Self-monitoring and self-management of oral anticoagulation (Review)

Main results We identified 18 randomized trials (4723 participants). Pooled estimates showed significant reductions in both thromboembolic events (RR 0.50, 95% CI 0.36 to 0.69) and all-cause mortality (RR 0.64, 95% CI 0.46 to 0.89). This reduction in mortality remained significant after the removal of low-quality studies (RR 0.65, 95% CI 0.46 to 0.90). Trials of self-management alone showed significant reductions in thromboembolic events (RR 0.47, 95% CI 0.31 to 0.70) and all-cause mortality (RR 0.55, 95% CI 0.36 to 0.84); selfmonitoring did not (thrombotic events RR 0.57, 95% CI 0.32 to 1.00; mortality RR 0.84, 95% CI 0.50 to 1.41). Self-monitoring significantly reduced major haemorrhages (RR 0.56, 95% CI 0.35 to 0.91) whilst self-management did not (RR 1.12, 95% CI 0.78 to 1.61). Twelve trials reported improvements in the percentage of mean INR measurements in the therapeutic range. No heterogeneity was identified in any of these comparisons. Authors’ conclusions Compared to standard monitoring, patients who self-monitor or self-manage can improve the quality of their oral anticoagulation therapy. The number of thromboembolic events and mortality were decreased without increases in harms. However, self-monitoring or self-management were not feasible for up to half of the patients requiring anticoagulant therapy. Reasons included patient refusal, exclusion by their general practitioner, and inability to complete training. P L A I N L A N G U A G E S U M M A R Y Self-monitoring and self-management of oral anticoagulation therapy Near patient or point-of-care testing devices have made it possible for people on long-term oral anticoagulation to monitor their blood clotting time measured as the international normalized ration (INR) in the home setting. Patients who self-test can either adjust their medication dose according to a pre-determined dose-INR schedule (self-management) or they can call a clinic to be told the appropriate dose adjustment (self-monitoring). Several published studies suggest these methods of monitoring anticoagulation therapy may be equal to or better than standard monitoring by a physician. In total, we found 18 randomized trials that compared self-monitoring and self-management with standard monitoring. The combined results of these trials showed a halving of thromboembolic events and all-cause mortality with self-monitoring and self-management and no reduction in the number of major bleeds. Self-management had similar reductions in thromboembolic events and mortality to the overall benefit, with no effect on major bleeds. Self-monitoring halved the number of major haemorrhages that occurred but did not significantly reduce the rates of thrombotic events or all-cause mortality. In conclusion, self-monitoring or self-management can improve the quality of oral anticoagulant therapy, leading to fewer thromboembolic events and lower mortality, without a reduction in the number of major bleeds. Self-monitoring and self-management are not feasible for all patients, which requires the identification and education of suitable patients. B A C K G R O U N D Oral anticoagulation therapy with vitamin K antagonists has been shown to reduce thromboembolic events (Connolly 1991; Corporative 1990; SPAF 1996; EAFT 1993; Ezekowitz 1992; Go 2003) in multiple clinical contexts. These include atrial fibrillation, treatment of deep-vein thrombosis, prosthetic heart valves, and acute myocardial infarction. Optimal anticoagulation with warfarin or other vitamin k antagonists like acenocumarole or phenprocoumon could potentially prevent more than half of the strokes related to atrial fibrillation and heart valve replacements with a relatively low risk of major bleeding complications (Buckingham 2002); however, much of this potential is still not obtained because of under and suboptimal use (Stafford 1998). The number of patients receiving oral anticoagulant drugs has been constantly increasing during the last decade. Reasons include improvements in clinical outcomes, increasing common disease indications for their use (Manotti 2001), and improvements in anticoagulant safety (Ansell 2001). In 1994, 250,000 patients in the United Kingdom were receiving anticoagulant therapy 2 Self-monitoring and self-management of oral anticoagulation (Review) Copyright © 2010 The Cochrane Collaboration. Published by John Wiley & Sons, Ltd. (Baglin 1994); 10 years later this number had increased to around 950,000 patients (Fitzmaurice 2005). Vitamin k antagonist (warfarin, acenocumarole, or phenprocoumon) treatment usually requires regular monitoring of prothrombin time (PT) with dose-adjustment by a specialized hospital service, primary care physician, registered nurse, nurse practitioner, or pharmacist (Hirsh 1998). Numerous obstacles to the use of warfarin exist; including practical, patient, physician, and healthcare system-related barriers. Due to the complex pharmacokinetics of warfarin, continuous monitoring and dose adjustments are required. Different values and preferences amongst physicians and patients about the relative importance of bleeding and thromboembolic events, non-adherence to drug treatment, non-adherence to clinical guidelines, drug interactions, and increased costs of monitoring and therapy all have significant roles to play in the management of anticoagulation therapy (Heneghan 2008). Vitamin k antagonists belong to the drug class known as coumarins. They produce their anticoagulant effect by interfering with the metabolism of vitamin k. There are various different types of coumarins but warfarin is the most prescribed drug. Warfarin has a high bioavailability (Breckenridge 1978) and is rapidly absorbed from the gastrointestinal tract so that maximal blood concentrations are reached 90 minutes after oral administration. Warfarin has a half-life of 36 to 42 hours; in the blood it is bound to plasma proteins (mainly albumin). It accumulates in the liver where the two isomers are metabolically transformed by different pathways (Ansell 2004). Another vitamin K antagonist is acenocumarole, which has a similar action to warfarin but differs in some pharmacological properties (for example it has a shorter half life and a lower risk of haemorrhage). The maximum activity of both drugs is reached within one or two days of treatment and the anticoagulant effect is maintained for approximately two days after stopping treatment with acenocumarole and between two and five days with warfarin. Phenprocoumon is another vitamin k antagonist that has traditionally been the oral anticoagulant of choice in Europe. It has similar actions to other vitamin k antagonists but has a half-life of 144 hours. As a result of their pharmacokinetic properties, these agents interact with many other drugs and their blood levels are affected by vitamin k intake in the diet, changes in metabolism, and concomitant illnesses, which makes the levels difficult to control (Greenblatt 2005). The pharmacodynamics of warfarin are subject to genetic and environmental variability (Hirsh 2001) such that there is considerable variation in the action of these drugs both between different individuals (inter-individually) and within the same individual (intra-individually). A ’therapeutic target range’ has been established to deal with this variability and is expressed as the international normalized ratio (INR). This INR was established as a standard way of reporting the prothrombin time (PT). Furthermore, using the INR formula (INR = patient PT/mean normal PT) the ratio between patient PT and normal PT is calculated to the power of the ISI (International Sensitivity Index), which is the conversion factor for the used thromboplastin against the WHO standard. The ‘therapeutic range’ for anticoagulants is narrow. INR values over 4.5 increase the risk of major bleeding and an INR less than 2 increases the risk of thromboembolism (Cannegieter 1995; Hylek 1996; Kearon 2003). The inter and intra-individual variability and the narrow target range requires frequent testing and appropriate adjustment of the drug dose. In addition, time within the therapeutic INR target range is highly dependent on the frequency of testing (Horstkotte 1998). Different values and preferences amongst patients and physicians have also been described with the former willing to accept a much higher risk of bleeding for an associated reduction in risk of stroke (Devereaux 2001). An economic model analysed the cost of suboptimal oral anticoagulation and showed the following. If 50% of those not receiving warfarin prophylaxis had optimal anticoagulation, 19,380 emboli would be prevented and 1.1 billion US dollars could be saved. If 50% of those currently receiving warfarin as part of routine medical care had optimal anticoagulation, 9852 emboli would be prevented and 1.3 billion US dollars could be saved (Caro 2004). Current models of oral anticoagulation management within the UK include the traditional hospital outpatient model and various forms of community-based models, all requiring patient attendance at a clinic (Fitzmaurice 2002). In other countries, such as Canada, a primary care physician monitors the INR and adjusts the warfarin dose (Sunderji 2004). The introduction of portable monitors (point-of-care devices) allows the patient to self-test at home with a drop of whole blood. Portable monitors for monitoring long-term oral anticoagulation were introduced in the 1990s. Portable monitors have proved to be reliable with regard to analytical accuracy, although INR measurements tend to be lower with the portable coagulometers compared to laboratory analysers (Christensen 2009; Poller 2006). Generally patients receive a structured educational programme given by the nurses or physicians responsible for their care. In addition, they receive training in self-testing, instructions to prevent bleeding and thromboembolic complications, and are made aware of the effects of diet and medicatio


