Grad Student Competition
Dairy cattle lameness scoring systems, measures of agreement and diagnostic test evaluation
| Date/Time: |
8/28/2026
09:30
|
| Author: |
Shankar P Poudel |
| Clinic: |
University of Minnesota |
| City, State, ZIP: |
Falcon Heights, MN 55108 |
S. P. Poudel, DVM, MS
1
;
E. Shepley, PhD
2
;
G. Cramer, DVM, DVSc
3
;
1Department of Veterinary Population Medicine, College of Veterinary Medicine, University of Minnesota, St. Paul, MN 55108
2Department of Veterinary Population Medicine, College of Veterinary Medicine, University of Minnesota, St. Paul, MN 55108
3Department of Veterinary Population Medicine, College of Veterinary Medicine, University of Minnesota, St. Paul, MN 55108
Introduction:
Lameness in dairy cows is a prominent welfare and economic problem in farms. Effective lameness management in the herd requires frequent monitoring and the implementation of preventive measures such as hoof trimming and foot bathing. The objective of our study is to evaluate the performance of two-camera based locomotion scoring technologies (A and B) in farms and compare them with visual locomotion scores (VLS).
Materials and methods:
The mobility scores data for A and B were collected on farms daily, between April 2025 and January 2026, and VLS data were collected biweekly during the same period. Commercially available camera-based systems scored cows while returning to their pen from the milking parlor with numerical scores of 0-100, while VLS scores were taken by the scorer on a 0-3 point scale. Scores were categorized as lame and severely lame using the threshold guidelines from companies A and B, with 2 as lame and 3 as severely lame for the VLS method. The agreement between VLS, A, and B, was evaluated on a random date when scores for both comparison groups existed. Concordance correlation coefficient (CCC) and Gwet’s Coefficient (AC1), sensitivity (SEN), and specificity (SP) were calculated for each technology and between technologies, using VLS as the gold standard.
Results:
A total of 4,147, 3,800, and 3,986 unique cow observations were available for A, B, and VLS, respectively. The prevalence of lameness ranged from 25-29% and from 2.3-2.9% for severe lameness. The bootstrapped CCC between A and B showed poor agreement (0.45; CI: 0.43, 0.47) with a mean bias of 10 between raw scores. The AC1 between A and B showed substantial agreement (0.62; CI: 0.59, 0.64) in lame cows. In severely lame cows, AC1 showed near-perfect agreement (0.97; CI: 0.97, 0.98). The AC1 values of VLS with system A and system B were 0.521 (CI: 0.49, 0.55) and 0.53 (CI: 0.50, 0.56), indicating moderate agreement in lame cows, while the AC1 values were near perfect agreement in severely lame cows, 0.97 (CI: 0.96, 0.98) and 0.96 (CI: 0.96, 0.97), respectively. The SEN of A compared to VLS for lameness was 47.2 (CI: 44.0, 50.5), and SP was 78.6 (CI: 77.0, 80.1). The SEN of B compared to VLS for lameness was 50.6 (CI: 47.3, 54.0), and Sp was 79.3 (CI: 77.7, 80.9). In severely lame cows, for A, the SEN was 34.1 (CI: 24.2, 45.2) and SP was 98.5 (CI: 98.0, 98.8), and for B, the Se was 38.4 (CI: 28.1, 49.5) and Sp was 98.1 (CI: 97.5, 98.5).
Significance:
The two-camera based systems showed comparable and moderate agreement with visual locomotion scoring for lame cows and near-perfect agreement for severely lame cows.
