The association between asthma drugs and severe life-threatening attacks

Objectives: To measure the association between asthma drugs and death or ICU admission due to asthma (severe life-threating attack of asthma [SLTA]), and to assess the possibility that these associations may not be causal but due to the prescription of these drugs to patients with more severe disease (confounding), Design: Retrospective cohort study of 655 asthmatics who attended an emergency department in 1986 to 1987 followed till death or May 1989. Methods: Outcome events were death or ICU admission due to asthma (SLTA), All hospital attendances were identified and patients classified at each according to drug exposure and a wide variety of measures of asthma severity. Incidence rates were computed as total outcome events divided by person-time contributed for each subject classified according to drug use and asthma severity. Rate ratio (RR) estimates for severe asthma outcomes associated with use as compound to nonuse of asthma drugs were calculated, Severity markers were identified and used to adjust the crude RR estimates, Results: One hundred five SLTAs (15 deaths, 90 ICU admissions) occurred in 66 patients, Like inhaled fenoterol, oral beta-agonists, theophylline, cromolyn, inhaled steroids, and oral steroids were all associated with an increased risk of SLTA. When adjusted progressively for measures of severity, these increased risks became insignificant except for cromolyn. Conclusion: Unadjusted RR estimates for severe asthma events comparing exposure to a particular drug with nonuse are overestimates due to confounding. Control with two severity markers (hospital admission in the last year, use of oral corticosteroid at the time of previous admission) removes some confounding but control for additional severity markers not available in previous studies reduces the effect estimates further, These results suggest that the problem of confounding is substantial in nonrandomized epidemiologic studies of asthma drugs, Previous studies reporting RR estimates are likely to be confounded.