Using structural-nested models to estimate the effect of cluster-level adherence on individual-level outcomes with a...

Area of Science:

• Biostatistics
• Epidemiology
• Public Health

Background:

• Estimating causal effects of adherence in randomized trials often involves instrumental variables to address unmeasured confounding.
• Common frameworks include regression with endogenous variables, principal stratification, and structural-nested modeling.
• Existing methods can yield ambiguous causal interpretations or rely on difficult-to-interpret assumptions.

Purpose of the Study:

• To develop a structural-nested modeling approach for estimating the causal effect of cluster-level adherence on individual-level binary outcomes in three-armed cluster-randomized trials.
• To extend the methodology to accommodate polytomous adherence and cluster-randomized trials with unequal individual selection probabilities.
• To provide a computationally feasible implementation of the proposed method.

Main Methods:

• Developed a structural-nested modeling (SNM) approach to address challenges posed by principal stratification in complex scenarios.
• Extended SNM to handle cluster-randomized trials with unequal probabilities of individual selection.
• Implemented the SNM approach using accessible programming methods.

Main Results:

• The developed SNM approach provides unambiguous causal interpretations.
• The method successfully accommodates cluster-randomized trials with unequal individual selection probabilities.
• The approach demonstrates good performance in simulations, though model misspecification can lead to issues with the estimating equation.

Conclusions:

• Structural-nested modeling offers a viable and interpretable framework for estimating causal effects of adherence in complex cluster-randomized trials.
• The extended methodology is applicable to trials with unequal individual selection probabilities.
• The approach was successfully applied to evaluate the impact of a water, sanitation, and hygiene intervention on pupil absence.