Separable effects for adherence.

Area of Science:

• Pharmacoepidemiology
• Causal Inference
• Biostatistics

Background:

• Comparing medication effectiveness is challenging due to differing patient adherence rates.
• Traditional per-protocol estimands, assessing effectiveness under sustained use, have limitations when sustained adherence is difficult to achieve.
• Existing methods struggle to disentangle medication effects from adherence behaviors.

Purpose of the Study:

• To introduce a novel class of estimands: separable effects for adherence (SEA).
• To propose methods for quantifying medication initiation strategies by isolating adherence mechanisms.
• To provide a more robust framework for evaluating treatment effectiveness in real-world scenarios with variable adherence.

Main Methods:

• Developed separable effects estimands that modify medications to hold adherence components constant.
• Utilized causal graphs to represent and interrogate assumptions about treatment component mechanisms.
• Described an algorithm for constructing causal graphs for separable effects and proposed semi-parametric weighted estimators.

Main Results:

• Separable effects estimands can quantify the effectiveness of medication initiation strategies.
• The approach allows for evaluation under the adherence mechanism of specific medications.
• Causal graphs facilitate reasoning about the assumptions required for identifying these effects.

Conclusions:

• Separable effects for adherence offer a valuable alternative to traditional estimands when sustained adherence is problematic.
• This framework enhances the ability to assess medication effectiveness and inform initiation strategies.
• The proposed methods and causal graph approach provide a rigorous foundation for pharmacoepidemiological research.