© 2004 Elsevier Ireland Ltd. All rights reserved. Background: Measuring survival from sudden out-of-hospital cardiac arrest (OOH-CA) is often used as a benchmark of the quality of a community's emergency medical service (EMS) system. The definition of OOH-CA survival rates depends both upon the numerator (surviving cases) and the denominator (all cases). Purpose: The purpose of the public access defibrillation (PAD) trial was to measure the impact on survival of adding an automated external defibrillator (AED) to a volunteer response system trained in CPR. This paper reports the definition of OOH-CA developed by the PAD trial investigators, and it evaluates alternative statistical methods used to assess differences in reported "survival." Methods: Case surveillance was limited to the prospectively determined geographic boundaries of the participating trial units. The numerator in calculating a survival rate should include only those patients who survived an event but who otherwise would have died except for the application of some facet of emergency medical care-in this trial a defibrillatory shock. Among denominators considered were: total population of the study unit, all deaths within the study unit, and documented ventricular fibrillation cardiac arrests. The PAD classification focused upon cases that might have benefited from the early use of an AED, in addition to the likely benefit from early recognition of OOH-CA, early access of EMS, and early cardiopulmonary resuscitation (CPR). Results of this classification system were used to evaluate the impact of the PAD definition on the distribution of cardiac arrest case types between CPR only and CPR + AED units. Results: Potential OOH-CA episodes were classified into one of four groups: definite, probable, uncertain, or not an OOH-CA. About half of cardiac arrests in the PAD units were judged to be definite OOH-CA events and therefore potentially treatable with an AED. However, events that occurred in CPR-only units were less likely to be classified as definite or probable OOH-CA events than those in CPR + AED units (43% versus 55%, odds ratio 0.78, 95% confidence interval 0.57-1.07). The study retained sufficient power to permit a statistical analysis of the alternative hypothesis that the CPR + AED method results in twice as many survivors as a CPR-only approach. The result is critically dependent on the denominator used for calculating survival rates; but the analysis does not require a denominator as the numerators will have identical Poisson distributions (counts for rare events) under the null hypothesis since randomization distributes the risk of cardiac arrest evenly between the two arms. Conclusion: Reported OOH-CA rates and survival rates vary widely, depending upon the definitions applied to events. Rigorous assessment of treatments applied to improve survival can be obscured by inappropriate definitions. Large-scale randomized interventions designed to improve survival from OOH-CA can be evaluated based upon the absolute numbersof patients surviving, rather thana changein the proportion surviving.