Authors' conclusions
Participants who self-monitor or self-manage can improve the quality of their oral anticoagulation therapy. Thromboembolic events were reduced, for both those self-monitoring or self-managing oral anticoagulation therapy. A reduction in all-cause mortality was observed in trials of self-management but not in self-monitoring, with no effects on major haemorrhage.

Self-monitoring and self-management of oral anticoagulation therapy Background
Point-of-care testing devices have made it possible for people on long-term oral anticoagulation to monitor their blood clotting time, measured as the international normalized ratio (INR). Patients who self-test can either adjust their medication dose according to a predetermined dose-INR schedule (self-management) or they can call a clinic to be told the appropriate dose adjustment (self-monitoring). Several published studies and systematic reviews have suggested these methods of monitoring anticoagulation therapy may be equal to or better than standard monitoring by a physician.

Study characteristics
This is an update of the original review published in 2010. We performed a new search and found 10 new studies (with 4227 participants) to add to the original review, which changed some of the findings.

Main results
In total, we found 28 randomised trials including 8950 participants that compared self-monitoring and self-management with standard monitoring. The quality of the evidence was generally low to moderate. The combined results of the 28 trials showed a halving of thromboembolic events with self-monitoring and self-management and no reduction in the number of major bleeds. Self-management had similar reductions in thromboembolic events and mortality to the overall benefit, with no effect on major bleeds. Self-monitoring halved the number of major haemorrhages that occurred but did not significantly reduce the rates of thrombotic events or all-cause mortality.

Conclusion
In conclusion, self-monitoring or self-management can improve the quality of oral anticoagulant therapy, leading to fewer thromboembolic events and lower mortality, without a reduction in the number of major bleeds. Self-monitoring and self-management are not feasible for all patients, which requires the identification and education of suitable patients.

B A C K G R O U N D Terminology
• Point-of-care testing (POC): diagnostic testing performed in a clinic, home, or other site of patient care (rather than in standard reference laboratory) • Point-of-care device: portable monitor used by a healthcare provider (physician, nurse, or other) or patient to determine a clinical measure • Self-monitoring: the trained participant uses point-of-care testing to perform the international normalized ratio (INR) test and inform his or her healthcare provider of the result. The physician or another healthcare provider adjusts the anticoagulant dose using the results obtained by the participant • Self-management: trained participant uses point-of-care testing to perform the INR test, interpret the result, and adjust the dosage of anticoagulant accordingly (adapted from Brown 2007)

Description of the condition
Oral anticoagulation therapy has been shown to reduce thromboembolic events in atrial fibrillation, treatment of deep-vein thrombosis, prosthetic heart valves, and acute myocardial infarction (Connolly 1991;Go 2003;SPAF 1996). Optimal anticoagulation with warfarin or other vitamin K antagonists like acenocoumarol or phenprocoumon could potentially prevent more than half of the strokes related to atrial fibrillation and heart valve replacements with a relatively low risk of major bleeding complications (Buckingham 2002;Hart 2007); however, much of this potential is still not obtained because of under and suboptimal use (Stafford 1998). The number of patients receiving oral anticoagulant drugs has been increasing. Reasons include improvements in clinical outcomes, increasing common disease indications for their use (Manotti 2001), and improvements in anticoagulant safety (Ansell 2001). In 1994, 250,000 patients in the UK were receiving anticoagulant therapy (Baglin 1994); 10 years later this had increased to around 950,000 patients (Fitzmaurice 2005). Vitamin K antagonist (warfarin, acenocoumarol, or phenprocoumon) treatment usually requires regular monitoring of prothrombin time (PT) with dose-adjustment by a specialised hospital service, primary care physician, registered nurse, nurse practitioner, or pharmacist (Hirsh 1998). Numerous obstacles to the use of warfarin exist; including practical, patient, physician, and healthcare system-related barriers. Due to the complex pharmacokinetics of warfarin, continuous monitoring and dose adjustments are required. Different values and preferences amongst physicians and patients about the relative importance of bleeding and thromboembolic events, non-adherence to drug treatment, drug interactions, and increased costs of monitoring have significant roles to play in the management of anticoagulation therapy (Heneghan 2008).

Description of the intervention
Vitamin K antagonists belong to the drug class known as coumarins and produce their anticoagulant effect by interfering with the metabolism of vitamin K. There are various different types of coumarins but warfarin is the most prescribed. Warfarin has a high bioavailability (Breckenridge 1978), and is rapidly absorbed from the gastrointestinal tract, with maximal blood concentrations reached 90 minutes after oral administration. Warfarin has a half-life of 36 to 42 hours; in the blood it is bound to plasma proteins (mainly albumin). It accumulates in the liver where the two isomers are metabolically transformed by different pathways (Ansell 2004). An anticoagulation effect generally occurs within 24 hours of treatment imitation, and peak effect for warfarin takes two to five days. Another vitamin K antagonist is acenocoumarol, which has a similar action to warfarin but differs in some pharmacological properties (for example, it has a shorter half-life Barcellona 1998). Phenprocoumon is another vitamin K antagonist that has traditionally been the oral anticoagulant of choice in Europe. It has similar actions to other vitamin K antagonists but has a half-life of 144 hours. As a result of their pharmacokinetic properties, these agents interact with many other drugs and their blood levels are affected by vitamin K intake in the diet, changes in metabolism, and concomitant illnesses, which makes the levels difficult to control (Greenblatt 2005). The pharmacodynamics of warfarin are subject to genetic and environmental variability (Hirsh 2001), such that there is considerable variation in the action of these drugs both between different individuals (inter-individually) and within the same individual (intra-individually). A 'therapeutic target range' has been established to deal with this variability and is expressed as the international normalized ratio (INR). This INR was established as a standard way of reporting the prothrombin time (PT). Furthermore, using the INR formula (INR = patient PT/mean normal PT) the ratio between patient PT and normal PT is calculated to the power of the ISI (International Sensitivity Index), which is the conversion factor for the used thromboplastin against the World Health Organization (WHO) standard. The 'therapeutic range' for anticoagulants is narrow. INR values over 4.5 increase the risk of major bleeding and an INR less than 2 increases the risk of thromboembolism (Cannegieter 1995;Hylek 1996;Kearon 2003). The inter-and intra-individual variability and the narrow target range generally requires frequent testing and appropriate adjustment of the drug dose. In addition, time within the therapeutic INR target range is highly dependent on the frequency of testing (Horstkotte 1998). Different values and preferences amongst patients and physicians have also been described with the former willing to accept a much higher risk of bleeding for an associated reduction in risk of stroke (Devereaux 2001). An economic model analysed the cost of suboptimal oral anticoagulation and showed the following. If 50% of those not receiving warfarin prophylaxis had optimal anticoagulation, 19,380 emboli would be prevented and 1.1 billion US dollars could be saved. If 50% of those currently receiving warfarin as part of routine medical care had optimal anticoagulation, 9852 emboli would be prevented and 1.3 billion US dollars could be saved (Caro 2004).

