Duration of Anticoagulation Following Venous Thromboembolism:  A Meta-analysis FREE

Venous thromboembolism (VTE) continues to be associated with significant morbidity and mortality.¹ The role of unfractionated or low-molecular-weight heparins in the initial treatment of VTE is well established and has been shown to help limit morbidity and mortality. Patients with VTE are also susceptible to recurrent events, and clearly there is a need for ongoing anticoagulation, which is usually accomplished with warfarin.^2- 5

What constitutes the optimal duration of this long-term therapy is still controversial. The advantages of prolonging the duration of treatment must be weighed against the risk of adverse effects—primarily bleeding. However, randomized controlled trials and even meta-analyses investigating the impact of different durations of anticoagulation have reported conflicting results.^2- 3,6- 11 While some trials have reported benefits with longer anticoagulation, other trials have found that extended anticoagulation protected against recurrent VTE only while treatment continued; once anticoagulation was stopped, the benefits were lost.² A systematic review of the incidence rate of recurrent VTE concluded that the recurrence rate decreased with time after the index event and that the recurrence rate reached a lower plateau at about 9 months after the index event independent of the duration of anticoagulation.¹² However, another meta-analysis suggested that long-term therapy regimens did reduce the incidence of recurrence.¹⁰

The objective of this meta-analysis was to evaluate the evidence regarding the duration of anticoagulation therapy for VTE. Our primary hypothesis was that extending the duration of anticoagulation by a finite amount would result in a decrease in the risk of recurrent VTE.

Trials investigating the effect of prolonged anticoagulation on the risk of VTE recurrence vary in terms of the measures of effect reported and the periods used for defining them. The measures of effect we chose to use were the incidence rate ratio (IRR) and the incidence rate difference. In addition, since different studies used different periods to determine the reported measure of effect, a standard schedule for defining the timing of events was necessary.^2,8,13 We therefore analyzed all of the available studies based on a hypothetical standard reference model (Figure 1).

In this model, patients would typically receive anticoagulation with heparin immediately upon diagnosis of VTE (time A). Patients would usually be prescribed some form of long-term anticoagulation thereafter. This would continue for a set amount of time, at which point anticoagulation would be discontinued (time B). In the standard model, it is uncertain whether long-term therapy that instead ends anticoagulation at a later time (time C) would improve outcome.

Recent studies have shown that once anticoagulation is stopped the incidence rate of VTE increases and then returns to a lower baseline level.¹² Therefore, if the analysis of outcomes in randomized trials is limited to the period in which only patients receiving long-term anticoagulation are still receiving therapy and if follow-up stops at the point when long-term therapy is stopped, only the incidence rate from time B to time C will be measurable (Figure 1). Measuring the incidence rate from B to C leaves out events occurring immediately after long-term anticoagulation is stopped, when the incidence rate for the group receiving long-term therapy transiently increases, resulting in a falsely low estimate of the risk of recurrent VTE in the group receiving long-term therapy.

Therefore, in assessing the impact of long-term anticoagulation of finite duration, it is important that patients be followed up for several additional months after those receiving long-term anticoagulation are discontinued. This additional follow-up time corresponds to the period from time C to time D in Figure 1. Thus, the proper period for determining the incidence rate is from B to D, rather than from B to C. For this discussion, we will refer to the period from B to D as the risk period during the study with follow-up. Note that the distance from C to D must be sufficiently long to capture the transient bump in the incidence rate of VTE events that occurs shortly after discontinuation of anticoagulation. For this discussion, we will refer to this period from time C to time D as the study’s additional follow-up time.

We can apply these same principles to develop a system for evaluating the risk of recurrent VTE while receiving lifelong treatment as compared with a finite duration of treatment. In this case, the starting point is still time B, but the end point is time C, since patients receiving lifelong anticoagulation never stop therapy. For the purposes of this discussion, we will refer to the period from B to C as the risk period while receiving lifelong therapy.

We therefore decided a priori that the time interval of interest for determining the primary outcome of the meta-analysis would be the risk period of the study with additional follow-up (B to D in Figure 1). We also decided a priori that a study’s additional follow-up time had to be a minimum of 3 months to capture the initial bump in the incidence of VTE events, based on previous reports and a meta-analysis.^2- 3,7,12 We performed a separate analysis on all studies of lifelong therapy vs fixed-duration therapy (B to C in Figure 1).

