Optimal M-Switch Surveillance Policies for Liver Cancer in Hepatitis C-Infected Population

Abstract

Hepatocellular carcinoma (HCC) is the most common type of liver cancer and the fastest-growing cause of cancer-related deaths in the United States. Most HCC cases are attributed to chronic hepatitis C virus infection, which affects nearly 3 million Americans and 170 million globally. Although surveillance for HCC in hepatitis C patients can improve survival, the optimal surveillance policies remain unknown. In this study, we develop a mixed-integer programming (MIP)-based framework to systematically analyze a rich set of policies and determine the optimal HCC surveillance policies with the maximum societal net benefit. Our MIP-based framework captures two problem features that make dynamic programming-based formulation computationally intractable. In particular, our proposed framework allows to (1) explicitly formulate M-switch policies that are practical for implementation, and (2) tailor surveillance policies for each subpopulation by stratifying surveillance intervals based on the observable disease states. We theoretically analyze the HCC surveillance problem, characterize when the surveillance policies should be adapted to populations with different disease progression rates, and quantify the trade-off between decreasing HCC incidence and increasing treatment outcomes. We carefully parameterize our model using clinical trial data, a previously validated simulation model, and published clinical studies. Our numerical analyses lead to three main results with important policy implications. First, we find that, in addition to cirrhotic patients, expanding surveillance to patients in earlier stage of hepatitis C infection improves the cost-effectiveness of HCC surveillance. Second, compared with the one-size-fits-all type routine policies, we find that it is cost-effective to stratify surveillance strategies based on the stage of hepatitis C infection with less frequent cancer surveillance in earlier stages of infection. Lastly, we find that a little flexibility in the policy structure as captured by M-switch policies is sufficient to capture almost as much benefit as complex fully dynamic policies.