How the intervention might work
Current models of oral anticoagulation management within the UK include the traditional hospital outpatient model and various forms of community-based models, all requiring patient attendance at a clinic (Fitzmaurice 2002). In other countries, such as Canada, a primary care physician monitors the INR and adjusts the warfarin dose (Sunderji 2004). The introduction of point-of-care devices allows the patient to self-test at home with a drop of whole blood. Portable monitors for monitoring long-term oral anticoagulation were introduced in the 1990s. Devices have proved to be reliable with regard to analytical accuracy, although INR measurements tend to be lower with the portable coagulometers compared to laboratory analysers (Christensen 2009;Poller 2006), and have proved to be reliable devices for monitoring INR when checked regularly (Barcellona 2009). Generally, patients receive a structured educational programme given by the nurses or physicians responsible for their care. In addition, they receive training in self-testing, instructions to prevent bleeding and thromboembolic complications, and are made aware of the effects of diet and medications. Patients who self-test can either adjust their therapy according to a pre-determined dose-INR schedule (self-management) or they can call a clinic to be told the appropriate dose adjustment (self-monitoring).

Why it is important to do this review
In some countries, such as Germany, self-monitoring and selfmanagement with portable monitors are established therapeutic methods. There are several available point-of-care devices and the most well known is the CoaguChek® monitor. Other available monitors are the ProTime® Microcoagulation System, INRatio® Monitor, Hemochron Junior Signature, and the TAS near-patient test system. Potential advantages of self-monitoring and self-management include improved convenience for patients, better treatment adherence, more frequent monitoring, and fewer thromboembolic and haemorrhagic complications (Taborski 1999). Near-patient testing devices have made self-testing of anticoagulation therapy with vitamin K antagonists possible. Guidelines generally do not endorse self-monitoring or self-manage-ment (Fitzmaurice 2001), despite several authors of trials suggesting this approach may be equal to or better than standard monitoring (Anderson 1993;Cromheecke 2000;Sawicki 1999). A recent study suggested that self-monitoring and self-management are cost-effective strategies for those receiving long-term oral anticoagulation (Regier 2006). In addition, newer oral anticoagulants that do not require monitoring have not been established in heart valve patients (Eikelboom 2013) and are not suitable for all because of the numerous exclusion and individuals who cannot tolerate these drugs (DiNicolantonio 2012). To establish the current strength of the available evidence, we updated our systematic review of the impact of patient self-monitoring or self-management on treatment with oral anticoagulation therapy.

O B J E C T I V E S
To evaluate the effects on thrombotic events, major haemorrhages, and all-cause mortality of self-monitoring or self-management of oral anticoagulation compared with standard monitoring.

Types of studies
Randomised controlled trials (RCTs) assessing the therapeutic effectiveness and safety of self-monitoring or self-management of oral anticoagulation therapy.

Types of participants
All patients, adults and children, on long-term anticoagulant therapy (treatment duration longer than two months), irrespective of the indication for treatment (for example, valve replacement, venous thromboembolism, atrial fibrillation).

Types of interventions
Self-monitoring or self-management of oral anticoagulation compared with: • control of and dosage by personal physician; • anticoagulation managed services (hospital anticoagulation service); • anticoagulation clinics (management conducted by registered nurses, nurse practitioners, or pharmacists using dosage-adjustment protocols).

Types of outcome measures Primary outcomes
• Thromboembolic events • All-cause mortality • Major haemorrhage (e.g. haemorrhage requiring hospital admission or transfusion) • Time in range, and proportion of measurements within the therapeutic range for each particular condition

Secondary outcomes
• Minor haemorrhage (e.g. bleeding after minor trauma, nose bleed) • Frequency of testing • Feasibility of testing: participant factors (e.g. physical limitations), and non-participant factors (e.g. inability to attend training) • Quality of life and general satisfaction with treatment

Search methods for identification of studies Electronic searches
The searches for the initial review were run in November 2007 (Appendix 1). We re-ran the searches on 27 November 2013 (Appendix 2). We updated the searches on 1 July 2015 (Appendix 2) with exception of CINAHL which was last searched on 27 November 2013 (an updated search of CINAHL was not mandatory): • Cochrane Central Register of Controlled Trials (CENTRAL) 2015, Issue 6, the Cochrane Library, • MEDLINE (Ovid, 1946 to June week 4 2015), • Embase (Ovid, 1980(Ovid, to 2015, and • CINAHL (EBSCO, 1982to November 2013.
We limited our searches to randomised controlled trials by using a maximally sensitive strategy (Dickersin 1994;Lefebvre 1996and Lefebvre 2011in 2015.

Searching other resources
We also searched for ongoing trials on the UK National Re-

Data extraction
Two review authors (EAS and IJO) screened studies for inclusion and retrieved all potentially relevant studies. Three review authors (JM, PA, CH) independently extracted data on study population, intervention, pre-specified outcomes, methodology, and quality from eligible trials. The authors were not blinded to any aspect of the studies (for example, journal type, authors' names, institution). We resolved disagreements by consensus. If needed, we sought additional information from trial authors. We used Cohen's kappa to assess agreement between the two authors on the selection of articles for inclusion. We extracted information on disease characteristics and training provided to the different groups. In the self-management group, we extracted information on the actions participants subsequently undertook. We extracted the characteristics of the population studied, including the number of, and reasons for, participants not entering the trial (for example, refusal or exclusion). Additionally, we sought information on the reasons for discontinuation by participants allocated to the intervention.
In the case of cross-over studies, the outcomes of interest are potentially confounded by the cross-over and we only used data from the first part of the trial (before cross-over).

Quality assessment
Three review authors (EAS, IJO, CH) independently extracted methodological information for the assessment of risk of bias. They used the following five components: method of randomisation, concealment of allocation, intention-to-treat, number of, and reasons for, participant losses to follow-up, and blinding. We performed a sensitivity analyses for study quality by including only those studies with clear methods of randomisation and concealment of allocation (high-quality studies). We also used the Grades of Recommendation, Assessment, Development and Evaluation (GRADE) (GRADE 2008) to assess the quality of the included studies.

