PharmPK Discussion - Additive or proportional error model

• On 12 Jul 2013 at 09:48:17, S Vijaykumar (VijaykumarS.-at-.DRILS.ORG) sent the message
  

  
  Hello All,
  
  I’d  grateful  for any opinions on understanding the choice of error model in POPPK analysis.
  
  Is it based on the range of concentration in the data
    (Or) the model which gives lowest global minima in the Base model
    (Or the model brings improvement in the diagnostic plots (like tight fitting in DV vs IPRED, CWRES
  vs IVAR etc).
  
  Thanks in advance.
  
  VIJAY KUMAR SRIPURAM
  Senior Research Associate
  Dr Reddy’s Institute of Life Sciences
  University of Hyderabad Campus
  Hyderabad, AP, India.
  
  Email: vijaykumars.at.ilsresearch.org,
  vijaykumarsripuram.at.gmail.com
  
  

  Back to the Top

• On 12 Jul 2013 at 13:53:15, Nick Holford sent the message
  

  
  Vijay,
  
  You wrote:
  
  "I’d  grateful  for any opinions on understanding the choice of error model in POPPK analysis.
  
  Is it based on the range of concentration in the data
    (Or) the model which gives lowest global minima in the Base model
    (Or the model brings improvement in the diagnostic plots (like tight fitting in DV vs IPRED, CWRES
  vs IVAR etc).
  
  I would use the model that gave the lowest objective function value in the Final model. The Base
  model is always more wrong than the Final model. I  never waste my time looking at the so called
  "diagnostic plots".
  
  Best wishes,
  
  Nick
  
  

  Back to the Top

• On 13 Jul 2013 at 08:44:45, Mukul Minocha sent the message
  

  
  Hi Vijay,
  
  As a general principle, if the data set contains wide range of concentrations you tend to go with
  proportional error model, on the other hand, an additive error model could be used if the
  concentration range is narrow.
  You might have an analytical assay (LC/MS/MS/HPLC or ELISA) where the error is proportional over
  most of the conc. range but constant near the limits of quantification, a combined proportional +
  additive error model could be used in that case (more often the case). More info can be found on
  ACCP website http://accp1.org/pharmacometrics/theory_gmp_model3.htm.
  
  An important diagnostic plot that you might want to check is RES (CWRES) vs PRED plot, for instance,
  a fan shaped pattern indicates a proportional error model is required. And finally, a decrease in
  OFV is an obvious choice for an appropriate residual error structure.
  
  Best
  
  Mukul Minocha
  CTM
  University of Maryland Baltimore
  
  

  Back to the Top

• On 15 Jul 2013 at 12:09:18, Vijay Kumar Sripuram sent the message
  

  
  Dear Mukul,
  
  Thanks for your suggestion.
  
  Before I received your inputs, I tried with all three basic error models. After comparing all plots
  and -2LL values, I am a bit surprised as it is exactly opposite to what expected.
  
  The additive error model results shown good improvement in diagnostic plots with little improvement
  in -2LL where as in multiplicative, the -2LL is found to be much improved though the plots are not
  as good as in additive one. And there no significance in additive +multilplicative one.
  
  In this regard please correct me if I am wrong in concluding the improvement in plots i.e.,
  
  Plots were inferred to be improved if the data points fitted between +2 and -2 for CWRES vs PRED and
  in case of DV vs PRED, the tightness & closeness to the regression line.
  
  Thanks and Regards
  
  VIJAY KUMAR SRIPURAM
  Dr Reddy's Institute of Life sciences
  University of Hyderabad
  Hyderabad, AP
  India.
  
  

  Back to the Top

Copyright 1995-2014 David W. A. Bourne (david@boomer.org)