Routine serum biomarkers and lightgbm modeling for the diagnosis of abomasal ulcers in culled dairy cows.

Abomasal ulcer (AU) is a common condition in dairy cows that significantly affects productivity and longevity. This study evaluated the diagnostic value of serum biomarkers and the fecal occult blood test (FOBT), and tested a Light Gradient Boosting Machine (LightGBM) model to predict ulcer occurrence. Postmortem analyses were performed on 304 culled dairy cows, from which blood and fecal samples were collected. Animals were classified according to the presence (n = 249) or absence (n = 55) of abomasal ulcers. No individual biomarker, including gastrin or pepsinogen, demonstrated consistent diagnostic value. The FOBT showed high specificity (90.9%) but low sensitivity (27.7%), while the LightGBM model showed limited discriminatory performance (AUC ROC = 0.60). The model identified creatinine, platelet count, GGT, mean corpuscular hemoglobin, and total protein as relevant features. These findings indicate that current serum biomarkers and routine blood variables are insufficient for accurate individual diagnosis of AU, underscoring the need for more specific diagnostic tools.