Quantitative data synthesis
For the analysis we used Review Manager (RevMan) Version 5.3.
For the statistical analysis we calculated risk ratios (RRs) and 95% confidence intervals (CIs) as summary statistics. We used a fixedeffect model with the Mantzel-Haenzel method to calculate the pooled odds ratio; and Peto's method to verify the results in uncommon outcomes. We examined heterogeneity amongst studies with the Chi 2 and I 2 statistics (Higgins 2003). Where significant heterogeneity existed, we used the random-effects model (DerSimonian 1986). We examined publication bias by constructing a funnel plot of precision (SE of the log RR) against RR for the endpoints of major haemorrhage and thromboembolic episodes. We performed a sensitivity analysis by excluding studies with high risk of bias and prespecified subgroup analyses according to clinical indication (mechanical valve replacement or atrial fibrillation), and self-monitoring or self-management therapy. We performed a post-hoc subgroup analysis according to who provided the control group care (specialist anticoagulation clinic, family physician). Meta-regression in STATA tested any subgroup interaction on the outcomes. The ratio of the average test frequency per individual patient/year between intervention and control groups was calculated and linear regression was used to assess the association with study duration. Pooling of the mean percentage of tests in the therapeutic range was not possible; results were summarised using means and ranges. We tested subgroup interactions using meta-regression (Intercooled STATA 9.1 for Windows).
To provide further insight into the adequacy of the total sample size across all trials, we calculated a posteriori the optimal information needed for our meta-analysis (Pogue 1997). To determine this optimal information size, we assumed a 2% risk of thromboembolism (median control event rate from trials in the review) and a 50% RR reduction with a power of 95% and a two-sided alpha = 0.01.

Results of the search
The search for the initial review retrieved 463 citations, which included 18 relevant trials. The updated searches in November 2013 and July 2015 identified 5136 new citations and identified an additional 10 trials (with 4227 participants) for inclusion. In total, we identified 5894 citations through database searching as well as one additional unpublished citation (Kaatz Unpublished). Of these, we excluded 758 duplicate records, leaving 5136 potentially relevant studies. A further 5067 citations were removed after being deemed irrelevant to our research question. We independently assessed 69 full-text articles for eligibility. Of these, 22 articles were excluded and 20 articles were secondary publications of primary studies already included in the review ( Figure 1).

Included studies
A total of 27 included publications provided data on 28 trials including 8950 participants (one publication, Gadisseur 2003 Self mge; Gadisseur 2003 Self monit, contained data on two trials that compared self-monitoring or self-management of oral anticoagulation with standard monitoring). Trials were published between 1989 and 2013 and were largely undertaken in Europe (five each in UK and Germany; three each in the Netherlands and Denmark; one in each of Ireland, France, Spain and Austria); seven were undertaken in United States and Canada; and one was conducted in Australia. In total, 4723 participants on long-term anticoagulation were included in our analysis. In the three remaining trials participants in the control group could use either primary care or specialist clinics (Christensen 2006;Dignan 2013;Siebenhofer 2007). Duration of studies varied from two months (White 1989) to more than 24 months (Körtke 2001; Matchar 2010); the mean duration was 12 months. Analysis of publication bias using funnel plots of major haemorrhage and thromboembolic events showed no evidence of asymmetry ( Figure 2, Figure 3).

Risk of bias in included studies
The reported risk of bias was generally low to moderate. The nature of the intervention made blinding of the allocation of intervention to the participants impossible, although blinding of study staff and outcome assessment was possible. We contacted nine authors of the 27 included trials for additional details of randomisation process, concealment of allocation, and blinding. The additional information provided generally raised our ratings of the quality of the trial, indicating that authors had met methodological criteria.
We also obtained valuable validity information from the ACP Journal Club structured reviews on two occasions. ACP reviews contact study authors when needed and are a valuable source of additional information for validity issues.
After the addition of extra information supplied by authors, 11 trials were judged to be of high risk of bias (Azarnoush 2009;Christensen 2011;Gardiner 2005;Khan 2004;Matchar 2010;Rasmussen 2012;Sidhu 2001;Soliman Hamad 2009;Thompson 2013;Verret 2012;White 1989) and were removed in the sensitivity analysis. These 11 trials did not perform intention-to-treat analyses and randomisation and/or allocation concealment was unclear. Overall, the available evidence was judged to be moderate according to the GRADE Working Group grades of evidence (Summary of findings for the main comparison). This was due to flaws in study design; most commonly there was an absence of information about the allocation concealment procedure or blinding and the number of events was less than 300 for the primary outcomes (Characteristics of included studies). The overall risk of bias graph and summary table are shown in Figure 4 and Figure  5.

Randomisation and allocation concealment
Twenty-one trials reported adequate information about the randomisation process (Christensen 2006 Figure 4 and Figure 5).  Figure 4 and Figure 5).

Follow-up
Of those assigned to the intervention, 25% (range 0% to 57%) stopped self-monitoring or self-management by the end of the trial. Nine trials used an intention-to-treat analysis (

Effects of interventions
See: Summary of findings for the main comparison Selfmonitoring or self-management of oral anticoagulation vs. standard care; Summary of findings 2 Self-monitoring of oral anticoagulation vs. standard care; Summary of findings 3 selfmanagement of oral anticoagulation vs. standard care

Primary endpoints Thromboembolic events
Twenty-six trials reported thromboembolic outcomes; however, eight trials showed no events in either the intervention or control arm (Christensen 2006 showed both to be significant (subgroup interaction P = 0.33).

All-cause mortality
Twenty-two trials reported information on mortality; 11 trials did not report any deaths in the intervention and control groups and are therefore excluded form the pooled analysis (

Major haemorrhage
Twenty-four trials reported major haemorrhage outcomes, four of which did not report any events (Christensen 2006;Cromheecke 2000;Gardiner 2005;White 1989 Of the participants assigned to the intervention 24.9% (range 0% to 57.3%) were unable to complete self-monitoring or self-management. The main reasons for the dropouts were: problems with the device, physical limitations preventing self-testing and problems with attending the training assessments or failing the assessment. Due to inadequate data, we were unable to rate the quality of the evidence.

Other outcomes
Thirteen studies evaluated quality of life outcomes. These included ease of use (Gardiner 2005) (Matchar 2010), reported that there were no adverse events resulting from physical interaction with the testing device. Due to inadequate data, we were unable to rate the quality of the evidence for quality of life and satisfaction.

Optimal information size
The calculated optimal information size needed for a reliable and conclusive treatment effect was 5150 in each arm. This assumed a 2% thromboembolic event rate in the control group, a 50% RR reduction, a power of 95%, and a two-sided alpha = 0.01. The current meta-analysis has approximately 4000 in each arm, which would give a 78% power using the same assumptions.
One of the main trials included in the meta-analysis showed a clear absence of correlation between the benefits observed and the degree of control (Menendez-Jandula 2005). We therefore questioned the influence of this study by performing a post hoc sensitivity analysis that removed the trial; the beneficial effects observed for all the major outcomes remained similar.

