AI & IRT
in Clinical Trials

Interactive Response Technology (IRT), also called IWRS (Interactive Web Response System) or RTSM (Randomization and Trial Supply Management), is the operational backbone of patient allocation and investigational product (IP) distribution in clinical trials.

At its core, IRT answers four questions in real time: Which treatment arm does this patient get? (randomization), What dose and kit should ship to this site? (supply management), Has the patient met criteria for the next visit? (dosing triggers), and Can we unblind this case for safety? (emergency unblinding).

Before IRT, these were manual, fax-based, or phone-based processes. A site called a central coordinator; the coordinator flipped through a randomization list; a pharmacy tech repackaged kits. That worked for small trials. It collapses under the weight of adaptive designs, large multi-country programs, just-in-time manufacturing, and the expectation that a site can randomize a patient at 2 a.m. local time.

The term IRT still lingers because it was the original label for phone- and web-based patient-interaction systems. But the scope has widened dramatically. RTSM better captures what modern platforms actually do: they manage not just randomization, but stratification algorithms, dynamic allocation, site inventory, depot-to-site logistics, temperature excursion handling, returns and destruction, and cross-system integration with EDC, CTMS, and eCOA.

The shift from IRT to RTSM mirrors the broader shift in clinical operations from documenting what happened to orchestrating what happens next. A modern RTSM platform does not just record a randomization; it predicts when a site will run out of kits, triggers a resupply shipment before the stockout, and adjusts the forecast when enrollment slows, all without human intervention.

Trial supply is where AI in RTSM pays the clearest dividend. The classic problem: overstock sites with expensive IP to avoid stockouts, and you waste drug and budget; understock them, and patients miss doses while you air-freight kits across continents.

Randomization seems algorithmically simple, assign patient N to arm X, until you layer on the constraints of modern trials: stratification by site, country, and biomarker; minimization to balance covariates; dynamic allocation that adjusts probabilities as the trial unfolds; and blinding that must survive regulator inspection.

AI contributes in two ways. First, imbalance detection: models monitor stratification variables in real time and alert the stats team if a site or region is drifting from target ratios before it becomes a protocol deviation. Second, simulation-driven design: Monte Carlo models test randomization schemes against plausible enrollment scenarios, including country-specific dropout curves and seasonal enrollment dips, so the chosen scheme is robust before the first patient is screened.

In adaptive trials, AI-driven RTSM platforms manage dose arm expansion and treatment switching rules in real time. When an interim analysis recommends adding a dose arm, the platform updates stratification, triggers new kit manufacturing batches, and rebalances supply forecasts, all while maintaining the blind.

An RTSM platform that sits in isolation is almost as bad as a spreadsheet. The value emerges when it talks to the rest of the trial stack:

• EDC integration: Enrollment events in the RTSM trigger screening and baseline CRF workflows; dosing records flow back as exposure data.
• CTMS integration: Site activation status in CTMS gates randomization eligibility; RTSM shipment logs feed into site-monitoring visit prep.
• eCOA integration: Patient-reported outcome schedules drive visit-window logic in the RTSM, ensuring kits are available before a patient reports for assessment.
• Safety integration: Emergency unblinding events in the RTSM trigger automated safety notifications to the pharmacovigilance team.
• Lab integration: Central lab result-driven stratification, e.g., biomarker-positive cohorts, is handled by the RTSM without manual site callbacks.

AI sits on top of these integrations as an orchestration layer: it detects when data across systems is inconsistent (a patient randomized but no screening CRF; a kit shipped but no dispensing log), surfaces the anomaly, and routes it to the right functional owner.

IRT/RTSM systems are GxP computerized systems and must meet 21 CFR Part 11 (U.S.) and EU Annex 11 requirements: validated for intended use, access-controlled, audit-trailed, and backed by a disaster-recovery plan.

For AI-enhanced RTSM, the bar goes higher. Regulators expect:

• Algorithmic transparency: The randomization algorithm, supply-forecasting model, and any dose-adjustment logic are documented in the validation package. Black-box models for patient allocation are unacceptable.
• Human override: Every automated action, a resupply trigger, a dose adjustment, an unblinding, is reviewable and reversible by an authorized user with a captured rationale.
• Audit trail integrity: The system logs who did what, when, and from which system, including API calls between RTSM and EDC/CTMS. Logs are immutable and regularly reviewed.
• Blinding preservation: AI must not inadvertently reveal treatment assignment through supply patterns, shipment timing, or kit labels. Blind review of supply algorithms is standard practice.