2.1. Experimental Design
An observational study was conducted using 40 Holstein cows per dairy from 4 commercial dairy herds (n = 658 quarters total) located in Tulare and Fresno County, California. Individual quarter milk samples were collected on the same day and just prior to drying off. Bulk-tank milk samples (n = 100) were provided by the UC Davis Veterinary Medicine Teaching and Research Center’s Milk Quality Laboratory. No knowledge of the cows, location of the dairy, or other management factors of these dairy herds were known, other than sample collection time (at AM or PM). Samples were collected between October 2020 and August 2021, and there were no exclusion criteria for individual or BTM samples.
Cows in Dairies 1 and 3 were housed in dry lots with headlocks. Cows in Dairies 2 and 4 were housed in free stall barns with headlocks. Dairies 1, 3, and 4 milked cows twice a day, whereas Dairy 2 milked cows 3 times a day. No procedures were undertaken that were different from the regular cow handling protocols of the dairies, and there were no humane endpoints for the cows on this study.
2.2. Individual Quarter and BTM Sampling
To collect each individual quarter sample, teat ends were wiped with ethanol-soaked gauze (3 mL), and foremilk was discarded. Fifty mL of milk was hand collected into separate tubes for each quarter. Milk samples were then placed on ice and transported to the lab. Then, 25 mL of milk were separated into two different vials, with 5 mL for analysis by DHIA using the Combi system and the rest for bench top analyses.
For BTM sampling, 50 mL of BTM milk were collected from the laboratory after bacteriological sampling had occurred. Twenty-five mL were aliquoted into separate vials, one for bench top analysis and the other for Tulare County Dairy Herd Improvement Association (DHIA). Only meters RT-10 adapter (RT-10; Dairy Quality Inc., Newmarket, ON, Canada), DeLaval Cell Counter (DSCC; DeLaval, Gurnee, IL, USA), PortaCheck Quick Test (PC; PortaCheck, White City, OR, USA), electrical conductivity meter (ECM; OHAUS, Parsippany, NJ, USA), and pH meter (HC; Hanna Instruments, Smithfield, RI, USA) were used for BTM analysis, as the samples were not obtained cow-side.
2.3. SCC Analyses
All individual quarter and BTM samples were mixed gently prior to measurement of SCC. The Combi, which was designated as the reference standard test, was owned and operated by DHIA, and was a combination of two modules, a Fourier Transform Spectrometer and a Flow Cytometer, for the determination of SCC and milk composition. Cells within the milk sample were stained with a fluorescent dye and exposed to a laser beam, and then light was refracted in accordance with the amount of somatic cells.
The DSCC was a portable machine that uses cassettes to predict SCC. The RT-10 for the iPhone 5/s used the same cassettes to predict SCC. The cassettes contained a fluorescent reagent (propidium iodide) to stain the cell’s nuclei. In the DSCC, the cassette’s counting window was exposed to a light emitting diode, a digital camera photographed a picture of the stained nuclei, and then the DSCC counted the nuclei. The DSCC had a measuring range of 10,000 to 4,000,000 somatic cells/mL. In the RT-10, the camera of the iPhone was used to count SCC. There can be a high rate of cartridge failure if the counting window was damaged or smudged when handled, which can affect the usefulness of the device.
The PC used test strips that estimated the number of somatic cells by measuring the esterase enzyme present within white blood cells. The test strip was mixed with milk and an activator solution [3-(N-tosyl-lalanyloxy)-indol] (~80 µL) in the sample well, which caused a color change to blue, and represented the amount of esterase in the sample. A color chart was used to score SCC results in categories of ≤100,000; 250,000, 500,000, 750,000, 1,500,000, ≥3,000,000 cells/mL.
The California Mastitis Test (CMT; ImmuCell, Portland, ME, USA) was a qualitative four-welled plastic paddle that can test a cow’s individual quarters for SCC. A reagent that broke down cell membranes and contained a pH indicator (bromocresol purple) caused the milk to gel in accordance with the concentration of SCC. Milk and reagent were added to the paddle and were rotated and tilted until the reaction was completed. Results were scored a Negative (<200,000 SCC), Trace (150,000–500,000 SCC), +1 (400,000–1,500,000 SCC), +2 (800,000–5,000,000 SCC), or +3 (>5,000,000 SCC) according to the standards set by the manufacturer.
