Improving the accuracy of long-term prognostic estimates in hepatitis C virus infection

Obtaining unbiased estimates of HCV prognosis is difficult because of potential biases associated with study design and calculation methods. We propose a new method for estimating fibrosis progression rates. A Markov model with fibrosis health states (F0-F4) was created. The maximum likelihood method was used to estimate stage-specific progression rates. We compared the standard method to the new method using two well-known cohort studies. The known stage distribution at the end of follow-up was compared with stage predicted by the Markov model using both methods of calculating transition rates. We also compared rates obtained using both methods to known fibrosis rates in a series of Monte Carlo simulations. For Kenny-Walsh's study (1999), transition rates between F0-F1, F1-F2, F2-F3, and F3-F4 were 0.042, 0.045, 0.097 and 0.070 fibrosis units/year (new method) and 0.045 units/year (standard method). The new method predicted fibrosis stage and known transition rates in Monte Carlo simulations more accurately. The standard method underestimates 30-year cirrhosis rates by up to 40%. The new (Markov maximum likelihood or MML) method allows accurate estimation of stage-specific transition probabilities from the many studies in which only a single biopsy is available. Application of the method supports the hypothesis that rates of fibrosis vary between stages.