Describing followup maturity

In all studies with a survival-type endpoint there will usually be a mixture of subjects in which the critical event has been observed and those in which it has not. Mature data are those in which most of the events that have been targeted in the design of the trial have been observed. There are a number of ways in which to summarize the reliability and maturity of follow-up. The numbers at risk at various stages along the Kaplan-Meier survival curve and the indication of the censored data on these curves, together with SEs at specific time points, are useful measures. However, they do not give single concise measures. Simple summaries which have been suggested for this purpose include the median follow-up time of those individuals still alive, the minimum and maximum follow-up times of these individuals and the proportion of patients who have experienced the event.

In all trials the following statistics should always be calculated and reported (see Table 9.5): the numbers at risk at various appropriate timepoints on the survival curves: the targeted total number of events, and observed number of events. In trials in which a large proportion of the patients are expected to experience the event by the time of analysis, for example in trials of advanced disease, the most informative statistic to report is the proportion of patients with an event. In contrast, in trials in which less than say 60 per cent of patients have experienced the event by the time of analysis, the most informative statistic is the median follow-up of survivors.

For all trials, to check that there is no imbalance in follow-up between groups being compared it can be useful to calculate a Kaplan-Meier 'follow-up' curve. To do this, we label those patients who have 'died' as actually being 'censored' on their date of death, and those patients who are still alive as having an 'event' on the date they were censored. The estimated median follow-up can be read off the curves in the same way as median survival.

Table 9.5 Statistics to report follow-up from randomized trials

Statistic to report follow-up

Which type of trial?

Numbers at risk

All trials

Targeted total number of events

All trials

Observed number of events

All trials

Proportion of survivors followed to a timepoint

All trials

of interest, e.g. two or five years

Proportion of patients with an event Advanced disease trials

Proportion of patients with an event Advanced disease trials

Months of follow-up

Patients at risk

CHART 338 193 83 34 18

Conventional 225 1 14 33 13 4

Patients at risk

CHART 338 Conventional 225

Fig. 9.2 (a) Survival in the CHART lung cancer trial. Reprinted with permission from Elsevier Science (The Lancet, 1997, 350, 161-5) (b) Follow-up in the CHART lung cancer trial.

Example. Saunders and colleagues report a randomized trial of continuous hyper-fractionated accelerated radiotherapy (CHART) versus conventional radiotherapy in patients with locally advanced non-small cell lung cancer. Approximately 600 patients were planned to be entered into this trial, and a total number of deaths of approximately 475 was targeted to achieve the power and type I error for the targeted difference. A total of 563 patients were actually entered into the trial and a total of 444 deaths were observed, i.e. when the results of this trial were reported, events had been seen in the large majority, 79 per cent (444/563), of patients. The Kaplan-Meier survival curves for the two groups, are shown in Fig. 9.2a and the follow-up curve in Fig. 9.2b.

It can be seen that beyond two years the numbers at risk become relatively small, especially in the conventional radiotherapy group, with the numbers alive and at risk at three and four years of thirteen and four patients, respectively. Comparison of the

Months of follow-up

Fig. 9.2 (a) Survival in the CHART lung cancer trial. Reprinted with permission from Elsevier Science (The Lancet, 1997, 350, 161-5) (b) Follow-up in the CHART lung cancer trial.

Kaplan-Meier curves gives a hazard ratio of 0.76 (95 per cent confidence interval 0.63 to 0.92; p = 0.004) indicating a 24 per cent reduction in the relative risk of death at any time with CHART.

0 0

Post a comment