Various statistical approaches have been used to estimate lung function decline over age among individuals with cystic fibrosis (CF). We investigated the impact of different statistical models and real-world scenarios on results and conclusions regarding rate-of-decline.
The sample included individuals with CF aged >6 years in the CF Foundation Patient Registry (2003-2016). Marginal, mixed-effects and joint longitudinal-survival models for estimating rate-of-FEV1 decline were implemented under scenarios that mimic attributes of different available types of data varied by sample size, duration and frequency of follow-up. We assessed the impact of linear and nonlinear trajectories and different correlation structures.
Overall, all models indicated an approximately linear rate of decline until age 30 (-1.4% predicted/year) with minimal difference between marginal and mixed models; nonlinear models fit better than linear models. Joint models suggested more severe FEV1 decline over time. Mixed model estimates had more variability between scenarios than marginal models. Duration of follow-up was the only scenario that impacted estimates. Rate of FEV1 decline was associated with mortality across scenarios (estimated hazard ratio, HR and 95% CI for death/lung transplant, HR: 0.67, 95% CI: 0.66-0.68).
Choosing an appropriate modeling strategy depends on the research question and data structure. Longer follow-up is best characterized with nonlinear terms. Association of rate-of-decline and survivorship should be assessed even in younger cohorts.