Patients who are not assessed for the primary outcome

In most trials, outcome measures are measured many months and sometimes years after randomization. It is therefore possible that some of these patients will be 'lost to follow-up' before the outcome measure has been observed. For example, if the analysis is focusing on the response at twelve weeks, then some patients may have died, others may not be assessed because they are too ill and do not turn-up, while, perhaps more rarely, others do not turn up because they are well. If these missing data are more common in one group than another then bias in the comparison can creep in; even if the numbers of patients with missing data are similar in the two groups the reasons for the missing data may be different in the two groups. This situation is perhaps more difficult than those above, because the data on these patients are just not available for inclusion in the analysis. Options in these circumstances include regarding patients with missing data as 'failures' on therapy, making the general assumption that this is the prime reason data are missing. This may not be appropriate for all outcome measures, such as quality of life, where imputation of missing values is possible, but potentially dangerous. We discuss how these issues may be addressed for different outcome measures of response (Section 9.4.5) and quality of life (Section 9.4.9). This difficulty emphasizes the importance of concentrating on an outcome measure that can be assessed in all randomized patients such as, for example, duration of survival (see Section 5.2)

0 0

Post a comment