Method: How to avoid decision errors resulting from unjustified use of a general threshold or an invalid linear score to utilise somatic cell counts in dairy cows?

Bovine somatic cell counts (SCCs) research and management programmes appear to neglect implications for decision−making of the substantial measurement error of SCC. Here, I suggest three frameworks for presenting somatic cell count values from routine collections of cow milk that circumvent possible managerial flaws or biases associated with the current use of a general threshold, such as 200 000 cells/mL, or a linear somatic cell score (SCS). The suggestions are applicable to any kind of dairy herd, require access to two consecutive SCC values from all cows in a homogeneous group, and apply standard statistical techniques. Suggestion #1 demonstrates the nature of pairs of SCC records with a simple assumption-free percentile analysis. Suggestion #2 presents the same data around an elliptical framework with 68 and 95% ellipsoidal prediction limits assuming that a bivariate normal distribution provides a valid description of paired log10-transformed SCC values from cows without mastitis or major udder infection. Visual inspection of the data cloud supports the identification of individual outliers, leverage points, or systematic trends in the study population. These deviations are plausible indications of new cases or the substantial reduction of udder inflammation at cow level, or systematic changes in SCC patterns within group. Suggestion #3 applies an errors-in-variables regression model to assess agreement between two consecutive log10-transformed SCCs at the cow level. This parametric model gives estimates of systematic deviation from perfect agreement (45° slope) at cow level, and automatically and objectively identifies likely outliers and leverage points by means of case-level residual diagnostics. Consequently, this third suggestion supplements visual inspection of the percentile and elliptical analyses with a tool for objectively ranking cow-level SCC deviations from a central (normal, random, or noisy) pattern of SCC changes and identifies distinct patterns (distributions) of SCCs. The three analytical frameworks all demonstrate that a dichotomising (single universal threshold) may not meaningfully identify distinct patterns of cow-level SCCs. Five categories seem sufficient and necessary to capture the main components of a complicated SCC pattern. An assumption of normal distribution may be valid for describing SCCs after log transformation in some herds, but the user must validate this assumption. Substantial upward and downward changes in SCCs (outliers) can be explained as new cases or reductions of udder inflammation, respectively, and such pathobiologically different mechanisms may not fit a single linear scale such as SCS. I discuss possible approaches to combine the three SCC frameworks with microbiological diagnoses.