Unsupervised clustering of serum lipase activity in cats: a data-driven approach to correlate clinical, laboratory, and ultrasonographic findings.

Abstract

BACKGROUND: Serum lipase activity has not been compared with ultrasonographic and clinical findings in cats. HYPOTHESIS/OBJECTIVES: Interpret serum lipase activity in sick cats using unsupervised, data-driven clustering. ANIMALS: Five hundred sixty-three hospitalized client-owned cats with serum lipase activity > 30 U/L (reference intervals, 8-26 U/L). METHODS: Retrospective study. Unsupervised k-means clustering of serum lipase activity (k = 3, chosen by elbow and silhouette diagnostics) grouped cats into 3 ranges. Least squares mean (LSmean) lipase values were: cluster 1: 53 U/L (range, 30-125 U/L); cluster 2: 229 U/L (range, 147-411 U/L); and cluster 3: 778 U/L (range, 441-1613 U/L). Clinical (n = 563), laboratory (n = 563), and ultrasonographic (n = 318) findings were compared across clusters using general linear models and Fisher's exact tests with multiplicity correction (Bonferroni). RESULTS: Ultrasonographic diagnosis of pancreatitis (USDx) was significantly (P = .0021) more common in cluster 2 (64%) compared to cluster 1 (47%). Pancreatic enlargement (59%), hypoechogenicity (56%), and hyperechoic mesentery (49%) were significantly more common in cluster 2 compared to cluster 1, but 40%-50% of cluster 2 cats and 25%-66% of cluster 3 cats did not have these US abnormalities. Only lethargy was significantly more common in cluster 2 cats. Significant differences in laboratory values were found for urea (highest in cluster 3), triglycerides (highest in cluster 2), and protein (lowest in cluster 3). CONCLUSIONS AND CLINICAL IMPORTANCE: Despite marked hyperlipasemia, approximately 50% of cats have no ultrasonographic pancreatic abnormalities or changes insufficient for an USDx. Future routine pancreatic size assessment might help detect subtle changes. Abdominal pain is not significantly more common at higher lipase activities.