A D D I T I O N A L S U M M A R Y O F F I N D I N G S [Explanation]
Self -m onitoring of oral anticoagulation vs. standard care Patient or population: Patients on long-term anticoagulant therapy (treatm ent duration longer than two m onths) irrespective of the indication f or treatm ent Settings: Prim ary care, specialist clinics (Europe, Am erica, Canada) Intervention: * The basis f or the assumed risk (e.g. the m edian control group risk across studies) is provided in f ootnotes. The corresponding risk (and its 95% conf idence interval) is based on the assum ed risk in the com parison group and the relative effect of the intervention (and its 95% CI). CI: Conf idence interval; RR: Risk Ratio GRADE Working Group grades of evidence High quality: We are very conf ident that the true ef f ect lies close to that of the estim ate of the ef f ect. M oderate quality: We are m oderately conf ident in the ef f ect estim ate: The true ef f ect is likely to be close to the estim ate of the ef f ect, but there is a possibility that it is substantially dif f erent. Low quality: Our conf idence in the ef f ect estim ate is lim ited: The true ef f ect m ay be substantially dif f erent f rom the estim ate of the ef f ect. Very low quality: We have very little conf idence in the ef f ect estim ate: The true ef f ect is likely to be substantially dif f erent f rom the estim ate of ef f ect * The basis f or the assumed risk (e.g. the m edian control group risk across studies) is provided in f ootnotes. The corresponding risk (and its 95% conf idence interval) is based on the assum ed risk in the com parison group and the relative effect of the intervention (and its 95% CI). CI: Conf idence interval; RR: Risk Ratio GRADE Working Group grades of evidence High quality: We are very conf ident that the true ef f ect lies close to that of the estim ate of the ef f ect. M oderate quality: We are m oderately conf ident in the ef f ect estim ate: The true ef f ect is likely to be close to the estim ate of the ef f ect, but there is a possibility that it is substantially dif f erent. Low quality: Our conf idence in the ef f ect estim ate is lim ited: The true ef f ect m ay be substantially dif f erent f rom the estim ate of the ef f ect. Very low quality: We have very little conf idence in the ef f ect estim ate: The true ef f ect is likely to be substantially dif f erent f rom the estim ate of ef f ect 1 Downgraded f rom high to low because of serious risk of bias and im precision of ef f ect estim ate. 2 Downgraded f rom high to m oderate because of serious risk of bias. 3 Downgraded f rom high to low because of serious risk of bias and substantial heterogeneity.

D I S C U S S I O N
To our knowledge the present review is the most comprehensive review to date. We identified 28 randomised controlled trials (RCTs) trials (8950 participants). Self-monitoring or self-management of oral anticoagulation leads to a significant 50% reduction in thromboembolism but no reduction in all-cause mortality. However, trials of self-management led to a significant reduction in all-cause mortality. Self-management did not reduce major haemorrhages nor did self-monitoring.

Quality of the evidence
The GRADE approach was employed to interpret result findings and the GRADE profiler (GRADEPRO) allowed us to import data from Review Manager to create 'Summary of findings' tables. The overall quality of the evidence for the effect of self-monitoring or self-management of oral anticoagulation on major haemorrhage was moderate; the quality was downgraded because of serious risk of bias across the studies. The quality of the evidence for trials of self-management was downgraded from high to low because of serious risk of bias and imprecision of effect estimate (i.e. large confidence intervals). The quality of the evidence for trials of selfmonitoring was downgraded from high to low because of serious risk of bias across the studies and strong suspicion of publication bias.
The overall quality of the evidence for thromboembolic events was moderate; downgraded because of serious risk of bias across the included studies. The quality of the evidence for trials of either self-monitoring or self-management were downgraded from high to moderate because of serious risk of bias. The overall quality of the evidence for mortality was moderate; downgraded because of serious risk of bias across the included studies. The quality of the evidence for trials of either self-monitoring or self-management were downgraded from high to moderate because of serious risk of bias. The overall quality of the evidence for minor haemorrhage of the evidence was low; downgraded because of serious risk of bias and substantial heterogeneity in the meta-analysis result. The quality of the evidence for trials of self-management was downgraded from high to low because of serious risk of bias and substantial heterogeneity. The quality of the evidence for trials of self-monitoring was downgraded from high to moderate because of serious risk of bias across the included studies. Due to inadequate data, we were unable to rate the quality of the evidence for the following outcomes: (i) Frequency of testing; (ii) Feasibility of testing; and (iii) Quality of life and satisfaction.

Comparison with other reviews
This systematic review provides information additional to a substantial body of evidence from previously published reviews  2005), reported no difference between participants undertaking self-management and those receiving usual care in the time spent in the therapeutic range and in the incidence of adverse effects. Bazian's review (which was less comprehensive) also did not show a difference between self-management and routine care (Bazian 2005). In the Bloomfield 2011 review, patients assigned to selfmonitoring or self-management had significantly lower total mortality, lower risk for major thromboembolism, and no increased risk in major haemorrhage. An individual patient data meta-analysis (Heneghan 2012), which included 11 trials with data for 6417 participants and 12,800 person-years of follow-up, reported a significant reduction in thromboembolic events in the self-monitoring group (Hazard Ratio (HR) 0.51; 95% CI 0.31 to 0.85), but not for major haemorrhagic events (HR 0.88, 95% CI, 0.74 to 1.06) or mortality (HR 0.82, 95% CI 0.62 to 1.09). In this review patients, younger than 55 years showed marked reductions in thrombotic events (HR 0.33, 95% CI 0.17 to 0.66), as did patients with mechanical heart valve (HR 0.52, 95% CI, 0.35 to 0.77). The greater reduction in mortality with self-management compared with self-monitoring observed in this review might be related to the higher frequency of thromboembolic events seen in the latter group. Also, reduced mortality might be affected by increased patient empowerment, whereby self-management influences other aspects of a patient's lifestyle (i.e. adherence to treatments) as they take on more of a locus of control for their condition. A 2015 Health Technology Assesment (HTA) systematic review (Sharma 2015) on the clinical effectiveness and cost-effectiveness of point-of-care tests of people receiving long-term vitamin K antagonist therapy reported that self-monitoring significantly prevented thromboembolic events and reduced all-cause mortality in people with artificial heart valves, and similarly to this current review, reported greater reductions in thromboembolic events and all-cause in those self-managing. In addition, the review reported net UK health and social care costs, which over a 10-year period were equivalent to standard monitoring costs. Intrinsic limitations to self-monitoring and self-management in-clude the reluctance of individuals to participate in self-management and the extensive training required to do so. Self-monitoring is not feasible for up to half of the patients requiring anticoagulation. Factors influencing patient participation within trials included problems with the device; physical limitations; attending training sessions; or failing the assessment. An additional problem with adoption in clinical practice will be the relatively high cost of the test strips. The reliability of self-testing devices can affect test results; however, available devices give INR results which are comparable with those obtained in laboratory testing (Ansell 2005). Self-monitoring and self-management are also associated with a rate of testing that is higher than with usual care. In effect self-managed warfarin dosing is analogous to self-adjusted insulin dosing according to a pre-specified sliding scale (Ansell 1996). Such self-managed treatment has been practiced for years by people with diabetes (Ansell 1996), and the use of self-monitoring or self-management offers independence and freedom to travel for selected patients. Our review has some potential limitations. First, our search was comprehensive, making serious publication bias less likely, but it remains a concern. Therefore, the results may represent an overestimate of the true effect of treatment. Second, variability in the quality of care in the control groups can affect the rate of testing and hence the benefit and safety of standard anticoagulation monitoring. Specialist programmes may improve outcomes by the same mechanism as self-monitoring or self-management, that is improving the time in therapeutic range and lessening the frequency of adverse outcomes. However, our post hoc subgroup analysis did not verify this effect. A further modifying factor is education and training. The two trials in which patients consented to participate and received education alone had better results than did those patients allocated to routine care (Gadisseur 2003 Self mge; Gadisseur 2003 Self monit; Khan 2004). Third, for all the major outcomes of this review, limitations in the published reports led to an absence of information about the allocation concealment procedure or blinding. However, several authors were successfully contacted and the additional information that they provided generally raised the assessed quality of the trials. This finding is in agreement with recent empirical evidence suggesting that authors fail to report concealment of randomisation and blinding (Devereaux 2004). Finally, for all the major outcomes there was a low numbers of events. Overestimates are likely in trials with fewer than 500 events and large overestimates of the effects are more likely in trials with fewer than 200 events. (Bassler 2010) Self-monitoring or self-management are likely to prevent thromboembolism to a greater extent than standard monitoring. The mechanism of effect is probably through increasing the number of INR values in range and therefore the longer time that patients are in the therapeutic range. Despite the limitations outlined above the apparent beneficial effects are large, and even smaller true underlying effects would probably justify widespread use of self-monitoring and self-management of oral anticoagulation in suitable candidates.

