Ensuring Measurement Agreement with JMP for Reliable Veterinary Diagnostics (2026-US-PO-2936)

Hi, my name is Tom Richardson. I'm an R&D scientist at IDEXX Laboratories. Today, my colleague Wenyu Su and I will present our work on measurement agreement analysis to help ensure reliable veterinary diagnostics. IDEXX is a leader in pet health diagnostics, providing innovative solutions for companion animals, livestock, poultry, water testing, and the dairy market. Within IDEXX, the companion animal group offers a wide range of diagnostic products and services, including in-clinic analyzers, reference laboratory testing, and veterinary software solutions.

To maintain confidence in diagnostic results, it is essential to carefully evaluate the accuracy and precision of our instruments. In this project, we used a JMP add-in to streamline measurement agreement analyses and faster data-driven support. The goal of this work was to evaluate measurement agreement between two instrument models to establish comparability and ensure consistent analytical performance. Do you want to move on to the next slide, Wenyu?

Sure.

In this study, we compared two instruments across three species: canine, feline, and equine. For each species, we evaluated 23 analytes, such as glucose, calcium, and cholesterol. Our instruments focus on testing plasma or serum samples for these analytes. Because each analyte is unique, we powered our experiment appropriately and focused our sample collection to target 100 samples per species per analyte, while also testing the dynamic range of that assay.

After collecting the concentration data, we wanted to perform six different analyses for each analyte. Across 2 instruments, 3 species, and 23 analytes, that adds up to 414 analyses in total with 300 data points per analyte. That's a lot of data. Completing all of these analyses manually would require a large amount of repetitive point-and-click work, making this process very time-consuming.

To address this challenge, we developed a JMP add-in that automates the workflow, saves significant time, and makes the analysis process much more efficient. If you want to back up one slide, Wenyu. The JMP add-in brings several commonly used methods into one workflow, including equivalence testing, least squares regression, orthogonal regression, Passing-Bablok regression, and Bland-Altman plots.

The add-in also gives us two options for equivalence testing. We can evaluate either the differences or the percent difference between the two instruments. After setting the acceptance threshold, the add-in can automatically generate all the analyses and graphs with a single click. This greatly reduces manual effort and improves efficiency and consistency in the analysis process. Now, I will hand it over to Wenyu. She's going to demonstrate the add-in and also discuss how we handle censored data that arise from the instrument's detection limits.

Thank you, Tom. Now, I will use an example data set to show how the JMP add-in works. This data set has 98 samples that cover the full concentration range of interest for an analyte. For each sample, it was measured by both instruments, making this a paired study. To run the add-in, click on the Add-Ins menu, choose Method Comparison add-in. In this case, we want to compare the 2 instruments, instrument number 1 and instrument number 2. I will select instrument number 2 as New Method, instrument number 1 as Current Method.

The add-in also provides two options for equivalence test. As Tom mentioned earlier, we can either perform an equivalence test based on difference or based on percentage difference. For this analysis, we want to perform an equivalence test based on the difference, so I will choose the first option. We can also adjust the alpha level for the equivalence test. For this demo, I will adjust the limit as default. Then click on OK.

Next, we need to provide a threshold for the equivalence test. For this analyte, the threshold is three. Then click on OK. All the relevant analysis results and the graphs will be given. The add-in provides the distribution of the difference, the equivalence test results. In addition to the least squares model, the add-in also provides the JMP output for orthogonal regression and Passing-Bablok regression.

This is very useful because the regular least squares model doesn't take into account the measurement variation in the X variable. In addition, the JMP add-in also generates two versions of Bland-Altman plots, one based on the difference, another one based on the percentage difference. This is very useful, especially when we have extremely small or extremely large values.

Looking at both plots will give us a more complete picture of the agreement between the two instruments. Overall, by utilizing the JMP add-in, we are able to eliminate repetitive point-and-click, reduce manual effort, so that we can spend more time in analyzing the results rather than generating them.

Another common issue we encountered with this data set is the censored data due to detection limit issue. In this case, for instrument number 2, we have a detection limit of 20. When the measurement is smaller than 20, or less than 20 is reported. For this case, sample 33 ran into this issue. To address this issue, we evaluated four approaches.

The first approach was to eliminate that data from the analysis. The second approach was to impute the detection limit, in this case, 20. The third approach was to impute half of the detection limit, in this case, 10. The fourth approach was to turn on the detection limit column property and then fit a generalized regression between the two variables to evaluate the agreement between the two instruments.

To do that, click on the column header of the instrument number 2 variable, click on Column Info. Then, under Column Properties, choose Detection Limit. Where is it? Here. Then, in this case, we have a lower detection limit of 20. Click on OK. Then go to Analyze menu, select Fit Model. We will use instrument number 2 as Y variable, instrument number 1 as X variable. Under Personality, choose Generalized Regression, and then click on Run button.

The intercept and the slope estimate obtained here will take into account the censored data into analysis. Now, let's compare the four approaches. As you can see, the intercept estimate from the fourth approach is exactly the same as the intercept estimate obtained by eliminating the censored data from the analysis, with only very minor difference in the 95% confidence intervals.

With the second and third imputation approach, we got slightly larger intercept estimates. Which suggests that the imputation approach might introduce bias into the analysis. The generalized regression with the detection limit column property provided improved estimation of the constant bias. Thus, we got a more reliable estimate.

Overall, by leveraging the capabilities in JMP, we were able to make our analysis more efficient and robust so that we can better evaluate the instrument performance, and we can deliver high-quality diagnostic results. With that said, I will hand it over to Tom to wrap up. Tom.

Thank you, Wenyu, for that great demonstration. That concludes our brief presentation and demonstration on Ensuring Measurement Agreement with JMP for Reliable Veterinary Diagnostics. Thank you so much for listening to our presentation. We hope you have a wonderful day.