Potential conflict of interest: Nothing to report.
Early prediction of response to therapy in genotype 1 chronic hepatitis C is difficult. Two predictive models, a pretreatment scoring model (PreT-SM) and a fourth week of therapy scoring model (4w-SM) were constructed in a cohort of 104 patients from a single center (estimation cohort) and validated in a cohort of 141 patients from four independent centers (validation cohort). Individual scores were calculated using variables independently associated with sustained virological response (SVR). Baseline viral load, aspartate aminotransferase/alanine aminotransferase ratio, serum cholesterol, and a numerical score for noninvasive estimation of liver fibrosis were included in the PreT-SM; HCV RNA clearance and PreT-SM scores were included in the 4w-SM. Receiver operating characteristic analysis revealed the area under the curve in the estimation cohort and in the validation cohort to be, respectively, 0.856 and 0.847 for the PreT-SM and 0.908 and 0.907 for the 4w-SM. Low scores were associated with SVR, high scores with non-SVR. The best cutoff scores from the PreT-SM (7 and 9.70) identified, respectively, 36% of patients with SVR and 41% of those with non-SVR from the validation cohort, with high accuracy (≥90% positive predictive value [PPV] and specificity). Similarly, cutoff scores of 3.20 and 5.60 from the 4w-SM identified, respectively, 71% of patients with SVR and 53% of those with non-SVR from the same cohort with high accuracy (PPV and specificity >92%). In conclusion, these models predicted response to therapy before or after 4 weeks of treatment in approximately 60% of genotype 1 patients and may be valuable for the management of this condition. (HEPATOLOGY 2006;43:72–80.)
Pegylated interferon in combination with ribavirin is the best therapy for chronic hepatitis C.1, 2 However, this treatment is not satisfactory, because many patients do not respond or relapse after treatment, adverse effects are frequent, and cost is high. Because new drugs against hepatitis C virus (HCV) are not available, more rational use of pegylated interferon and ribavirin is necessary to improve the efficacy and safety of therapy and to reduce expenses.
Tailoring treatment to fulfill individual requirements is emerging as a promising strategy. Using this approach, it has been shown that treatment duration of genotype 2 or 3 infected patients can be reduced from 24 weeks3, 4 to 12 or14 weeks in subjects with rapid virological response.5, 6
The problem is more complex in genotype 1 infection. The currently recommended treatment of pegylated interferon and ribavirin administration for 48 weeks7–9 does not seem adequate, because a 24-week course of therapy is sufficient to achieve sustained virological response (SVR) in a significant proportion of patients,3, 10 whereas extension of treatment to 72 weeks may be necessary in others.11 The large variability of responsiveness to treatment among these patients makes tailoring therapy difficult. Therefore, accurate prediction of response before or very early during therapy may be beneficial.
Although HCV genotype is the main determinant of response to therapy, additional factors such as age, sex, race, body weight, duration of infection, viral load, biochemical abnormalities, extent of liver fibrosis, and compliance with treatment are involved as well.12, 13 Consideration of these factors might provide insight into the possible efficacy of treatment in patients infected with the same genotype.
We report two models for early prediction of virological response to therapy in patients with genotype 1 infection. Models were constructed at two time points—immediately before treatment and during the fourth week of therapy—by computing the relative influence of factors other than genotype on the virological response observed in a cohort of patients treated at a single center. Models were then assessed in an external cohort of patients with chronic hepatitis C who received identical therapy.
Two cohorts of treatment-naïve Caucasian patients with genotype 1 chronic hepatitis C who received pegylated interferon α-2b and ribavirin combination therapy were retrospectively studied. One cohort (estimation group) included 104 consecutive Caucasian patients who initiated therapy between August 2001 and February 2003 at a single center (Hospital Clínic, Barcelona, Spain). The other cohort (validation group) included 141 patients consecutively treated at four different hospitals located throughout Spain (Hospital La Princesa, Madrid; Hospital Nuestra Señora del Mar, Barcelona; Hospital Valdecilla, Santander; and Hospital de Valme, Seville). All of the hospitals adhered to similar diagnostic and therapeutic protocols for the management of chronic hepatitis C according to Spanish regulations and recommendations from the Spanish Association for the Study of the Liver. Eligible patients had documented chronic hepatitis C as indicated by detectable HCV RNA in serum and persistently abnormal alanine aminotransferase for at least 6 months. None of the patients had evidence of recent HCV infection. Patients with HIV infection, current hepatitis B virus infection, and other potential causes of liver disease or contraindications for anti-HCV therapy were not included. Patients gave oral informed consent to receive therapy and gave permission for use of their medical records.
