Verification of adverse events can be problematic, and issues regarding the reliability of such assessments have been raised (Sanazaro and Mills, 1991; Thomas et al., 2002). For some types of adverse events, such as ADEs, scales having high interrater reliability, such as the Naranjo algorithm, have been developed (Naranjo et al., 1981). Overall, reliability in identifying adverse events can be expected to be higher with the use of triggers than with chart review because the evaluation relates to a discrete event. Nonetheless, greater standardization in the verification of adverse events is important—for example, using highly structured definitions of events, as is the case for nosocomial infections, or tools similar to the Naranjo algorithm.
The size of patient safety databases at the state and regional levels will quickly become far too great for any individual to oversee their contents. Data mining will therefore be necessary to uncover patterns, test hypotheses, and even recognize whether individual new reports have been seen before.
Much clinical information is contained in clinical notes and incident reports. Natural language processing can be used to analyze such data. Research is needed to develop natural language processing tools for patient safety applications.
New methods are needed for promoting and speeding up the dissemination of knowledge and tools related to patient safety to aid and support health care administrators, care providers, and patients.
Existing knowledge and tools regarding patient safety need to be incorporated into audit criteria used to determine whether a health care organiza-