Article Text

Download PDFPDF
Original article
Risk of death and stroke associated with anticoagulation therapy after mitral valve repair
  1. Nana Valeur1,
  2. Charlotte Mérie2,
  3. Morten Lock Hansen2,
  4. Christian Torp-Pedersen3,
  5. Gunnar H Gislason2,
  6. Lars Kober4
  1. 1Department of Cardiology, Bispebjerg Hospital, Copenhagen, Denmark
  2. 2Department of Cardiology, Gentofte Hospital, Gentofte, Denmark
  3. 3Department of Health, Science and Technology, Aalborg University, Aalborg, Denmark
  4. 4Department of Cardiology, The Heart Centre, Rigshospitalet, University of Copenhagen, Copenhagen, Denmark
  1. Correspondence to Dr Nana Valeur, Bispebjerg Hospital, Bispebjerg Bakke 23, 2400, Copenhagen, Denmark; nv{at}


Objective Guidelines generally recommend oral anticoagulation to be considered the first 3 months after mitral valve repair based on small studies and consensus. However, in several studies no benefit of anticoagulation has been found.

Methods From the national registries we identified all Danish patients who underwent mitral valve repair during the period between 1997 and 2012. Medication, hospitalisation and mortality data were studied. The association of use of vitamin K antagonists (VKAs) at discharge and risk of stroke/death was evaluated by means of Cox regression, landmark analyses and propensity matched models.

Results 2188 patients without prior VKA use, stroke or death day 7 after discharge were included and median follow-up was 4.9 years (0–13.7). 859 (39%) were discharged on VKAs and 523 (24%) experienced death or stroke, 60 of these occurred within the first 3 months and 24 between 3 and 6 months. Compared with patients without post-discharge VKA, patients on VKA had a lower risk of death/stroke at 3 months (HR=0.28, CI (0.13 to 0.62), p=0.002) and in the time period from 3 to 6 months (HR=0.85, CI (0.35 to 2.07), p=0.72). Risk of significant bleeding complications within 3 months were comparable in the two groups with 23 (2%) among patients without VKA and 6 (1%) among VKA-treated.

Conclusion VKA treatment after mitral valve repair is associated with a markedly lower risk of adverse events as stroke or death without excess major bleeding risk during the first 3 months following surgery.

Statistics from

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.


Mitral valve repair is considered less thrombogenic than mitral valve replacement.1 ,2 Even though oral anticoagulation may be considered for the first 3 months after bioprosthetic mitral valve replacement (class IIA indication, level of evidence C),3 ,4 according to current guidelines, only a limited number of patients are discharged with anticoagulation.5 ,6 Also, guidelines are based on expert consensus and few small retrospective studies with diverging results and set-ups evaluating mitral valve replacement rather than repair.

We examined whether use of vitamin K antagonists (VKAs) after mitral valve repair is associated with any differences in risk of thromboembolic events, major bleeding and death by conducting a nationwide Danish cohort study.


Data from national registries, including the National Patient Registry and the Registry of Medicinal Product Statistics (Prescription registry) were used. Data were linked on individual levels using the Danish personal registration number system in which all residents of Denmark are registered with a unique personal number.7 Vital status was obtained from the Central Population Register, recording all deaths within 14 days.

Data from the Danish National Patient Registry were used to select all patients without previous cardiac surgery having mitral valve repair surgery performed in the period between 1 January 1997 and 31 December 2012. The Danish National Patient Registry contains information on all hospital admissions, diagnoses and procedures in Denmark since 1978. Diagnoses were classified according to the International Classification of Diseases, Eighth Revision (ICD-8) and International Statistical Classification of Diseases, 10th Revision (ICD-10) and operations classified according to the NOMESCO Classification of Surgical Procedures (NCSP). We registered patients having a mitral valve repair as the following operation procedures: KFKB00-10 and KFKC00-60. Previously, the use of diagnoses from the Danish National Patient Registry has proven valid with positive predictive values of 74%–97%.8 ,9

Information on patient medication was obtained from The Danish Register of Medicinal Product Statistics containing information on all claimed prescriptions in Denmark (coded according to International Anatomical Therapeutic Chemical classification),10 date of drug dispensation, strength and quantity dispensed. Because of partial reimbursement of drug expenses by the Danish healthcare authorities, all pharmacies are required to provide information that ensures complete and accurate registration. This registry has previously been described and found accurate.11 By Danish law ( scientific projects using information from existing registries does not require approval from a scientific ethical committee. Thus, ethical approval and informed consent was not required from the Danish Ethical Committee, since this study was retrospective and conducted in the public organisation (Statistics Denmark) using encrypted personal data. The study was approved by the Danish Protection Agency with reference number GEH-2014-013; I-Suite number:02731.

