Logo: to the web site of Uppsala University

Introduction: In recent years, the application of peripheral blood eosinophil (PBE) as a biomarker for predicting exacerbation in chronic obstructive pulmonary disease (COPD) patients has been widely discussed. Previous analyses have described a continuous relationship between baseline PBE counts and exacerbation rate; however, with some limitations.

Aim: To establish a repeated time to event (RTTE) model linking the hazard of moderate or severe COPD exacerbation risks to the different treatment regimens with time-varying PBE count as a covariate.

Method: Data from IMPACT trial, a phase III randomized clinical trial (n=10355) that compared once-daily triple therapy (fluticasone furoate (FF)/umeclidinium (UMEC)/vilanterol (VI)) with dual therapies (FF/VI and UMEC/VI) were used. Different parametric hazards (exponential, Weibull, Gompertz) were explored for developing the RTTE model. Additionally, a surge function was included to explain the early high exacerbation rate. The potential covariates, including PBE count, were investigated through exponential relation to the underlying hazard.

Results: A Gompert hazard with the addition of a surge function for each arm described best the data, with a baseline hazard estimate of 0.00336 days-1 for VI with proportional reductions of 0.647 and 0.77 for FF and UMEC, respectively. As a result, the estimated hazard of FF/UMEC/VI was the lowest (0.001674 day-1), compared to an estimated hazard of 0.002174 day-1 and 0.002687 day-1  for FF/VI and UMEC/VI, respectively. The estimated shape parameters were -0.0007113 day-1 and -0.001287 day-1 for FF/VI and UMEC/VI, respectively. The surge of the hazard reached its peak on day 16, increasing the hazard by 35.2 %. The effect of time-varying PBE correlated positively to the hazard (a covariate effect of 0.283, centered of 160  cells/μL), with a 2.8% decrease and a 2.9% increase of hazard at a PBE of 60 cells/μL and 260 cells/μL, respectively.

Conclusion: The developed RTTE model captured the observed data well across treatment arms. The time-varying PBE count improved significantly the model fit compared to the baseline PBE count, describing an increased hazard with increasing PBE counts. This project work has the potential to develop simulation studies (the occurrence of exacerbations from different scenarios with and without ICS at different PBE counts), a categorical events model (a separate hazard for different exacerbation severity), and a model assessing exacerbation interval.