Systematic review and meta-analysis of refractometry for diagnosis of inadequate transfer of passive immunity in dairy calves: Quantifying how accuracy varies with threshold using a Bayesian approach.
calves
BRIX or classical refractometer (REF) devices
measuring the serum immunoglobulin G concentration in neonatal calves
detection of calves with serum IgG < 10 g/L
Abstract
Inadequate transfer of passive immunity (TPI) is associated with increased risk for calfhood disease and increased risk of mortality and morbidity. Accurately diagnosing calves and herds with inadequate TPI is of primary importance and brix (BRIX) or classical refractometer (REF) devices are more practical for this purpose than measuring the serum immunoglobulin G concentration in neonatal calves. We previously reported a systematic review and meta-analysis for quantifying the pooled accuracy of BRIX and REF for detecting calves with serum IgG < 10 g/L noting that sparse data were available especially because studies did not report the same thresholds. We updated the previous systematic review using different methods that accounted for the test results distribution in calves with or without inadequate TPI. With this approach, all reported cut-offs for a specific study are used in that Bayesian approach that quantifies how accuracy varied among all reported thresholds. Five new manuscripts were included, which represented 4 new studies since the initial study was performed. A total of 11 REF and 9 BRIX studies were available. The meta-analytic methods allowed reporting variation of the true and false positive rate across and among all reported cut-offs. Pooled points estimates (95 % Bayesian credible intervals) for sensitivity (Se) and specificity (Sp) of REF < 5.5 g/L were 86.1 % (68.5-97.9%) and 76.2 % (65.9-88.4%) whereas BRIX < 8.4 % was associated with Se of 91.6 % (77.2-99.5%) and Sp of 88.2 % (65.4-99.8%). Interestingly, the accuracy (Se + Sp-1) was generally higher for BRIX than for REF at the reported cut-offs. Besides the benefit of providing pooled estimates for all reported and unreported BRIX and REF thresholds, the general framework used in this study could potentially be used in many veterinary diagnostic tests studies that reported multiple thresholds accounting for potentially different tests distributions in population with and without the target condition.