Causal Effects on Nonterminal Event Time With Application to Antibiotic Usage and Future Resistance

Area of Science:

• Epidemiology
• Biostatistics
• Infectious Diseases

Background:

• Comparing antibiotic treatments for future resistance is complex due to differential patient survival.
• Existing methods for causal inference in semi-competing risks settings may exclude clinically relevant patient subpopulations.

Purpose of the Study:

• To develop a new causal inference framework for estimating the effects of antibiotic treatments on antibiotic resistance.
• To define a more inclusive principal stratum, the infected-or-survivors (ios), for causal effect estimation.
• To introduce the feasible-infection causal effect (FICE) for a broader patient population.

Main Methods:

• Utilized a semi-competing risks approach with a nonterminal event time for resistant infection.
• Defined the novel 'infected-or-survivors' (ios) principal stratum.
• Developed large-sample bounds and derived FICE identification using bivariate frailty illness-death models.
• Employed an expectation-maximization algorithm and Monte Carlo procedure for estimation.

Main Results:

• The proposed 'infected-or-survivors' (ios) subpopulation is more inclusive than previous definitions.
• The feasible-infection causal effect (FICE) provides a clinically relevant causal estimand.
• The methods were applied to detailed clinical data from a hospital setting.

Conclusions:

• The new framework and feasible-infection causal effect (FICE) offer a more comprehensive approach to studying antibiotic resistance.
• The methods are applicable to real-world clinical data, aiding in treatment comparison and resistance prediction.