We evaluated the benefit of post-discharge VKA treatment in patients having mitral valve repair performed with respect to comorbidity, procedures, medication, age and sex, focusing on post-operative mortality, thromboembolic events and bleeding complications, considering medication with VKA, as well as medication for heart failure (HF), ischaemic heart disease (IHD) and arrhythmia. Diabetes was defined according to use of antidiabetic therapy as done previously.12 Only three patients received another anticoagulant than warfarin (Dabigatran) and excluded from the analyses.

Outcome measures

Our primary end point was the combination of stroke or death. Mortality, stroke and major bleeding were used as secondary end points. We evaluated the risk of major bleeding requiring hospital admission with gastrointestinal bleeding/ulcer, intracranial and other serious bleeding as airway bleeding or bleeding from the urinary tract. End points as well as individual consideration of major bleedings were evaluated in a short-term interval (7 days–3 months), in longer intervals (7 days–1 year) and overall (7 days to end of follow-up).

Statistical analyses

Patient characteristics were compared between those initiating VKAs and those who did not. Categorical variables were compared by using χ2 test and Student's t test or the analysis of variance test for continuous data.

The primary analyses were performed according to the two VKA groups, patients on VKA on day 7 after discharge or not, and all analyses started 7 days after discharge to give patients time to claim their prescriptions on VKAs. Therefore patients who experienced a stroke or died within 7 days after discharge, as well as patients with missing data were excluded from the analyses. For sensitivity, all analyses were repeated allowing patients to claim prescriptions up to day 14. We further explored these associations according to VKA as a time-dependent variable (described below).

The risk of stroke, death and bleeding associated with VKA was estimated by means of Cox proportional hazards models adjusted for relevant covariates (age, gender, IHD, HF, diabetes mellitus, mitral valve repair + coronary artery bypass (CABG), renal impairment, chronic obstructive pulmonary disease (COPD), history of stroke and atrial fibrillation). All variables from table 1 were entered in the Cox models, but only those with significant importance in one of the models were retained in the analyses. The proportional hazard assumptions were tested using Schoenfeld residuals and evaluated by log(−log (survival)) curves, and absence of interactions between model covariates were tested, and found valid unless otherwise reported. HR and CIs are given.

Table 1

Baseline characteristics according to vitamin K antagonist (VKA) treatment following mitral valve repair

To adjust for baseline differences between VKA groups, we performed a sensitivity analysis with propensity score-matched subgroups based on relevant variables (age, gender, IHD, HF, diabetes, mitral valve repair+CABG, renal impairment, COPD, history of stroke, atrial fibrillation), using the Greedy matching program (

To assess if risk of our chosen outcomes changed over time, landmark analyses were performed for three periods (day 7–3 months, 3–6 months, 6–12 months) in Cox models incorporating VKA treatment at baseline in one set of analyses, and as a time-dependent variable together with predefined covariates in sensitivity analyses. Incidence rates for the same time periods were performed by splitting the time periods and calculating number of events per 1000 person-years (py) in each time band, stratified by VKA treatment comparing use versus no use. Patients were censored at time of event, emigration or end of 2012. In order to control for competing risk, additional sensitivity analysis comparing cumulative incidence of stroke in relation to VKA groups were tested with Gray's test for short-time and long-time treatments.

A two-sided p value of less than 0.05 was considered statistically significant. All analyses were performed by use of SAS software (V.9.2; SAS Institute), and Stata software (V.11.1; StataCorp).


From 1997 to 2012, 3080 patients underwent mitral valve repair, from these we excluded 892 patients (29%) (figure 1). Of these, 628 patients (21%) received VKA therapy prior to the operation; 116 patients had concomitant mechanical aortic valve replacement performed and 7 had a reoperation during hospitalisation resulting in a mechanical mitral valve; 120 died or had a stroke during admission or within the first 7 days after discharge.

