An improved approach to modelling patient reclassification: HbA_1c as an example

External quality assurance (EQA) programs must assess whether the analytical performance of commonly used in vitro diagnostic (IVD) devices is fit for their intended clinical use. We developed a method for this assessment based on reclassification of results at clinical decision limits and applied it to results from a haemoglobin A_1c (HbA_1c) EQA program.

Using a commutable EQA program to provide analytical performance characteristics, we simulated 10,000 measurements of the HbA_1c value of each of 6,326 subjects in the National Health and Nutrition Survey (NHANES) by common IVD devices and the NHANES device. A parametric empirical Bayes approach was used to account for regression to the mean. We estimated the rates at which the IVD devices would reclassify results and compared them with the estimated reclassification rate for the NHANES device, calculating the difference between them as the 'incremental reclassification rate' (IRR). IRRs were 'optimal' if <1.5 %, 'desirable' if <3.0 %, and 'minimal' if <4.7 %.

EQA performance data were available for five HbA_1c devices: Abbott Afinion, Bio-Rad D series, Roche Cobas, Sebia Capillarys, and Siemens DCA. The median IRRs were optimal for all methods for diabetes diagnosis (0.11-0.22 %) and for diabetes monitoring (-0.51-1.01 %), with negative IRRs representing lower reclassification rates than the NHANES device. For prediabetes diagnosis, the Afinion failed the minimal standard (8.62 %), but median IRRs were desirable for the other methods (1.43-2.69 %). EQA providers should consider reporting IRR for suitable analytes to help identify poorly performing IVD devices.