prediction score for clinical disease progression was constructed in the derivation cohort. Table 4 shows the results of univariate and multivariate Cox regression analyses. In contrast to mortality, the association between clinical disease progression and platelet count was nonlinear. Also, HCV genotype 3 versus non 3 remained an independent predictor of cirrhosis-related events and was thus included in the final model (table 4, model B). All variables included in this model remained statistically significantly associated with clinical disease progression after bootstrapping and multiple imputation analyses.

Table 4

Cox regression analyses for clinical disease progression in the derivation cohort

The risk score for clinical disease progression is represented by: Rc=(5.2 * age in years) − (2.8 * platelet count per 109/L)+(0.00517 * (platelet count per 109/L)2)+(358.2 * log10(AST/ALT)) + (83.7 for male patients) + (60.6 in case of HCV genotype 3). The C statistic of this model was 0.80 (95% CI 0.76 to 0.83). According to the interquartile ranges of Rc, the 5-year cumulative event rate was 1.6% (95% CI 0.0 to 4.7) in the low-risk group (Rc<−91.9), 13.8% (95% CI 8.1 to 19.5) in the intermediate-risk group (−91.9≤Rm≤61.0) and 53.1% (95% CI 41.3 to 64.9) in the high-risk group (Rm>61.0) (p<0.001) (figure 3A).

Figure 3

The risk score for clinical disease progression is represented by Rc=(5.2 * age in years) – (2.8 * platelet count per 109/L)+(5.17 – 3 * (platelet count per 109/L)2)+(358.2 * log10(AST/ALT))+(83.7 for male patients)+(60.6 in case of HCV genotype 3). (A) Cumulative clinical disease progression rate according to three risk groups based on the IQRs of Rc in the derivation cohort. The low-risk group included 72 (25%) patients with Rc<−91.9, the intermediate-risk group included 146 (50%) patients with Rc≥−91.9 and ≤61.0, and the high-risk group included 72 (25%) patients with Rm>61.0. Clinical disease progression rates were determined by Kaplan–Meier analysis and compared by the log rank test. (B) Absolute risk for clinical disease progression at 2.5, 5 or 7.5 years as a function of the clinical disease progression risk score, which was estimated with the baseline event-free survival derived from the final Cox model in the derivation cohort (absolute risk of clinical disease progression risk = 100* (1 – (baseline event-free survival(t)  ×)); with x=exp(Rc/100)).

The individual 2.5-year, 5-year and 7.5-year risk for clinical disease progression according to the clinical disease progression risk score is illustrated in figure 3B. The baseline event-free survival, as derived from the final Cox model, was 0.9139, 0.8330 or 0.7509 at 2.5-years, 5-years or 7.5-years, respectively (under the condition that Rc=0). Among the 54 (19%) patients with an estimated annual event rate below 1% during the first 5 years of follow-up (Rc<−127), no cirrhosis-related events were observed (table 5). In these patients the observed cumulative rate with clinical disease progression was 2.5% (95% CI 0.0 to 7.4) after 7.5 years (see online supplementary table S3).

Table 5

Predicted against observed 5-year clinical disease progression rate

The mortality risk score correlated strongly with the clinical disease progression risk score (Pearson's R=0.93, p<0.001). However, the accuracy of the mortality risk score for the prediction of clinical disease progression was lower (C statistic=0.78, 95% CI 0.74 to 0.82).

Prediction of clinical disease progression in the validation cohort

In the validation cohort, 103 patients showed clinical progression of their liver disease, with an overall 5-year event-free survival probability of 77.3% (95% CI 72.2 to 82.4) (figure 1B). The C statistic of the risk score for clinical disease progression was 0.74 (95% CI 0.68 to 0.79) in the validation cohort. The observed cumulative incidences of events corresponded very well to those predicted (table 5 and see online supplementary table S4).

Discussion

In our multicentre follow-up study among patients with chronic HCV infection and advanced hepatic fibrosis or cirrhosis we found that, during a median follow-up of 8.1 years, 25% of the patients who did not attain SVR died and 42% experienced clinical disease progression. Readily available and objective variables were used to develop a reliable risk score for mortality, including the readily available clinical parameters age, sex, platelet count and AST/ALT ratio. This score accurately predicted the long-term mortality risk among patients with chronic HCV infection and advanced hepatic fibrosis who did not attain SVR. Importantly, the predictive accuracy was validated in a large and independent international cohort of patients with chronic HCV infection and cirrhosis.23 Prediction of clinical disease progression was optimised with a comparable but separate risk score, for which the weighing of the variables was slightly adjusted. As expected, the risk score for clinical disease progression included similar variables as liver-related morbidity is closely related to mortality. Inclusion of HCV genotype was anticipated as well, as HCV genotype 3 has previously been associated with fibrosis progression and HCC.32 ,33 Although the differences between the scores are modest, separate scores led to the highest predictive accuracies for the respective outcomes.

