OBJECTIVE: To address the common issue of incomplete follow-up data in cost-effective analysis, we compared the actual average cumulative medical costs with the estimated costs using a set of statistical methodologies applicable to censored cost data. METHODS: A study cohort with monthly recorded concomitant medication costs was selected from the population of a randomised clinical trial. Among a total of 70 subjects, a pattern of 25% non-informative censoring was applied prior to the endpoint of interest (the first event between death and 1-year follow-up visit). Statistical methods applied to deal with censored data included naïve estimators from complete case analysis (CCA) and available-case analysis (ACA), as well as Lin’s, inverse-weighted and regression-based multiple-time-interval estimators. The covariate considered in regression models was continuous variable age. Bootstrapping with 10,000 replications was used to obtain the standard deviation (SD). RESULTS: The actual average total cost per subject was £268.7 (SD: £52.0). Estimations from the 5 methods given by mean (SD) were: CCA: £467.7 (£110.8), ACA: £246.0 (£49.8), Lin’s estimator: £301.6 (£60.6), inverse-weighted estimator: £294.5 (£15.6), regression-based method: £245.5 (£49.2). CONCLUSIONS: By ignoring subjects with incomplete cost data, CCA overestimated the average cumulative cost as subjects with shorter survival tend to cumulate higher costs. Lin’s and inverse-weighted non-parametric estimators that make no assumption for the distribution of cost data slightly overestimated the average total cost. Regression-based method gave better results for both mean and SD than one-sample estimators (Lin’s and inverse-weighted) as it considered one cost-related factor (age) as covariate. The multiple-time-interval strategies (Lin’s, inverse-weighted and regression-based) effectively assess cost information from censored subjects by treating them as uncensored in some of the time intervals. However, in this situation, small-size dataset and light censoring make ACA the best estimator.