1UniversitätsSpital Zürich Universitäres Herzzentrum Zürich, Schweiz; 2UniversitätsSpital Zürich Klinik für Kardiologie Zürich, Schweiz; 3Universität Bremen Faculty 03: Mathematics and Computer Sciences Bremen, Deutschland
Background
Takotsubo Syndrome (TTS) is a heterogeneous condition with outcomes varying from spontaneous recovery in within days or few weeks to serious complications, including in-hospital death. Self-organizing maps (SOMs) are unsupervised neural networks able to organize data according to their similarities and identify clusters.
Purpose
The aim of present study was to establish a new classification of TTS-subgroups based on SOMs and to assess outcomes of different subgroups.
Methods
Patients with TTS were enrolled from a multicenter registry. Clinical parameters such as age, sex, systolic blood pressure on admission, heart rate on admission, left ventricular ejection fraction on admission, white blood cell levels on admission, history of hypertension, history of diabetes mellitus, InterTAK classification, and TTS morphological type were recorded. Based on patterns and similarities in clinical features, patients were classified into clusters using SOMs algorithm. Overall long-term mortality was assessed for each subgroup.
Results
A total of 1’613 patients were included in the study population. Using SOMs, patients were classified into 4 groups (S1, S2, S3, S4). Overall survival at 10-years follow-up was significantly different among groups (Log Rank p < 0.001). Adjustment with Cox-regression analysis was performed using the group with better prognosis (S1) as reference. This showed that patients in the groups S2 (adjusted HR: 2.77; 95% CI: 1.51 to 5.08), S3 (adjusted HR: 3.85; 95% CI: 2.22 to 6.65), and S4 (adjusted HR: 4.76; 95% CI: 2.66 to 8.53) had a significantly higher risk of mortality at 10-years follow-up.
Conclusions
A novel classification in subgroups of patients with TTS can be achieved with implementation of SOMs based on clinical variables easily obtainable in the acute phase. Thus, it could potentially serve as a useful tool to better characterize clinical profiles and predict prognosis in patients with TTS.