When estimating EC50 for a single drug, does one need all of the raw data at each drug concentration or simply the average response at each concentration? How does one decide what model or equation to choose (e.g. four versus five parameter logistic curve)? When obtaining the EC50, is there a best approach for determining the 95% confidence interval?
Dear David,
Does one need all of the raw data at each drug concentration or simply the average response at each concentration? the answer is the second choice.
How does one decide what model or equation to choose (e.g. four versus five parameter logistic curve)? The answer is to use the following link:
http://www.graphpad.com/guides/prism/6/curve-fitting/index.htm?reg_doseshift.htm
When obtaining the EC50, is there a best approach for determining the 95% confidence interval?The recommended approach will be the one described in the following link (using excel):
http://researchguides.library.vanderbilt.edu/c.php?g=69346&p=855555
Hoping this will be helpful,
Rafik
You need to first ask yourself if the between subject variability is of your interest, or you simply want to know the central tendency of the concentration vs response profile. We usually apply population PKPD analysis approach which takes into account all the individual raw data collected. In this way, one can differentiate between subject and within subject variability. The common population PKPD software package is called NONMEM. Once you have the model developed and the parameter fitted, you will get the EC50 typical value, between subject variability in terms of %CV, and SE for both two parameters, and from there you can calculate CI.
You can estimate this using a non-linear mixed effects model, typically with an Emax model- which imposes a sigmoidly distribution of dose-response.
Thank you Rafik, Jun, & Ben for your thoughts. Much appreciated! It seems like many curve fitting strategies in applied statistics, we need to apply the proper approach and then compare our results in terms of model fit to the data and scientific judgment.
Hi David, its actually not that bad- but you probably need access to software to run your data with. Do you have access to a statistician?
Ben-
I am a statistician and have a junior biostatistician working with me. We have been using SigmaPlot version 13 to get EC50's using a 5-parameter logistic model for curve fitting but not sure whether 95% CIs are available with that software. I have and am somewhat famiiiar with NONMEM for PKPD, as was mentioned by Jun Shi.
Thanks!
Hi David.
In my opinion you should use a especific software for PD modeling, for example MONOLIX.
First, select different models (Hill equation with 3 or 4 parameters, logistic equation, etc), second notes the results using the -2*log-likehood criterion for each model, and third, use the excelent graphic interface for statistical diagnostic combined with the population analysis (implemented with SAEM algorithm) and select the best model for the data. Moreover, you can export the graphics to sigmaplot (If you prefer).
I recomend you a population approach (non linear mixed effect modeling) for data analysis. Of course, is more complex than individual analysis, but provides more information in relation to residual, interinvidual or interoccasion variability.
I´m a monolix user, but you can use other excelent softwares as nonmen, phoenix NLME or P-metric.
Kind regards
Thank you very much Juan! I guess a similar strategy can be used when assessing and comparing efficacy and EC50s or ED50s for different drugs, even if we have drug combinations. Most of what I have read about pertains to dose-response curves and probit analysis in Finney's textbook (1971).
-David
I recomend two books for you:
"Kenakin: A Pharmacology Primer (Fourth Edition) ", perhaps the best book of pharmacodynamic theory. (I have the third edition, and I love this book!)
other book is "Bonate: Pharmacokinetic-Pharmacodynamic Modeling and Simulation 2nd" (a reference textbook for PK/PD modeling).
Juan Ma
- Badges
- Science topic
- Similar topics
- Chemistry
- Pharmacology
- Pharmaceuticals
- Drugs