FDA Assesses Antipsychotic's Link to Certain Tumors

To find the reports, the team used a technique dubbed the Multi-Item Gamma Poisson Shrinker (MGPS). The technique was developed through a cooperative research development program of the FDA and outside statisticians and information-systems experts.

“We are just finishing that agreement and now moving this tool into the hands of our safety evaluators to use on a more routine and regular basis,” Seligman explained.

The FDA's AERS database receives an estimated 420,000 reports a year, Seligman continued. Since its inception in 1968, the database has compiled more than 2.5 million reports for more than 8,000 drugs and biological products.

“We only have a staff of 24 safety evaluators who are monitoring these reports,” he added, “and so part of the motivation for developing this tool was to help us have another way—other than the human mind—to sift through all of these data and look for potential statistical associations that need subsequent careful review and follow-up.”

The MGPS is a complex Bayesian computer-based algorithm that calculates relationships between a drug and a particular adverse event to determine if the event occurs on a higher-than-expected basis. The resulting calculation (a“ reporting rate”) can be adjusted for estimated patient-exposure rates (how many patients have taken the specific medication for how long a period) and for background rates of a given event in a specific population. The result is also adjusted for common demographic variables that could confound the reporting rate. The reporting rate for the adverse event in people taking the drug in question is then compared with reporting of the same adverse event over the entire 35 years of data for all medications in the database.

A reporting rate that is statistically significantly higher than the rate for the whole database is referred to as a “signal.”

Researchers both inside and outside the FDA believe the method is reliable enough to provide safety evaluators with consistent signals that deserve further scrutiny.

Using MGPS to mine AERS has its limitations. Adverse events reported to AERS are widely accepted as representing only a fraction of the true number of such events. This underreporting, along with incomplete reports, makes it difficult to determine causality in a drug/adverse-event combination.

In addition, numerous factors may bias reporting of particular drug/adverse-event combinations. For example, the recent FDA warnings regarding adverse events associated with antipsychotics and antidepressants likely increased awareness of those events among both prescribers and patients, possibly leading to an increase in reporting.

Nevertheless, “validated data-mining techniques offer great promise to enhance pharmacovigilance practices,” Doraiswamy and Szarfman wrote in an article in the September 2004 Pharmacotherapy.

The FDA's Seligman agreed, noting, “Safety evaluators need tools that will allow them to compare drugs within classes, drugs across classes, and multiple drugs in combination that may be associated with a particular adverse event. They need tools that will also allow them to combine different sets of adverse-event terms that might all be associated with a particular outcome. In the case of risperidone, for example, we looked at various terms, such as `elevated prolactin,' `hyperprolactinemia,' and `galactorrhea,' as well as `benign pituitary tumor' or `pituitary adenoma.'”

In the end, the finding of a risperidone/pituitary tumor signal is not all that surprising, Seligman told Psychiatric News.

“Given that you have here a drug that is well known to cause elevated prolactin levels, and there is pretty solid evidence of induction of pituitary adenomas in rodent models,” the finding appears to confirm the validity of the new data-mining technique. In this case, the method successfully identified an increased frequency of reporting of an adverse event that would, in theory, not only be plausible, but also predictable.

However, Seligman cautioned, the technique cannot confirm or rule out a causal association between a specific drug—or group of drugs—and an adverse event. Finding a signal simply suggests the need for follow-up.

The FDA-approved labeling of risperidone does include a precaution regarding the drug's documented tendency to induce hyperprolactinemia. The label notes that increased prolactin levels can increase rate of growth in about one-third of human breast cancers. In addition, the label acknowledges that “an increase in pituitary gland, mammary gland, and pancreatic islet cell hyperplasia and/or neoplasia was observed in the risperidone carcinogenicity studies conducted in mice and rats.”

Doug Arbesfeld, a spokesperson for Johnson and Johnson Pharmaceutical Services, whose Janssen unit markets Risperdal, told Psychiatric News, “We take seriously all reports concerning patient safety, and we agree with the study's authors that their findings are preliminary.” He added, “It is something that we are aware of and that we are looking into.” ▪