Previous clinical trials investigating annual bleeding frequency during oral anticoagulation therapy demonstrated ranges from 1.7% to 3.0% for major and approximately 0.6% for fatal bleedings 
. Clinical trials are confounded by varying degrees of selection bias in the study populations.
Randomized studies have reported frequencies of less than half that described in observation studies, which is probably related to differences in the entry criteria for patients 
. In a large Italian inception-cohort study, 1.1% major and 0.25% fatal bleeding rates were found in 2745 patients during 2011 patient years 
. In our study, comprising 2731 patients with over 5044 treatment years, the risk of bleeding was approximately 4 times higher (3.9%). Here all patients on treatment were recorded and all patient records reviewed, while in the Italian study patients with expected difficulty in obtaining appropriate follow-up were excluded. Furthermore, the mean age in our study was 69 years and 56% of the patients were older than 70 years, compared with a mean of 63 years and 35% older patients in the Italian study. In some studies, age was considered a risk factor for bleeding, 
while others could not find such a relationship 
Due to the long observational period and large cohort in this study, the greater risk of bleeding conferred by older age is highly significant and this association was independent of other possible determinants studied. The age and sex distribution in our study reflects the general population on oral anticoagulants in Sweden and, therefore, is relevant for clinicians and their administration of warfarin treatment 
In Sweden, warfarin treatment is monitored in both specialized anticoagulation clinics and primary healthcare centres. One topic under debate is whether the skill of the monitoring site has an impact on treatment safety as related to bleeding complications [17–19]
In our study we could not find any advantage with regard to bleeding complications in the specialized clinics. The target INR (2.0–3.0) was the same in both settings. High INR has been suggested to confer a greater risk of bleeding 
, while low INR reduces the risk 
. The majority of major (64%) and fatal bleedings (81%) occurred at therapeutic or subtherapeutic INR levels (i.e. below 3.0). Thus, there are other factors contributing to increased bleeding susceptibility that need to be elucidated. It has also been suggested that the bleeding risk is higher during the first period of warfarin treatment 
. Due to the non-inceptive design of this study, we can not evaluate this issue as a possible contributor to the increased risk of haemorrhage in patients treated for venous thromboembolism. The mean duration of treatment was shorter in this group (1.08 years) compared with the group with atrial fibrillation (2.04 years). However, differences in frequency of bleeding complications in different indications may also have been influenced by how patients with high-risk features for bleeding complications are dealt with. Treatment with oral anticoagulants may have been withheld in atrial fibrillation while in venous thromboembolism the treatment time has instead been reduced and in patients in need of a heart valve prosthesis a biological prosthesis has been implanted instead of a mechanical prosthesis.
An association with bleeding has been reported for hypertension 
and cancer 
. In addition, a history of peptic ulcer disease was reported as a risk factor 
. In the multivariate model we confirmed the observation for a history of previous peptic ulcer disease as an independent determinant with a threefold increased risk, but no association with hypertension, diabetes, or cancer was found. Moreover, patients with a history of previous peptic ulcer disease had a very high risk of gastrointestinal bleeding during oral anticoagulation, the risk being 13 times higher with an absolute risk of 12.1 per 100 treatment years. These patients must be carefully examined before the initiation of oral anticoagulation. Eradication of helicobacter pylori could possibly diminish this extremely high bleeding risk, although this merits further studies.
Even though oral anticoagulation is beneficial in many situations, bleeding complications constitute a severe clinical problem and may outweigh the benefit of warfarin. The site of INR monitoring seems less important. The data from this study indicate a need for caution in treatment of the elderly. The clinician's reluctance to treat the oldest patients with warfarin when faced with less aggravating indications is hereby supported.