Reliability of hospital data for population-based studies of air pollution.

The use of hospital databases for research into the respiratory effects of air pollution has been questioned. In an attempt to address that issue, reabstracts of 1,279 discharge records from 14 Montreal hospitals were compared with the universal health insurance database of Quebec. Agreement levels on discharge diagnoses were 94.9% for asthma; 75.5% for all other respiratory diagnoses combined, including upper airway infections, pneumonia, and coronary obstructive pulmonary disease (COPD) (90% after ignoring disagreements between closely related respiratory diagnoses); and 93.1% for a nonrespiratory comparison group. Factors associated with misclassification included use of nonurgent admissions; delays in hospital admission from emergency rooms; and differences in levels of diagnostic agreement between hospitals, age groups, and outcome groups. These should be taken into account in air pollution epidemiologic research in which databases of the kind commonly maintained in North American health care systems are used.