Usual Approaches to Monitoring Safety

Many adverse events that occur during the course of a trial are expected. Some may be due to the disease itself. Others may be caused by the product under study, by comorbidities, or by drug interactions. For studies of drugs or biolog-icals, sometimes a particular class of product may be known to cause specific adverse events. Early phase studies should have already elucidated the mechanism of action of the drug and its pharmacokinetics, so that common adverse events attributable to the drug should have already been identified. In Phase 3 trials of diseases that are not life-threatening, serious adverse events will usually occur only rarely. In life-threatening disease, adverse experiences that reflect the process of the disease are common, but their rates are expected to be similar in the treated and control groups. Of particular concern are unanticipated serious or life-threatening events that arise during the course of the trial.

Standard statistical guidelines can specify methods for monitoring anticipated adverse events. When prior experience provides a background rate, a formal statistical monitoring plan can detect rates in excess of expectation. For example, many injected vaccines cause injection site reactions, such as soreness or redness. Similarly, bleeding is common in trials of antithrombotic agents. If historical data provide an expected event rate, the investigators can design a monitoring plan to identify an unacceptably high rate. Although it is generally preferable to assess safety by comparing the event rates by treatment group, sometimes it is also useful to compare the rates within a trial to the rates expected from historical data. A DSMB should know if the rates in the trial in both the treatment and control arms are meaningfully above expectation. As we show below, for very rare events, a few occurrences in the treated group and none in the control can lead to concern.

For unanticipated events, a DSMB risks reacting to a falsely "discovered" endpoint [Jennison and Turnbull (2000) and Mehrotra and Heyse (2004)], for the event may have occurred in the treatment arm purely by chance. Therefore, DSMBs typically wish to dampen a rush to judgment about the product or intervention. In prevention trials among healthy volunteers or in trials of diseases or conditions not usually accompanied by many types of serious adverse events (e.g., relief of minor headache pain), one analysis might compare the total number of adverse events to a standard rate known from historical data (a hurdle) or to the comparator (a placebo or active control). The observation that the patients in the test group had more adverse events than expected or more than those in the comparator might be disconcerting.

Trials in diseases with many associated adverse events (e.g., Type 2 diabetes) may compare only the aggregate of all serious adverse events to those that lead to hospitalization or are life-threatening. For diseases in which many serious adverse events occur over a short time period (e.g., sepsis), such an approach will generally not be useful, for nearly all patients will experience serious adverse events. Instead, focusing on particular serious adverse events will likely increase the specificity of the comparison.

A DSMB that encounters a worrisome excess of adverse events in a Phase 3 clinical trial typically responds by monitoring that event more frequently or spending more time discussing it. Sometimes, the DSMB waits until the difference between the number of events in the treated and control group hits some nominal level of significance. The statisticians often complain that such an approach is biased, because it includes the event that led to the monitoring, nonrigorous, because it does not preserve the Type 1 error rate, and is woefully lacking in power. We have rarely seen a DSMB institute a formal plan for subsequent monitoring of that event.

0 0

Post a comment