CO35 | OPERATIONAL TOOL FOR VETERINARY SURVEILLANCE OF MILK: STATIC-LONGITUDINAL INTEGRATION OF FARM ANALYTICAL DATA

Giulio Grossi¹, Lukas Ebner², Laura Lomellini¹ | ¹Azienda Sanitaria dell'Alto Adige, Bolzano, Italy; ²Federazione Latterie dell'Alto Adige, Bolzano, Italy.

Objective. This study describes the development and application of the Veterinary Surveillance Index (ISV), a decision-support tool designed to support the prioritisation of veterinary assessment in dairy farms in South Tyrol. The ISV integrates static indicators of farm-level criticality with longitudinal signals derived from temporal changes in routinely collected bulk-tank milk parameters.

Materials and Methods. The study included 946,029 repeated measurements from 4,371 dairy farms represented by pseudonymised farm codes in 126 municipalities. The observation period extended from January 2023 to June 2026, with about 60 measurements per farm-year. Variables included somatic cell count, total bacterial count, fat, protein, water, pH, casein, cryoscopic index, inhibitors and clostridial spores. A reproducible Python-based pipeline normalised raw data and generated four static components: udder health, milk hygiene, cryoscopy and inhibitors, weighted 35, 30, 20 and 15 points, respectively. Expert-informed provisional cut-offs classified the composite score as low (<35), moderate (35–<60) or high (≥60). Longitudinal priority was based on worsening in udder-health, hygiene and cryoscopic domains: one worsening domain determined intermediate priority and at least two determined high priority. Recent inhibitor positivity also determined high longitudinal priority, while any inhibitor positivity triggered high integrated priority. Static and longitudinal outputs were combined through a hierarchical, monotone rule. Longitudinal features were calculated for the full observation period and for the preceding 12-, 6- and 3-month windows. Sensitivity analysis varied each domain weight by ±20%, with renormalisation to 100. Rules were implemented in Python and a local web-based Streamlit application, with outputs reviewed by veterinarians.

Results. Static classification was low for 4,180 farms (95.6%), moderate for 176 (4.0%) and high for 11 (0.3%); 4 farms (0.1%) were not evaluable. Longitudinal priority was routine for 3,049 farms (69.8%), intermediate for 765 (17.5%), high for 118 (2.7%) and not evaluable for 439 (10.0%). Integrated priority was routine for 3,318 farms (75.9%), intermediate for 843 (19.3%) and high for 206 (4.7%); 4 farms remained not evaluable. Integration increased priority in 898 of 4,367 evaluable farms (20.6%), including 195 reclassified to high priority. Among high-priority farms, the most frequent worsening signals involved milk hygiene (92; 44.7%), udder health (82; 39.8%) and cryoscopy (73; 35.4%); signals were not mutually exclusive. Of the 206 high-priority farms, 197 (95.6%) had an adequate information base and documented milk-delivery activity in 2026. Across ±20% domain-weight perturbations, static farm rankings remained highly stable (Spearman ρ = 0.9973–0.9999), while integrated-priority agreement with the baseline was 99.47–99.68%; 98.1–100% of baseline high-priority farms were retained and no farm changed by two priority levels.

Conclusions. The ISV converted routinely collected analytical data into a transparent and reproducible prioritisation framework. Longitudinal integration identified an operationally relevant group not captured by static classification alone, while final priority assignment was highly robust to moderate changes in domain weights. The ISV supports, but does not replace, veterinary judgement and should not be interpreted as a validated predictor of food-safety risk. Prospective validation against official-control findings, confirmed non-compliances and other relevant external outcomes is required.