Value of central event adjudication.

Stroke welcomes Letters to the Editor and will publish them, if suitable, as space permits. They should not exceed 750 words (including references) and may be subject to editing or abridgment. Please submit letters in duplicate, typed double-spaced. Include a fax number for the corresponding author and a completed copyright transfer agreement form (available online at To the Editor: The Perindopril Protection Against Recurrent Stroke Study (PROGRESS) Collaborative Group notes that the End Point Adjudication Committee (EPAC) process had no discernible impact on that trial's statistical conclusions. 1 However, we caution that this result is not necessarily representative and that an EPAC is a wise investment for most studies. This claim is supported by the following points. First, the rejection rate of 9.9% seen in PROGRESS is low by our experience. Traditionally, one asks investigators to " cast a wide net " and send in any possible event, even if they believe it unlikely to be a true event, to ensure that no true events escape detection (ie, no false-negatives). This process leads to more spurious events, but that is rectified by the EPAC process to remove them (ie, no false-positives). This process ensures the best overall data. As mentioned by the authors, this rate of rejection can also be affected by the end points chosen. A " softer " end point such as unstable angina typically has a much larger rejection rate than a " harder " end point such as central nervous system hemorrhage. Furthermore, if local investigators submitting potential stroke end points are neurologists, the rate of rejection is lower than in cardiology-based multicenter trials in which typically 20% to 30% of stroke end points are rejected by central neurology adjudicators. Second, failing to perform the adjudication process will result in more noise in the end point data. Even if this noise is balanced by treatment arm, this could significantly impact the study by reducing the statistical power. (For example, Arm A has 75 true events and is contaminated by 25 false-positives versus Arm B has 50 true events and is also contaminated by 25 false-positives. A statistical test using the investigator-assigned results would likely not result in a statistically significant result, but the adjudicated results would.) Another way to look at this issue is at the rejection rate and whether it is differential by treatment arm. This appears not to have been an issue for PROGRESS, …