Therapy and Diagnostic Procedures.
After diagnosis, patients in both groups received identical treatment with pegylated interferon α-2b (initial dose 1.5 μg/kg/wk) and ribavirin (1–1.2 g/d, according to body weight lower or greater than 75 kg). Therapy was administered for at least 24 weeks and up to 48 weeks. Patients were visited at 4-week intervals during the initial 24 weeks of therapy, at 8-week intervals during the next 24 weeks of therapy, and at weeks 12 and 24 after treatment withdrawal. Biochemical determinations (fasting glucose, cholesterol, triglycerides, alanine aminotransferase [ALT], aspartate aminotransferase [AST], γ-glutamyltransferase, bilirubin, alkaline phosphatase, total protein, and albumin) and hematological determinations (hemoglobin, red cell, white cell and platelet counts, and differential white cell count) were performed at local laboratories within 1 week before initiation of therapy and at each visit during and after therapy via automated procedures.
The normal range values of biochemical and hematological parameters were identical in the five hospitals involved in the study. HCV RNA serum concentration was measured at baseline, and qualitative HCV RNA was determined at least at weeks 4, 12, 24, and 48 during therapy and at week 24 after therapy. HCV RNA determinations were performed locally using identical commercial reagents (Amplicor HCV RNA and Amplicor Monitor HCV RNA 2.0; Roche Diagnostic Systems, Basel, Switzerland), following the manufacturer's recommendations. The hepatitis C virus genotype was determined using a modified restriction fragment length polymorphism procedure at Hospital Clínic, Barcelona, as previously reported,14, 15 and a second-generation reverse hybridization line probe assay (Inno-LIPA HCV II; Innogenetics, Zwijndrecht, Belgium) at the remaining hospitals.
Hepatic fibrosis was evaluated by a single pathologist at each hospital according to Scheuer16 as no fibrosis (F0), portal fibrosis (F1), periportal or portal septa with preserved hepatic architecture (F2), architectural distortion but no cirrhosis (F3), and definite cirrhosis (F4). Liver biopsy was not performed or was insufficient for appropriate analysis in 11 patients (10.6%) from the estimation group and in 8 patients (5.7%) from the validation group.
The baseline features of patients in the estimation and validation groups were compared to assess similarity between the two cohorts. All available demographic, biochemical, hematological, virological, and histopathological data were analyzed. A recently described scoring index17 designed to identify the presence of significant hepatic fibrosis was calculated for each patient and was included in the analysis. This fibrosis prediction index scores age, serum cholesterol, γ-glutamyltransferase activity, and platelet count. The χ2 test was used for comparison of categorical variables, and the Student t test was used for continuous variables after logarithmic transformation of those variables with deviation from normality.
Two scoring models, a pretreatment scoring model (PreT-SM) and a fourth week of therapy scoring model (4w-SM) for prediction of SVR response before therapy or at week 4 of therapy, respectively, were constructed using data from the estimation group and subsequently evaluated in the validation group. To construct the PreT-SM, a univariate analysis of baseline data was performed to identify factors associated with SVR in patients from the estimation group. Response was defined by the presence or absence of detectable HCV RNA during and after therapy. Patients with negative HCV RNA at the end of therapy and after 24 weeks on no therapy were classified as having SVR, whereas those with positive HCV RNA at the end of therapy, as well as those with virological relapse, were classified as having non-SVR. Variables associated with virological response were further analyzed via multivariate logistic regression with forward stepwise selection to identify variables independently associated with response. The identified variables and their corresponding regression coefficients were used to calculate PreT-SM individual scores. To evaluate the power of the PreT-SM for prediction of response, receiver operating characteristic (ROC) curves were constructed, and the area under the curve as well as the sensitivity, specificity, and positive and negative predictive values were calculated separately in both groups. All calculations were performed using SPSS version 10.0 software (SPSS, Chicago, IL).
The 4w-SM was constructed following the same procedure using data available after 4 weeks on therapy from patients in the estimation group. PreT-SM values, the percent change of serum aminotransferases in relation to baseline values, and clearance of HCV RNA from serum were included in the analysis. The power of the 4w-SM for prediction of response was assessed in the estimation and in the validation groups via ROC analysis.
Characteristics of the Patients at Baseline and Results of Therapy.
The main baseline features of the estimation and validation study groups are summarized in Table 1. The median serum cholesterol level was slightly higher in the validation group compared with the estimation group, but the difference was marginally significant. The mean serum HCV RNA level was slightly higher in the validation group compared with the estimation group (P = .024), but there were no significant differences between the two groups in the proportion of patients with low, intermediate, or high serum HCV RNA levels. Differences were not observed concerning demographic, biochemical, or hematological features, and the severity of liver fibrosis—either via direct reading of liver biopsies or as estimated via fibrosis prediction index—was similar as well. Thus, the estimation and validation groups had similar features at baseline.
