IATDMCT 2026, a joint meeting with the French Society of Pharmacology and Therapeutics (SFPT), ran September 27-30 in Rennes, France. The congress's final published count was 943 participants from 41 countries, with 719 abstracts presented according to the pre-closing remarks.
The recurring questions: when can a model be trusted, and why do validated approaches still stall in practice?
1. Validation was the center of gravity. A tacrolimus analysis found poor identifiability consistently predicted failure on external validation. A beta-lactam best-practice statement held that models should be validated in the population where they’re used. InsightRX’s work showed that a VPC can flag a correct model as misspecified when clinicians adapt dosing to what they observe.
2. AI drew measured skepticism. Nearly all surveyed researchers use AI, but only about a third consistently disclose it. A pharmacovigilance analysis described the first adverse reaction reports linked to AI use.
3. Machine learning is building on population PK, not replacing it. The strongest work used ML to correct PopPK parameters, including a multicenter study that improved vancomycin clearance and day-2 AUC prediction on external validation.
4. The implementation gap is operational. Ward pharmacist involvement was independently associated with day-1 vancomycin AUC target attainment, with no increase in AKI. An amikacin study found modeling improved accuracy, but its authors called it hard to implement in hospital practice.
5. Renal function estimation for vancomycin is converging. Several groups tested updated eGFR equations against vancomycin clearance with similar results, and no estimator clearly outperformed the rest. InsightRX’s poster took the same question from the model side.
InsightRX is taking two of these threads forward: how to validate models when clinical decisions shape the data, and how renal function should enter vancomycin dosing.