- Top of page
- Authorship contributions
- Financial disclosure
Vitamin K antagonists (VKA) are effective in primary and secondary prevention of thromboembolism, but the associated risk of bleeding is an important limitation. The majority of bleeds are clinically mild. In this study, we assessed whether these minor bleeds are associated with major bleeding, when controlling for other important risk indicators, including the achieved quality of anticoagulation. For this, 5898 patients attending a specialized anticoagulation clinic were retrospectively studied for 1 year after initiation of VKA therapy. The risk of major bleeding was estimated using a multivariate piecewise exponential model with time-varying exposure for occurring minor bleeds. In patients with a minor bleed (N = 1015) subsequent major bleeding occurred more frequently than in patients without a minor bleed (N = 4883), with an incidence rate of 2·3 [95% confidence interval (CI) 1·4–3·7] vs. 1·2 per 100 person-years (95% CI 0·9–1·7). The adjusted relative risk of subsequent major bleeding after a minor bleed was 2·9 (95% CI 1·1–7·2, P = 0·024). The percentage of time that a patient had an International Normalized Ratio (INR) above 5 was also independently associated with major bleeding, with a 2·2-fold increased risk in patients with at least 9% of time above INR 5 (95% CI 1·3–4·0, P = 0·006). Minor bleeds alert for subsequent major bleeding, independent of the quality of anticoagulation.
Vitamin K antagonists (VKA) are widely used in primary and secondary prevention of thromboembolism. Although these drugs are effective, the associated risk of bleeding is an important limitation (Schulman et al, 2008). Both effectiveness and risk of bleeding are related to the actual level of anticoagulation, expressed by the International Normalized Ratio (INR) (Veeger et al, 2005; Schulman et al, 2008). The intensity of anticoagulation varies widely within and between patients, due to environmental and genetic factors (Penning-van Beest et al, 2001; Kucher et al, 2004; Wittkowsky & Devine, 2004; Palareti et al, 2005; Wilkinson, 2005; Wittkowsky, 2005; Ansell et al, 2008), as well as patient compliance (van der Meer et al, 1997; McCormick et al, 2001). In this respect, the individual time spent below, within and above the INR target range can be used to identify patients on VKA who are at high risk of (recurrent) thromboembolism and bleeding (Veeger et al, 2005, 2006).
Due to these variabilities, continuous monitoring of the level of anticoagulation is mandatory. In the Netherlands, a unique nationwide network of 61 specialized anticoagulation clinics is responsible for the VKA management of all outpatients. In these specialized clinics, experienced physicians and dosing assistants are involved in optimizing VKA therapy, using computerized dosing programmes and protocols for dose adjustments in case of under- and over-anticoagulation.
In assessing the risk of bleeding while using VKA, the primary focus is on major bleeding (Schulman et al, 2008). An important risk factor for major bleeding is the achieved level of anticoagulation. A previous major bleed is also identified as a risk factor for subsequent major bleeding, although not consistently in all studies (The Stroke Prevention in Atrial Fibrillation Investigators, 1996; White et al, 1996; Beyth et al, 1998; Kuijer et al, 1999; Shireman et al, 2006). Such an association has not yet been demonstrated for minor bleeding, which might be due to over-anticoagulation or a pre-existing mild bleeding disorder, which is enhanced by VKA. In either case, the occurrence of a minor bleed during VKA therapy could indicate a possible high-risk situation, necessitating a renewed risk-benefit assessment of the VKA therapy in those patients.
In this study, we hypothesized that after a minor bleed, patients are at increased risk for subsequent major bleeding.
- Top of page
- Authorship contributions
- Financial disclosure
In total, 5898 consecutive patients were included in our study. Their characteristics and anticoagulation data are summarized in Table I. Cumulative treatment time was 3892 patients-years, during which 117854 INR measurements were performed.