Our risk scores represent the first validated predictive tools to specifically assess long-term prognosis among patients with advanced but compensated HCV-induced liver disease. They will be useful for counselling patients regarding their prognosis and possibly to determine the benefit that might be expected from a repeated attempt to eradicate chronic HCV infection. Furthermore, the scores might be useful for the allocation of costly and extensive semi-annual surveillance with ultrasound for cirrhosis-related complications. Current guidelines state that such surveillance is only cost effective for patients with chronic HCV infection if the annual HCC incidence exceeds 1.5%.16 Careful allocation is relevant, especially as the population of patients with HCV-induced cirrhosis is rapidly growing.2 ,3

As it was specifically designed to assess long-term mortality in patients with compensated cirrhosis, the proposed mortality risk score represents a valuable prognostic tool in addition to the Child–Turcotte–Pugh (CTP) score and the Model for End-stage Liver Disease (MELD) which are mainly used for prediction of short-term mortality in patients with advanced or decompensated cirrhosis. Both these scores were originally developed to assess the mortality risk of invasive shunting procedures among patients with cirrhosis with severe portal hypertension.34 ,35 Although the CTP score has shown predictive value for non-operative mortality in patients with cirrhosis, it was never actually validated to be used for this purpose.36–39 In addition, the CTP score contains subjective components and lacks statistical weighting of the included variables. The more sophisticated MELD is free of these limitations and has replaced CTP for the allocation of donor liver grafts after it was validated to predict mortality in patients with end-stage liver disease.34 ,40 ,41 However, this mainly concerns prediction of short-term mortality due to the high mortality rates in these validation studies as 2–21% of the patients in the study by Kamath et al and 12% of the patients in the study by Wiesner et al died within 3 months.40 ,41 In comparison, among our derivation and validation cohort only 1% of the patients had died within 1 year of follow-up. The performance of MELD for long-term clinical outcome could not be reliably assessed in our cohort as baseline creatinine and INR were available for only a minority of the patients (23%). However, others indicated the predictive accuracy of MELD was limited among patients with compensated cirrhosis and for the assessment of mortality beyond 3 years.42 ,43 In contrast to the CTP and MELD score, our risk scores do not include bilirubin or albumin as these markers were not independent predictors of long-term clinical outcome. Deteriorating liver function leading to death in patients with end-stage liver disease is often accompanied by elevated bilirubin and lowered albumin levels; however in patients with compensated advanced liver disease, who were included in this study, the liver function can remain stable over long periods of time. Our scores are thus meant to be used for the guidance of patients with compensated HCV-induced cirrhosis and preserved liver function. For patients with decompensated cirrhosis the MELD and CTP score remain the most useful tools for the assessment of mortality.

In a recent post-hoc analysis of the HALT-C trial, another objective prediction score was suggested to assess the likelihood of disease progression in patients with chronic HCV infection and advanced hepatic fibrosis who did not attain SVR.44 While the prospective nature of this study is important, the risk score was never validated. Several other differences compared with our study are relevant to discuss as well. First, the HALT-C score was not solely based on clinical events. In fact, increases in CTP score ≥7 contributed to 66% of the composite endpoints studied, while only 48% of these patients actually progressed to having liver failure with a delay of up to 3.1 years. As we are lacking data regarding increases in CTP score, it would not be valid to assess the performance of the HALT-C model in our cohort. Second, although early detection of HCC is one of the key reasons for surveillance, this important complication was excluded from the combined endpoint in HALT-C analyses. The endpoints assessed in our study might thus be considered to be clinically more relevant. Third, age was no independent predictor of disease progression in the HALT-C study, possibly because this trial included a selected population with narrow age range.45 However, higher age has been repeatedly associated with worse clinical outcome.5–7 ,23 ,46 ,47 In contrast, our derivation cohort included all consecutively treated patients, including many patients treated outside of clinical trials. Consequently, our risk score was based on patients who are representative of the general Western population with HCV-induced advanced liver disease.

Importantly, the risk scores performed very well in the validation cohort, despite several differences between the derivation and validation cohort, which indicates that our risk scores are consistent and robust. First, patients included in the validation cohort were older compared with the patients in the derivation cohort, and all patients in the validation cohort had cirrhosis, whereas the derivation cohort also included patients with advanced fibrosis. Second, in contrast to the derivation cohort, patients in the validation cohort were required to have abnormal ALT or AST levels. However, the proportion of patients with HCV-induced cirrhosis and normal aminotransferase levels is likely to be small.48 Third, almost half of the patients in the validation cohort were not treated with interferon-based antiviral therapy, while interferon-based treatment was an inclusion criterion in the derivation cohort. A proportion of the untreated patients in the validation cohort are thus likely to have attained SVR with antiviral therapy, and these patients might have a beneficial prognosis. However, as only 5% of the treated patients with cirrhosis in the validation cohort cleared the virus, this is not likely to have influenced current results. As the efficacy of the earlier interferon-based therapies has been limited among patients with advanced hepatic fibrosis, those patients without SVR in current long-term follow-up studies can be considered to represent the natural history.23 ,49 Therefore, our risk scores also have the potential to perform well among untreated patients. Caution is currently required when treating patients with specific comorbidities which contraindicate interferon therapy and also influence prognosis. Of course, further validation by independent groups and impact and implementation analyses will need to be conducted.50 Assessing the predictive accuracy of our risk scores in patients with advanced hepatic fibrosis due to causes other than chronic HCV infection would be relevant as well. However, differences in the natural history prior to the stage of decompensated cirrhosis might limit generalisation of our risk scores to all chronic liver diseases.

