The Triangulation WIthin a STudy (TWIST) framework for causal inference within pharmacogenetic research.

Area of Science:

• Pharmacogenetics
• Causal Inference
• Biostatistics

Background:

• Pharmacogenetic studies often rely on treated individuals, making strong assumptions to estimate genetically moderated treatment effects (GMTE).
• Violations of these assumptions can lead to biased estimates of GMTE.
• Existing methods may not fully leverage available data, particularly information from untreated populations.

Purpose of the Study:

• To review and extend methodological underpinnings for estimating genetically-driven treatment effect heterogeneity.
• To present a robust framework, 'Triangulation WIthin a STudy' (TWIST), for causal analysis within a single dataset.
• To provide guidance on selecting appropriate estimation strategies for GMTE.

Main Methods:

• Review of existing pharmacogenetic approaches for GMTE estimation.
• Development of a robust GMTE estimator incorporating untreated individuals.
• Clarification of Mendelian randomization and modified confounder adjustment for partial assumption violations.
• Introduction of the TWIST framework for triangulating evidence within a single dataset.
• Re-analysis of UK Biobank data (CYP2C19/Clopidogrel/stroke and APOE/statin/CAD).

Main Results:

• Demonstration that incorporating untreated individuals provides a robust GMTE estimate when standard assumptions are violated.
• Identification of conditions under which Mendelian randomization and modified confounder adjustment yield consistent GMTE estimates.
• Illustration of the TWIST framework's utility in strengthening causal analysis within a single study.
• Empirical evidence from UK Biobank data supporting the proposed methods.

Conclusions:

• The TWIST framework enables robust causal inference in pharmacogenetics by triangulating evidence from different estimation strategies within a single dataset.
• The proposed methods offer improved efficiency and reliability in estimating genetically moderated treatment effects, especially when standard assumptions are challenged.
• This work provides a decision framework to guide the selection and combination of estimators for accurate pharmacogenetic analyses.