Generalized Confidence Intervals for the Mean of Medical Cost Data with Zeros

In this paper, we consider the interval estimation for the means of medical cost data. Medical cost data may contain zero values, and the nonzero values can often be modeled by a lognormal distribution. Under the model, we proposed three generalized confidence intervals approaches for the mean: a Jeffrey based approach, an Agresti-Coull based approach, and a Wilson based approach. Our simulation studies suggested that the Jeffrey based generalized confidence interval outperformed the other methods and that it has highly accurate coverage probability and fairly low bias, even in small sample settings.