Development of a Web App to Convert Blood Insulin Concentrations among Various Immunoassays Used in Horses

SIMPLE SUMMARY: In horses, the hormone insulin is frequently measured in blood samples, as increased concentrations can lead to a painful condition of the hoof called laminitis. The early detection of an increased insulin concentration is essential for animal welfare. There is considerable disagreement between different measurement methods (assays) for insulin concentration, however, meaning that the threshold for diagnosis must be determined for each assay individually. To alleviate this requirement, we derived formulas from previous assay comparisons, to convert the values from one assay to another, and made them available through a free web app. Over a wide range of commonly used insulin assays, veterinarians can now compare their measurements to previously published values. Scientists can also use the app to compare publications using different assays. ABSTRACT: The measurement of the blood insulin concentration, and comparison to cut-offs, is essential in diagnosing insulin dysregulation, a common equine endocrinopathy. However, different insulin assays provide disparate results. We aimed to ease comparison between assays by compiling original and published data into a web app to convert insulin measurements from one assay to another. Data were available for ADVIA Centaur insulin chemiluminescent immunoassay (CLIA), Beckman Coulter insulin radioimmunoassay (RIA), Immulite 1000 CLIA, Immulite 2000 CLIA, Immulite 2000 XPi CLIA, Mercodia equine insulin enzyme-linked immunosorbent assay (ELISA), and Millipore porcine insulin RIA. Linear models were fitted for 13 assay pairs using non-decreasing splines, and integrated into this app. Assay comparisons including data from several studies showed a lower performance. This indicates technical variation between laboratories, which has not been described before, but is relevant when diagnostic measurements and cut-offs are provided by different laboratories. Nevertheless, the models’ overall high performance (median r(2) = 0.94; range 0.57–1.00) supports their use to interpret results from diagnostic insulin measurements when the reference assay is unavailable, and to compare values obtained from different assays.