Landmark And Survival Analyses

A landmark analysis of an event endpoint assesses the proportion of patients in whom an event has occurred by a specific point in time. It does not take into account how soon before that time point the event occurred. This type of analysis is most appropriate when the time during the trial at which an endpoint occurs is of far less clinical significance than whether or not the endpoint occurs at all. Typically, this situation arises in relatively short trials in which the risk of an endpoint event potentially affected by the study intervention is largely limited to a short period after enrollment. Trials of treatments for acute sepsis syndrome and acute myocardial infarction often use landmark mortality at 28 or 30 days as a primary endpoint, thus essentially considering all deaths before this time point as equally bad outcomes.

A time-to-event analysis (frequently called a survival analysis, even when the endpoint does not involve mortality) takes into account the timing of an endpoint event (1). The outcome of a patient who experiences an adverse outcome event (e.g., death, tumor progression) early is considered (i.e., is ranked as) worse than that of a patient who experiences the event later.

An example of a trial with a time-to-event endpoint is the study of Avonex® (interferon beta-1a) in primary relapsing, remitting multiple sclerosis. The primary endpoint was the time to progression in disability (2). In this study, all subjects had some degree of disability at the onset, and all were expected to progress in disability eventually. Delay in the progression of disability was clinically meaningful and desirable.

Usually, time-to-event analyses are displayed graphically by using KaplanMeier survival methods, and nonparametric rank tests are used to test for differences between groups (3). These methods facilitate meaningful graphic presentation and analysis of survival data when the length of follow-up on patients enrolled early in the trial is substantially longer than the follow-up on patients enrolled more recently.

Although use of cutpoints may make a measurement-based endpoint more like an event, use of time-to-event analyses may make an event-based endpoint more like a measurement. For interventions that only delay events or that both delay and prevent events, a time-to-event analysis is likely to be more sensitive to differences in effect than a landmark analysis. However, particularly in short-term studies such as those described above, a time-to-event analysis may detect differences of little clinical meaning. Theoretically, for example, a new treatment that reduces mortality in an acute disease might appear inferior to the control group if the deaths in the treatment group occurred a few days earlier than the deaths in the control group.

When a landmark analysis is chosen for the primary endpoint, it is important that the time point selected for the analysis be late enough so that it captures most of the events potentially affected by the intervention(s), but not so late that many endpoint events (e.g., deaths) are captured that are likely unrelated to the disease or treatment under study. If the time point is too early, not only might sensitivity be lost, but also the validity and reproducibility of the endpoint findings may be impaired by the fact that minor variations in the time of occurrence of important, treatment-related events could result in the event not being counted at all.

Although a single analysis of the primary endpoint is desirable (see below), it is often advisable to plan prospectively and perform landmark analyses as well as time-to-event analyses at several time points. The results will usually correlate strongly; if they do not, the outcome data should be explored for unexpected relationships to time.

0 0

Post a comment