Lameness in dairy cattle, primarily caused by hoof lesions, remains a widespread issue in North America with significant welfare, economic, and sustainability impacts. Although emerging technologies incorporating machine learning offer promising solutions for the detection of hoof lesions and lameness, inconsistent outcome definitions for hoof lesions and lameness may limit model generalizability and on-farm application. This scoping review aims to examine how hoof lesions and lameness outcomes are defined and classified across studies, identify inconsistencies in outcomes definitions, and provide guidance for improving standardization to improve the effectiveness of machine learning applications. A systematic search was conducted in October 2024 across 4 databases: Scopus, PubMed, Agricola, and CAB Digital Library. Of the 1,149 identified references, 1.7% (20/1149) of studies met the inclusion criteria and were selected for data extraction. Among these, 55% (11/20) developed outcomes based on locomotion scores, 35% (7/20) used hoof lesion data, and 10% (2/20) incorporated both. Of the studies using only locomotion scores (n = 12), 66.7% (8/12) used a 5-point scale, 16.7% (2/12) used a 4-point scale, and 16.7% (2/12) used a 3-point scale. Of these, 50% (6/12) created binary outcomes, 33.3% (4/12) created 3-level outcomes, and 8.3% (1/12) used 5-level outcomes. Among the studies using hoof lesion outcomes (n = 8), 62.5% (5/8) assessed multiple lesion types, while 37.5% (3/8) focused on differentiating stages of digital dermatitis. Additionally, 37.5% (3/8) did not provide any descriptions of the hoof lesions used. Descriptions of the hoof lesions also varied among those that did provide criteria. Reporting of population characteristics was limited: country of data collection was reported in 95% (19/20) of studies, breed(s) in 70% (14/20), and lactation-related information in 65% (13/20). Bedding type, flooring, and outdoor access were reported in 30% (6/20), and production system type in only 10% (2/20). Furthermore, only 10% (2/20) of studies validated their models using data from herds external to the original study population. To improve the generalizability and real-world utility of machine learning applications in dairy cattle management, future research must provide clear outcome definitions, detailed reporting of study and population characteristics, perform external validation, and evaluate the impact of outcome classification.
Swartz et al. (Sun,) studied this question.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: