A sequential, multiple assignment, randomized trial design with a tailoring function

Area of Science:

• Clinical Trial Methodology
• Biostatistics
• Machine Learning in Healthcare

Background:

• Sequential Multiple Assignment Randomized Trials (SMARTs) are adaptive clinical trial designs.
• Current SMART designs often rely on binary tailoring variables, limiting flexibility.
• Developing dynamic treatment regimens (DTRs) requires efficient estimation methods.

Purpose of the Study:

• To introduce a new SMART design utilizing a continuous tailoring function instead of a binary variable.
• To enable simultaneous development of tailoring variables and estimation of DTRs.
• To offer a more flexible and efficient alternative to existing SMART designs.

Main Methods:

• Application of tree-based regression learning and Q-learning for DTR development.
• Comparison with balanced randomized SMART and typical SMART designs.
• Utilizing a continuous outcome for second-stage treatment decisions in SMARTs.

Main Results:

• The proposed SMART design with a tailoring function efficiently estimates DTRs.
• This design is more flexible across various scenarios compared to traditional SMARTs.
• It removes the necessity for a predefined binary tailoring variable.

Conclusions:

• SMARTs employing a tailoring function offer enhanced flexibility and efficiency in clinical trial design.
• This methodology facilitates the development of personalized treatment strategies.
• The approach advances clinical trial methodology for adaptive treatment selection.