Comparing two assays

We want to change the platform we use in our laboratory for a particular quantitative assay. Before we introduce it, we need to check that the new test is going to perform reliably.

We know the reproducibility of both platforms (i.e. if we tested the same sample over an over, what spread of result values we might expect) in terms of the mean and standard deviation across various ranges of known values within the limits of the assay. These are normally distributed.

Lacking any statistical expertise in our department, our intention is to test a number of samples on both platforms (around 100, approx. 60% of which will be positives and 40% should give a completely negative result). Each sample will only be tested once on each platform. We intend to accept any result on the new platform as being equivalent to that on the old one if the New platform result + 2*SD of New platform > Old platform result OR New platform result - 2*SD of New platform < Old platform result.

Does this seem a reasonable way of reassuring ourselves that the two platforms give comparable results? How confident could we be about the conclusion if 100% of our results 'matched' (as defined above)? What if only 80% match?

Thanks! I've found their article and it looks perfect. I was uneasy about our proposal because of the interference created by any variability in the old platform result but their method seems to account for this.


TS Contributor
There is another approach to comparing measurement systems called the Youden plot. Both work well, so use whichever better suits your needs. The Bland-Altman plot will be easier to explain to non-technical people.