The necessary inclusion criteria were established before the literature search. Included studies were randomized controlled trials that involved a difference in duration of anticoagulation between study groups and studies that primarily involved first episodes of VTE or that involved anticoagulation with 1 or more agents, recurrence of VTE as a measured result of the study, and similar initial therapy in both study groups at time of diagnosis. Excluded studies were those enrolling only pure populations of high-risk patients, such as those with protein S or C deficiency. It was believed that these populations would represent a group clinically distinct from patients without these high-risk profiles, and as such inclusion would result in significant clinical heterogeneity. In addition, there is less clinical controversy about how to treat these patients. Two independent reviewers assessed each article for inclusion and exclusion criteria, with adjudication by a third reviewer in cases of disagreement. Studies published in abstract form were excluded because of the concern that non–peer-reviewed data might introduce bias into the report.^14- 15

A computerized search of PubMed, EMBase Pharmacology, and the Cochrane database of clinical trials was conducted using the terms venous thromboembolism or pulmonary embolism or deep vein thrombosis or DVT or PE; coumarin or heparin or low-molecular-weight heparin or anticoagulant(s) or anticoagulant treatment or anticoagulant therapy or duration of anticoagulant therapy or anticoagulation. In addition, the clinical trial Web sites http://www.clinicaltrials.gov and http://ctr.glaxowellcome.co.uk were searched. A hand search of the citations of review articles and meta-analyses on VTE and its treatment was conducted to identify potentially relevant articles not captured by the search strategy. The search was restricted to articles pertaining to adult humans and presented or translated into English and published from 1969 through 2004. All articles were reviewed and scored for quality by 2 independent reviewers and adjudicated by a third reviewer in cases of disagreement.

Articles were assigned quality scores using 2 different scales. The first was the Jadad scale, which measures quality by focusing on 3 dimensions of internal validity: randomization, blinding, and withdrawals.¹⁶ The second was a quality scale specific for VTE, ranging from 0 to 1, which measured quality based on 7 characteristics: randomized treatment allocation (relative weight, 0.19), outcome assessor blinded to treatment (0.18), objective diagnostic criteria applied at entry (0.18), objective diagnostic criteria for recurrence (0.17), withdrawal of patients described (0.11), blinding of physician and/or patient to treatment (0.09), and whether risk factors were described (0.08).

Data regarding VTE, major bleeding, and person-time at risk were extracted by 2 independent reviewers using a standardized form, with adjudication by the remainder of the investigators in cases of disagreement. For each study, incidence rates in events per person-year at risk were calculated for the outcomes of recurrent VTE, death from VTE, and major bleeding (defined as clinically overt and associated with a decrease in hemoglobin level of at least 2 g/dL or the need to transfuse 2 or more units of red blood cells if the hemorrhage was intracranial, retroperitoneal, or required discontinuation of anticoagulation). When person-time data were not presented (either directly or indirectly through a Kaplan-Meier survival curve), the person-time of the interval was estimated by multiplying the number of participants present at the beginning of the interval by the duration of the interval and subtracting person-time for events occurring within the interval. For this calculation it was assumed that events were equally likely throughout the interval unless the data were available in the report. Life-table analysis was also used to calculate person-time at risk when the published data permitted.^17- 20 Subsequently, letters were sent to the corresponding authors to verify the number of events, person-months at risk, and study design features. In addition, follow-up data on patients that had been lost to follow-up at the time of publication of 1 study but that were collected subsequently were made available, and these were used in the analysis.⁹

A pooled IRR and rate difference was calculated for each outcome of interest for each time interval using a random-effects model and the inverse variance method (DerSimonian and Laird) using STATA version 8.2 (Stata Corp, College Station, Tex). We chose a priori to use the pooled estimate of the IRR and the rate difference as our measures of effect and to analyze the results based on 95% confidence intervals (CIs) and to report 2-sided P values. There was no adjustment for multiple comparisons. Statistical heterogeneity between and within groups was measured using the Q statistic.

Risk factors of the study populations were extracted from the reports using a standard definition. This definition was based on the definitions used in different trials. There was significant variation in the definition of what constituted idiopathic VTE, permanent risk factors, and temporary risk factors in different trials. Data were abstracted from each study and then reclassified according to the standard definition. Idiopathic in the revised context was the absence of temporary risk factors, cancer, and prior VTE.

