Sequential Randomization Tests with E-Values Enable Continuous Clinical Trial Monitoring

Researchers have introduced e-RT, a randomization-based e-value method designed for continuous clinical trial monitoring across binary, event-only, and continuous endpoints. In randomized trials where repeated interim analyses of accumulating data must keep Type I error controlled, sequential analyses using e-values allow continuous assessment of evidence while controlling their own error rates. The newly developed e-RT approach is intended primarily as a non-binding ancillary review trigger for data safety monitoring boards, operating alongside prespecified primary stopping boundaries.

Methodologically, at each data update, endpoint data are reviewed before the current randomized assignment enters a betting-martingale wealth calculation. Because the assignment cannot be predicted from observed data under the null hypothesis, the resulting wealth process functions as a nonnegative martingale. Consequently, Ville’s inequality controls the monitor’s Type I error at any point in time, providing valid surveillance regardless of the wager policy so long as wagers remain predictable.

Under the proposed framework, a formal charter prespecifies the role of the alert, which can function as a process-only review, a standalone stopping rule, or a designated path to an efficacy stop. To examine its operating characteristics, the authors evaluated e-RT in simulations against calibrated group-sequential designs matched on sidedness and Type I error. In addition, the researchers illustrated the application of the methodology by reanalyzing data from two completed mechanical ventilation trials.

Simulation results showed that the calibrated one-sided design achieved 11.6 percentage points higher power than the design-calibrated binary e-RT variant, though e-RT signaled earlier and maintained a null crossing rate below the nominal 5%. When a treatment effect was present, an e-RT alert occurred before 91.8% of scheduled stops, providing a median lead of 776 patients. In the process-only role, this early alert left the primary analysis’s Type I error completely unchanged.

The investigations also evaluated stopping rules and adjustments, observing that permitting every threshold crossing to stop the trial increased the null stop rate to 7.2%. Implementing a confirmation-bar repair reduced the null stop rate back to 5.6% while preserving statistical power, whereas an alpha-partition repair remained bounded by construction at a modest cost to power. The reanalyses of completed mechanical ventilation trials utilized deidentified data sourced from the National Heart, Lung, and Blood Institute biorepository, known as BioLINCC, underscoring e-RT as a transparent surveillance tool to alert monitoring boards between scheduled interim analyses.

Subject of Research: Sequential randomization tests using e-values for continuous clinical trial monitoring

Article Title: Sequential randomization tests using e-values for clinical trial monitoring across endpoint types

Article References: Zampieri, F. G., Albuquerque, A. M., & Ramdas, A. (2026). Sequential randomization tests using e-values for clinical trial monitoring across endpoint types. BMC Medical Research Methodology. https://doi.org/10.1186/s12874-026-03048-z

Image Credits: AI Generated

DOI: 10.1186/s12874-026-03048-z

Keywords: clinical trials, e-values, sequential randomization tests, Type I error, interim monitoring, data safety monitoring boards, biostatistics, Ville's inequality, Sequential, randomization, tests, clinical