Predicting Paediatric Tobramycin Pharmacokinetics with Five Different Methods Joseph Standing, Elizabeth Greening, Victoria Holden, Susan Picton, Nicola Young, Henry Chrystyn, Mats Karlsson Uppsala Universitet, Sweden St James’s University Hospital, Leeds, UK University of Huddersfield, UK Predicting Paediatric Tobramycin Pharmacokinetics with Five Different Methods “Children are not small adults” – Differences (Kearns 2003 NEJM) “Children are just small adults” – Similarities (Anderson 2008 Ann Rev PT) • When and how can adult PK data help with paediatric analysis? Predicting Paediatric Tobramycin Pharmacokinetics with Five Different Methods • Aminoglycoside • Mainly Gm-ve activity • Blind therapy in feb. neutropaenia (in Leeds) • Once daily dosing (Maglio 2002) Predicting Paediatric Tobramycin Pharmacokinetics with Five Different Methods • • • • Log P = -7.3 (DiCicco 2002) Freely soluble in water Renal elimination Narrow therapeutic index – Peaks >10mg/L (efficacy) – AUC <100mg.hr/L (toxicity) Predicting Paediatric Tobramycin Pharmacokinetics with Five Different Methods • ADULT INDEX (Aarons 1989 BJCP): – 97 adults – 322 observations – 16-85yrs, – CrCl 10-166mL/min (Cockroft Gault) – Median 2 samples per dose Predicting Paediatric Tobramycin Pharmacokinetics with Five Different Methods • PAEDIATRIC INDEX: – 112 children, – 650 observations – 1-16yrs – CrCl 16-173mL/min (Anderson 2008) – Median 2 samples per dose Predicting Paediatric Tobramycin Pharmacokinetics with Five Different Methods • PAEDIATRIC TEST: – 54 children – 110 observations – 1-12yrs – CrCl 29-101mL/min (Anderson 2008) – 2 samples per dose Predicting Paediatric Tobramycin Pharmacokinetics with Five Different Methods Predict PAEDIATRIC TEST with: 1. 2. 3. 4. 5. ADULT INDEX PAEDIATRIC INDEX Pooled ADULT/PAEDS INDEX PAEDIATRIC INDEX with NWPRIOR PAEDIATRIC INDEX with TNPRIOR Priors in NONMEM • Use of prior knowledge (Gisleskog 2002) • NWPRIOR = Normal / Wishart-1 – Fixed and random effects – No prior on residual variability • TNPRIOR = Normal / Normal – Priors on all parameters Aims • Evaluate adult data to predict paediatric PK • Choose model to recommend dosing in children Overview • • • • • Introduction Aims Method Results Conclusions Method • PK Model (NONMEMVI FOCEI) –2 compartment –CL scaled to CrCl –VD and VP scaled to wt –Q scaled to wt0.75 –BOV on F for each dose –Proportional residual error (Aarons 2005 BJCP Editorial) Method 1. Analyse each index dataset 2. Take final parameter estimates, run PAEDIATRIC TEST MAXEVAL = 0 OFV Measures overall fit Method 3. Calculate patient averaged % prediction errors (PRED-OBS) x 100 PRED (IPRED-OBS) x 100 IPRED 4. Reduce paediatric index to half, quarter, eighth original size Overview • • • • • Introduction Aims Method Results Conclusions Results • PAEDIATRIC TEST OFV with params from each method (MAXEVALS=0) – Adults: – Paeds: – Pooled: – NWPRIOR: – TNPRIOR: 316.5 295.6 304.1 312.8 297.5 Results Test Data Versus Paediatric Index Population Prediction Test Data Versus Paediatric Index Individual Prediction (log scale as proportional residual error) (log scale as proportional residual error) 100 Observed tobramycin conc (mg/L) Observed tobramycin conc (mg/L) PAEDIATRIC TEST predicted with PAEDATRIC INDEX 10 1 1 10 Population predicted conc (mg/L) 100 100 10 1 1 10 100 Individual predicted conc (mg/L) Patient averaged prediction error: Patient averaged individual prediction error: 9.2%(-5.2,23.6) 3.9%(1.7,6.2) Results – Interim Summary • PAEDIATRIC INDEX best at predicting PAEDIATRIC TEST • What happens when paediatric data are less informative? – 56, 28 or 14 children in INDEX Results • Reduced no. children in index (56, 28, 14) • Mean OFV from 5 random samples Adult only TEST OFV: 316.5 OFV of Paediatric Test Data versus Number of Children in Index Data 335 330 325 Test OFV 320 Paediatric Index 315 Pooled Index 310 NWPRIOR Index TNPRIOR Index 305 300 295 290 0 20 40 60 80 Number of children in index dataset 100 120 Results - Summary • Bias in adult predictions • As imprecision rises (with fewer children), adult bias becomes less important How to predict whether it is worth including adult data? Results • For each INDEX dataset: • Case deletion diagnostic (CDD) – 14 children, remove 1 – Estimate remaining 13 – Evaluate OFV for removed child – Repeat for all, sum OFVs Results CDD result: 1. TNPRIOR: 2. Paeds: 3. NWPRIOR: 4. Pooled: 227.0 254.5 332.8 368.1 OFV of Paediatric Test Data versus Number of Children in Index Data 335 330 325 320 Test OFV • Paediatric Index 315 Pooled Index 310 NWPRIOR Index TNPRIOR Index 305 300 295 290 0 20 40 60 80 Number of children in index dataset 100 120 Results Final Model • Pooled paediatric INDEX and TEST: All Paediatric Data Individual Predictions versus Observations (log scale) All Paediatric Data Population Predictions versus Observations (log scale) 100 Observed tobramycin conc (mg/L) Observed tobramycin conc (mg/L) 100 10 1 10 1 0 0 0 1 10 Population predicted tobramycin conc (mg/L) 100 0 1 10 Individual predicted tobramycin conc (mg/L) 100 Results All Paediatric Data Conditional Weighted Residual Error versus Population Predictions 5 5 4 4 3 3 2 2 1 1 0 -1 0 14 CWRES CWRES All Paediatric Data Conditional Weighted Residual Error versus Time After Dose 0 -1 0 -2 -2 -3 -3 -4 -4 60 -5 -5 Time after dose (hr) Population predicted tobramycin conc (mg/L) Overview • • • • • Introduction Aims Method Results Conclusions Conclusions • Best prediction of paed PK was with paed PK! • Whether to add adult data depends on relative informativeness (CDD could help with this) • Model for dose recommendation (+ TDM) developed Acknowledgements Patients who took part • Leon Aarons - adult data from: http://www.rfpk.washington.edu/ • Uppsala colleagues • Pfizer for postdoc funding (JS) References • • • • • • • Aarons L, Vozeh S Wenk M, Weiss P, Follath F. British Journal of Clinical Pharmacology, 1989;28:305-14. Raw data from: http://www.rfpk.washington.edu/ Aarons L. 2005. Physiologically based pharmacokinetic modelling: a sound mechanistic basis is needed. British Journal of Clinical Pharmacology, 60:581-3. Anderson BJ & Holford NHG. Annual Review of Pharmacology & Toxicology, 2008;48:12.1-12.30. DiCicco M, Duong T, Chu A, Jansen SA. 2002. J Mat Res B Appl Biomater, 65:137-49. Gisleskog PO, Karlsson MO, Beal SL. 2002. Journal of Pharmacokinetics and Pharmacodynamics, 29:473-505. Kearns GL, Abdel-Rahman SM, Alander SW, Blowey DL, Leeder JS, Kauffman RE. 2003. New England Journal of Medicine, 349:1157-67. Maglio D, Nightingale CH, Nicolau DP. 2002. International Journal of Antimicrobial Agants, 19:341-8. Martindale. 2007. Martindale, the complete drug reference. 35th Edition, Pharmaceutical Press, London, UK. Extra Slides NWPRIOR Degrees of Freedom Give DOF for each ETA prior SE(s2) = s2 (2/(N-1))½ s2 = variance (ETA) N = DOF CrCl Estimation from SeCr • Aarons used Cockroft Gault: = 150-age*wt SeCr (+/-10% m/f) • “Anderson Holford” in children = CPR SeCr Results