Secondary and subgroup analyses were performed based on a priori decisions. Secondary analyses were conducted to (1) estimate the pooled IRR and rate difference of VTE for lifelong vs fixed-duration therapy, (2) obtain a more precise estimate of the risk of major bleeding, (3) determine the risk of death from VTE, and (4) assess the impact of low-quality studies and publication bias. We used meta-regression to analyze the relationship between outcomes and study quality, level of anticoagulation, duration of anticoagulation in the group receiving short-term therapy, percentage of patients with temporary risk factors, percentage of patients with permanent risk factors, and percentage with idiopathic VTE. Publication bias was assessed using the Begg rank correlation method and the Egger weighted regression method. The influence of individual studies on the pooled effect-size estimate was analyzed by performing influence analysis, in which the pooled estimates were recalculated omitting 1 study at a time. Cumulative meta-analysis which included the cumulative evidence at the time each study was published was also performed. In each case the α for significance was set at .05.

Sixty-seven articles were selected for full review (Figure 2). A total of 11 trials were identified that fulfilled the a priori inclusion and exclusion criteria.^{2- 3,6,9,17- 19,21- 24} Also identified were 4 studies that did not have a period of follow-up after cessation of treatment in the group receiving long-term therapy (Table 1).^8,13,20,25 These correspond to studies that measure recurrence rates from time B to time C (Figure 1) but could not be included in the primary analysis because the studies’ additional follow-up time was zero. These studies were included in the secondary analysis of the risk of recurrence with lifelong vs fixed-duration therapy, since they met the other inclusion criteria.

The median duration of short-term therapy was 1.75 months (interquartile range [IQR], 1.0-3.0 months; n = 15); this corresponds to the period from time A to time B in Figure 1. The median duration of long-term therapy was 6.0 months (IQR, 3.0-10.5 months; n = 15); this corresponds to the period from A to C. The median difference between short-term and long-term therapy was 4.5 months (IQR, 2.0-9.3 months; n = 15); this corresponds to the period from B to C. After long-term therapy was stopped, patients were followed up for a median of 12.8 months (IQR, 9.0-18.0 months; n = 11); this corresponds to the period from C to D. There was significant variation between studies in terms of the patient population studied and their risk factor profiles (Table 2). Study quality was generally good, and there was a high degree of correlation between the Jadad and disease-specific quality score (Spearman rank, R = 0.82; P<.001). In the 11 studies that followed up patients for at least 3 months following cessation of therapy in the group receiving long-term therapy, the median duration of short-term therapy was 1.5 months (IQR, 1.0-3.0 months), and the median duration of long-term therapy was 6.0 months (IQR, 3.0-6.0 months).

There were 136 recurrent VTE events during 2645 person-years among patients in the group receiving long-term therapy and 192 recurrent VTE events during 2501 person-years among those in the group receiving short-term therapy (weighted incidence rate, 0.052 vs 0.072 events/person-year). Using a random-effects model, there was a decrease in the risk of recurrent VTE when longer durations of anticoagulation were used (Table 3 and Figure 3). Meta-regression demonstrated no direct relationship between outcome and year of publication (P = .39). Cumulative meta-analysis indicated that by 1995, with the publication of the study by Schulman et al,⁹ there was evidence that prolonging anticoagulation decreased the risk of recurrent VTE (P<.001). As additional studies were subsequently published, the finding of benefit remained consistent.

An influence analysis was performed limited to those studies (n = 8) for which reliable estimates of the person-time at risk were available (Table 4).^{2- 3,6,8- 9,13,21- 25} Three trials, all performed more than 15 years ago, did not have precise data on person-time at risk.^17- 18,20 The risk of recurrence during study and follow-up for the group of studies with reliable estimates of person-time at risk was similarly reduced when long-term anticoagulation was used. Meta-regression demonstrated that as the duration of anticoagulation in the group receiving short-term therapy increased, the incremental benefit of prolonging anticoagulation decreased, both in terms of relative risk (P = .04) and rate difference (P = .005) (Figure 4). Twelve studies provided sufficient data to estimate the risk of recurrent VTE with lifelong vs fixed-duration therapy. This corresponds to the period from time B to time C in Figure 1.^{2- 3,6,8- 9,13,17- 18,20,22,24- 25} There were 40 recurrent VTE events during 2050 person-years among patients while receiving lifelong anticoagulation and 231 recurrent VTE events during 1968 person-years among patients receiving short-term anticoagulation (weighted incidence rate, 0.020 vs 0.126 events/person-year). Using a random-effects model, there was a decrease in the risk of recurrent VTE with lifelong anticoagulation (Table 3 and Figure 5). On the relative-risk scale there was no indication of heterogeneity, but there was evidence of heterogeneity on the rate-difference scale (P<.001).

