This section presents three examples. The first shows when a single rare, unanticipated adverse event could serve as a sentinel for further monitoring. In the second, the sentinel is a small number of unanticipated events of a specific type. The third case provides an example where an increased rate of an adverse event could trigger further formal monitoring.
Example 4.2.1 (A single sentinel event.) A death occurred early in the course of a clinical trial of a childhood vaccine. The trial was so small that any death was unexpected. Consequently, the death caused serious concern for the DSMB; members wanted to be confident that the death did not indicate an unanticipated problem with the vaccine.
Example 4.2.2 (A small number of events comprising the sentinel.)
In a clinical trial randomizing subjects to either a new or a control vaccine, three subjects in about 900 in the new vaccine group and one of 300 in the control vaccine group experienced a specific serious adverse event. The p-value for the difference in these rates is 0.6. The FDA was consulted about the findings and was worried not about the difference between the two rates, but about the overall event rate itself, because the anticipated serious adverse event rate had been roughly 1 in 10,000, substantially lower than the observed rate of 1/300.
Example 4.2.3 (A sentinel event rate.) The Women's Health Initiative (WHI) was composed of four trials. Here we discuss the PERT trial (Progesterone-Estrogen Replacement Therapy), the component that studied the hypothesis that postmenopausal treatment with estrogen and progesterone reduces the rate of myocardial infarction. This trial, designed to last 12 years, had many endpoints and several major hypotheses. The investigators and the DSMB published their monitoring plan for safety and efficacy [Freedman et al. (1996)]. The investigators anticipated that estrogens would reduce the rate of myocardial infarction, hip fracture, and colorectal cancer, but increase the rate of invasive breast cancer, pulmonary embolism, and endometrial cancer. Stroke was a "monitored" endpoint, that is, one that the DSMB was to look at carefully over the course of the trial, but no a priori hypothesis was established concerning the direction of the effect.
Early in the trial, the DSMB noted a higher incidence of stroke and pulmonary embolism, and a lower rate of breast cancer in the PERT group. Quite surprisingly and contrary to the primary hypothesis of the study, the PERT group also had an early higher incidence of myocardial infarction than the placebo group. The trends for pulmonary embolism, stroke, and myocardial infarction continued throughout the trial, and the breast cancer rates became higher in the PERT than in the placebo group.
The DSMB of the WHI recommended stopping the PERT trial early because interim results showed:
• An excess of invasive breast cancer on the treated arm. This excess exceeded the prespecified boundary for harm for breast cancer.
• A myocardial infarction rate in the treated arm substantially higher than in the placebo arm. The monitoring plan for the WHI included an O'Brien-Fleming bound for benefit on myocardial infarction. Because the plan had no a priori bound for harm, the observed excess did not cross any prespecified monitoring boundary.
• Accounting for both positive and negative effects on seven monitored endpoints, an adverse effect of treatment that was consistent with the pre-specified boundary for overall harm.
The National Heart, Lung, and Blood Institute, the sponsor of the trial, stopped the PERT trial soon after the DSMBs recommendation [Writing Group for the Womens Health Initiative Investigators (2002)].
Monitoring the PERT trial was complicated by the fact that the WHI included an estrogen-alone (ERT) trial, which studied women who had a hysterectomy prior to randomization. Early in the course of the study, the DSMB had observed more myocardial infarctions in both hormone-treated groups (ERT and PERT) than in their respective placebo groups. In response to these observations, the DSMB requested periodic updates on the overall event rates and based the decision to continue the study on its best judgment at each meeting. The DSMB could have regarded the excess rate of myocardial infarction as a sentinel rate.
Was this article helpful?