A second potential limitation is that not all patients underwent an EP study. As noted previously, the clinical characteristics of patients having an EP study versus those not having an EP study did not differ in clinically relevant characteristics. Moreover, patients undergoing an EP study had similar event rates for VT or VF, CHF hospitalization or death, in follow-up to those patients not undergoing an EP study. Third, the findings concerning the predictive value of the EP test for VT or VF are specific to the EP testing protocol used. Protocols using more than three extrastimuli, which may induce VT in a greater proportion of patients (35- 37) or other variations could possibly have led to different findings. As noted previously, the MUSTT study, the largest other study prospectively after patients with remote MI, used a virtually identical protocol to the MADIT II study protocol. In addition, we found that the induction of VT tended to be inversely related to the occurrence of VF in follow-up, no matter which inducibility definition was used. Fourth, concerning the tendency of noninducible patients to have an increased likelihood of VF, the classification of events as VT or as VF relies on arbitrary definitions and does involve potentially subjective electrogram data interpretation. Thus, one should not overinterpret the tendency toward more VF in the noninducible group, a trend that was not statistically significant by multivariate analysis. Finally, these data pertain to the relationship between inducibility and subsequent VT or VF episodes. The occurrence of VT or VF treatment by an ICD should not be equated with a mortal event because some episodes undoubtedly would have terminated spontaneously if the ICD had not intervened. The magnitude of this effect is observed when comparing the incidence of a first ICD therapy for VT or VF, 27% at two years (26), with the difference in total mortality between ICD and conventional groups, 6% at two years (38).