Table 1. Comparison of Baseline Features in the Estimation and Validation Groups
Estimation Group (n = 104)
Validation Group (n = 141)
NOTE. Qualitative variables are expressed as the median (25th–75th percentile); categoric variables are expressed as n (%).
Abbreviation: GGT, γ-glutamyltransferase.
Liver biopsy was not performed or was inadequate for diagnosis in 11(11%) patients from the estimation group and in 8 (6%) patients from the validation group.
Fibrosis prediction index was calculated according to Forns et al.17
SVR was observed in 50 (48%) of 104 patients in the estimation group and in 77 (55%) of 141 patients in the validation group, whereas 54 patients (52%) in the estimation group and 64 patients (45%) in the validation group did not have SVR. This difference was not statistically significant. Thus, the final result of therapy in the estimation and in the validation groups was similar as well.
Prediction of Virological Response at Baseline.
Baseline features of patients from the estimation group with and without SVR were compared via univariate analysis. As shown in Table 2, significant differences were observed concerning age, AST/ALT ratio, γ-glutamyltransferase activity, plasma cholesterol, leukocyte and platelet count, serum HCV RNA, and severity of fibrosis. Logistic regression analyses were performed using variables associated with response at univariate analysis as input variables to identify factors independently associated with response. Variables that represented a different expression of the same concept (e.g., the fibrosis prediction index and the histological severity of fibrosis or serum HCV RNA titer and the proportion of patients with low, high, or intermediate viremia) were introduced into separated regression models.
Table 2. Univariate Analysis of Baseline Features in Patients With and Without Sustained Virological Response in the Estimation Group
Sustained Virological Response (n = 50)
No Sustained Virological Response (n = 54)
NOTE. Qualitative variables are expressed as the median (25th–75th percentile); categorical variables are expressed as n (%).
Abbreviation: GGT, γ-glutamyltransferase.
Liver biopsy was not available or was insufficient for diagnosis in 7 (14%) sustained responders and in 4 (7%) nonsustained responders.
Several models designed to predict the individual probability of response were calculated using the regression coefficients of variables independently associated with virological response. The power of each model was then assessed via ROC analysis. Table 3 shows the results of the PreT-SM, which produced the greatest area under the curve (0.856) in the estimation group (Fig. 1A). The PreT-SM was calculated as follows:
The median (25th–75th percentile) value of the PreT-SM was 8.44 (7.05–9.68) in the estimation group, 7.11 (5.97–8.20) in patients with SVR, and 9.46 (8.44–10.13) in those without SVR (P < .000) (Fig. 1C).
Table 3. Best Multiple Logistic Regression Model for Prediction of Nonsustained Virological Response at Baseline in the Estimation Group (n = 103)
In the validation group, the area under the curve was 0.845 (Fig. 1B). The median (25th–75th percentile) value of the PreT-SM in this group was 8.10 (7.12–9.53), and corresponding values in patients with and without SVR were 7.43 (6.14–8.11) and 9.42 (8.47–10.25), respectively (P < .000) (Fig. 1D).
Individual score values from the PreT-SM in the estimation and validation groups, according to response, are shown in Fig. 2. Score values lower than 7 were associated with SVR, whereas score values higher than 9.70 were associated with non-SVR. The positive predictive value (PPV), negative predictive value, sensitivity, and specificity of this model at low and high cutpoints for prediction of SVR and non-SVR, respectively, are presented in Table 4. Fifty-five (53%) patients from the estimation group and 81 (57%) patients from the validation group showed PreT-SM scores between 7 and 9.70, which had low predictive value.
Table 4. Prediction of Virological Response by the PreT-SM at Low and High Score Values
Prediction of Virological Response at Week 4 of Therapy.
In the estimation group, clearance of HCV RNA at week 4 was observed in 27 (56%) of 48 patients with SVR and in 1 (4%) of 54 without SVR (P = .000). Serum HCV RNA was not determined at this time in 2 patients. The median (25th–75th percentile) decrease of ALT activity with respect to baseline in patients with and without SVR was 63.6% (53.2%–75.5%) and 35.7% (14.9%–60.5%), respectively (P = .000). These variables, along with the PreT-SM, were entered into a logistic regression model to identify independent factors associated with virological response (Table 5), and, as shown below, the two variables independently associated with sustained response were used to calculate the fourth week predictive score:
ROC analysis revealed the area under the curve of the 4w-SM in the estimation group (n = 101) to be 0.908 (Fig. 3A). The median (25th–75th percentile) value of the 4w-SM was 4.59 (2.54–6.01) in the estimation group, and the corresponding values in patients with and without SVR were 2.49 (0.57–4.11) and 5.82 (4.96–6.38), respectively (P < .000) (Fig 3C).
