RE: Some big shapes this morning…..at last!28 Jan 2026 19:37
Going back to the trial design I thought this overview could be useful:
Key Comparisons
Accuracy in MTD Selection: BOIN generally outperforms 3+3 by more accurately identifying the true MTD. In simulations across various scenarios BOIN achieves a 12-16% higher percentage of correct selection (PCS) of the MTD, and in cases where the MTD is at higher doses, it can nearly triple the PCS compared to 3+3.
Efficiency and Patient Allocation: BOIN is more efficient, often resulting in shorter trial durations and lower costs by adapting in real-time and allocating more patients to the MTD (especially at higher doses or with higher target DLT rates). The 3+3 design can prolong trials due to its conservative escalation and fixed cohort requirements, leading to fewer patients treated at effective doses.
Safety and Risk Management: While 3+3 minimizes overdose risk by being overly cautious (e.g., lower chance of assigning >60% of patients above the MTD), it increases under-dosing, where many patients receive suboptimal doses. BOIN balances this better with built-in overdose controls and dynamic adjustments, reducing under-dosing without significantly raising overdose risks, thus improving ethical standards by minimising patient exposure to ineffective or overly toxic doses.
Simplicity and Flexibility: Both are relatively easy to implement—BOIN requires no complex programming or simulations beyond basic setup, making it accessible like 3+3. However, BOIN offers more flexibility, such as variable cohort sizes (e.g., 4 patients) and adjustable target toxicity rates, whereas 3+3 is fixed to cohorts of 3 (or 3+3) and standard targets.
In summary, BOIN represents an evolution over 3+3, providing better performance across most metrics while maintaining ease of use. The 3+3 design is still widely used for its familiarity and safety focus in conservative settings but is increasingly seen as outdated for not fully utilising trial data.