BRIDGE Trials Calculator

BRIDGE trials reuse participant-level data from completed trials of AI models, reducing the enrollment requirement for subsequent trials. By transforming trials into adaptive, modular studies, our proposed design makes Level I evidence generation feasible for every model iteration, thereby accelerating cost-effective translation of AI into routine care.

See our paper for more details.
Our implementation and case studies can be found on Github.

Guide

⚡ Try example values

BRIDGE trial sample size calculations require:

Trial parameters: Type I error rate α, power 1-β, case-to-control ratio k, and superiority margin. Alternatively, a pre-computed baseline sample size can be provided.
Prior trial statistics: Number of patients enrolled in the prior trial and the case-to-control ratio used. Alternatively, a CSV with prior trial metadata can be uploaded.
Model concordance: The proportion of individuals identified as high-risk by the new model who were also classified as high-risk by the legacy model, P(R_L=1 | R_N=1).
Anticipated incidence: Expected incidence rates p_T and p_C in both concordant and discordant strata.

Trial parameters

I need to compute my baseline sample size

I already know my baseline sample size

Type I error rate, α

Power, 1 - β

Ratio of case
to control, k

Superiority
margin, δ

Anticipated incidence

p_T (concordant)

p_C (concordant)

p_T (discordant)

p_C (discordant)

Baseline sample size

Ratio of case to control, k

Prior trial statistics

I only have N and k from my prior trial

I have a CSV with full prior trial metadata

Number of patients enrolled in the prior trial

Prior trial ratio of case to control, k_prior

Metadata CSV must include the binary columns: sample_id, legacy_model_prediction, new_model_prediction, treatment

📁 Click to upload CSV file

Concordance

Legacy and new model concordance

P(R_L=1 | R_N=1)

Value must be between 0 and 1

BRIDGE validity checklist

Population consistency

Patient characteristics distributions are similar across trials within concordant and discordant strata.
Inclusion/exclusion criteria and recruitment processes are consistent.
Underlying disease prevalence, risk factor distributions, and treatment responses are stable.

Outcome consistency

Outcome definitions and measurement methods are identical.

Intervention consistency

Clinical intervention triggered by legacy and updated AI model recommendations are similar.

Implementation consistency

Compliance with AI model recommendations are similar across trials.
All shared input variables between the legacy and updated AI models are measured identically.

BRIDGE Trial Sample Size Calculator