Table 5. Multiple Logistic Regression Model for Prediction of Nonsustained Virological Response at Week 4 of Therapy in the Estimation Group (n = 102)
95% CI for Odds Ratio
NOTE. Two patients had missing data.
As determined by qualitative polymerase chain reaction at week 4 of therapy. For all calculations, negative HCV RNA was entered as “1” and positive HCV RNA was entered as “2”.
Percent decrease of ALT at week 4 of therapy with respect to baseline level.
Complete data were available in 120 patients from the validation group. The area under the ROC curve of the 4w-SM in this group was 0.905 (Fig. 3B). The median (25th–75th percentile) value of the 4w-SM in the validation group was 4.07 (1.52–5.63), and the corresponding values in patients with and without SVR were 1.86 (1.01–3.51) and 5.64 (4.68–6.45), respectively (P < .000) (Fig 3D).
Individual score values resulting from the 4w-SM in the estimation and validation groups, according to response, are shown in Fig. 4. Score values lower than 3.20 were associated with SVR, whereas score values higher than 5.60 were associated with non-SVR. The predictive value of this model at low and high cutpoints is presented in Table 6. Thirty-three patients (33%) from the estimation group and 40 patients (33%) from the validation group had 4w-SM scores between 3.20 and 5.60, which had low predictive value.
Table 6. Prediction of Virological Response by Scores From the 4w-SM at Low and High Cut-Points and by HCV RNA Serological Status at Weeks 4 and 12 of Therapy
The 4w-SM score values lower than 3.20 predicted SVR more accurately than clearance of HCV RNA at week 4, whereas score values higher than 5.60 predicted non-SVR more accurately than persistence of detectable HCV RNA at week 4. These differences were more evident in the validation cohort. Score values below 3.20 predicted SVR more accurately than HCV RNA clearance at week 12, while persistence of detectable HCV RNA at week 12 was more sensitive than the 4w-SM for prediction of non-SVR (Table 6).
Prediction of response to antiviral therapy in chronic hepatitis C is often difficult. Measurement of HCV RNA decline early during therapy is frequently used as a predictor of response because, according to data from two large pivotal trials, a decline of serum HCV RNA at week 12 of therapy lower than 2 log10 from baseline is highly predictive of non-SVR.2, 18 In these studies, HCV RNA was measured in central, highly committed laboratories using carefully processed serum samples. However, in daily clinical practice, due to the intrinsic variability of laboratory techniques19 and less stringent handling of serum samples,20 quantification of HCV RNA may be less accurate and prediction of response might be less reliable. In addition, the value of HCV RNA decline at week 12 as a predictor of SVR is rather low.2, 18 Thus, alternative methods for prediction of response warrant investigation.
Infection with genotype 2 or 3, low viral load, and absence of advanced hepatic fibrosis have consistently been identified as independent predictors of SVR.1, 2, 21 In addition, response may be related to other factors such as age, sex, race, body weight, duration of infection, biochemical markers of liver disease, degree of hepatic inflammation, and severity of hepatic fibrosis.12, 13, 21, 22 Recently, insulin resistance has also been associated with impaired response.23 Despite this information, early prediction of response in individual cases remains difficult, particularly when potentially favorable and unfavorable factors coincide in the same patient.
Few attempts have been made to explore the possible advantages of combining several factors for prediction of response. For instance, an analysis of patients treated with standard interferon α-2b plus ribavirin showed that the probability of SVR was 80% in patients presenting with four favorable factors, whereas the response rate in patients with four unfavorable factors was only 20%. Although relatively accurate, the applicability of this approach is very limited, because less than 10% of candidate patients presented a clearly depicted profile for prediction of response.24
In the current study, we report two scoring models for prediction of virological response in genotype 1–infected patients. To fulfill recently proposed criteria for prediction of clinical states in individual patients,25 our models were constructed using data from a cohort of 104 patients from a single center and validated in an external cohort of 141 patients from four different centers. Variables independently associated with virological response, with their regression coefficients, were used to formulate scoring indexes. The calculated individual scores should summarize the relative influence on response to therapy that factors other than HCV genotype have in each subject.