Table I. Patients’ characteristics and anticoagulation data.
| ||Total N = 5898|
|Age (years)*; P < 0·001||68 (37–86)|
|Diabetes mellitus (%); P = 0·04||9|
|Malignancy (%); P = 0·04||4|
|Concomitant drugs (%)|
| Low dose aspirin||3|
| Beta blockers||5|
| Calcium antagonists||7|
| Antiarrhythmic drugs||8|
| Lipid lowering drugs||6|
| Oral contraceptives/HRT||1|
|Follow-up time (months)||9·1 (1·4–11·8)|
|Number of INRs||20 (7–35)|
|Time between INRs (d)||13 (3–28)|
|Time with INR > 5·0 (%)||3 (0–23)|
|ITTR (%)||43 (13–75)|
Acenocoumarol was the predominantly used VKA (97%). The remaining patients used phenprocoumon (2%) or switched VKA during their treatment period (1%). VKA was combined with low dose aspirin in 3% of patients. Median treatment time was 9·1 months (90% central range, 1·4–11·8), during which 20 INR measurements were performed (median, 90% central range, 7–35). Based on all the actual measurements, median INR was 3·0 (90% central range, 1·6–5·7). At a patient level, the median percentage of time spent above INR 5·0 was 3% (90% central range 0–23). When considering the patients that were classified as ‘over-anticoagulated’ (i.e. a-priori defined as the upper quartile of patients with a highest amount of time above INR 5·0), these patients were at least 9% of their treatment time above INR 5·0 (median percentage of time above INR 5·0 = 15% (90% central range 9–36).
Minor bleeding. A total of 1501 minor bleeds occurred in 1015 of 5898 patients (17%), 716 patients had a single minor bleed (71%), 194 patients two minor bleeds (19%), and 105 patients multiple recurrent minor bleeds (10%). The median INR at the time of the minor bleed was 3·3 (90% central range 1·9–6·5). The absolute risk, expressed as the incidence rate per 100 person-years, was 29·6 (95% CI, 27·8–31·4). Haematomas (40%), nose bleeds (27%), conjunctiva bleeds (11%) and haematuria (10%) were the most frequently observed minor bleeds (Table II). There was a difference in the occurrence of minor bleeds between women and men, mainly due to a difference in the number of haematomas and nose bleeds, with 418 and 241 in women vs. 177 and 167 in men. This resulted in a higher observed risk of minor bleeding in women than men (incidence rates 37·1, 95% CI, 34·1–40·2 in women vs. 23·1, 95% CI, 21·0–25·4 in men; rate ratio 1·6, 95% CI 1·4–1·8, P < 0·001).
Table II. Type of minor bleeds.
|Minor bleeds||N (%)|
| Haematoma||595 (39·6)|
| Nose bleed||408 (27·2)|
| Conjuctiva bleed||163 (10·9)|
| Haematuria||143 (9·5)|
| Gastrointestinal bleed||68 (4·5)|
| Haemoptoe||32 (2·1)|
| Vaginal bleed||28 (1·9)|
| Muscular bleed||12 (0·8)|
| Other||52 (3·5)|
A medical intervention, including temporary interruption, was required for 73 minor bleeds in 64 patients. These minor bleeds were classified as clinically relevant.
Major bleeding. Major bleeding occurred in 56 patients (0·9%), with an incidence rate of 1·4 per 100 person-years (95% CI, 1·1–1·9). Nine major bleeds were fatal. The types of major bleeding are presented in Table III. The level of anticoagulation at the time of major bleeding, expressed as the median INR, was 3·4 (90% central range 1·9–9·6).
Table III. Major bleeding.
|Type of event*||Minor bleeding||Total|
|Yes n = 1015||No n = 4883||n = 5898|
|Major bleeding n (%)||18 (1·8)||38 (0·8)||56 (0·9)|
| Non-fatal gastrointestinal||5||8||13|
| Fatal gastrointestinal||1||1||2|
| Non-fatal intracranial||1||4||5|
| Fatal intracranial||2||5||7|
| Other major bleeding†||9||17||26|
|Incidence rate (95% CI) per 100 patient-years||2·3 (1·4–3·7)||1·2 (0·9–1·7)||1·4 (1·1–1·9)|
Prediction of major bleeding. A higher risk of major bleeding was observed in the patients who had suffered a minor bleed. The incidence rate in these patients was 2·3 per 100 person-years (95% CI, 1·4–3·7), whereas in patients without a prior minor bleed the incidence rate was 1·2 per 100 person-years (95% CI, 0·9–1·7). The incidence rate of major bleeding after a minor bleed was 3·9 per 100 person-years (95% CI, 2·3–6·2). In this, follow-up time was restricted to the post-minor bleed period. Notably, in the 64 patients in whom a clinically relevant bleed had occurred, for which they received a medical intervention, including temporary interruption, major bleeding did not occur.
