Dose-Finding & Late-Onset Toxicity

TITE-BOIN Calculator

Real-time dose decisions for late-onset toxicity trials (Yuan, Lin, Li, Nie & Warren, 2018) — lets patients still being followed contribute time-weighted partial safety information, so accrual doesn't have to pause for complete follow-up on everyone. Runs entirely in your browser.

Current Dose: Enrolled Patients

Enter every patient treated at the current dose, whether their follow-up is complete or still pending.

φ — target

φ₁

φ₂

DLT assessment window (days)

Elimination cutoff P(p>φ)

Patients (days followed so far, DLT status)

Recommendation

Computed from the real time-weighted effective sample size, not a rounded approximation.

Enter patient data and click Get Recommendation.

How This Calculator Works

TITE-BOIN (Yuan, Lin, Li, Nie & Warren, 2018) extends BOIN to trials where toxicity can appear late relative to how quickly patients are accruing — common with immunotherapy and molecularly targeted agents. Waiting for every enrolled patient to complete their full DLT assessment window before making the next dose decision can slow accrual substantially; TITE-BOIN allows patients still being followed to contribute partial information instead.

The Decision Statistic

Every patient contributes a weight to the effective sample size n_eff:

Patient had a DLT: weight = 1 (in both n_eff and the DLT count x — the event already occurred)
Patient completed the full window, no DLT: weight = 1 (in n_eff only)
Patient still pending, no DLT yet: weight = (days observed) / (assessment window) — capped at 1

This is the standard linear/uniform weighting (assuming toxicity is equally likely to occur at any point within the assessment window). The same escalation and de-escalation boundaries λ_e and λ_d from standard BOIN are then applied directly to x/n_eff in place of the ordinary observed proportion.

Elimination Rule

The same overdose-control rule as BOIN — P(p > φ | data) > cutoff — computed via a Beta(1+x, 1+n_eff−x) posterior. Because the Beta distribution's parameters don't need to be integers, this generalizes cleanly to the fractional n_eff that time-weighting produces, without rounding.

References

Yuan, Y., Lin, R., Li, D., Nie, L., & Warren, K.E. (2018). Time-to-event Bayesian optimal interval design to accelerate phase I trials. Clinical Cancer Research, 24(20), 4921–4930.

Liu, S. & Yuan, Y. (2015). Bayesian optimal interval designs for phase I clinical trials. Journal of the Royal Statistical Society: Series C, 64(3), 507–523.