Detecting adverse drug events through data mining

In the decade since the release of the Institute of Medicine (IOM) report To Err Is Human: Building a Safer Health System, the health care field has steadily increased its understanding of adverse drug events (ADEs).1 Initially, research in the area focused on determining the rate of adverse events in hospitalized patients.2 An important question throughout this initial stage was to determine the best method for collecting information about ADEs. One particular focus was to determine whether various reporting methods detected different types of ADEs. The explored reporting methods included volunteer reporting and prompted reporting from physicians, nurses, pharmacists, and patients, along with manual review of the medical chart.3 These ADE studies invariably found that medical record review was superior in terms of the number of events detected. Unfortunately, reviewing medical records for specific "trigger events" is time-consuming and thus has limited applicability to clinical practice. These trigger events include specific medication orders, abnormal laboratory test values, and nurses' or physicians' notes suggestive of an ADE.4 An additional challenge to successfully detecting ADEs is that each reporting method detects a unique subset of events.5