TheAssociation Between AsthmaDrugs andSevereLife-Threatening Attacks*

estimates. Results: Onehundred five SLTAs(15deaths, 90ICUadmissions) occurred in66patients. Likein¬ haled fenoterol, oralp-agonists, theophylline, cromolyn, inhaled steroids, andoral steroids were all associated with an increased risk ofSLTA.Whenadjusted progressively for measures ofseverity, these increased risks becameinsignificant except forcromolyn. Conclusion: Unadjusted RRestimates for severeasthma eventscomparing exposure toaparticular drug with nonuse are overestimates duetoconfounding. Control with twoseverity markers (hospi¬ tal admission inthelast year, use oforal corticosteroid atthetimeofprevious admission) removes some confounding butcontrol foradditional severity markers notavailable inprevious studies re¬ ducestheeffect estimates further. Theseresults suggest thattheproblem ofconfounding is substantial innonrandomized epidemiologic studies ofasthmadrugs. Previous studies reporting RR estimates arelikely tobeconfounded. (CHEST 1996; 110:1446-51)