Session Title: Category 5g. VIRAL HEPATITIS: g. HEPATITIS C - CLINICAL (THERAPY)
Presentation Date: Apr 15, 2010
PREDICTIVE MODEL OF RESPONSE TO PEGINTERFERON IN CHRONIC HEPATITIS C: IMPACT OF MUTATIONS IN ISDR AND CORE REGION REVEALED BY CLASSIFICATION AND REGRESSION TREE ANALYSIS
M. Kurosaki1*, N. Sakamoto2, M. Iwasaki3, M. Sakamoto4, Y. Suzuki5, N. Hiramatsu6, F. Sugauchi7, A. Tamori6, N. Izumi8
1Division of Gastroenterology and Hepatology, Musashino Red Cross Hospital, 2Tokyo Medical and Dental University, 3Seikei University, Tokyo, 4University of Yamanashi, Yamanashi, 5Toranomon Hospital, Tokyo, 6Osaka University Graduate School of Medicine, Osaka, 7Nagoya City University Graduate School of Medical Sciences, Nagoya, 8Musashino Red Cross Hospital, Tokyo, Japan. *firstname.lastname@example.org
Background and aims: Pretreatment prediction of sustained virological response (SVR) aids in the selection of chronic hepatitis C patients for peg-interferon (PEG-IFN) plus ribavirin (RBV) therapy. This study aimed to develop a predictive model for SVR.
Methods: Genotype 1b chronic hepatitis C patients treated with PEG-IFN plus RBV at 6 hospitals in Japan (n=800) were randomly assigned to the model building (n=506) and validation (n=294). Predictive models of SVR were determined by classification and regression tree analysis (CART) using IBM-SPSS modeler software. Reproducibility was validated. For the external validation, 524 patients treated at another 29 hospitals in Japan were used.
Results: Two models were developed. The first model (model-1) was based on routine biochemical parameters. Younger age (< 50), lower alpha-fetoprotein (< 8ng/ml), higher platelet (≥120x109/L), lower gamma-glutamyltransferase (< 40 IU/l), and male gender were selected as predictive factors which divided patients into 3 groups. The rate of SVR was 75%, 44%, and 23% for the high, intermediate and low probability group respectively. Model-1 had sensitivity of 50% and specificity of 85%. The reproducibility of the model was confirmed by validation (r2=0.92 to 0.93). Another model (model-2) was based on routine biochemical parameters and HCV mutations. The number of mutations in ISDR (≥2) was selected as the best predictor of SVR followed by younger age (< 60), arginine at position 70 of HCV core, higher low-density-lipoprotein cholesterol (≥120mg/dl), and absent fibrosis (F0-1) which divided patients into 3 groups. The rate of SVR was 75%, 64%, and 32% for the high, intermediate and low probability group respectively. Model-2 had sensitivity of 60%, specificity of 85%, and area under ROC of 0.74. Poor adherence to drugs lowered the rate of SVR in low probability group (18%) but not in high probability group (77%).
Conclusions: CART analysis highlighted the impact of HCV mutations on pretreatment prediction of SVR. Decision tree models are useful for predicting the probability of response to therapy, and have potential to support clinical decisions regarding the selection of patients for therapy. Strict adherence to drugs is especially important for patients with low predicted probability of response.