Methods: We mailed surveys to 515 randomly selected physicians in the American College of Obstetricians and Gynecologists' Collaborative Ambulatory Research Network; 253 surveys (49.1%) were returned. Participants were presented with an abstract from a single randomized trial finding positive results for a fictitious new drug. Participants were randomly assigned to receive one version of an accompanying hypothetical scenario, which varied between participants on three dimensions: journal prestige (high vs. moderate), total number of trials published on this new drug (1 vs. 5), and level of conflict of interest ("no conflict": study funded by NIH, presented by independent physician at Grand Rounds; "presenter conflict": study funded by NIH, presented by pharmaceutical representative; "researcher conflict": study funded by pharmaceutical company, presented by independent physician). Participants answered questions about their likelihood of prescribing the new drug as a first-line therapy.

Results: There were few significant differences between conditions. However, physicians reported that they would be less likely to prescribe the new drug if the journal prestige were lower or conflict of interest (researcher or presenter) were present; the opposite changes significantly increased prescribing likelihood. Contrary to normative behavior, physicians in the presenter conflict condition were significantly more likely to prescribe the new drug when the study presented was one of five trials rather than the sole trial published (p=.02). If physicians accounted for selective presentation of studies, the opposite result would have been expected. Moreover, when participants were asked to guess how favorable the results of this study were toward the new drug, relative to the other trials published so far, the proportion saying that it was no more or less favorable or that they had "no way of guessing" was 66% in the no conflict condition, 55% in the presenter conflict condition, and 57% in the researcher conflict condition (p=n.s.).

Conclusions: Physicians may not consider that the studies they receive from pharmaceutical representatives are biased samples of the literature. If physicians do not recognize how bias enters the process, they cannot adjust for it.