Statistical Calculators › Group Sequential, Adaptive, and Interim Analysis › Group Sequential Design for Survival Endpoints
← All Calculators

Group Sequential, Adaptive, and Interim Analysis

Group Sequential Design for Survival Endpoints

Two-look group sequential planning for a two-arm time-to-event endpoint using the log-rank test. Calculates the required number of events, total sample size, interim timing, and efficacy boundaries using a Lan-DeMets O’Brien-Fleming-type alpha-spending design.

Trial Design Inputs

This implementation uses the exponential survival / constant-hazard special case of ’s GST3 two-sample survival group sequential framework.
Hazard ratio is experimental / control. Values below 1 indicate the alternative is in the lower-hazard direction.
Constant exponential event hazards and constant dropout hazard are assumed. Set dropout to 0 for no loss to follow-up.

Design Results

Event-driven group sequential design with the final analysis at 100% information.
Enter design assumptions and click Calculate Group Sequential Design.

Methodology

This calculator implements a two-sample log-rank group sequential design in the constant-hazard special case of the survival group sequential framework described by for GST3. The design uses two analyses: one interim analysis at a specified information fraction and one final analysis at 100% information. Information is treated as proportional to the number of observed events, which is the standard approximation for the log-rank statistic under proportional hazards.

Group sequential boundaries

The efficacy boundary uses the two-sided Lan-DeMets alpha-spending approach with an O’Brien-Fleming-type spending function. At information fraction t, cumulative two-sided alpha spent is

α(t) = 2 − 2Φ(z1−α/2 / √t)

The interim critical value is obtained from the alpha spent at the interim information fraction. The final critical value is then solved from the canonical joint normal distribution of the two correlated log-rank Z statistics, with correlation √t between the interim and final statistics. This preserves the requested overall two-sided type I error.

Required events

For equal or unequal randomization, the log-rank information is approximated by D·p(1−p), where D is the total number of events and p is the experimental-arm allocation fraction. Under a proportional-hazards alternative, the stage-k Z statistic has mean

μk = log(1 / HR) √(D p(1−p) tk)

The required maximum number of events is the smallest continuous solution giving the requested power; the displayed event target is rounded up to the next whole event.

Converting events to subjects

For constant event and dropout hazards with uniform accrual, the expected event probability for a subject is calculated using the Kim-Tsiatis survival-design formulation. For an event hazard λ, dropout hazard δ, accrual duration R, and total study duration T = R + follow-up, the average event probability is

P(event by T) = [ λ / (λ+δ) ] × [ 1 − { exp(−k·F) − exp(−k·T) } / {Rk} ],   k=λ+δ

The treatment and control event probabilities are weighted by the planned randomization proportions. Required total enrollment is the required event count divided by this weighted event probability. Expected timing of the interim analysis is obtained by solving the same event-accumulation equation for the interim event target.

Validation example

Zhu, Ni & Yao (2011) describe a two-arm survival group sequential example with median survival 9 months versus 6 months, 1:1 allocation, uniform 18-month accrual, 30 months of maximum follow-up, two-sided α=0.05, 85% power, and one interim analysis at 50% of the events using a Lan-DeMets spending function chosen to approximate O’Brien-Fleming. With the constant-hazard/no-dropout implementation used here, the calculation reproduces approximately 220.81 required events, rounded to 221 events, and approximately 228.44 subjects, rounded to 229 subjects. The expected interim occurs at approximately 17.20 months.

References

Zhu, L., Ni, L., & Yao, B. (2011). Group Sequential Methods and Software Applications. The American Statistician, 65(2), 127–135. doi:10.1198/tast.2011.10213.

Kim, K., & Tsiatis, A. A. (1990). Study duration for clinical trials with survival response and early stopping rule. Biometrics, 46(1), 81–92. doi:10.2307/2531632.

Lan, K. K. G., & DeMets, D. L. (1983). Discrete sequential boundaries for clinical trials. Biometrika, 70(3), 659–663.

Advanced User Manual, Section 7.2.5.4, Survival (Time-to-Event) Analysis (GST3) — Sample Size Determination, describing the two-sample log-rank survival group sequential framework and Kim-Tsiatis event calculations.