RESULTS
The total study sample was 106 encounters. In 63 (59%) of these encounters, corticosteroid were prescribed; Table 1 summarizes encounter characteristics according to steroid administration. Fifty-three patients were represented by the 106 encounters, 11 patients (21%) were never given a steroid over all their hospitalizations and 27 patients (51%) were always administered a steroid, while the remainder (28%) had hospitalizations with and without steroids given. Twenty-nine (55%) of our study sample was hospitalized once, 19% was hospitalized twice, and 17% experiences ≥3 hospitalizations.
Our analyses revealed that >98% of documented reasons for steroid use was for poor improvement in clinical response. In our overall sample, encounters in which patients were given steroids had less improvement at midpoint (average percent change from admission in FEV1pp=4.9, SD=11.3) than encounters in which patients were not given steroids (average percent change in FEV1pp=20.1, SD=24.6; p-value<0.001; Table 1).
Because of the potential for confounding by indication, we generated a propensity score matched (PS-matched) sample, which controlled for the non-equivalent likelihood of steroid use in the overall sample. The PS-matched sample included 25 matched encounters, representing 36 patients total (Table 2, Figure 2). In the PS-matched sample, there was no longer a significant difference between groups for the change in FEV1pp at midpoint (p-value=0.661). FEV1pp did not differ between groups at baseline, at admission, or at discharge in either the overall sample or the PS-matched sample.