In addition, we performed a sensitivity analysis on the risk of major bleeding after a minor bleed, excluding the haematomas. This is especially relevant given that misclassification of minor bleeds might have occurred to some degree. When excluding haematomas, in patients with a minor bleed the incidence rate of major bleeding was 2·6 per 100 person-years (95% CI, 1·4–4·6). In the patients with a haematoma this was 1·9 (95% CI, 0·7–4·2), not statistically significantly different (rate ratio 1·4 (95% CI, 0·5–4·0, P = 0·53). When taking only the time after the minor bleed into consideration, the incidence rates were 4·3 (95% CI, 2·2–7·5) and 3·3 (95% CI, 1·2–7·2), respectively (rate ratio 1·3; 95% CI, 0·5–3·7, P = 0·63).
Furthermore, also in the ‘over-anticoagulated’ patients (highest quartile, more than 9% of treatment time above INR 5·0), major bleeding was more frequently observed. The incidence rate in these patients was 1·9 per 100 person-years (95% CI, 1·2–3·0), whereas in the patients in the lower three quartiles this was 1·2 (95% CI, 0·9–1·7).
The adjusted relative risks associated with minor bleeding and over-anticoagulation were estimated using a multivariable piecewise exponential model. Both over-anticoagulation and prior minor bleeding were considered as time-dependent risk indicators, in which exposure was assessed for each 30-d period.
As shown in Fig 1, a minor bleed was an independent risk indicator with a 2·9-fold increased risk of major bleeding during the following month (HR = 2·9, 95% CI, 1·1–7·2, P = 0·024). ‘Over-anticoagulation’ was associated with a 2·2-fold increased risk of major bleeding (HR = 2·2, 95% CI, 1·3–4·0, P = 0·006). Additional independent risk indicators identified by multivariate analysis were diabetes mellitus, malignancy, use of diuretics (time-dependent) and age above 70 years. Model evaluation resulted in acceptance of the exponential model (Lagrange multiplier test; P = 0·12), indicating that the hazard of major bleeding was constant over the 1-year interval.
Figure 1. Hazard ratios of major bleeding, from survival analysis using a piecewise exponential model. The presented variables were included in the final model.
Download figure to PowerPoint
After excluding the haematomas, piecewise exponential survival analysis resulted in a crude hazard ratio for the risk of major bleeding after a minor bleed of 4·1 (95% CI, 1·5–11·4, P = 0·007). After adjustment this reduced to 3·8 (95% CI, 1·4–10·6, P = 0·010).
An additional analysis was performed using an exposure period of 2 months. The crude hazard ratio for prior minor bleeding over the 2-month period was 3·5 (95% CI 1·7–7·7, P = 0·001). The adjusted hazard ratio was 3·3 (95% CI 1·5–7·0, P = 0·002), indicating that the increased risk lasts over the 1-month period.
- Top of page
- Authorship contributions
- Financial disclosure
To our knowledge, this study is the first to specifically address minor bleeding as potential risk indicator for major bleeding. We showed that the occurrence of a minor bleeding was associated with a 2·9-fold increased risk of major bleeding in the following month. This increased risk was independent of the amount of over-anticoagulation.
The absolute risk of major bleeding in our study was 1·4 per 100 person-years. Taking into consideration that our cohort consisted of unselected patients in whom VKA therapy was managed for 1 year by a specialized anticoagulation clinic, this absolute risk is in line with the rates reported previously (van der Meer et al, 1993; Palareti et al, 1996; van Es et al, 2002; Linkins et al, 2003; Ansell et al, 2008; Schulman et al, 2008; Palareti & Cosmi, 2009). Our rate of minor bleeding of 29·6 per 100 person-years is more difficult to compare. In a recent large clinical trial in AF patients the rate of minor bleeding was 18·2 (Connolly et al, 2009). Two other large trials reported higher incidence rates for combined major and minor bleeding, ranging from 30 to 47 per 100 person-years (Olsson, 2003; Albers et al, 2005).
