Presentation on theme: "Automated Surveillance for Adverse Drug Events at Duke University Health System Peter M. Kilbridge, M.D. Washington University School of Medicine AHRQ."— Presentation transcript:
Automated Surveillance for Adverse Drug Events at Duke University Health System Peter M. Kilbridge, M.D. Washington University School of Medicine AHRQ Technology and Patient Safety September 26, 2007
2 How can we measure adverse events? Most organizations: Voluntary reporting is the only mechanism available Anecdotal Misses the majority of events Chart review: Very resource-intensive Variable in effectiveness (e.g., implicit vs. explicit review methodologies) Not comprehensive Computerized surveillance: effective, but rarely employed Specialized IT requirements Still requires significant clinical resources
3 How automated ADE detection works: Computerized surveillance of patient data searching for evidence suggesting that an ADE has occurred Rules engine uses combinations of data to detect potential ADEs and fire signals Signals that fire are investigated by study clinicians to determine causality:whether they represent true ADEs, and if so, grade severity and nature of ADEs. Same rules engine for 3 hospitals
4 Examples of Surveillance Rules Antidote ordered/dispensed Toxic serum drug level Physiologic parameters changing + active order for a medication Heparin AND rapidly falling platelet count Nephrotoxic medication AND rising Cr Hypoglycemia AND D25, D50 ordered INR > 4 AND active order for warfarin...
5 Scoring Causality and Severity Causality: Probability that a signal represents a true Adverse Drug Event Naranjo algorithm (Clin. Pharmaco. Ther. 1981;30:239-245): Score must be probable or defininte to count as an ADE. Severity: Duke severity scoring system, 0-6 scale Like NCC-MERP, 3 and over equals harm to the patient
7 Example: Coagulation-Related ADE Alert: warfarin and an INR > 4 Investigation: Adult patient previously admitted for A. fib, discharged on warfarin. Patient returned to the ED 10 days later feeling unwell; while in the ED, vomited 200cc of bright red blood. INR was 12.3. Patient required FFP and vitamin K, and was transferred to the ICU.
8 Sometimes we find something else: Alert: Naloxone Investigation: Patient administered Midazolam 2mg IV and Fentanyl 50 mcg for upper GI, plus ? sprays cetacaine for procedure. Later found unresponsive, hypotensive, with respiratory compromise. Naloxone given with no response. Methemoglobin level =13.7. Patient administered methlyene blue 90mg IV with reduction of methemoglobin level to 1.3.
15 Measures to address the problem at problem hospital: Clean the rooms of C. diff patients with bleach; switching wipes used on wards to hypochlorite-based product Sent letters to MDs, hospital personnel that alcohol foam shouldn't be used for hand hygiene when C. diff a concern; isolation signs updated to include same information List of attendings and rooms of patients with C. diff to get a better handle on the problem; feedback as available Result:
16 C. difficile colitis, 2 hospitals: Intervention
18 Pediatrics: ADE vs. Voluntary Reporting (12 mo)
19 Automated ADE Surveillance: Challenges Looking where the light is: we are limited by available data types Performs consistently across large populations, but is not comprehensive Resource requirement for evaluations How best to use the data on ADE incidence Used as a primary measure of medication safety? Is it used to design, implement, monitor safety improvements?
20 Automated ADE Surveillance: Next Steps At Washington University / St. Louis Childrens Hospital: Use Event Detector expert system to detect ADEs in pediatric inpatients across SLCH Surveillance of pediatric patients with chronic disease in the ambulatory setting and across transitions in care (AHRQ R18 Award): Cancer, Sickle Cell Disease, Cystic Fibrosis Data from clinic notes (text analysis), pharmacy, laboratory, ED, inpatient and ambulatory EMRs New trigger types for these particular patient populations