Evaluation of the sample needed to accurately estimate outcome-based measurements of dairy welfare on farm

Dairy welfare assessment programs are becoming more common on US farms. Outcome-based measurements, such as locomotion, hock lesion, hygiene, and body condition scores (BCS), are included in these assessments. The objective of the current study was to investigate the proportion of cows in the pen or subsamples of pens on a farm needed to provide an accurate estimate of the previously mentioned measurements. In experiment 1, we evaluated cows in 52 high pens (50 farms) for lameness using a 1- to 5-scale locomotion scoring system (1 = normal and 5 = severely lame; 24.4 and 6% of animals were scored ≥3 or ≥4, respectively). Cows were also given a BCS using a 1- to 5-scale, where 1 = emaciated and 5 = obese; cows were rarely thin (BCS ≤2; 0.10% of cows) or fat (BCS ≥4; 0.11% of cows). Hygiene scores were assessed on a 1- to 5-scale with 1 = clean and 5 = severely dirty; 54.9% of cows had a hygiene score ≥3. Hock injuries were classified as 1 = no lesion, 2 = mild lesion, and 3 = severe lesion; 10.6% of cows had a score of 3. Subsets of data were created with 10 replicates of random sampling that represented 100, 90, 80, 70, 60, 50, 40, 30, 20, 15, 10, 5, and 3% of the cows measured/pen. In experiment 2, we scored the same outcome measures on all cows in lactating pens from 12 farms and evaluated using pen subsamples: high; high and fresh; high, fresh, and hospital; and high, low, and hospital. For both experiments, the association between the estimates derived from all subsamples and entire pen (experiment 1) or herd (experiment 2) prevalence was evaluated using linear regression. To be considered a good estimate, 3 criteria must be met: R² > 0.9, slope =1, and intercept = 0. In experiment 1, on average, recording 15% of the pen represented the percentage of clinically lame cows (score ≥3), whereas 30% needed to be measured to estimate severe lameness (score ≥4). Only 15% of the pen was needed to estimate the percentage of the herd with a hygiene score ≥3, whereas 30% to estimate the prevalence of severe hock lesions. Estimating very thin and fat cows required that 70 to 80% of the pen be measured. In experiment 2, none of the pen subsamples met our criteria for accurate estimates of herd prevalence. In conclusion, we found that both a higher percentage of the pen must be sampled to generate accurate values for relatively rare parameters and that the population measured plays an important role in prevalence estimates.