Abstract: FR-PO918
Estimating Causal Effects of Glucocorticoid Use on Infection Rates in Glomerular Diseases Using a Novel Marginal Structural Proportional Rates Model
Session Information
- Glomerular Diseases: Potpourri
October 25, 2024 | Location: Exhibit Hall, Convention Center
Abstract Time: 10:00 AM - 12:00 PM
Category: Glomerular Diseases
- 1402 Glomerular Diseases: Clinical, Outcomes, and Therapeutics
Authors
- Zee, Jarcy, University of Pennsylvania Perelman School of Medicine, Philadelphia, Pennsylvania, United States
- Liu, Qian, The Children's Hospital of Philadelphia, Philadelphia, Pennsylvania, United States
- Glenn, Dorey A., The University of North Carolina at Chapel Hill Kidney Center, Chapel Hill, North Carolina, United States
- Denburg, Michelle, The Children's Hospital of Philadelphia, Philadelphia, Pennsylvania, United States
- Schaubel, Douglas Earl, University of Pennsylvania Perelman School of Medicine, Philadelphia, Pennsylvania, United States
Background
Estimation of time-varying glucocorticoid treatment effects on infections have been limited to risks of first infection events, thus ignoring recurrent infections. Given the common recurrences of infections in patients with glomerular disease, and the need to account for time-dependent confounders, novel statistical methods are required to estimate causal treatment effects on recurrent outcomes.
Methods
We developed a marginal structural proportional rates model (MSPRM) to estimate the causal effect of a time-varying treatment on recurrent outcomes in the presence of time-dependent confounding. We derived estimation and inference procedures and conducted simulations to demonstrate model performance. We used the MSPRM to estimate the effect of time-varying glucocorticoid use on recurrent infection-related acute care events (hospitalizations or ER visits) accounting for proteinuria and serum albumin over time in the Cure Glomerulonephropathy (CureGN) study. Baseline confounders included age, sex, race, Hispanic ethnicity, glomerular disease type, eGFR, comorbidities, tobacco use, and time from biopsy to CureGN enrollment.
Results
Out of N=2496 CureGN study participants, N=405 had at least one infection, with a total of 590 infections. The MSPRM estimated that glucocorticoid use increased rates of infection by 1.70 (95% CI 1.32-2.18) times compared to no use (Figure). Based on simulations, the MSPRM has little bias compared to standard proportional rates models, which suffer from confounding and/or overadjustment bias.
Conclusion
The MSPRM can estimate causal time-varying treatment effects on recurrent time-to-event outcomes using weights to balance time-dependent confounders. Including recurrent events can better capture disease burden and morbidity compared to only considering the first event, and our proposed model provides a semi-parametric method to analyze these longer-term outcomes.
Funding
- Private Foundation Support