Abstract
Standard antimicrobial doses fail a measurable fraction of critically ill patients: in the DALI point-prevalence study, 16% of infected patients did not keep free β-lactam concentrations above the MIC for even half of the dosing interval, and those patients had lower odds of a good outcome. Bayesian model-informed precision dosing (MIPD) combines a population pharmacokinetic model with patient covariates and one or two measured concentrations to forecast exposure and propose a dose. This narrative review examines whether the approach has changed outcomes at the bedside. For vancomycin, Bayesian AUC-guided dosing has become the method preferred by the 2020 consensus guideline and has been associated with less nephrotoxicity. For β-lactams and ciprofloxacin the picture is different. Three randomised trials with 889 analysed patients (TARGET, DOLPHIN and AutoKinetics) found no improvement in organ dysfunction, length of stay or mortality, and in the largest target attainment did not improve either. The trials share weaknesses: late dose advice, unvalidated models, capped doses and unselected populations. Bayesian tools have refined how exposure is estimated; they have not yet been tested in a form that could redefine therapy. The design that would test them is described.