Regression discontinuity designs in epidemiology: causal inference without randomized trials

Area of Science:

• Epidemiology
• Biostatistics
• Public Health

Background:

• Regression discontinuity design (RDD) is a quasi-experimental method for estimating causal effects.
• RDD leverages threshold-based assignment to treatment, creating quasi-randomization for individuals near the cutoff.
• While prevalent in economics, RDD adoption in epidemiology remains limited.

Purpose of the Study:

• To describe RDD, its implementation, and causal inference assumptions.
• To demonstrate RDD's generalizability to survival and nonlinear models common in epidemiology.
• To apply RDD to determine optimal timing for antiretroviral therapy (ART) initiation in HIV patients.

Main Methods:

• The study describes the regression discontinuity design methodology.
• It adapts RDD for survival analysis and nonlinear models.
• Data from a South African HIV cohort (2007-2011) was analyzed using RDD.

Main Results:

• Patients initiating ART just below the 200 cells/microL CD4 count threshold had a 35% lower hazard of death.
• Hazard ratio for mortality was 0.65 (95% CI: 0.45-0.94) for early vs. deferred ART initiation.
• RDD provided a causal estimate of early ART's effect on mortality.

Conclusions:

• Regression discontinuity design is a valuable tool for causal inference in epidemiology.
• RDD can be effectively applied to survival data and complex epidemiologic models.
• Early ART initiation for HIV patients near the CD4 threshold significantly reduces mortality risk.