Abstract
The problem of evaluating the long-term effects of recombinant human deoxyribonuclease (rhDNase) on forced expiratory volume in one second (FEV1) in cystic fibrosis (CF) patients was considered. A two-stage mixed effects model, incorporating relevant predictive variables, captured the diverse patterns of decline of FEV1 for patients with different demographic characteristics. Based on the results of modeling the dropout process, it is clear that the probability of early dropout was closely related to patient's responsiveness to rhDNase treatment. Failure to consider the existence of informative censoring severely biased the estimates of the rate of decline and affected the interpretation of the results.
Acknowledgments
The authors wish to thank Drs. A. DeVault, V. De Gruttola, D. Giltinan, K. Doksum, C. Johnson, S. Selvin, G. Thomson, and X. Tu for their fruitful discussions and support during the study.