Group sequential testing under instrumented difference-in-differences approach.

Area of Science:

• Epidemiology
• Biostatistics
• Observational Studies

Background:

• Unmeasured confounding poses a significant challenge to causal inference in observational studies.
• Instrumented difference-in-differences (iDiD) offers a method to address confounding by using instrumental variables and difference-in-differences.
• Existing methods may not provide valid inference in the presence of unmeasured confounders.

Purpose of the Study:

• To propose a novel group sequential testing method for causal inference using the instrumented difference-in-differences (iDiD) framework.
• To ensure valid statistical inference even when unmeasured confounders are present.
• To enable earlier detection of treatment effects in ongoing observational studies.

Main Methods:

• Developed a group sequential testing approach integrated with the iDiD method.
• Estimated average or conditional average treatment effects at sequential time points using accumulated data.
• Derived the joint distribution of test statistics under the null hypothesis using M-estimation and utilized alpha-spending functions for sequential boundaries.

Main Results:

• The proposed iDiD group sequential method provides valid inference in the presence of unmeasured confounders.
• Evaluated on synthetic data and a real-world healthcare database (Clinformatics Data Mart Database).
• Successfully detected a significant adverse association between rofecoxib and acute myocardial infarction much earlier than its market withdrawal.

Conclusions:

• The novel group sequential testing method enhances causal inference from observational data by addressing unmeasured confounding.
• This approach allows for timely identification of potential safety concerns or treatment effects.
• The method demonstrates practical utility in pharmacovigilance and clinical research.