Experience collecting interim data on mortality: an example from the RALES study

Table 1 Randomized Aldactone Evaluation Study (RALES) comparing spironolactone to placebo: observed and projected number of deaths and summary statistics at the interim analyses

			Observed
	Data	Meeting				Estimated	z-value		Observed
Look	cutoff	date	Placebo	Treatment	Relative	information	P
no.	(d-mo-y)	(d-mo-y)	deaths	deaths	risk	time^*	Critical	Observed	(two-sided)
Interim analysis cut-offs with the sweeps as they occurred
1	09-8-96	24-8-96	70	52	0.76	0.12	6.38	1.58	0.11
2	10-3-97	17-3-97	136	109	0.83	0.24	4.43	1.69	0.092
3	14-8-97	25-8-97	224	175	0.80	0.34	3.67	2.55	0.011
4	26-3-98	30-3-98	304	241	0.81	0.48	3.04	3.02	0.0026
5	14-8-98	24-8-98	351	269	0.78	0.57	2.79	3.75	0.00018
Estimated interim analysis cutoffs without the sweeps^†
4a	26-3-98	30-3-98	281	222	0.81	0.45	3.16	2.93	0.0034
			(275-289)	(214-228)	(0.76-0.85)			(2.33-3.69)
5a	14-8-98	24-8-98	333	256	0.79	0.55	2.81	3.59	0.00034
			(325-341)	(248-264)	(0.74-0.83)			(2.87-4.33)
Interim analysis cutoffs that would have occurred if we had known the true numbers and times of deaths
1b	09-8-96	24-8-96	81	59	0.75	0.14	5.88	1.80	0.072
2b	10-3-97	17-3-97	189	140	0.76	0.26	4.24	2.75	0.0060
3b	14-8-97	25-8-97	257	201	0.80	0.38	3.46	2.82	0.0048
4b	26-3-98	30-3-98	330	254	0.79	0.51	2.95	3.56	0.00038
5b	14-8-98	24-8-98	383	283	0.76	0.60	2.72	4.46	0.000008

This three-panel table shows (a) the data observed in the RALES trial at each of the five interim analyses; (b) simulated data for the fourth and fifth interim analysis had the sweeps not occurred; and (c) the data that would have been observed had all the deaths been reported within 24 h of their occurrence, as specified by the protocol. ^*No. of events that have occurred divided by the no. of events that will occur up to the planned end of the trial. Because the number of events that will occur is unknown at the time of a DSMB meeting, one must estimate it on the basis of data observed thus far and assumptions made about the future. ^†Observed values are median number of events (in 1000 simulated iterations of the trial), with ranges shown in parentheses. The 1000 iterations of the simulation assuming neither sweep showed a probablility of 18% and 100% that the data would have first crossed the prespecified stopping boundary at the fourth and fifth interim analyses, respectively. These numbers refer to looks 4a and 5a (data not shown).

ISSN: 1745-6215