Implications for practice
Self-monitoring or self-management by patients can improve the quality of oral anticoagulation therapy compared to standard monitoring. Patients spend more time within the therapeutic range resulting in decreases in thromboembolic events with no increase in harms.

Implications for research
For the results to be generalisable to the population at large, there is a need for population-based studies that collect data on adverse event rates, time in range, and factors that impinge on successful self-monitoring and self-management (Nagler 2014). The nature of this intervention lends itself to a registry to guarantee its safety and effectiveness in clinical practice. Future studies should set out to understand why people decide to use self-management (or not) and should incorporate consumer knowledge about self-management, triggers to seek care, self-efficacy or self-confidence to selfmanage, and perceived or actual support. Further studies should explore components of the intervention that affect the feasibility of self-monitoring and self-management and identify means to improve uptake and effectiveness. In addition, given the low rates of adverse events in trials of self-monitoring, studies comparing its use to newer oral anticoagulants, which do not require monitoring, are warranted.
Participants 206 adult patients who had undergone valve replacement with a mechanical prosthesis, with or without myocardial revascularisation

Self-monitoring vs standard monitoring
Randomised to standard monitoring of INR at a laboratory including at least one monthly assay at a medical analysis laboratory (n = 103), or self-testing using either the CoaguChek® system (n = 55) or the INRatio® system (n = 48). Self-testing was performed weekly, and in addition once monthly INR measurements were carried out at the laboratory on the same day as the self-measurement. Only the results of the monthly tests for each group were compared. Education relating to VKA therapy was provided, the same to all participants in all allocation groups. The target INR and target range were determined for each participant on the basis of the type of surgery, and according to their risk factors for thromboembolic disease (target ranges were between 2 and 3. Intention to treat analysis Low risk ITT analysis was performed.
Reporting of losses of follow-up Unclear risk < 20% losses to follow-up.

Blinding
Low risk Blinded data collectors.

Methods
Single centre, open-label, randomised controlled trial Participants 100 ambulatory patients aged > 18 years (mean age 63 years in intervention group, 69 years in control group) receiving oral anticoagulation therapy for > 8 months. The study was based in the Center of Self-managed Oral Anticoagulation (Denmark) Exclusion criteria included: previous self-management.

Interventions
Self-management vs usual care Randomisation to a) self-management (n = 50), in which participants were trained to self-monitor using a Coaguchek ® coagulometer to measure INR once a week and also to adjust their anticoagulant dosage accordingly b) usual care (n = 50), in which conventional management included at least monthly INR testing at a hospital or physician centre and dosage adjusted by the physician In both groups an additional INR analysis was performed monthly and the participant contacted if INR was < 1.5 or > 4.5 After 6 months intervention, the control group began training to self-manage their anticoagulation therapy Methods Single centre, randomised, controlled cross-over trial.

Participants
Participants were 50 consecutive outpatients who were receiving long-term anticoagulation (mean age 42 years). The study was based in the departments of cardiology and internal medicine of the Academic Medical Centre (Amsterdam, The Netherlands)

Self-management vs usual care
The intervention group used home self-testing using Coaguchek ® to self-monitor prothrombin time and self-dosing testing performed once a week The conventional management was done by the anticoagulation clinic INR testing was also performed in all participants at 1-2 week intervals by the central laboratory; these results were not made available to participants or managing physicians After three months patients crossed over the alternative management strategy

Participants
Participants were ambulatory adults aged > 18 years attending an anticoagulation clinic, receiving anticoagulation therapy for > 6 months, judged as capable of self-management, and with satisfactory INR control (n = 56) The study was based in six general practices that used the Birmingham model of anticoagulation management (West Midlands, UK)

Self-management vs usual care
Participants were randomised to: a) self-management (n = 30): self-testing using Coaguchek ® device and self-adjustment of dosing. Testing was performed every 2 weeks or after 1 week following dosage adjustment b) conventional management (n = 26) in practice clinics. Participants Participants (n = 320) were adults aged 18 to 75 (mean 57) years having received anticoagulation therapy for > 3 months, requiring long-term anticoagulation therapy, and attending one of two anticoagulation clinics (The Netherlands) Exclusion criteria included: antiphospholipid syndrome; life-threatening illness; < 1 year life expectancy; diminished understanding or physical limitations preventing participation

Self-monitoring or self-management vs usual care
Participants were randomised to one of four groups, using a 2-step partial Zelen design Group A (n = 52): self-testing using Coagucheck ® monitoring device. Group B (n = 47): self-testing using Coagucheck ® and self-dosing. Group C (n = 60): received education alone and routine care. Group D (n = 161): received only routine care. Group D did not provide informed consent for randomisation into the study and were unaware of study participation For the purposes of this review, the results of the trial are presented as Gadisseur 2003 self-management, which is group B versus group D, and separately as Gadisseur selfmonitoring, which is group A versus group D

Methods
Single centre, randomised controlled trial. Participants (n = 320) were adults aged 18 to 75 (mean 57) years having received anticoagulation therapy for > 3 months, requiring long-term anticoagulation therapy, and attending one of two anticoagulation clinics (The Netherlands) Exclusion criteria included: antiphospholipid syndrome; life threatening illness; < 1 year life expectancy; diminished understanding or physical limitations preventing participation

Participants
The study enrolled 320 participants. Mean age 57 years who were receiving long-term anticoagulation. The study was based in two Dutch anticoagulation clinics

Self-monitoring or self-management vs usual care
Participants were randomised to one of four groups, using a 2-step partial Zelen design Group A (n = 52): self-testing using Coagucheck ® monitoring device. Group B (n = 47): self-testing using Coagucheck ® and self-dosing. Group C (n = 60): received education alone and routine care. Group D (n = 161): received only routine care. Group D did not provide informed consent for randomisation into the study and were unaware of study participation For the purposes of this review, the results of the trial are presented as Gadisseur 2003 self-management, which is group B versus group D, and separately as Gadisseur selfmonitoring, which is group A versus group D

Random sequence generation (selection bias)
Low risk Computer-generated list.
Allocation concealment (selection bias) Low risk Concealment of allocation was not reported, contact with author led to information on appropriate method of concealment Intention to treat analysis High risk ITT analysis was not performed Reporting of losses of follow-up High risk Loss to follow-up was not reported Blinding Low risk Participants were not blinded to the intervention (except group D who were unaware of trial participation). Health care providers were blinded to the intervention

Gardiner 2005
Methods Single centre, randomised controlled trial.

