Active Clinical Trials for Personalized Medicine.

Area of Science:

• Biostatistics
• Clinical Trial Design
• Machine Learning

Background:

• Individualized treatment rules (ITRs) personalize medical care based on patient characteristics, enhancing treatment efficacy.
• Randomized clinical trials (RCTs) are a common source for estimating ITRs, but are costly and not optimized for this purpose.

Purpose of the Study:

• To propose a cost-effective method for estimating optimal individualized treatment rules (ITRs).
• To leverage active learning strategies within ongoing clinical trials to improve ITR estimation efficiency.

Main Methods:

• Developed an active learning approach to identify and recruit the most informative patients for ITR estimation.
• Utilized simulation studies and real-world clinical trial data to validate the proposed method.

Main Results:

• The proposed active learning method significantly outperforms existing approaches in estimating ITRs.
• Derived theoretical risk bounds that support the observed empirical advantages of the active learning strategy.

Conclusions:

• Active learning offers a cost-effective and efficient strategy for estimating individualized treatment rules from clinical trial data.
• This approach has the potential to significantly improve patient care through more precise treatment personalization.