BackgroundPatients with unprovoked venous thromboembolisms (VTEs) can be sub-classified based on the different phenotypes using a latent class analysis (LCA), which might be useful for selecting individual management strategies. MethodsIn the COMMAND VTE Registry-2 database enrolling 5197 VTE patients, the current derivation cohort consisted of 1556 patients with unprovoked VTEs. We conducted clustering with an LCA, and the patients were classified into subgroups with the highest probability. We compared the clinical characteristics and outcomes among the developed subgroups. ResultsThis LCA model proposed 3 subgroups based on 8 clinically relevant variables, and classified 592, 813, and 151 patients as Class I, II, and III, respectively. Based on the clinical features, we named Class I the younger, Class II the older with a few comorbidities, and Class III the older with many comorbidities. The cumulative 3-year anticoagulation discontinuation rate was highest in the older with many comorbidities (Class III) (39.9 %, 36.1 %, and 48.4 %, P = 0.02). There was no significant difference in the cumulative 5-year incidence of recurrent VTEs among the 3 classes (12.8 %, 11.1 %, and 4.0 % P = 0.20), whereas the cumulative 5-year incidence of major bleeding was significantly higher in the older with many comorbidities (Class III) (7.8 %, 12.7 %, and 17.8 %, P = 0.04). ConclusionThe current LCA revealed that patients with unprovoked VTEs could be sub-classified into further phenotypes depending on the patient characteristics. Each subclass phenotype could have different clinical outcomes risks especially a bleeding risk, which could have a potential benefit when considering the individual anticoagulation strategies. Clinical trial registrationURL: http://www.umin.ac.jp/ctr/index.htmCOMMAND VTE Registry-2: Unique identifier, UMIN000044816COMMAND VTE Registry: Unique identifier, UMIN000021132