Machine Learning Clustering Identifies Achalasia-Spectrum Phenotypes in Distal Esophageal Spasm.

BACKGROUND/AIMS: : Distal esophageal spasm (DES) is a rare, heterogeneous esophageal motility disorder with variable treatment responses. Machine learning methods are well suited to distinguish DES phenotypes that could inform therapeutic decisions and outcomes. This exploratory study aims to identify and characterize unique DES phenotypes using unsupervised machine learning clustering. METHODS: : Adults undergoing high-resolution esophageal manometry (HRM) for esophageal symptoms between 2005-2025 were included. Uniform manifold approximation and projection (UMAP) was used for dimensionality reduction of HRM metrics. DES patients were classified as within-cluster or outside-cluster according to density contours. Patient characteristics, manometric parameters, functional lumen imaging probe findings, and treatment outcomes were compared between clusters. RESULTS: : Among 5360 patients, 42 (0.8%) had DES, 2400 (45%) normal motility, and 107 (2.0%) type III achalasia. UMAP identified 33 DES patients clustering alongside type III achalasia patients, while 9 DES patients remained outside this cluster. The primary differentiating characteristic between clusters was premature supine contraction frequency (median 8 vs 0, P < 0.001). Within-cluster patients demonstrated higher Esophageal Hypervigilance and Anxiety Scale scores (30 vs 20.5, P = 0.010). All 3 within-cluster patients who underwent peroral endoscopic myotomy achieved symptom resolution, compared to partial response in outside-cluster patients. CONCLUSIONS: : Machine learning clustering successfully identified a DES phenotype characterized by high premature contraction frequency that clusters with type III achalasia. Premature contraction frequency may serve as a clinical marker for selecting DES patients for lower esophageal sphincter-directed therapies. Given the rarity of DES, larger validation studies are needed to confirm these findings.