Meta-regression did not demonstrate any relationship between outcome and level of anticoagulation, percentage of patients with temporary risk factors, or percentage of those with idiopathic VTE (Table 3). There was no evidence that study quality impacted the risk of recurrent VTE, as measured by meta-regression with either the Jadad or the disease-specific study quality score. There was no evidence of publication bias using the Begg rank correlation method. There was evidence of possible publication bias as measured by the Egger weighted regression method, but this was only evident on the relative-risk scale and not the risk-difference scale. In addition, this was only present when evaluating recurrence during the study with follow-up. Examination of funnel plots demonstrated that if studies had been omitted because of publication bias, the direction of the bias favored publication of studies that reported less benefit with long-term anticoagulation.

Influence analysis for the risk period during the study with follow-up demonstrated that a study by Schulman et al⁹ had the greatest impact on the pooled estimate (Table 4). When only recent (within the last 10 years) high-quality studies (Jadad score ≥3 and/or disease-specific quality score ≥0.9) were included (n = 6), the risk of recurrence was still less with long-term anticoagulation (Table 4).

Influence analysis of recurrence while receiving lifelong therapy indicated that a study by Schulman et al had the greatest impact.⁸ The results were not significantly different when a study limited to patients with second episodes of VTE was included (Table 4).²⁶

Data on death from recurrent pulmonary embolism following cessation of treatment in the group receiving short-term therapy were available in 10 studies.^{2- 3,6,8- 9,13,18,21,23,25} There were 12 deaths during 3663 person-years among patients receiving long-term anticoagulation and 14 deaths during 3437 person-years among those receiving short-term anticoagulation (weighted incidence rate, 0.0017 vs 0.0027 events/person-year, random-effects rate difference, −0.001 [95% CI, −0.004 to 0.001]; P = .58; pooled IRR, 0.92 [95% CI, 0.43 to 1.94]; P = .82). There was no indication of heterogeneity or publication bias. Meta-regression demonstrated no relationship between outcome and study quality, level of anticoagulation used, duration of anticoagulation in the group receiving short-term therapy, percentage of patients with temporary risk factors, percentage of those with permanent risk factors, and percentage with idiopathic VTE.

Data on major bleeding during treatment were available in 7 studies.^{2- 3,8,13,18,22,25} There were 18 bleeding events during 1571 person-years among patients receiving prolonged anticoagulation and 9 bleeding events during 1497 person-years among patients who had finished short-term anticoagulation (weighted incidence rate, 0.011 vs 0.006 events/person-year; random-effects rate difference, 0.005 [95% CI, −0.002 to 0.011]; P = .14; pooled IRR, 1.80 [95% CI, 0.72 to 4.51]; P = .21). There was no indication of heterogeneity and no evidence of publication bias. Meta-regression demonstrated no relationship between outcome and study quality, level of anticoagulation used, duration of anticoagulation in the group receiving short-term therapy, percentage of patients with temporary risk factors, percentage of those with permanent risk factors, and percentage with idiopathic VTE.

The optimal duration of anticoagulation for VTE is that which minimizes the aggregate of harm that occurs from recurrent VTE and major bleeding. Our meta-analysis helps to synthesize and supplement the existing body of evidence regarding the optimal duration of therapy by more precisely quantifying the impact of prolonging anticoagulation and by highlighting methodological issues with respect to meta-analysis and clinical epidemiology that are particularly important for studying VTE.

