When optimizing the dosing of drugs in the paediatric population, population pharmacokinetic (PK) models can be helpful. These models estimate population values of PK parameters based on data from all patients in the study and individual values of PK parameters based on data and measurements from each patient. In clinical practice, it is more practical to use simpler models that are common for the entire paediatric population because using more complex models or models separated based on age groups can be more difficult to use and may frequently result in errors when using them.
In the doctoral dissertation, we first identified studies in which authors developed population PK models of gentamicin in the paediatric population and the impact of the most commonly studied covariates on gentamicin PK parameters. In the next step, we retrospectively collected and analyzed data on therapeutic monitoring of gentamicin concentrations in the paediatric population. We then optimized selected literature population PK models of gentamicin using PRIORapproach.Foreachmodel,weevaluatedtheaccuracyandprecisionofpredictinggentamicinconcentrationsinourdataandusedthebestmodeltotesttheimpactofsomelessstudiedcovariatesongentamicinclearance(CL).Inasystematicliteraturereview,weexamined17studiesindetailwhereauthorsusedmodelingwiththeNONMEMsoftwarepackage.Basedonthecollecteddata,wealsoformulatedrecommendationsregardinggentamicindosingforeachagegroupwithinthepaediatricpopulation.Theproposeddosingintervalforextremelyprematurenewbornswas48hours,forprematurenewborns36hours,andforfull−termnewbornsandinfants36or24hours.Comparedtoinitialdoses,lowerdoseswereproposedbasedonthedevelopedmodelsforprematurenewbornsandinfants,comparabledosesforfull−termnewbornsandinfants,andhigherdosesforchildren.Inaddition,weidentifiedlessfrequentlystudiedcovariatessuchasfat−freemass(FFM),concomitanttherapywithotherdrugs,bodytemperature,andcriticalillness,whichsuggestanimpactongentamicinPKandshouldbefurtherinvestigated.Inaretrospectiveclinicalstudy,weincludedpaediatricpatientswhoweretreatedwithgentamicinandforwhomatleastonemeasurementofplasmagentamicinconcentrationwasavailable.Therapeuticgentamicinconcentrationsweresignificantlymorefrequentlyoutsidethetherapeuticrangeinnewbornsthaninolderagegroups.Forlessfrequentlystudiedcovariates(criticalillness,fluidbalance,concomitanttherapywithotherdrugs,etc.),weassessedtheircorrelationwiththeappropriatenessoftheminimumgentamicinconcentration(cmin).Theproportionofinadequatecminwassignificantlyhigherincriticallyillpatientsandinpatientsreceivingconcomitanttherapywithvancomycinanddopamine,whilesignificantassociationswithorganismhydrationstatuswerenotdemonstrated.Sincetheretrospectivelycollecteddatawerequitesparseregardingthenumberofgentamicinconcentrationmeasurements,weusedliteraturemodelsindevelopingthepopulationPKmodel.WeselectedsevenliteraturepopulationPKmodelsofgentamicindevelopedontheentireorasbroadapaediatricpopulationaspossible.WeoptimizedthemodelsusingthePRIOR approach within the NONMEM software environment. In general, the optimized models better described our data than the literature models. As expected, the accuracy and precision were lower for a priori predictions based solely on variables than for a posteriori predictions that also consider individual gentamicin concentration measurements. Based on visual predictive check and assessment of accuracy and precision, the optimized model by Wang 2019 (the chosen population PK model) was found to be the most suitable model for predicting gentamicin concentrations in our data.
Using the selected population PK model, we evaluated the accuracy and precision of predicting gentamicin concentrations when dosing without a model or with a model in a way that includes only one or more concentrations. We demonstrated that predicting gentamicin concentrations is significantly improved when dosing based on a model and when using all available gentamicin concentration measurements for an individual patient to estimate parameters. We also evaluated the difference in gentamicin PK between intensively (critically ill) and non-intensively (non-critically ill) treated children using the selected model. The CL and Vss values in all age groups were lower in critically ill patients than in non-critically ill patients, however, the coefficient of variation between critically and non-critically ill children did not differ in any significant way. Based on the realization that there are differences in gentamicin PK between critically and non-critically ill patients, we used simulations to investigate how these differences are reflected in the plasma concentration profile of gentamicin for each age group, and what is the probability of achieving a cmin of gentamicin ⡤1 mg/L after the first dose and in steady-state at different dosing intervals. In critically ill preterm and full-term neonates, infants, and toddlers, plasma concentrations of gentamicin were higher due to lower CL and Vss, and the probability of achieving cmin ⡤1 mg/L was lower, suggesting the expedience of using longer dosing intervals in critically ill children.
At the end, we checked whether any of the covariates, which have been less studied so far and were evaluated in our patient population, significantly affect gentamicin CL, and those with a significant impact (co-treatment with dopamine, fluid balance evaluated as a daily percentage of accumulated fluid, critical illness assessed using the PELOD 2 scoring system and the number of failing organs, and treatment with induced hypothermia) were included in the selected model. By taking these covariates into account, we further improved the predictive performance of the selected model, but at the same time, the model became quite complex and would require further validation before being used in routine clinical practice. Nevertheless, based on the results obtained, we conclude that instead of developing a new PK model, external validation of literature models adapted to the studied population is a reasonable solution.
|