The ECM used electrodes to measure the resistance or density of milk, which was electrically positive. This model had a range of 0.00–19.99 S/m, and an accuracy of ±2.5% fs.
The HC used an electrode to measure the pH of milk. Changes in milk pH were due to compositional changes, such as extracellular fluid components and blood, which led to an increase in pH. This meter had a range of 0–14 pH, and an accuracy of ±0.05 pH.
Individual quarter temperatures were taken 1 h after sampling using a dual infrared temperature thermometer (IR5; Klein Tools, Lincolnshire, IL, USA). The device used dual laser beams as a focal point for the temperature sensor on the front part of the tool for individual quarter temperature determination. The emissivity level was set to 0.98, the emissivity value used to measure the temperature of human skin [
18]. This meter was aimed at the caudal area of each quarter. Temperature readings were able to be taken from a safe distance (~1.5 m) from the cow due to the dual laser to lessen personal risk and cow stress.
A common practice for dairies to decide which cows to treat at dry-off is the previous month’s SCC (PSCC) [
19] by using machines such as the Combi. However, if the days between the previous test date and the days since last test (DSLT) are far apart, PSCC may change as infections may be hard to detect in between the SCC test and the dry-off [
20]. Dairy 1 PSCC and DSLT data were used to evaluate observed SCC because they had the lowest herd SCC and they treated all cows at dry-off.
2.5. Statistical Analysis
The unit of interest in this study was mammary gland individual quarter and BTM sample SCC. We defined cell counts ≥ 200,000 SCC cells/mL as an intramammary infection based on [
22]. Positive intramammary infection tests were also defined by manufacturer’s directions as SCC ≥ 200,000 cells/mL for DSCC and RT-10 meters, category ≥ Trace for CMT, SCC category ≥ 250,000 cells/mL for the PC meter. Positive intramammary infections were defined as ECM meter readings ≥ 5.0 mS/cm [
16], pH ≥ 6.6 for the HC meter [
23], Temperatures ≥ 35 °C for IR5 [
24], and of SCC ≥ 200,000 cells/mL for PSCC [
17].
To determine how well meters were able to predict individual quarter SCC measured by the Combi, SCC predicted by each meter was regressed on Combi SCC using general linear models [
25]. The independent variable was Combi SCC, and each regression contained one of the dependent variables, RT-10, DSCC, PC, CMT, ECM, HC, IR5 SCC, dairy, and cow. Dairy and cow were not significant contributors to the prediction of SCC in any of the regressions, and the residual errors were normally distributed.
Dairy 1 was the only dairy that tested all cows using DHIA each month. So, to evaluate the practice of using previous test date SCC (PSCC), Combi SCC were regressed on DSLT and PSCC using Dairy 1 data only using general linear models [
25]. Combi SCC had to be averaged by cow for each monthly test date since PSCC is a composite sample from the mammary gland. Since DSLT was not a significant contributor to Combi SCC, it was dropped from the regression, so the dependent variable was Combi SCC, with independent variable PSCC.
To evaluate how well meters predicted BTM SCC, the SCC estimated by the meters RT-10, DSCC, ECM, and HC were each regressed on Combi SCC using general linear models [
25]. The dependent variable was Combi SCC, and each regression contained one of the independent variables RT-10, DSCC, ECM, and HC SCC.
For individual quarter and BTM meter predictions of SCC, tests for goodness of fit, and the coefficient of determination (R
2), mean bias (MB%), error due to mean absolute error (MAE), and partitioning of the mean square predictive error (MSPE) due to central tendency (CT%), unequal variation (UEV%), error due to random variation (RV%), and error due to slope ≠ 1 (%) were evaluated [
26].
Determination of diagnostic sensitivity (SE), specificity (SP), prevalence, accuracy, likelihood ratios positive (LR+), negative (LR−), and predictive values positive (PPV) and negative (NPV) for each meter for both individual quarter and bulk-tank milk samples was completed using a diagnostic test evaluation calculator [
27]. Disease prevalence was calculated as the total number of positives using the Combi divided by the total number of milk samples.