Sample size for comparison of changes in the presence of right censoring caused by death, withdrawal, and staggered entry.

In estimating and comparing the rates of change of a continuous variable between two groups, the unweighted averages of individual simple least-square estimates from each group are often used. Under the linear random effects model, these statistics are maximum likelihood estimates for the expected rates of change when all individuals have complete observations. However, death and withdrawal often cause observations on the variable of interest to be right censored for some participants, which makes any subsequent measurements impossible (to be referred to as right censoring). In this situation, the unweighted averages are no longer efficient in comparison with the generalized least-square estimates. Relationship between sample size, frequency of measurement, and right censoring are described for the different estimation procedures. Using realistic estimates of the random effect parameters, we illustrate that if there were 8% right censored observations each year due to participants' death or loss to follow-up, the sample size requirements for a proposed 3-year controlled clinical trial of alpha 1-protease inhibitor replacement therapy could be more than doubled if the unweighed rather than the generalized least-square estimates were used.