Figure 1

Flowchart of patients with a MVR. AVR, aortic valve repair; MVR, mitral valve repair; reopr, reoperated.

In general, the 892 patients excluded were older than the included patients with more comorbidity.

The final study population comprised 2188 patients who had never received VKA prior to mitral valve repair. Of these 751 patients (34%) were on VKA therapy on day 7 after discharge. Baseline characteristics according to anticoagulation status at day 7 are presented in table 1. Patients on VKA treatment were predominantly male; they tended to be younger and were less likely to have a history of HF, COPD, stroke or diabetes.

A small number of patients had both mitral valve repair and biological aortic valve replacement, equally distributed in the two VKA groups. Patients who also had CABG were less frequently represented in the VKA group. Median duration of VKA therapy was 156 days (5-95 percentiles 49-4262 days).

Clinical outcomes

Combined end point (death/stroke)

Among the 2188 patients, 523 (24%) experienced the primary outcome (death and/or stroke) during a median follow-up of 4.9 years (0–13.7). A total of 60 patients (3%) experienced the primary outcome within the first 3 months and the risk was significantly lower among patients on VKA compared with those without at 3 months and long-term follow-up, in both univariate and multivariable analyses at 3 months (multivariable HR=0.28 (0.13 to 0.62; p=0.002) and long-term (multivariable HR=0.81 (0.67 to 0.98; p=0.03).


A total of 427 patients (20%) died during long-term follow-up and 44 (2%) died within 3 months from discharge. Both long-term and short-term (3 months) mortality were significantly lower in patients receiving VKA compared with those without VKA (multivariable HR 0.29 (0.11 to 0.73; p=0.009; multivariable HR 0.78 (0.63 to 0.97; p=0.03)), respectively.

Incidence rates were highest early after discharge and decreased to a stable level after 3 months. Mortality rates were significantly lower in patients treated with VKA, the rate during the first 3 months after discharge was 41 (24–70) per 1000 py compared with 143 (100–205) per 1000 py for patients without VKA. From 3 months to 1 year the event rate in patients treated with VKA was 24 (15–38) per 1000 py versus 28 (19–43) per 1000 py in patients not treated with VKA. Sensitivity analysis with VKA as time-dependent variable resulted in HR 0.47 (0.29 to 0.77; p=0.003) for 3 months, and HR 0.86 (0.69 to 1.08; p=0.19) for long-term follow-up.


In total, 172 patients (8%) suffered from a stroke, 21 (1%) during the first 3 months after discharge where risk of stroke was higher in patients without VKA therapy (HR 0.22 (0.05 to 0.97); p=0.05 (table 2)). Rate of stroke in patients treated with VKA was 17 (7–42) per 1000 py and 78 (47–130) per 1000 py in patients without VKA.

Table 2

Death, stroke and bleedings at different time points according to treatment with vitamin K antagonists (VKA)

The incidence rates of stroke also decreased significantly 3 months after discharge where rates of stroke were comparable in the two groups (HR 0.90 (0.63 to 1.30; p=0.59)). From 3 months to 1 year event rate in warfarin patients was 22 (14–36) per 1000 py versus 21 (13–35) per 1000 py in patients not treated with warfarin.


The risk of significant bleeding complications did not differ between the exposure groups. Evaluating bleeding risk with VKA as a time-dependent variable showed no differences in risk in the first year (HR 0.49 (0.20 to 1.21), but it showed a trend towards increasing bleeding risk over time (figure 2).

Figure 2

Forest plot of HRs according to vitamin K antagonist (VKA) treatment in different time intervals.

Atrial fibrillation

Incidence of atrial fibrillation was highest early after discharge. During index hospitalisation, 387 were registered with atrial fibrillation. From day 7 to day 30 after discharge, 156 patients were hospitalised with atrial fibrillation. A total of 54 patients (35%) at day 7 and 111 patients (71%) at day 30 were on VKAs. Among the 156 incident atrial fibrillation patients, 1 of 111 patients with atrial fibrillation in the VKA-treated group had a stroke compared with 4 of 45 patients with atrial fibrillation without VKA. From day 30 to 3 months, another 103 patients were diagnosed with atrial fibrillation and 90 of these patients (87%) eventually received VKA (3 months), 1 of 90 (1%) with atrial fibrillation from the VKA group developed a stroke, whereas 1 of 13 (8%) with atrial fibrillation developed stroke in the non-VKA group.

