chapter 06 · AI lens

AI in bioequivalence.

AI is operational at the bioanalytical and pre-study end of BE; supportive at the modelling end (PBPK); and not yet acceptable as primary registration evidence. Inspection risk varies sharply by use case.

/ 10

AI in bioequivalence.

Where AI is changing things now & 2026–2030

AI is operational at the bioanalytical and pre-study end of BE; supportive at the modelling end (PBPK); and not yet acceptable as primary registration evidence. Inspection risk varies sharply by use case — site selection low, model-based biowaiver justification high.

/ AI 01 · production

Cohort enrichment & site selection.

Replaces manual site screening; augments volunteer-pool matching to inclusion criteria. Standard 2024+ via Medidata, ICON, Syneos platforms. Not yet explicitly covered in BE guidance. Inspection risk low — data selection, not data generation.

/ AI 02 · emerging

PBPK model qualification for biowaiver.

Augments BCS biowaiver justification with mechanistic evidence. FDA 2024 MIDD guidance permits as supportive; EMA cautious. Model qualification standards unspecified — the regulatory boundary is unclear. Risk high if a model fails post-approval.

/ AI 03 · production

Bioanalytical peak detection.

Replaces manual chromatogram review; augments operator variability reduction. Standard in LC-MS/MS instruments (Sciex, Waters, Thermo). Already under ICH M10 v2 scope discussion.

/ AI 04 · production

Pop-PK parameter estimation.

Augments BE analysis supporting long-acting injectables and population studies. Established via NONMEM, Monolix, Phoenix. ICH E9 statistical guidance framework, not BE-specific. No AI-specific friction.

/ AI 05 · pilot

RWE curation for synthetic controls.

Augments external control arm selection for synthetic BE studies. Pilot stage (FDA-led, EMA cautious). No formal BE-specific framework. Medium risk — data quality and representativeness contested.

/ AI 06 · research

AI formulation optimisation.

Augments excipient selection and release-mechanism design for modified-release. Research-stage; not yet operational in registration submissions. Considered a formulation development tool, not registration-quality. Low risk for 2026–2030.