23 Mechanism-based modeling / systems pharmacology Evaluating proposed mechanisms of action, resistance and synergy Design mechanistically optimized monotherapy and combination dosage regimens.
24 Mechanism-based modeling of antibiotic action and resistance Resistance often limits access to target site.Time course & mechanisms of activity and resistance.Efflux pumps,Beta-lactamase activityError-prone replicationBulitta JB et al. Curr Pharm Biotechnol, 2011:
30 Prospective ‘validation’ based on external in vivo data Model predictions at 106 CFU/mL inoculumLog10 (CFU/mL)A: 30 minB: 5 hTime (h)fT>MIC40%60%93%75%100%Time > MICBacteriostasis target ~35% fT>MICNear-maximal cell killing target ~65% fT>MICLog10 CFU per lung at 24 hCFU: Colony-forming unit.Craig WA. Clin Infect Dis 1998, 26:1-12.Andes D & Craig WA. IJAA 2002; 19:261-8.Neutropenic mouse lung infection model (at 24 h) The model quantitatively predicted thePKPD target values for cephalosporins.Bulitta JB, et al. Current Pharmaceutical Biotechnology 2011; 12:Targets in patients: Ambrose PG et al. CID 2007, 44:
31 Achieve therapeutic target goals most precisely in an individual patient Account for MIC / pathogen, renal function, other diseases, etc.
32 Different Bayesian updating methods to individualize PK parameters in an unstable critically ill patientAvailable in Pmetrics, Best DoseMichael Neely, Roger Jelliffe et al.Bulitta JB et al. Curr Pharm Biotechnol, 2011:
34 Optimal Dose Selection programs for antibiotics Roberts JA et al., Lancet Infect Dis 2014; 14:
35 Software choices for antimicrobial PK/PD Carefully defining the objective should always be the first step.A variety of powerful software tools are available and accessible also to beginner and intermediate users. No single tool does it all.Very significant improvements in software usability, efficiency and robustness of algorithms were achieved over the last years.Model estimation time is usually no longer a real limitation, even for complex models with >30 parameters. (Parallelized estimation!) In the future, a semi-automated code generator will be very helpful.Performing a Monte Carlo simulation to optimize empiric dosage regimens is very helpful. However, this is NOT the same as selecting an optimal dosage regimen for an individual patient.Softwares for optimal dosing of individual patients are available and are being enhanced for different devices (incl. smart phones).Communication / explanation of results by a skilled modeler is critical.
39 Exposure response – continuous outcome data Cefotaxime vs K. pneumoniae in neutropenic lung infection model (after 24 hours of therapy)Cmax / MICAUC / MICTime > MICLog10 CFU per lung at 24 hSoftware tools:Many nonlinear regression tools.Suggestions:ADAPT (maximum likelihood, free, user-friendly).WinNonlin (as commercial package).Many other tools equally capable.Modeling approach:Amount of data Usually significantType of output data ContinuousSignal Often strongTime-course data No (or not used)Between subject variability Yes (but not used in analysis)Recommended algorithm Maximum likelihood or(Weighted least squares)Craig WA. Clin Infect Dis 1998, 26:1-12. Pictures from: Drusano GL. Nat Rev Microbiol 2004; 2:
40 Exposure response – Dichotomous (Yes/No data) AUCProbability of cure or toxicityAfibrile on day 7Nephotoxicity for Q24h dosingNephotoxicity for Q12h dosingMIC: 4 mg/LMIC: 1 mg/LMIC: 0.25 mg/LSoftware tools:Statistical packages for logistic regression and CART analysis.Parametric hazard models to describe time-dependent risks (eg. of death or adverse events).Suggestions:Systat for logistic regression.(Other tools equally capable.)Population modelling tools for parametric hazard models:NONMEM, S-ADAPT, Monolix, etc.Modeling approach:Amount of data Less (especially for tox.)Type of output data Dichotomous (e.g. Live/Dead, Yes/No)Signal Often weakerTime-course of risk No (or usually not used)Between subject variability Yes (but not used in analysis) Recommended algorithm Maximum likelihood or Bayesian algorithmsDrusano & Louie. Antimicrob Agents Chemother 2011; 55:
41 Optimize empiric dosage regimens via Monte Carlo simulation Individual patient’s PK and MIC are unknown.Can incorporate influential covariates (e.g. renal impairment).Bulitta JB et al. Curr Pharm Biotechnol, 2011:
42 Achieve Target Goals Pmetrics www.lapk.org Population PK model Sequential Multiple-Model (MM) Bayesian updatingPmetricsIndividual patient dataPopulation PK modelInteracting Multiple-Model (MM) approach.Here, PK parameters can change over time (unstable patients!)Individual patient dataUnstable patients!Slide kindly provided by Dr. Roger Jelliffe.