nav[aria-label="Primary Navigation"] { padding: 0; & ul { list-style: none; width: 100%; display: flex; flex-direction: row; justify-content: start; align-items: start; gap: 30px; padding: 0; & li { margin: 0; } & ul li { list-style: none; } } }

How to calculate ICH precision calculations using Variance component analysis

JayC008

I am trying to perform variance component analysis (using JMP 18.0.2) to calculate %CV values for ICH precision data, more specifically to compute (a) Repeatability, (b) Intermediate precision and (c) reproducibility.

The study was performed by 2 operators, measuring 3 replicates on 2 different occasions, across 5 different sample concentration.

Would really appreciate if someone can help me to do the proper analysis to get the ICH precision data. I have tried using Variability/Attribute Gauge chart and performed nested analysis using REML as well as Bayesian analysis. Which of these should i use for the analysis? I also realize the outcomes are impacted by the model type (nested or crossed). SHould i use REML or Bayesian? Should i use nested or crossed? Should Operator be nested under occasion or Occasion nested under operator?

Any help would be much appreciated.

Thank you.

Best regards

Jai

Variance component analysis_Trial analysis.jmp

Accepted answers

All comments

Ressel

I'm wondering regarding the structure of your dataset. The rows are not uniquely identified by combinations of Occasion, Operator and Repeat. In other words, how can Operator #1 measure the same Repeat 5 times at a single Occasion?

statman

The way you describe the situation, it appears Operator, Occasion and Samples are all crossed (every operator sees every occasion and every sample) and repeats would be nested (I don't know why you call them replicates?). So your model should be factorial for Operator, Occasion and Sample and repeats nested (the nested components will be for precision repeatability). While you can use a range chart to assess stability of the repeated measures and to quantify that component of variation and you can have an X-bar chart to plot the precision repeatability (this will be the basis for the control limits) against the other sources (this would debiased without normalizing the concentrations), control charts will not be useful for assessing the other 3 components due to the presence of interactions.

On another note, you have purposely varied Sample by changing the concentrations. Are these reference standards? If so analysis interpretations change. This will bias the variance component. You should normalize this before analysis. Perhaps take the actual measured value minus the theoretical concentration?

MRB3855

Hi @JayC008 : You mention ICH; exactly which ICH guidance are you trying to follow?

JayC008

Hi @MRB3855 : I am trying to compute precision values for Repeatability, Intermediate precision and reproducibility as per ICHQ2(R2) guidelines.

JayC008

Hi @Ressel : Thanks for your comments. I am trying to establish precision for repeatability, intermediate precision and reproducibility for an assay method. (Based on ICHQ2 recommendation, for repeatability assessment one would need a minimum of nine determinations covering reportable range for the procedure (e.g. 3 concentrations/ 3 replicates each). Concentration range for my assay is 2.5e5 to 2e6 (sample concentrations). These samples were prepared in triplicates (repeats). There were 2 operators that were taking 3 measurements (triplicates) for each of these sample concentrations. The same procedure was repeated on a second occasion. Hence 2 occasions, 2 operators, 3 repeats for 5 sample concentrations that cover the range of the analytical assay.

The question is, how do i get precision CVs for repeatability (repeat measurements across the 5 different concentrations), intermediate precision (across operators and occasions) and reproducibility (across operators).

stan_koprowski

Hi @JayC008,

I might be misunderstanding your asks here but perhaps you are seeking additional guidance on some of the data calculations and how to perform them within JMP?

Please clarify if that is not what you are asking as I don’t think your initial choice of data calculations/ platform is where I would start.

For others following along I have included a link to the ICHQ2(R2) guideline.

Based on what I read in section 3.3.2.4 you should be able to use the distribution platform and or the tabulate platforms within JMP for the calculations of recommended data.
Others might have a quick option /platform for generation of the confidence interval using Clopper-Pearson clopper-pearson confidence interval for proportion or with a formula.

3.3.2.4 Recommended Data
“The standard deviation, relative standard deviation (coefficient of variation), and an appropriate
100(1-α) % confidence interval (or justified alternative statistical interval) should be reported.
The observed interval should be compatible with the corresponding precision acceptance
criteria, unless otherwise justified.
Additionally, for multivariate analytical procedures, the routine metrics of RMSEP encompass
accuracy and precision.”

Method validation will need additional documentation of procedures beyond the help and scope of the JMP User community discussion board.

Also, one should demonstrate the Quantitation Limit and Detection Limits for the analytical procedure under investigation.

cheers,

Stan

peter_t

I would suggest to combine (concatenate) Occasion and Operator into one column describing each setup of the method. Here is what your data looks like for the highest sample concentration.

In this way you can estimate variation between setups, and variation between Repeats within setup. I would do this using Analyze -> Fit Model, where Setup (concatenated Occasion and Operator) is modelled as a Random effect.

In the output from this analysis you can find the SD (Sqrt Variance Component) for between setups, within a setup, and the total (sum) of these. The 95% confidence intervals associated with each of these SDs can be calculated as the square root of 95%Lower and 95%Upper. In ICH Q2 terms Total variation is Intermediate precision, and residual is Repeatability.