We are developing an HPTLC method for simultaneous estimation of five antidiabetic drugs. While performing a linearity study, through regression fitting model of WINCATS software, the R-square value for all the drugs are better in polynomial regression equation compared to linear model..and giving 0.99 and above values....Can polynomial regression equation be used in method validation?
What could be the justification?
Dear Udaykumar,
Yes, you can use polynomial regression equation but normally XY Scatters with smooth line and markers shall be used.
Yes. As long as you can PROVE to the auditor/reviewer that your analytical method is predictable. Remember, an AAS curve is sigmodal but most chemists use the linear portion of this curve. Also, a linearty curve is a subset of the polynomial equation.
Using polynomial equations is pretty normal in the pharmaceutical field. Especially when you work with biotech/biologics or detectors with limited range of linearity like ELSD. I have never had compliance issues after submitting analytical methods with polynomial curves.
As hinted above though, you could adjust your sample concentrations to fit within the linear portion of the curve if that's possible for you.
As for the second part of the question, how to justify it. Run a method qualification by doing 5 solutions with concentration ranging from at least 80-120% of your target. Include all your analytes. Run those in triplicates and calculate the %RSD and the regression coefficient. Ideally, R squared has to be 0.99 or more, but for many methods 0.98 is fine. After that, perform accuracy from spiked samples: Spike your diluent/placebo/matrix with your analytes in at least 80-120% range from the target and analyze in triplicates. Calculate recovery (expected ~95-105% for assays) and %RSD. If you don't have matrix other than your diluent, than you can calculate the accuracy from your linearity solutions. Have another colleague run the method on a different day to confirm your results (you can run just linearity curve and one sample at 100% target). This will prove that you have data for your curve and that the method is accurate for quantitative purposes. Then you can propose this to a more regulated environment like GDP or GMP lab to do full method validation.
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Regression