Participants
Participants (n = 84) were adults aged > 18 (mean 58) years who had received anticoagulation therapy for > 8 months and had a record of good compliance The study was based in an anticoagulation clinic in University Hospital (London, UK)

Self-monitoring vs usual care
Participants were randomised to: a) self-testing (n = 44) using the Coagucheck ® monitoring device once per week b) control (n = 40), receiving usual care by visiting the hospital anticoagulation clinic for testing every 4 weeks The intervention group attended two training sessions.

Bias Authors' judgement Support for judgement
Random sequence generation (selection bias) Low risk Random sequence was generated using variable block sizes and stratification Allocation concealment (selection bias) Low risk Sealed opaque envelopes.
Intention to treat analysis Low risk ITT analysis was performed Reporting of losses of follow-up Low risk No participants were lost to follow-up.

Blinding
High risk Participants were not blinded to the intervention. It is unclear whether study or medical staff were blinded to the intervention

Khan 2004
Methods Single centre, randomised controlled trial.

Participants
Participants (n = 125) were adults aged > 65 (mean 73) years with atrial fibrillation receiving oral anticoagulation for > 12 months previously for atrial fibrillation Exclusion criteria included: inability to use the Coagucheck system due to general frailty, poor hearing or eyesight, impairment of hand function, dementia, residence in care home The study was based in a university based anticoagulation service (Newcastle, UK) Interventions

Self-monitoring vs usual care
Participants were randomised to: a) Group A (n = 44) used home weekly self-testing using the Coagucheck ® monitoring device. Weekly dosage adjustment was advised by telephone by the study co-ordinator b) Group B (n = 41) received education alone and clinical care c) Group C (n = 40) received usual care. Methods Single centre, randomised controlled trial.
Participants Participant (n = 600) were adults (mean age 63 years) receiving life long-term oral anticoagulation after mechanical heart valve replacement The study was based in Ruhr University (Germany).

Self-management vs usual care
Participants were randomised to: a) self-management (n = 305): self-testing using Coagucheck ® system, and self-adjustment of dosage. In addition, monthly INR measurements were reviewed by the anticoagulation clinic b) control (n = 295): outpatient cardiologic check up and coagulation controls every 6 months. It is unclear if these participants adjusted their anticoagulation dosage themselves

Matchar 2010
Methods Multi-centre, randomised, parallel-group, controlled trial Participants Participants (n = 2,992) were adults with atrial fibrillation, a mechanical heart valve, or both, requiring long-term warfarin therapy and competent in self-testing

Self-monitoring vs usual care
Randomisation was within strata of anticoagulation duration (< 3 or ≥ 3 months) and indication for warfarin (atrial fibrillation with or without mechanical heart valve) within each site Participants were randomised to a) self-testing (n = 1465): participants measured their INR once a week and recorded the results via an interactive voice-response reporting system with web-based local monitoring. If the participant reported a measurement outside the assigned INR range or reported having been hospitalised, the system directed the participant to contact study

Random sequence generation (selection bias)
Low risk Centralised telephone randomisation.
Allocation concealment (selection bias) Low risk The sequence of randomisation was concealed until the participant was assigned to a group Intention to treat analysis Low risk ITT analysis was used Reporting of losses of follow-up Unclear risk 11.9% of participants were lost to followup; reasons not reported Blinding Low risk Blinded outcome assessors.

Participants
Participants were individuals requiring oral anticoagulation therapy (n = 54)

Self-monitoring vs usual care
Participants were randomised to self-testing with computer aided decision making, two different algorithms (computer algorithm group A n = 19, computer algorithm group B n = 18), or to usual care (monitoring and treatment by physicians) (n = 17) Outcomes Time to therapeutic range, time in therapeutic range, INR.

Trial identification
Study duration Mean 28 week follow-up

Rasmussen 2012 (Continued)
Oral anticoagulant used Warfarin Notes Authors reported that there were insufficient data to provide valid measurements of thromboembolic events and severe bleeding. in this study it is difficult to estimate the individual effects of self-testing and of computer dosage calculation

Bias Authors' judgement Support for judgement
Random sequence generation (selection bias) Low risk Randomisation was performed by computer.
Allocation concealment (selection bias) Unclear risk Allocation concealment was not reported.
Intention to treat analysis Unclear risk ITT not reported.

Reporting of losses of follow-up Unclear risk
Losses to follow-up were not reported.

Blinding
Low risk Study investigators were blinded to computer algorithm group A vs group B allocation; statisticians were blinded to allocation

Ryan 2009
Methods Single centre, randomised controlled cross-over study Participants Participants were individuals receiving ongoing warfarin therapy for > 2 months and who had internet access, and were able to use a home INR meter (n = 162) Interventions Self-monitoring vs usual care Participants were randomised to supervised self-testing or to usual care (conventional clinic management) for 6 months; subsequently the allocation was reversed for a further 6 months In the self-testing group, participants initially self-tested INR twice weekly. Once the INR was therapeutic for 2-3 consecutive readings, the interval between tests was increased to a maximum of every 2 weeks. Participants accessed a web-based system to enter signs and symptoms and INR and receive instant automated guidance on dose and testing; if INR deviation was serious the participant was asked to take a bolus dose of warfarin (< 1.5) or hold their warfarin (> 5.0) and/or to log in later the same day for additional instructions. If a participant reported a symptom suggestive of a bleed or an embolus they were told to seek immediate medical advice. The research pharmacist accessed the caregiver interface of the program at least once daily to review participant problems. Any participant who failed to test their INR or log in to the program as scheduled was contacted by telephone the same day. All new dosage recommendations were reviewed and adjusted if necessary. All extreme INRs (<1.5 or

Bias Authors' judgement Support for judgement
Random sequence generation (selection bias) Low risk Computer-generated randomisation sequence Allocation concealment (selection bias) Low risk Allocation implemented via sealed envelopes Intention to treat analysis High risk ITT was not performed.
Reporting of losses of follow-up Low risk 30 participants (19%) withdrew; reasons reported Blinding High risk Participants and study staff were not blind to the intervention allocation. It is not reported whether data analysts were blind to allocation

Participants
Participants (n = 179) were individuals (mean age 55 years), receiving long term oral anticoagulation. The study was based in 5 referral centres (Germany) Interventions

Self-management vs usual care
Participants were randomised to a) self-management (n = 90): home self-testing and self-dosing using a Coagucheck ® Participants randomised to the self-management group received a structured educational program comprising three consecutive weekly teaching sessions of 60-90 minutes

Bias Authors' judgement Support for judgement
Random sequence generation (selection bias) Low risk Computer program.
Allocation concealment (selection bias) Unclear risk Possibly adequate or not used.
Intention to treat analysis Low risk ITT analysis was used Reporting of losses of follow-up Low risk 14/179 (7.8%) of participants were lost to follow-up; reasons not reported Blinding Low risk Participants were not blinded to the intervention; data collectors were blinded

Sidhu 2001
Methods Single centre, randomised controlled trial.

