Learn-As-you-GO (LAGO) trials: optimizing treatments and preventing trial failure through ongoing learning

Area of Science:

• Biostatistics
• Public Health
• Clinical Trial Design

Background:

• Adapting intervention packages during trials is common in public health but challenges standard statistical validity.
• Existing methods for adaptive trials (LAGO) are limited, particularly for continuous outcomes.

Purpose of the Study:

• To extend the Learn-As-you-GO (LAGO) methodology to continuous outcomes in adaptive public health intervention trials.
• To establish conditions for valid statistical inference when intervention packages are modified during a trial.
• To develop new statistical tools for analyzing adaptive trials with continuous outcomes.

Main Methods:

• Developed new mathematical theory for LAGO trials with continuous outcomes, distinct from binary outcome methods.
• Derived point and interval estimators for intervention effects under adaptive designs.
• Ensured validity of hypothesis tests for overall intervention effects.
• Created confidence sets for optimal intervention packages and confidence bands for mean outcomes.

Main Results:

• Established conditions for valid statistical inference in LAGO trials with continuous outcomes.
• Provided novel estimators and valid hypothesis testing procedures for adaptive intervention studies.
• Developed methods to identify optimal intervention packages and characterize outcome variability.

Conclusions:

• The presented theory and methods enable valid analysis of public health trials where intervention packages are adapted.
• This work supports the design and analysis of complex, adaptive large-scale intervention trials.
• The findings are crucial for implementation science research requiring flexible intervention strategies.