Source: Kilbridge PM, Noirot LA, Reichley RM, et al. Computerized surveillance for adverse drug events in a pediatric hospital. J Am Med Inform Assoc. 2009; 16(5): 607– 612; doi: 10.1197/jamia.M3167Investigators from St. Louis Children’s Hospital, St. Louis, MO (SLCH) performed a prospective evaluation of a rules-based computer surveillance program to detect inpatient adverse drug events (ADEs). The data monitored included patient-specific demographic, encounter, laboratory, and pharmacy information from hospital information systems for all admissions, excluding children with cancer, from February 1, 2008 to July 31, 2008.The computer program generated alerts based on combinations of the knowledge-based rules and the patient-specific data. The rule set included modifications specific for pediatric settings plus rules for medication-induced electrolyte abnormalities such as hyper- and hypokalemia requiring intervention. Alerts were stored and made available to two study pharmacists who independently reviewed all alerts according to an established methodology for: a) the occurrence of an adverse event, b) the likelihood that the event was caused by a drug, and c) the severity of the harm to the patient.During the six-month study period, SLCH had 6,889 non-oncology admissions representing 40,250 patient-days, for which the program generated 1,226 alerts, of which 160 were true ADEs, a rate of 4 ADEs per 1,000 patient days. Of the 160 ADEs, 135 resulted in temporary harm, 20 resulted in prolonged hospitalization, 4 resulted in permanent harm, and one resulted in death. The most common alerts associated with an ADE (true positives) were hypokalemia (66), hypomagnesemia (19), nephrotoxicity (18), and naloxone administration (9). The most commonly implicated drugs with ADEs were diuretics, antibiotics, immunosuppressants, narcotics, and anticonvulsants. The average age of patients suffering ADEs was 6.3 years and most ADEs occurred in intensive care units (cardiac 35%; pediatric 27%; and newborn 7.5%).Only three of the ADEs had been reported by physicians. The overall positive predictive value (PPV) of the rule set was 13%, with a wide range of PPVs for individual rules (0–100%). The authors conclude that automated surveillance is effective for detecting ADEs in hospitalized children.The detection and prevention of ADEs is an important and growing area of both inpatient1 and ambulatory2 patient safety and informatics research. Until recently, the identification of ADEs has been largely manual, retrospective, cumbersome, and time-consuming. Even when facilitated through electronic data collection methods, voluntary reporting misses many events. In response to these challenges, an important avenue of informatics research has arisen to find ways to use computers to automate the task of detecting ADEs, to make it real-time and proactive (ie, to prevent them) through the development of “triggers” or rules that associate specific findings (history, physical examination, laboratory results) and/or pharmacy orders with medical errors or adverse events.The true potential of triggers lies in linking them to “live” electronic patient data for clinical decision support that provides real-time knowledge-based alerts to providers (for patient care), and to administrators and researchers (for understanding and mitigating vulnerabilities in specific populations and clinical environments).In evaluating their computer program, the authors found ADE rates comparable with those found by Kaushal3,4 in a similar academic setting. Not surprisingly, they found 70% of ADEs occurring in pediatric/neonatal intensive care units, with over half of the data trigger events related to electrolyte abnormalities and a large proportion of ADEs due to electrolyte-wasting medications (diuretics, antimicrobials, and anti-rejection drugs). Most ADEs were of low severity.The average PPV of rules used by the program (13%), combined with the wide variances of PPVs of individual rules and their frequencies with regard to the study population, point to areas for further work in designing appropriate trigger-data combinations. In addition, just as pediatric inpatients differ from adult inpatients, so there may be differences between the study population and other pediatric populations (inpatients in community hospitals, oncology patients, etc.), that may impact the types of rules and data required to create effective triggers.As the authors point out, the development of effective triggers depends on several factors including a rule or rule set that matches the vulnerabilities of a specific population within a specific clinical environment or workflow, and a sufficient range (and quality) of clinical data. This is a rich area of applied clinical informatics research that will require collaboration among pediatric clinicians, pharmacists, laboratory medicine personnel, and informaticians.