Despite a tremendous increase of direct oral anticoagulants (DOACs) prescriptions in recent years, only few data is available analysing prescribers' adherence to Summary of Product Characteristics (SmPC). We aimed to assess adherence to registered indications, contraindications, special warnings/precautions, and potential drug-drug interactions for three DOAC compounds (dabigatran, rivaroxaban, and apixaban) in six databases of five European countries (The Netherlands, United Kingdom, Spain, Denmark, and Germany). We included adult patients (≥18 years) initiating DOACs between 2008 and 2015. For several SmPC items, broad definitions were used due to ambiguous SmPC terms or lacking data in some databases. Within the study period, a DOAC was initiated in 407 576 patients (rivaroxaban: 240 985 (59.1%), dabigatran: 95 303 (23.4%), and apixaban: 71 288 (17.5%)). In 2015, non-valvular atrial fibrillation was the most common indication (>60% in most databases). For the whole study period, a substantial variation between the databases was found regarding the proportion of patients with at least one contraindication (inter-database range [IDR]: 8.2%-55.7%), with at least one special warning/precaution (IDR: 35.8%-75.2%) and with at least one potential drug-drug interaction (IDR: 22.4%-54.1%). In 2015, the most frequent contraindication was “malignant neoplasm” (IDR: 0.7%-21.3%) whereas the most frequent special warning/precaution was “prescribing to the elderly” (≥75 years; IDR: 25.0%-66.4%). The most common single compound class interaction was “concomitant use of non-steroidal anti-inflammatory drugs” (IDR: 3.0%-25.3%). Contraindications, special warnings/precautions, and potential drug-drug interactions were present in a relevant number of new DOAC users. Due to broad definitions used for some SmPC terms, overall proportions for contraindications are prone to overestimation. However, for unambiguous SmPC terms documented in the databases sufficiently, the respective estimates can be considered valid. Differences between databases might be related to “true” differences in prescription behaviour, but could also be partially due to differences in database characteristics.