Latent class analysis identifies novel coeliac disease subgroups with distinctive clinical features: a multicentric study

Background: Coeliac disease (CD) is a lifelong, immune-mediated disorder characterised by a wide clinical spectrum, which often challenges traditional classifications like the Oslo definitions. We aimed to recognise novel CD subtypes based on clinical features using latent class analysis (LCA) and validate these phenotypes with a predictive supervised model. Methods: In this multicentric retrospective study, 2478 adult CD patients from 19 Italian centres (2011–2021) were analysed. Clinical, laboratory, endoscopic, and histological data were collected. LCA was applied to categorical symptom variables, including gastrointestinal, haematological, neuropsychiatric manifestations, and fatigue, to identify latent clusters, with the optimal number of classes determined by the Bayesian Information Criterion. The newly derived classes were compared with the traditional Oslo classification using Chi-squared tests, while multinomial logistic regression (MLR) was employed to validate the associations between latent classes and additional features. Results: LCA identified four classes, namely Class 1 (predominant lower gastrointestinal symptoms); Class 2 (upper gastrointestinal manifestations); Class 3 (mainly asymptomatic or nonspecific cases); and Class 4 (microcytic anaemia with asthenia). Comparison with the Oslo classification revealed partial overlap, indicating potential misclassification under current criteria. MLR confirmed significant associations, with female sex strongly linked to Class 2 (OR 2.7, 95 % CI 1.52–4.78). Autoimmune comorbidities (OR 1.96, 95 % CI 1.18–3.25) and severe histological damage (B2 Corazza-Villanacci classification, OR 9.12, 95 % CI 1.86–44.63) were also more frequent in Class 2. Conclusions: LCA may offer a novel, data-driven approach to refine CD phenotyping that should be validated in future studies.