We’ve received follow-up insights on our 12-2-15 post, Reference Intervals in Multiple Sites:
From Dr. Emmanuel Favaloro, Institute of Clinical Pathology and Medical Research:
I suspect that everyone will have a slightly different perspective on this question, as informed from personal experience. We run a network of 27 labs, all using Stago instrumentation, but either Satellite, Compact or Evolution instruments. We used to run a single lab, and defined our original reference ranges based on several hundred normal samples. After many many years of spiking ourselves with needles I have decided that anything that avoids this barbaric practice is good. Ethically I am hesitant to draw blood from my colleagues unless there is no alternative.
I am based in Australia, so we do not have to strictly comply with CLSI guidelines, just good lab practice; and provide evidence for the outcomes, Similar results in EQA data from the different labs, plus regular inter-lab sample testing for comparability on cross-lab similarity suffice for this.
We no longer draw 20 normal individuals for determination of MNPT. We did not draw any normal individual samples for our last process of generating a common reference range for PT (or APTT) for our network of 27 labs. We essentially used a ‘transference’ method, and use similar transference methods to also inform on MNPT and ISI. For MNPT, we identified in 2008 that use of 20 individuals does not generate an accurate MNPT, see Favaloro et al, Pathology, 2008; 40, 296–306. Essentially, different sets of 20 normal samples will yield different MNPTs using the same test process, same lab, same reagent, and same instrument.
So, we have adopted a kind of ‘transference’ method to assess any change to MNPT based on comparison of old vs new test reagent (Favaloro et al, Clin Lab Sci 2012;25:13–25). We used a similar ‘transference’ method to transfer our single lab reference ranges (developed using >250 individual samples) to the other labs in our network. We used additional information based on factor sensitivity. We know that a single reference range is acceptable, as there is a high level of concordance in the network as evidenced by EQA and by ongoing inter-lab comparison data. I haven’t found the time to publish this yet. Maybe in another lifetime.
And from Dr. Thomas Exner, Haematex Research:
I think reference intervals for PT have been thoroughly well-defined over the years since introduction of INRs. Couple of points though… I always had problems getting appropriate normal donors because lab staff did not resemble the healthy patient group in age or other physiological characteristics. I recall Doug Triplett recommending synthetic phospholipids for use in coagulation reagents because the rabbit brain phospholipids varied seasonally. Perhaps there is a seasonal variation in vitamin K uptake in humans which might affect baseline PTs. So should a normal range be established mid-season?
From Dr. Ali Sadeghi-Khomami, Precision BioLogic Inc:
Hi George, CLSI recommends non-parametric (percentile) approach for establishing a new RI. For parametric (SD) RI analysis of multiple sites, I will add the following requirements to the mentioned list: dealing with outliers and analysis of variance to ensure homogeneity of data sets obtained from all sites as well as ensuring that data follow a normal distribution. Choosing between MNPT +/-2SD and 3SD depends on imprecision of a particular PT assay and the level of confidence required for diagnosis. What I have seen most from well known PTs were plus/minus 2SD but this is totally a trade off between false normal or abnormal diagnosis. Hope this helps.
And from Donna Castellone, 12-8-15: We do this at 3 sites, up to 9 analyzers one of 2 ways:
When it is a new analyzer and new lot, we collect 120 normal, IRB consented, apparently healthy, equal numbers of males and females. We run 120 on one analyzer, deemed as our standard, and then subsets of40 of each at each of the sites. We then pool results, and use +/- 2.5 SD as the interval.
Other times we may just do transference using 30 samples. We also look at sensitivities of factors and method correlation of “normal patients” (which is always questionable!)
We have to do this method to be in compliance with NYS and CAP. It is not a perfect science, a little bit of eyeballing and best fit stuff. We are reluctant to change ranges as clinicians don’t like tha. Hope this helps. Happy Holidays to everyone