The PreT-SM included baseline serum HCV RNA concentration, the AST/ALT quotient, serum cholesterol, and a numerical score designed to estimate the severity of hepatic fibrosis.17 The histological stage of liver fibrosis as determined via liver biopsy was not considered, because the histological fibrosis stage did not emerge as an independent predictor of response in the multivariate analysis. Although liver biopsy remains the gold standard for measuring hepatic fibrosis in chronic hepatitis C, histology may not be appropriate for built-up models for prediction of response to therapy. First, some patients do not accept liver biopsy, or liver specimens may not be adequate for diagnosis. Second, at least four different histological scoring indexes have been proposed to estimate the severity of liver fibrosis,14, 26–28 and models using a specific histological index may not be widely applicable if pathologists at other centers use a different index. Third, sampling error may be a problem, particularly in patients with more advanced fibrosis.29 Finally, interobserver variability may be important30 and may cause significant deviation of scores if histological data are included.
Interestingly, our study showed a strong association between virological response and the severity of liver fibrosis estimated using a noninvasive scoring index,17 suggesting that noninvasive procedures may be valid to circumvent the limitations of histology. The possible value for prediction of response of other noninvasive approaches for estimation of liver fibrosis such as other numerical scoring models31, 32 or transient elastography33, 34 warrants investigation.
Multivariate analysis at the fourth week of therapy disclosed that two variables—the pretreatment predictive score, and clearance of HCV RNA, which was determined using a commercially available qualitative technique—were independently associated with virological response. In this study, qualitative determination of HCV RNA at week 4 was chosen to assess response because qualitative assays have less variability among different laboratories than quantitative assays19 and because previous studies have consistently shown that clearance of HCV RNA after 4 weeks of therapy is strongly associated with SVR and may therefore have a high predictive value.18, 35–37 Unfortunately, the high negative predictive value for SVR of HCV RNA decline at week 122, 18 was not known at the time when most of our patients were treated. Thus, HCV RNA concentration at week 12 was not generally available, and a comparison of the 4w-SM and the widely used 12-week stopping rule was not possible.
Our scoring models were constructed using readily available clinical data and had a high predictive value. Low scores from both models were predictive of SVR, whereas high scores were predictive of non-SVR. In the validation group, a PreT-SM score of 7 had a 90% PPV for SVR and identified more than one third of sustained responders with 95% specificity. In other words, patients with a baseline score below 7 had a very high chance of developing SVR. At the other end of the spectrum, the PPV for non-SVR of a PreT-SM score of 9.70 was 90%, with 41% sensitivity and 96% specificity. Thus, patients with a pretreatment score higher than 9.70 had a very low chance of developing SVR after a standard 48-week treatment. This information, which can easily be obtained before treatment, may facilitate decisions in nearly one half of candidates to therapy and may also be relevant for development of alternative strategies.
The information provided by the 4w-SM may be clinically relevant as well. In the validation group, a score of 3.20 had 92% PPV, 71% sensitivity, and 93% specificity for SVR. It should be noted that this model predicted SVR more accurately than clearance of HCV RNA at week 4. Thus, patients with a 4-week score below 3.20—who represent 37% of all patients from our validation group—had a very high chance of developing SVR after a standard 48-week treatment. A low 4w-SM score may be appropriate to select candidates for evaluation of less aggressive therapies in future trials.
On the contrary, the PPV for non-SVR of a 4w-SM score of 5.60 was 97%, with 53% sensitivity and 98% specificity. Thus, more than one half of nonsustained responders can accurately be identified as soon as the fourth week of therapy. This model was more accurate than persistence of detectable HCV RNA at week 4 of therapy for prediction of non-SVR. This information may be valuable to consider discontinuation of therapy after 4 weeks of treatment in more than one half of nonsustained responders.
Our models have some limitations. Response cannot be predicted in the subset of patients with intermediate score values that were found in approximately one half of the total patient population at baseline and in 33% of the cases after 4 weeks of treatment. In these patients, therapy should be administered for at least 12 weeks, at which time a decision may be made based on measurement of HCV RNA decline. It must also be noted that this study was performed in treatment-naïve, Caucasian, genotype 1–infected patients who received weight-adjusted doses of pegylated interferon α-2b plus ribavirin for 48 weeks; therefore, our findings may not apply to other populations of patients, nor can they be extended to other therapeutic schedules. Furthermore, our models may not be applicable in subjects undergoing therapy for high cholesterol serum levels.
In conclusion, although new antiviral agents are under development, major changes in the treatment of chronic HCV infection are not expected in the near future. Therefore, new strategies for better use of currently available drugs are necessary. In this way, tailored treatments on an individual basis may be useful to improve efficacy and minimize side effects and cost. The predictive models reported here can be useful to this end.