Predictors of late asthmatic response - Logistic regression and classification tree analyses

To identify predictors of the late asthmatic response (LAR), we reviewed da ta from 60 asthmatic subjects who had undergone allergen challenge over the past 5 yr (33 females, age 31.4 +/- 6.7 yr [mean +/- SD], FEV1 90% +/- 14% predicted). Variables considered likely predictors of LAR included baselin e FEV1, PC20 methacholine (PC20), sputum eosinophil percent, and the decrea se in FEV1 within 20 min of allergen challenge. A LAR (FEV1 greater than or equal to 15% fall between 3 and 7 h after challenge) was documented in 57% of subjects. A variety of logistic regression methods revealed a significa nt inverse association between LAR and PC20 (odds ratio [OR] = 0.14 [95% CI = 0.03-0.66]) and a positive association between LAR and the decrease in F EV1 at 20 min (OR = 1.18 [1.04-1.33]). Classification tree analysis reveale d that a threshold of 0.25 mg/ml for PC20 was most predictive of LAR; LAR d eveloped in 87% of those with PC20 less than or equal to 0.25 mg/ml (n = 23 ) and in 38% of those with PC20 > 0.25 mg/ml (n = 37). Notably, in subjects with PC20 > 0.25 mg/ml, the incidence of LAR increased from 38% to 57% if the allergen-induced decline in FEV1 at 20 min was greater than or equal to 27%. Surprisingly, baseline FEV1 and percent eosinophils in induced sputum were not significantly associated with LAR. We conclude that a threshold v alue of 0.25 mg/ml for PC20 methacholine is a good predictor of LAR. Measur ing the PC20 methacholine may be useful as a screening method to improve th e efficiency of identifying asthmatic subjects with a LAR.