Participants
Participants (n = 100) adults aged < 85 years (mean age 61 years) who had undergone a heart valve operation and had been prescribed life-long anticoagulation

Self-management vs usual care
Participants were randomised to a) self-management (n = 51): home self-testing using the Coagucheck ® and self-dosing. INR testing performed once a week, participants were encouraged to perform more Reporting of losses of follow-up High risk 33.3% of participants were lost to followup in intervention group and 2% in the conventional management group Blinding Unclear risk Participants were not blinded to the intervention. It was not reported if study or medical staff were blinded to the intervention

Participants
Participants (n = 195) were adults aged > 60 (mean 69) years, with an indication for long-term oral anticoagulation Exclusion criteria included: previous participation in an anticoagulation self-management program; severe cognitive or terminal illness The study was based in 3 departments specialising in the treatment of participants

Self-management vs usual care
Participants were randomised to a) self-management (n = 99): home self-testing using the Coagucheck ® and self-dosing. INR testing performed once a week, adjusting anticoagulant dosage accordingly. Participants were asked to contact the training centre in case of difficulties b) usual care (n = 96): anticoagulant dosage adjusted by usual attending physicians in general practice or at a hospital based specialised anticoagulation clinic Outcomes Primary outcome: composite of all thromboembolic events requiring hospitalisation and all major bleeding complications Secondary outcomes: frequency and duration of hospitalisation; mortality; recurrence of stroke; numbers of INR values above 4.5 or lower than 1.7; treatment-related quality of life; cost-effectiveness Trial identification

Notes
Participants assigned to the self-management group participated in four consecutive weekly instruction sessions of 90 to 120 minutes each, in groups of three to six participants. Participants assigned to the control group participated in a single 90-miute session including basic theoretical information

Bias Authors' judgement Support for judgement
Random sequence generation (selection bias) Low risk Computer-based system.
Allocation concealment (selection bias) Low risk Allocation was done by a central statistical office by fax and without awareness of participant data. The sequence of randomisation was concealed until the participant was assigned to a group Intention to treat analysis Unclear risk ITT analysis was used for primary outcome; per protocol analyses used for other outcomes Reporting of losses of follow-up Low risk 19/195 (9.7%) participants lost to followup; reasons reported Blinding Low risk Blinded outcome assessors.

Participants
Participants were individuals who had undergone elective mechanical aortic valve replacement and who were computer competent (n=58) Interventions

Self-management vs usual care
Participants were randomised to self-management using CoaguChek devices (n = 29) or usual care (conventional clinic care) (n = 29) In the intervention group, participants received training and self-tested under the supervision of hospital ward staff until discharge, after which time participants self-tested and notified INR dose via the study website. For 4 weeks advice was given from the clinic; subsequently the participant's data were evaluated by the clinic every 3 months In the comparison group, participants received conventional care by the thrombosis clinic At one year all participants completed a quality of life questionnaire and were evaluated by a study physician

Participants
Participants (n = 139) were adults aged > 18 (mean 60) years, receiving warfarin for at least one month before randomisation and requiring anticoagulation for at least the subsequent year, and competent to manage their own anticoagulation therapy Exclusion criteria included: known hypercoagulable disorder, mental incompetence, a language barrier or an inability to attend training sessions Based in a tertiary care setting or by referral as an outpatient at the University of British Colombia (Canada) Interventions

Self-management vs usual care
Participants were randomised to: a) self-management (n = 69): home self-testing using Protime Participants in the self-management group were trained by a pharmacist in a 2-3 session, then required at a second pharmacist appointment to demonstrate competency in selftesting and self-dosing. In a first 2-yo 3-hour visit participants received education from a pharmacist

Random sequence generation (selection bias)
Low risk Computer-generated randomisation code.
Allocation concealment (selection bias) Low risk Randomisation code concealed.
Intention to treat analysis Low risk ITT analysis was used.
Reporting of losses of follow-up Low risk 10% of participants were lost to follow-up; reasons reported Methods Multicentre, randomised controlled trial.

Participants
The study enrolled 202 participants, mean age 64 years, with permanent non-valvular atrial fibrillation in long term anticoagulation. The study was based in 33 centres (Germany) Interventions Self-management vs usual care Self-testing using the Coagucheck ® monitor and self-adjusted dosing (regimen not reported). Usual care by family doctors (regime not reported).

Reply
In response to comments by Dominique Roberfroid submitted 2nd April 2012: Dear Dominique, In terms of mortality we report a reduction in all-cause mortality of 36% (RR 0.64, 95% CI 0.46 to 0.89) and as stated for selfmanagement of 45% (RR 0.55, 95% CI 0.36 to 0.84) over a follow-up period of 2 years. [1] It is correct that the study by Koertke et al [2] provides a substantial amount of data to the mortality analysis for self-management. In terms of the GRADE of the paper we therefore judged the evidence to be of moderate quality around the reported effects, particularly due to an absence of information on allocation concealment and also because there was a few number of events. However, in terms of study quality we confirmed in a subsequent publication, [3] and with direct communication with the authors for this Cochrane review that randomization and allocation concealment were clear in this study.
In terms of the baseline difference in this trial these imbalances could have led to differences in the outcomes. We stated in the discussion that the 36% reduction in mortality from all causes was largely influenced by one study. In addition we applied the logic of early stopping of randomized controlled trials to determine whether our meta-analysis could be considered definitive. It is not, the calculated optimal information size needed to reliably detect a plausible treatment effect is 2,300 patients per group for thromboembolic events alone.
Further to the publication of this review a large RCT was published in the US. [4] We have analysed this in our subsequent publication 'Self-monitoring of oral anticoagulation: systematic review and meta-analysis of individual patient data' [3] including individual patient data from 11 trials of 6,417 participants including the Koertke data. This allowed us to verify the trial methods as well as undertake time-to-event outcomes analysed with hazard ratios (HR) with 5 years of follow-up data, which take into account the number of people randomized and timing of events, and the time until last follow-up for each patient not experiencing an event. In this study [3] we reported a significant reduction in thromboembolic events in the selfmonitoring group (HR 0.51; 95% CI 0.31 to 0.85) but a non-significant reduction in death for the self-monitoring group (HR 0·82, 95% CI 0.62 to 1.09) and a non-significant effect in terms of death for the self-management group alone (HR 0.75, 95% CI, to 0.42 to 1.33). Therefore, the protective effect is lower than previously estimated in our previous review. [1] Reasons for this could include lower quality trial methodology but could also results as improvements in the control group care occur over time. The increase in the number of trials and participants allows us to improve the confidence around our estimated and we will incorporate this additional data into our updated review.