Sensitivity analyses

Using the propensity-score matching, 1224 patients were matched in two groups of 612 treated with VKA at discharge and 612 who were not. Overall C-statistics for propensity-score models was 0.71. Baseline characteristics were comparable between patients treated with VKA and those who were not (see online supplementary eTable 1). As in the primary analyses VKA treatment was associated with beneficial effect with respect to mortality and the combined end point during the first 3 months (figure 2).

The Kaplan–Meier curves for the combined end point and for death alone of the entire population and of the propensity-matched cohorts are given in figure 3.

Figure 3

Kaplan–Meier curves of the combined end point mortality/stroke and of mortality according to Vitamin K antagonist (VKA) treatment, in the entire population and in the propensity-matched population.

During day 7 to day 14, another 239 patients (11%) initiated VKA therapy and we tested the robustness of the results by including VKA therapy day 14 for patients without stroke or death before day 14. The association between VKA and risk of the primary outcome (mortality and/or stroke) in a propensity-matched population remained significant in multivariable analyses (HR 0.21 (0.08 to 0.56; p=0.002) for 3 months, HR 0.38 (0.22 to 0.67; p=0.0008) for 1 year and for long-term follow-up HR 0.68 (0.53 to 0.67; p=0.002)).

The results from the landmark analyses are depicted in HRs given in figure 2 for the time intervals 0–3 months, 3–6 months and 6–12 months.

During the first 3 months, 44 patients died, 21 had strokes and 34 had significant bleeds. Prolonging the observation period from 3 to 6 months, of the remaining 2104 patients, 11 died, 13 had strokes and 14 had significant bleeds (40 patients had follow-up shorter than 3 months). Looking even further, at the 6 months to 1 year interval, of 2045 patients, 29 died, 18 had strokes and 15 had significant bleeds. The cumulative incidence of stroke confirmed that VKA was associated with a lower incidence of stroke in the first 3 months (p=0.02; Gray's test), a difference that disappeared over long time (p=0.75).

Multivariable analyses of death and/or stroke with VKA as time-dependent variable are presented in figure 2. The HRs for different time intervals are close to the original results. Thus, a statistically significant effect of VKA was found during the first 3 months with regards to the combined end point, mortality and a strong trend for stroke. For the following periods HR was more or less stable around 0.70 for the use of VKA to protect against both stroke and death.

Bleeding risk was higher with VKA treatment than without VKA, but not significant during the first year after the operation.


Our study showed that VKA treatment after mitral valve repair was associated with significantly lower rate of death and stroke during the first 3 months after surgery compared with patients not receiving VKA treatment. There may be further advantage of VKA treatment up to 6 months, but this is not clear from our data because of few end points after the first 3 months. After 1 year the lower risk associated with VKA treatment disappeared. Second, major bleeding during the first year after operation was not higher in the patients treated with VKA compared with those not treated with VKA.

Guidelines only recommend oral anticoagulation to be considered after mitral valve repair for 3 months, but a large number of these patients do not receive anticoagulation. In our study, only 36% were in treatment at day 7 after discharge. Similarly, in a European study,6 63% of patients and, in an American study,5 an average of 46% of patients discharged from mitral valve repair were prescribed oral anticoagulation at the discretion of the surgeon.

Assumptions regarding the effect of anticoagulation after mitral valve repair are made from a few retrospective studies including bioprosthetic operations4 ,13 where definitions of adverse events and complications differed, that is, thromboembolic events defined as hospital-registered stroke alone14 ,15 or as stroke, transient ischaemic attack, peripheral or coronary embolism registered by patients questionnaires.2 ,16 In these studies VKA reduced the risk of thromboembolic events in mechanical and biological valve replacement.16–18 In a report of 951 patients with mitral valve repair, anticoagulation was recommended for 2 months but was not evaluated.19 We evaluated risk of death, stroke and bleeding according to registration and hospital diagnoses as in a recent study by Schwann et al.15

In contrast to our conclusions Schwann et al and another recent study15 ,20 and a few older studies of mitral valve repair or bioprosthetic replacement21 ,22 found no significant differences in 30 days stroke or mortality rates according to postoperative anticoagulation strategy. Schwann et al15 evaluated 249 patients (144 with mitral valve repair), of whom the 192 (77%) discharged on VKA were older with more comorbidity and more frequently treated with bioprosthetic replacement. They found no differences in mortality or stroke. Mortality rate at day 30 was 1.2% in both groups, whereas our mortality rate at day 90 was 2.3% in patients without VKA and 0.8% in patients with VKA at discharge. Compared with Schwann et al our population was larger and focused only on mitral valve repair. Also, we excluded patients treated with VKA before the operation assuming these patients had another risk profile. These differences might explain the difference in impact of early treatment with VKA.

