On estimators of medical costs with censored data

In the assessment of cost-effectiveness of alternative medical technologies, it is necessary to estimate the mean total cost per patient over the relevant patient population. Where information about costs comes from a clinical trial with censored data, care is needed to estimate mean total costs. We examine the theoretical connections between the two most widely used of a growing range of nonparametric estimators of costs under censoring. By clarifying the relationships between these simple methods we hope to make them more accessible and to facilitate the take-up of more sophisticated techniques. Recommendations are offered regarding the most appropriate of the available methods, but also on the potential for greater efficiency through parametric modelling.