© John Wiley & Sons, Inc.
FIGURE 21-2: Survival times from the date of surgery.
If you exclude all participants who were censored in your analysis, you may be left with analyzable
data on too few participants. In this example, there are only six uncensored participants, and removing
them would weaken the power of the analysis. Worse, it would also bias the results in subtle and
unpredictable ways.
Using the last-seen date in place of the death date for a censored observation may seem like a
legitimate use of LOCF imputation, but because the participant did not die during the observation
period, it is not acceptable. It’s equivalent to assuming that all censored participants died immediately
after the last-contact date. But this assumption isn’t reasonable, because it would not be unusual for
them to live on many years. This assumption would also bias your results toward artificially shorter
survival times.
In your analytic data set, only include one variable to represent time observed (such as Time in
days, months, or years), and one variable to represent event status (such as Event or Death),
coded as 1 if they are have the event during the observation period, and 0 if they are censored.
Calculate these variables from raw date variables stored in other parts of the data (such as date
of death, date of visit, and so on).
Looking at the Life-Table Method
To estimate survival and hazard rates in a population from a set of observed survival times, some of
which are censored, you must combine the information from censored and uncensored observations