The Impact of the Difference in Natural History Models on the Cost-Effectiveness of Antiviral Agents for Patients with Genotype 1 Chronic Hepatitis C

Abstract

Objectives: Complication costs can typically be modelled by Uniform and Pert distributions but it remains important to obtain an accurate estimation of variance since unknown variance is more problematic than unstable variance. The objective was (i) to compare the central tendency of the Uniform and Pert distributions for their respective ability to decrease complication cost variance and (ii) illustrate use of the CLT to ensure that the complication cost for each patient is a random sample from the assumed distribution for 100 theoretical patients. The CLT is used since it is incorrect to simply multiply a random number from the cost distribution by the number of patients. MethOds: A random set of costs (Normal distribution; n = 500; mean = 100; SD = 25) was generated (MS EXCEL). A Uniform distribution was fitted using the data range and a Pert distribution was fitted using the minimum, most likely (mean of the generated data), and maximum values. The mean and SD of both distributions were simulated (@Risk; Palisade) and compared. The "most likely" value of the Pert distribution was varied to include [mean +SD] and [mean -SD] to evaluate the effect of imprecise "most likely" estimates on the simulated variance. Monte-Carlo simulation was used with 10000 iterations. Results: Mean of the Uniform distribution was 99 (SD = 44), while the Pert distribution returned SD's of ~28 for all "most likely" scenarios. Mean simulated total cost for 100 patients using the CLT was 9969 (SD = 288.33; minimum = 8788; maximum = 11100; 95% CI = 9400 -10500). cOnclusiOns: The Pert distribution is robust to varied "most-likely" value estimates and decreases estimated variance over the Uniform distribution. Application of the CLT ensures that each patient's complication cost is independent of the next and that the spread of costs is not exaggerated.Objectives: This paper proposes a data-driven method to strengthen predictions from discrete choice experiments (DCEs). DCEs are used as predictive tools when there is little or no observational data available to predict uptake of a service or good, however, there is little evidence assessing how well DCEs predict behaviour. This review is the first to quantitatively assess the external validity of published DCE studies, before setting out a modelling framework for DCE estimates to be adjusted such that they better reflect real-world choices. An application to predicting demand for new HIV prevention products is presented. MethOds: We conduct a systematic review and meta-analysis to summarise how well DCE predictions reflect real-world choices. We synthesise results using bivariate mixed-effects logistic regression to produce pooled sensitivity and specificity estimates. A probabilistic approach is proposed to incorporate information to model uptake, and applied to a dataset predicting demand for new HIV prevention products. Results: Five studies were included in the meta-analysis. In total we consider 716 observations, where choices were correctly predicted 80% of the time; 53% of incorrect...

Cost-effectiveness analysis of sofosbuvir plus ribavirin in patients with genotype 2 chronic hepatitis C: an analysis with real world outcomes from a multicentre cohort in Japan

“…Relevant transition probabilities after antiviral therapy were extracted from a natural history model for HCV in Japan by Ishida et al ,16 which was developed as part of long-term research on economic evaluation of viral hepatitis control and prevention funded by the Health Labour Sciences Research Grant. The transition probabilities that were not reported by Ishida et al , eg, CC to HCC, were set by either Tanaka et al ,17 which contained data corresponding with the initial age of this analysis, or Cardoso et al ,18 which was used in a previous study on CEA in Japan 5…”

“…The range for the SVR in one-way SA was set from phase III trials for each drug 8 32–34. Basically, the range of each transition probability referred to the reported value by Ishida et al 16. For variables whose range is not reported, it was set as ±25% of each variable.…”