The Childhood Asthma Management Program (CAMP) Trial is a multicenter, masked, randomized control trial conducted in 1995. The study included children aged 5 to 12 who had mild to moderate asthma, evaluated by screening criteria. Children were randomly assigned to one of three treatment groups: Treatment A (Budesonide), Treatment B (Nedocromil), and Treatment C (placebo). Our dataset was a teaching dataset provided by the National Heart, Lung and Blood Institute, which contained observations on 2/3rds of subjects from the original CAMP trial with statistical techniques used to anonymize the data.
This study aimed to investigate long term differences in the annual rate of growth in pulmonary function as measured by Forced Expiratory Volume Per Second (FEV1). Therefore, our primary research question assessed whether there are differences in annual rate of growth in pulmonary function between children assigned Budesonide, Nedocromil, and placebo. Our secondary research question was is the association in annual rate of growth in pulmonary function and treatment confounded by age or gender i.e., Did the randomization work?
Graphical methods were used to assess model assumptions. A multilevel model approach was implemented to model growth in FEV1 measurements over the study period on two levels: individual growth and average growth of the sample. Full Maximum Likelihood and Restricted Maximum Likelihood estimation were used to obtain unbiased parameter estimates, variance components, and goodness of fit statistics. A forward elimination approach was taken to layer variables one by one and the significance of variables was evaluated using likelihood ratio tests on nested models. The final model will only include variables that add to the predictive power of the model or are required to answer the research question. Once an optimal model is determined, error covariance structures will be compared and chosen based on goodness of fit statistics and model parsimony. All statistical modeling was conducted using SAS PROC MIXED using SAS 9.4, and SAS OnDemand for Academics.
Graphical analysis presented evidence that FEV1 measured after bronchodilation changed linearly with time and therefore it was appropriate to model to annual rate of linear growth. The interclass correlation coefficient was 0.404, meaning approximately 40% of the variation in FEV1 measurements was explained by differences in subjects. Post treatment FEV1 was strongly associated with linear time (R_ϵ^2: 0.893) as was expected. Model comparisons concluded there was no confounding by baseline age or gender, meaning randomization was successful. The optimal model regressed post treatment FEV1 on months since randomization and treatment group. The optimal error covariance structure was heterogenous autoregressive. The baseline measurement for the control group was 1.86 (1.79, 1.93) liters of air per second. There was no difference in baseline measurements for FEV1 between treatment groups. The annual growth rate for the control group was 0.23 (0.22, 0.24) liters of air per second per year. There was no difference in growth rates between Treatment A and placebo. There was a significant difference in annual FEV1 growth between Treatment B and placebo. FEV1 grew 8% more annually in Treatment Group B compared to placebo, averaging a change of 0.249 (0.224, 0.273) liters of air per second annually.
