Use administrative databases with caution

There were 2 striking results in the article by Sandra Dial and colleagues: a sharp increase in the number of cases of community-acquired Clostridium difficile-associated diarrhea in 2003 and 2004 in Quebec and a low rate of antibiotic exposure in the weeks preceding the infection.1 The authors used the provincial hospital discharge summary (MED-ECHO) database to identify the cases on the basis of the first hospital admission for which C. difficile infection was listed as the primary diagnosis. This was considered to be a clinically relevant case definition of community-acquired C. difficile infection.

However, MED-ECHO coding rules state that the primary diagnosis is “the most important condition encountered by the patient during his hospitalization. In most cases, it is closely related to the reason for admission. In a patient with multiple conditions, the physician should assign the primary diagnosis to the condition requiring the greatest use of medical resources during the hospital stay.”2 Consequently, in patients with severe C. difficile- associated diarrhea acquired during a hospital stay and requiring intensive care or a prolonged hospital stay, this condition is more likely to be recorded as the primary diagnosis than is the reason for admission.

This potential for misclassification has 2 major implications for the study by Dial and colleagues.1 First, one cannot exclude the possibility that the increase in the number of so called community-acquired C. difficile-associated diarrhea in 2003 and 2004 reflects an increase in the number of nosocomial cases, given that the spread of the NAP1 strain in Quebec was associated with more severe disease.3^,4 Second, as the index date of the potentially misclassified cases was defined as the admission date, antibiotics given during hospital stay were not included in the study. This would lead to an underestimation of the antibiotic exposure in these cases. The use of administrative databases requires a good understanding of the coding rules, which are not always suitable for epidemiologic analyses.