• Nandini Dendukuri, Jane McCusker, François Bellavance, Sylvie Cardin, Josée Verdon, Igor Karp, and Eric Belzile.
• Technology Assessment Unit, McGill University Health Center, Montreal, Canada. nandini.dendukuri@mcgill.ca
• Med Care. 2005 Mar 1; 43 (3): 266-75.

BackgroundEmergency department (ED) use in Quebec may be measured from varied sources, eg, patient's self-reports, hospital medical charts, and provincial health insurance claims databases. Determining the relative validity of each source is complicated because none is a gold standard.ObjectiveWe sought to compare the validity of different measures of ED use without arbitrarily assuming one is perfect.SubjectsData were obtained from a nursing liaison intervention study for frail seniors visiting EDs at 4 university-affiliated hospitals in Montreal.MeasuresThe number of ED visits during 2 consecutive follow-up periods of 1 and 4 months after baseline was obtained from patient interviews, from medical charts of participating hospitals, and from the provincial health insurance claims database.MethodsLatent class analysis was used to estimate the validity of each source. The impact of the following covariates on validity was evaluated: hospital visited, patient's demographic/clinical characteristics, risk of functional decline, nursing liaison intervention, duration of recall, previous ED use, and previous hospitalization.ResultsThe patient's self-report was found to be the least accurate (sensitivity: 70%, specificity: 88%). Claims databases had the greatest validity, especially after defining claims made on consecutive days as part of the same ED visit (sensitivity: 98%, specificity: 98%). The validity of the medical chart was intermediate. Lower sensitivity (or under-reporting) on the self-report appeared to be associated with higher age, low comorbidity and shorter length of recall.ConclusionThe claims database is the most valid method of measuring ED use among seniors in Quebec compared with hospital medical charts and patient-reported use.

23 keywords...

hide…