Objectives: The aim of the present study was to verify whether artificial neural networks (ANNs) might help to elucidate the mechanisms underlying the increased prevalence of cardiovascular events (CV) in primary Sjögren's syndrome (pSS).
Methods: 408 pSS patients (395 F: 13 M), with a mean age of 61 (±14) years and mean disease duration of 8.8 (±7.8) years were retrospectively included. CV risk factors and events were analysed and correlated with the other pSS clinical and serological manifestations by using both a traditional statistical approach (i.e. Agglomerative Hierarchical Clustering (AHC)) and Auto-CM, a data mining tool based on ANNs.
Results: Five percent of pSS patients experienced one or more CV events, including heart failure (8/408), transient ischaemic attack (6/408), stroke (4/408), angina (4/408), myocardial infarction (3/408) and peripheral obliterative arteriopathy (2/408). The AHC provided a dendrogram with at least three clusters that did not allow us to infer specific differential associations among variables (i.e. CV comorbidity and pSS manifestations). On the other hand, Auto-CM identified two different patterns of distributions in CV risk factors, pSS-related features, and CV events. The first pattern, centered on "non-ischaemic CV events/generic condition of HF", was characterised by the presence of traditional CV risk factors and by a closer link with pSS glandular features rather than to pSS extra-glandular manifestations. The second pattern included "ischaemic neurological, cardiac events and peripheral obliterative arteriopathy" and appeared to be strictly associated with extra-glandular disease activity and longer disease duration.
Conclusions: This study represents the first application of ANNs to the analysis of factors contributing to CV events in pSS. When compared to AHC, ANNs had the advantage of better stratifying CV risk in pSS, opening new avenues for planning specific interventions to prevent long-term CV complications in pSS patients.