Table 3

Description and classification of field-defined and natural language processing systems for automated detection of harm*

Automated methodData source usedEvents identifiedComments
Complications Screening Program (CSP)8–10 46ICD-9 CM codesAdverse drug events, adverse surgical outcomes, infections, and miscellaneous complications such as fallsA computerised method for identifying potentially preventable complications of hospital care.
Health Evaluation through Logical Processing (HELP)11–14Electronic Medical Record: specifically including pharmacy, laboratory, radiology and surgery recordsAdverse drug events, adverse medical device events, infectionIntegrated electronic medical record of the LDS Hospital in Salt Lake City, Utah, which contains an interactive modular knowledge base that continually analyses information
Patient Safety Indicators (PSI)15–17 46Administrative data: billing information, ICD-9 CM diagnosis codes and procedure codesAdverse eventsA fully automated method developed by the Agency for Healthcare Research and Quality
Computer algorithms18–21Electronic Medical Record: components specific to the particular program: see online appendix 1Adverse events, adverse drug events, infectionSpecific, named computer programs
Lab signal detection tools22–26Laboratory DatabaseAdverse drug eventsAutomated tools search for key words or word combinations that signal potential or actual harm—for example, detection of elevated potassium levels
ICD-9 CM or billing code detection tools27–30Administrative data: ICD-9 CM or billing codesAdverse drug events, infections, surgical complicationsAutomated tools scan for diagnosis, discharge, or billing codes that signal potential or actual harm—for example, evidence of antibiotic exposure following a postoperative infection
Tools using computerised triggers31–45 50Electronic Medical Record: multiple sources such as pharmacy, laboratory, and microbiology databasesAdverse events, adverse drug events infectionAutomated tools using multiple triggers to signal actual or potential harm—for example, detection of elevated potassium levels (laboratory database) combined with certain medication administration (pharmacy database). Among the various tools included in this category, there are four named systems: Dynamic Pharmaco-Monitoring System, Nosocomial Infection Marker, Event Detector, New York Antimicrobial Resistance Project.
Natural language processing systems 33 34 47–49Free text in the Electronic Medical Record: discharge summaries, radiology reports, chart notesAdverse events, infectionSophisticated programs that ‘read’ free text via the application of computer logic
  • * Multiple detection strategies were used in several studies, including those that combined two or more field-defined systems,46 two natural language-processing systems,47 and both a field-defined and natural language-processing system.33 ,34