Marginal structural model to evaluate the association between cumulative osteoporosis medication and infection using...

Area of Science:

• Pharmacovigilance
• Epidemiology
• Biostatistics

Background:

• Suboptimal persistence to osteoporosis (OP) treatment is common.
• Treatment discontinuation/switching can introduce time-varying confounding.
• This confounding may bias the observed association between OP medications and infection risk.

Purpose of the Study:

• To assess the association between bisphosphonate (BP) treatment and infection incidence in postmenopausal women.
• To evaluate the impact of time-varying confounding on this association using a marginal structural model (MSM).

Main Methods:

• Utilized a US insurance database to analyze postmenopausal women.
• Employed a marginal structural model (MSM) to account for time-varying confounding.
• Stabilized weights were estimated by modeling treatment and censoring processes.

Main Results:

• Bisphosphonate (BP) treatment showed opposite associations with overall infection risk in unweighted Cox models versus weighted MSMs (IRR 1.15 vs. 0.79).
• BP treatment was consistently associated with a lower risk of serious infection in both models (IRR 0.79 vs. 0.71).
• Similar results were observed when current and past treatments were assessed simultaneously.

Conclusions:

• Discrepancies in effect estimates for overall infection suggest susceptibility to time-varying confounding.
• Analyses of composite outcomes with varying disease severity may be more prone to confounding.
• Marginal structural models (MSMs) are crucial for accurately assessing treatment effects in the presence of time-varying confounding.