There are certain limitations regarding the present study. Both cohorts included patients from tertiary referral centres, in which patients might have more advanced disease compared with the population with HCV-induced cirrhosis as seen in secondary care hospitals. However, patients with the most advanced liver disease were generally not treated with interferon-based therapy and were therefore not included in our study. The studied cohorts here are thus likely to be representative of the general population with chronic HCV infection and cirrhosis. As discussed above, for those patients with impaired liver function, the MELD and CTP score are probably more relevant. As we aimed to construct objective scores, which can be reliably assessed in daily practice, we did not consider several subjective factors which were previously found to be associated with clinical outcome.4 Diabetes mellitus was not included, as this was not prospectively assessed and fasting glucose, haemoglobin A1c and/or insulin levels were not available for the majority of patients in our study. However, the need for patients to come in fasting would decrease the practicality of the risk score. Limited data concerning the effect of antidiabetic therapy on clinical outcome also supports the exclusion of insulin resistance as an objective risk factor at this time. Nevertheless, the presence of diabetes mellitus was significantly associated with mortality (HR 2.59, 95% CI 1.44 to 4.64, p=0.001) and showed a trend for an association with clinical disease progression (HR 1.67, 95% CI 1.00 to 2.78, p=0.051) when added to our final Cox models. These results could not be confirmed in the validation cohort due to a lack of data on insulin resistance. Self-reported alcohol use and histological degree of hepatic fibrosis, which requires invasive liver biopsy, were not considered as these variables are not objective. However, active alcohol use influences the AST/ALT ratio, and might thus be indirectly accounted for in our risk scores. It can, nevertheless, be anticipated that the clinical outcome is worse than estimated by our risk scores among patients who continue to abuse alcohol. As expected, in our study a history of alcohol abuse as indicated by the treating physician was also independently associated with mortality (HR 1.8, 95% CI 1.06 to 3.21, p<0.032) and clinical disease progression (HR 2.1, 95% CI 1.34 to 3.33, p<0.001). Of course, all patients in our derivation cohort were treated with interferon-based therapy, which is not generally initiated among those with active alcohol abuse. In such patients abstinence should be advocated to improve their prognosis, also because this might enable anti-HCV therapy.51 ,52 As the AST and ALT levels fluctuate, regression dilution bias might have had an impact.53 However, as we included the ratio between these two strongly correlating markers, this might be of lesser concern for the risk scores presented here. Missing data in the derivation cohort was another expected limitation as our study has a retrospective nature and included patients from 1990 onwards. Also, blood works from before 6 months prior to the start of antiviral therapy were not considered. Importantly, the platelet count and AST/ALT ratio were missing at random, which is substantiated in the online supplementary materials (supplementary figure S2 and supplementary table S2), and we have performed imputation analyses. As for HCV genotype, the association between gender and clinical outcome was confounded in our cohort. Female patients were older and had a higher AST/ALT ratio compared with male patients. As expected, multivariate analyses confirmed the association between clinical outcome and HCV genotype or gender. Interestingly, liver stiffness was recently associated with clinical outcome in patients with cirrhosis and is objectively and non-invasively measured by transient elastography.54 Future studies need to evaluate whether the predictive accuracy of our risk score could improve when extended with liver stiffness.

In conclusion, we have developed and validated risk scores for long-term mortality and clinical disease progression in patients with chronic HCV infection and advanced liver disease who failed to attain SVR. The risk scores are based on objective and readily available laboratory markers and patient characteristics, so that they can be easily and reliably reproduced in daily practice.

Acknowledgments

We are grateful to the EUROHEP study group for providing their clinical follow-up data on patients with HCV-induced cirrhosis. Furthermore, we thank Lucille Maarschalkerweerd, Lena A. van Santen, and Melek Polat-Utku for their involvement in the data collection for Rotterdam (The Netherlands), E. Jenny Heathcote for her involved in designing the study and monitoring the data collection for Toronto (Canada), Elizabeth Lee, Bobbi Jo Quigley, Sharlene Camaya and Yvonne Oliveira for their involvement in the data collection for Toronto (Canada), Carola Mix, Markus Cornberg and Frank Grünhage for their involvement in the data collection for Hannover (Germany), and Lorenz Kuske for his involvement in the data collection for Bern (Switzerland).

References

View Abstract

Supplementary materials

  • Supplementary Data

    This web only file has been produced by the BMJ Publishing Group from an electronic file supplied by the author(s) and has not been edited for content.

    Files in this Data Supplement:

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.

Linked Articles