Sequential Analysis (Group Sequential Design)
Group Sequential Design · Also known as: sequential testing, group sequential design, interim analysis, Sıralı Analiz (Sequential Testing / Group Sequential Design)
Sequential analysis is a framework for conducting hypothesis tests with pre-planned interim looks at accumulating data, allowing a study to stop early for efficacy or futility while controlling the overall Type I error rate. The group sequential approach was formalised by Pocock (1977) and O'Brien and Fleming (1979), and remains the standard for confirmatory clinical trials and rigorous A/B experiments.
Read the full method
Sign in with a free account to read this section.
Method map
The neighbourhood of related methods — select a node to explore.
+4 more
When to use it
Use sequential analysis when you are collecting data sequentially over time and wish to monitor accumulating evidence without inflating the Type I error rate. It is the standard approach for confirmatory randomised controlled trials, large-scale A/B tests, and any study where early stopping for benefit or harm is ethically or commercially important. Prerequisites: the number of interim looks must be specified in advance; the outcome must be continuous or binary; observations should be independent across participants. The method works for both cross-sectional and longitudinal outcome structures with a minimum of roughly 20 subjects before the first interim look.
Strengths & limitations
- Enables ethically justified early stopping when a treatment effect is clear, reducing exposure of participants to inferior conditions.
- Controls the overall Type I error rate precisely across all interim looks, maintaining nominal significance.
- Flexible alpha-spending functions (e.g. Lan–DeMets) allow unplanned or unequally spaced interim analyses without inflating error rates.
- Well-established in regulatory contexts — accepted by the FDA and EMA for confirmatory trials.
- The analysis plan, including the number and timing of interim looks, must be locked in before data collection; post-hoc changes compromise error control.
- Sample size calculations are more complex than for fixed designs, typically requiring specialised software.
- Early stopping for efficacy can yield unstable, over-optimistic point estimates of the treatment effect (the 'winner's curse').
- Not appropriate when the study must run to completion for reasons other than hypothesis testing (e.g. safety monitoring alone).
Frequently asked
What is the difference between O'Brien–Fleming and Pocock boundaries?
O'Brien–Fleming uses very strict (large) boundaries at early interim looks and progressively relaxes them as the study approaches its maximum sample size. This makes accidental early stopping unlikely unless the evidence is overwhelming, and the final boundary is close to the unadjusted significance level. Pocock uses the same boundary at every look, which means each individual test uses a more conservative threshold but the study has a higher probability of stopping early. In practice O'Brien–Fleming is preferred for confirmatory trials because it preserves most of the nominal significance level for the final analysis.
Does sequential testing require a larger total sample size?
Not necessarily. If a true effect exists and is large, early stopping can substantially reduce the expected total sample size. However, the maximum sample size — the number needed if the study runs to completion — is slightly larger than for a fixed design, because the boundary adjustments use a small portion of the alpha budget at interim looks.
Can I add an interim look that was not in the original plan?
Only with an alpha-spending function, which can accommodate unscheduled looks by allocating alpha based on the information fraction accumulated so far. Adding an unplanned look without formal alpha spending is a protocol violation that inflates the Type I error rate and will typically be rejected by regulators.
How do I estimate the treatment effect after early stopping?
The naive point estimate is biased upward when the study stops early due to a boundary crossing. Bias-adjusted estimators — such as the median unbiased estimate or a shrinkage approach — are recommended alongside confidence intervals that properly account for the sequential stopping rule. Always report these alongside the raw estimate.
Sources
- O'Brien, P.C. & Fleming, T.R. (1979). A Multiple Testing Procedure for Clinical Trials. Biometrics, 35(3), 549–556. DOI: 10.2307/2530245 ↗
- Jennison, C. & Turnbull, B.W. (1999). Group Sequential Methods with Applications to Clinical Trials. CRC Press. ISBN: 978-0849303166
How to cite this page
ScholarGate. (2026, June 1). Group Sequential Design. ScholarGate. https://scholargate.app/en/statistics/sequential-analysis
Which method?
Set this method beside its closest kin and read them side by side — the library lays the books on the table; the choice is yours.
- Adaptive Clinical Trial DesignExperimental design↔ compare
- Bayesian Power AnalysisStatistics↔ compare
- One-sample t-testStatistics↔ compare
- One-way ANOVAStatistics↔ compare
- Simulation-Based Power AnalysisStatistics↔ compare