With respect to quantification of benefits from prolonged anticoagulation, recent studies have reported that prolonging the duration of anticoagulation prevented embolic events only if anticoagulation was continued.^2- 3,6,24 The power of these studies to detect a difference was limited, since recurrent VTE is a rare event. By pooling the results of many studies, this meta-analysis suggests that long-term anticoagulation reduces the risk of recurrent VTE. This is consistent with the findings of previous meta-analyses.¹⁰

The risk of recurrent VTE was greatly reduced if anticoagulation was continued (pooled IRR, 0.21; 95% CI, 0.14 to 0.31). If anticoagulation was discontinued in the long-term group, the risk of recurrence during the period of the study with additional follow-up was still diminished, but the magnitude of the effect size was less (pooled IRR, 0.69; 95% CI, 0.53 to 0.91). While there are insufficient data to determine precisely whether 3 months, 6 months, or a longer period of anticoagulation would be optimal, there is evidence that the incremental benefit of prolonging anticoagulation decreases as the duration of anticoagulation increases (Figure 4). Based on the limited available evidence, it appears that 6 or more months of treatment for patients at higher risk may be warranted, recognizing that increasing the duration of anticoagulation beyond 6 months results in a relatively modest incremental risk reduction.

This explains why some previous randomized controlled trials have demonstrated no effect. The number needed to treat to prevent 1 VTE event with long-term anticoagulation would be approximately 50 (95% CI, 25 to 1000); the average number of patients per study was only about 350, or 175 in each group (Table 1).

This meta-analysis also demonstrates the importance of precisely defining the relevant at-risk periods for clinical trials. The discordance between randomized controlled trials that reported very large benefits^8,25 and others that reported only small or no benefit^2- 3,9 can be explained by subtle but important differences in the periods used for defining recurrence rates. Once we take this into account, the relative risk reduction across studies is fairly consistent. It also allows quantification of the potential benefits of alternative therapeutic strategies, such as lifelong anticoagulation. The effect size with lifelong therapy is much larger; the number needed to treat to prevent 1 VTE event with lifelong anticoagulation would be approximately 9 (95% CI, 7 to 14).

However, recurrent VTE is only part of the story; bleeding risk also must be considered. In this meta-analysis, the risk of bleeding associated with prolonged anticoagulation did not reach statistical significance. The risk of bleeding, after increasing immediately following initiation of therapy, subsequently decreases to a steady but lower level.²⁷ This is consistent with our findings, since the time frame for defining incidence rates in this study is after anticoagulation has already been sustained for several months, placing patients in a lower-incidence period.

The relative amount of harm that is associated with each adverse outcome also must be considered. A large intracranial hemorrhage must be weighted more heavily than recurrent VTE of the lower extremity. Meta-analysis cannot answer these types of questions directly but rather serves to provide some of the information necessary to help make good decisions.

This investigation also highlights methodological issues that are relevant to meta-analysis in general. We chose a priori to base our primary analysis on a random-effects model. We did this because we believed there was a high probability that trials would vary in terms of their baseline risk for recurrence (clinical heterogeneity), and a random-effects model would provide the most conservative estimate. Our analysis found no evidence of significant statistical heterogeneity in terms of relative risk, but there was heterogeneity in terms of rate difference with lifelong therapy vs fixed-duration treatment.

It is probable that the significant differences in the baseline prevalence of risk factors explains much of the observed between-study variation in rate difference.^9,13,22 While meta-regression demonstrated no relationship between study outcome and particular risk factors, the limitations of meta-analysis methods in this regard are important to emphasize. One consequence of analyzing data at the study level rather than the individual-patient level is that study-level characteristics are assumed to apply to all members in a given study. Thus, interpretation of individual-level questions, such as what particular risk factors are associated with a high baseline risk for recurrence, are difficult to answer with meta-analysis. Any interpretation of the results is therefore best performed with caution and in the spirit of an exploratory data analysis.^28- 29

In summary, this meta-analysis indicates that long-term anticoagulation in patients with VTE does reduce the risk of recurrence. The magnitude of the risk reduction is greatest while receiving treatment, but even if treatment is stopped there is benefit. The incremental benefit of prolonging anticoagulation decreases as the duration of anticoagulation increases but persists to at least 6 months. The bleeding risk, both in absolute and relative terms, is very low and fairly constant between populations. This suggests that physicians should focus on risk stratification and that in certain high-risk populations lifelong anticoagulation may be needed, while for low-risk populations a shorter course of therapy may be best.