Only one of five key Xanax trials deemed positive by FDA

In a new study, researchers found that four of the five original studies on the effectiveness of alprazolam (Xanax) did not show it to be better than placebo. Two of the negative studies remain unpublished, while two others were presented as positive upon publication despite the drugs’ failure to beat the placebo. Only one of the five studies considered by the FDA to approve the drug actually showed a positive result.

The conclusion? Publication bias significantly inflates the apparent effectiveness of alprazolam XR, the researchers write.

Xanax is the most prescribed benzodiazepine in the United States, according to researchers. It is also the most commonly abused benzodiazepine.

The article was authored by Rose Ahn-Horst of Harvard Medical School, Massachusetts General Hospital and McLean Hospital, and former FDA reviewer Erick Turner of Oregon Health & Science University and Veterans Affairs Affairs Portland Health Care System. It was published in Psychological Medicine.

Turner used FDA data to highlight the influence of publication bias and the lack of effectiveness of other medications, including antidepressants and esketamine.

Because the FDA requires pharmaceutical companies to submit data on all trials that will be used for regulatory approval, it is possible to review data from all pivotal studies conducted rather than just those that end up being published.

In this case, researchers found that the FDA had received data on five Phase 2 or Phase 3 trials of Xanax extended-release alprazolam XR for panic disorder. The FDA considered four of the studies negative, demonstrating that the drug did not beat placebo on all primary endpoints. They approved the drug based on the fifth study, which was considered positive.

Study 1, conducted between 1986 and 1989, was presented as a positive result in its published version. However, it was considered negative by the FDA statistician because it only reached statistical significance on five of the seven primary endpoints. The published version omitted the negative results, making the study positive.

Study 2, conducted between 1988 and 1990, was published as a positive study and the FDA recognized all seven endpoints as statistically significant. However, data from one of the three study sites was destroyed prior to FDA analysis by the study investigator, and the FDA reported that the validity of the reported data could not be determined. verified. This is the positive study used by the FDA to approve the drug. Notably, the FDA had initially required the usual two positive studies, but according to the researchers, it relaxed that requirement in this case.

Study 3, conducted between 1990 and 1991, ended in complete failure, concluding that the drug was no better than a placebo on any of its main outcomes. All five results were not significant. Yet this study was published as a positive trial after researchers omitted the primary endpoint and presented only preliminary results using a different statistical method than they had used in the FDA filings.

Study 4, conducted between 1990 and 1991, was not significant on all five primary outcome measures. The study was never published.

Study 5, conducted between 1994 and 1995, was considered a negative trial by the FDA, and the FDA did not include more information about its statistics in this case. The study was never published.

Analyzing the published literature, researchers found an effect size of 0.47 in favor of Xanax (alprazolam). However, when they added the unpublished data, they found that the effect size dropped to 0.33.

According to the researchers:

These findings arguably change the risk-benefit ratio of prescribing this benzodiazepine, particularly in light of recent attention to their contribution to the opioid crisis and the availability of safer alternatives.

****

Ahn-Horst, R.Y. and Turner, EH (2023). Unpublished trials of alprazolam XR and their influence on its apparent effectiveness in panic disorder. Psychological Medicine, 1-8. https://doi.org/10.1017/S0033291723002830 (Link)