Statistical methods to control for confounders in rare disease settings that use external control

Area of Science:

• Biostatistics
• Clinical Trial Design
• Pharmacology

Background:

• Drug development for rare diseases often involves small patient cohorts.
• External control groups can be larger but introduce confounding challenges.
• Statistical methods for controlling baseline confounding in small sample trials are not well-defined.

Purpose of the Study:

• To evaluate statistical methods for controlling baseline confounding in rare disease clinical trials with small sample sizes.
• To compare the performance of matching, weighting, targeted maximum likelihood estimation (TMLE), and cardinality matching.
• To identify robust methods suitable for pediatric rare disease settings.

Main Methods:

• Extensive simulations were conducted to assess method performance.
• Commonly used matching and weighting techniques were examined.
• Targeted maximum likelihood estimation (TMLE) and cardinality matching were evaluated.
• Performance was assessed under various model specifications and treatment allocation ratios.

Main Results:

• Coarsened exact matching (CEM) and TMLE demonstrated robust performance across different model specifications.
• CEM is most suitable when the number of external controls significantly exceeds the treated group.
• TMLE performed better with less extreme imbalances in treatment allocation ratios.
• Bootstrap methods proved useful for variance estimation in small samples post-matching.

Conclusions:

• TMLE offers a flexible and robust approach for controlling confounding in small rare disease trials.
• CEM is a viable option when large external control groups are available.
• Bootstrap variance estimation is recommended for small sample matching procedures.