Methods for clustered encouragement design studies with noncompliance and missing data.

Area of Science:

• Biostatistics
• Clinical Trial Design
• Causal Inference

Background:

• Encouragement designs estimate intervention effects when direct randomization is impossible.
• Clustered encouragement designs (CED) randomize at the cluster level but assess compliance at the unit level.
• Noncompliance and missing data pose challenges in encouragement designs.

Purpose of the Study:

• To reanalyze data from a chronic heart failure study to determine the effect of physician adherence on patient outcomes.
• To propose causal inference methodology for randomized CED with all-or-none unit-level compliance.
• To extend current methods for nonignorable missing data and utilize multiple imputation.

Main Methods:

• Utilized a clustered encouragement design (CED) framework.
• Applied causal inference methods for all-or-none unit-level compliance.
• Employed multiple imputation techniques to handle nonignorable missing data.

Main Results:

• Developed novel causal inference methodology for CED.
• Extended approaches to account for nonignorable missing data.
• Demonstrated the application of multiple imputation in causal inference for CED.

Conclusions:

• The proposed methods provide a robust framework for analyzing CED studies with complex data issues.
• Physician adherence is a critical factor in improving patient outcomes in chronic heart failure.
• This methodology can be applied to various settings with noncompliance and missing data.