Our finding regarding the quality of anticoagulation as risk indicator is consistent with previous studies, where the association between quality of anticoagulation and major bleeding was clearly demonstrated (Samsa et al, 2002; Reynolds et al, 2004).
Some limitations must be considered. Firstly, our study cohort consisted of patients treated more than 10 years ago. This could have resulted in ‘outdated’ findings, and thereby limiting the generalisability of our findings. However, both the type of VKAs used and the strategy that was applied to achieve optimal anticoagulation in our study are still used in current practice, making our findings still applicable for current VKA therapy.
Unfortunately, at the time of data collection, information on important genetic factors related to the individual sensitivity to VKA (Hermida et al, 2002; Gage & Lesko, 2008) was not available. Another limitation is the potential underreporting of bleeding complications in a study in which complications are collected during the course of regular treatment. However, we only included ‘new’ patients and not those already on VKA at the time of the start of the study and we had access to detailed information regarding all bleeding complications during the 1 year of follow-up, including the minor bleeds. Furthermore, minor bleeds were assessed by the experienced staff of the anticoagulation clinic responsible for monitoring the VKA therapy and/or reported spontaneously by the patients themselves. Therefore, in our opinion underreporting was very limited, which is also indicated by the observed bleeding risk that is consistent with previous reports (van der Meer et al, 1993; Palareti et al, 1996; Go et al, 2003; Linkins et al, 2003; Olsson, 2003; Albers et al, 2005; DiMarco et al, 2005; Hylek et al, 2007; Schulman et al, 2008; Connolly et al, 2009; Palareti & Cosmi, 2009). This is also supported by the sensitivity analysis excluding haematoma, which is most prone for misclassification. In this respect, the associated risk after a haematoma was lower than after other types of minor bleed, although not significant. We can only speculate that this difference can be attributed to the ‘smaller’ haematomas. However, as additional information on the size of haematomas was not available, this could not be evaluated.
A second issue regarding generalisability is the intensity of anticoagulation used in the Netherlands. In particular, to avoid under-anticoagulation, the Dutch anticoagulation clinics use a therapeutic range which is 0·5 INR wider towards a higher intensity than the conventionally employed target ranges in other countries (e.g. INR 2·0–3·5 vs. INR 2·0–3·0). The target ranges that were used were 0·5 INR higher (e.g. INR 2·5–3·5 vs. INR 2·0–3·0). Despite the higher target ranges, the percentage of time within the target range as observed in our study (median ITTR 43%) was comparable to those reported in previous studies (Fihn et al, 2003; Ansell et al, 2008). The percentage of time with an INR above 5·0 was limited to only 3% of treatment time. Furthermore, with an absolute risk of major bleeding of 1·4 per 100 person-years, in the setting of a specialized anticoagulation clinic this strategy did not lead to excessive major bleeding.
Therefore, in our opinion, the observed association of both poor anticoagulation and minor bleeding with major bleeding is still relevant for VKA therapy today.
In conclusion, minor bleeding is associated with subsequent major bleeding. The occurrence of a minor bleed should increase the awareness towards a potentially high-risk situation. In this, the achieved level of anticoagulation should also be considered, not just the INR at the time of the minor bleed, but especially the amount of over-anticoagulation during the preceding treatment period. In our study, we used the percentage of time above INR 5·0 during a 30-d period.
In line with this awareness, a renewed risk-benefit assessment should be performed. Whether a lower intensity of anticoagulation might be more suitable for these high-risk patients, without compromising the efficacy of VKA therapy to prevent thromboembolic complications is speculative. In this, also VKA as the anticoagulant drug of choice could be reconsidered. For some patients, no anticoagulant agent, aspirin, or – in the near future – an oral II or Xa inhibitor might be a more suitable option.