In their recent paper, Ioannidis and colleagues [1] concluded that early initiation of zidovudine therapy offers a benefit that decreases over time. We believe that the data presented do not support that conclusion. Our major concern is the number of trials selected in which the number of patients lost to follow-up was high compared with the number of primary end point events (Table 1). Ioannidis and colleagues stated that "although these deviations and withdrawals do not appear large enough to markedly influence the results, they may affect the precision of the data." When the ratio of the number of patients with an event to the number of patients lost to follow-up is less than 1, we believe that the results might indeed be inverted, depending on the hypothesis made about the patients lost to follow-up. A particular bias may be introduced if the losses differ by treatment group [2]. Patients who drop out and those who continue to participate in the study may have different risk factors. These differences may place them at a higher (or lower) risk for the event of interest. In most of the studies presented, the number of patients lost to follow-up is particularly high and strikingly more frequent in studies with shorter follow-up periods. The fact that studies with more losses relative to events are also those with high significance levels adds to suspicions about the quality of these clinical trials. The only two studies (Concorde and Veterans Administration [VA]-298) for which the ratio of end points to events is greater than 2 are those with longer follow-up periods. Of note, both these studies lack statistical significance for the primary end point. Because individual patient data were unavailable, Ioannidis and colleagues should have tested the sensitivity of the results to inclusion or exclusion of studies with large numbers of losses compared with events. It is only after completion of those complementary analyses that a conclusion can be reached about the efficacy of early zidovudine treatment.