As in other studies of both aortic12 and mitral valve operations2 ,16 we found a significant decline in risk of death and stroke during the first months after surgery. In the study of Russo et al,2 patients with mitral valve repair represented the lowest risk of stroke compared with biological or mechanical replacement. The risk within 30 days was 1.5% comparable with ours (1% during the first 3 months) and returned to risk of stroke of the general population after 180 days, similar to our findings with an incidence rate of 143 per 1000 py during the first 3 months without VKA decreasing to 28 per 1000 py from 3 to 12 months.

We evaluated the risk of bleeding defined as hospitalisation for gastrointestinal bleeding/ulcer, intracranial bleeding and other serious bleeding (nose or urinary tract). The risk of serious bleeding has been addressed previously15 ,16 and like most reports we only registered few bleeding events, with no excess bleeding risk among patients treated with VKA during the first year after mitral valve repair.

Atrial fibrillation is frequent before and also de novo after mitral valve operation; in several studies incidences of 25%–40% were found during median follow-up of 5 years.14 ,23 We found a lower frequency of atrial fibrillation after discharge likely due to different registration. The impact of VKA treatment did not differ when censoring patients who developed atrial fibrillation.


Our data suggest that patients will benefit from treatment with VKA at least for the first 3 months, and probably 6 months after mitral valve repair surgery, emphasising current guidelines. However, our results should ideally be validated in a randomised setting, also testing whether long-term anticoagulation therapy is beneficial.


Since this is an observational study our results can only be hypothesis generating. Also, we have no detailed information on clinical data apart from diagnose codes. Thus, we do not have full knowledge of how many developed silent atrial fibrillation or bleeding complications not requiring hospitalisation, which may bias our results.

As we only had data on whether patients claimed a prescription on VKA, we have no information on whether the medication was taken or on the quality of anticoagulation. From large database studies evaluating international normalized ratio levels and time in therapeutic range between 2 and 3 in atrial fibrillation patients,24–26 the time in therapeutic range was in average less than 60%, and anticoagulation out of range was associated with increased risk of thromboembolic and haemorrhagic events.27 ,28


VKA treatment after mitral valve repair is associated with a significantly lower rate of death and stroke during the first 3 months (possibly 6 months) after discharge without a significantly higher bleeding risk up to 1 year.

Key messages

What is already known on this subject?

  • Anticoagulation is recommended for the first 3 months after bioprosthetic mitral valve replacement. The recommendation is surmised from a few small retrospective studies with diverging results and in practice, a large proportion of patients do not receive anticoagulation after mitral repair.

What might this study add?

  • Retrospective data on 2188 patients with mitral valve repair showed a lower risk of death/stroke at 3 months (HR=0.28, CI (0.13 to 0.62), p=0.002) in 751 patients receiving post-discharge vitamin K antagonists (VKAs) compared with 1437 patients without post-discharge VKA. Risk of significant bleeding complications was comparable in the same time period.

How might this impact on clinical practice?

  • This study may ensure better adherence of guidelines and should stimulate the conduction of a randomised trial of anticoagulation following mitral valve repair.



  • Contributors The manuscript as well as the conflict of interest policy has been reviewed and approved by all of the authors for submission. NV has generated most of the data and the first manuscript draft. GHG, LK, CM and CT-P has all participated in generating data and reviewing the manuscript, MLH reviewed and discussed the manuscript.

  • Competing interests None declared.

  • Ethics approval Danish Protection Agency with reference number GEH-2014-013; I-Suite number:02731.

  • Provenance and peer review Not commissioned; externally peer reviewed.

  • Data sharing statement All data are available for all statistical researchers in Denmark. On request from reviewers, we will supply with further possible data related to the study population.

Linked Articles