A meta-analysis and systematic review of the efficacy and safety of anticoagulants as cancer treatment

Impact on survival and bleeding complications

Abstract

BACKGROUND.

Preclinical evidence suggests that anticoagulants, in particular the low-molecular-weight heparins (LMWH), exert an antitumor effect, whereas clinical trials have reported conflicting results. The authors conducted a comprehensive, systematic review and meta-analysis of the evidence from randomized controlled trials (RCTs), to evaluate the impact of anticoagulants on survival and safety in cancer patients without venous thromboembolism.

METHODS.

A comprehensive systematic literature review of RCTs was performed without language restrictions through May 2006 with subsequent updates to the end of 2006, including an exhaustive search of electronic databases, major conference proceedings, article references, and content experts. Two reviewers extracted data independently. Primary study outcomes were 1-year overall mortality and all bleeding complications. Major and fatal bleeding complications were secondary outcomes.

RESULTS.

Across all 11 studies that were identified, anticoagulation significantly decreased 1-year overall mortality with a relative risk (RR) of 0.905 (95% confidence interval [95% CI], 0.847–0.967; P = .003). The RR for mortality was 0.877 (95% CI, 0.789–0.975; P = .015) for LMWH, compared with an RR of 0.942 (95% CI, 0.854–1.040; P = .239) for warfarin, resulting in an absolute risk difference (ARD) of 8% for LMWH and an ARD of 3% for warfarin. Improved survival with anticoagulation may be dependent on tumor type. Major bleeding episodes occurred less frequently in patients who received LMWH (ARD, 1%) compared with patients who received warfarin (ARD, 11.5%; P < .0001). Overall, fatal bleeding occurred rarely (ARD, 0.32%; P = .542).

CONCLUSIONS.

Anticoagulants, particularly LMWH, significantly improved overall survival in cancer patients without venous thrombosis while increasing the risk for bleeding complications. However, given the limitations of available data, the use of anticoagulants as antineoplastic therapy cannot be recommended until additional RCTs confirm these results. Cancer 2007. © 2007 American Cancer Society.

Cancer patients have a substantially higher risk of venous thromboembolism compared with noncancer patients.1–6 Patients with cancer frequently have laboratory evidence of hemostatic activation,7–10 and patients who are diagnosed with an idiopathic venous thrombosis are at increased risk of subsequently developing cancer.11–14 Considerable experimental data indicate that anticoagulants, in particular low-molecular-weight heparin (LMWH) and unfractionated heparin (UFH), exert an antitumor effect.15–32

The antineoplastic effect of anticoagulation first was evaluated clinically in the late 1960s and early 1970s with promising results, although the trials either were nonrandomized or had substantial methodological shortcomings.33–37 Zacharski et al. were the first to perform a randomized controlled trial (RCT) that studied the impact of warfarin on survival in several different cancer types, including cancers of the lung, colon, head and neck, and prostate. Their reports indicated an improvement in overall survival only in patients with small cell lung cancer (SCLC).38, 39 Results from the subsequent Cancer and Leukemia Group B study evaluating warfarin versus no warfarin therapy in patients with SCLC indicated that warfarin significantly increased the complete and partial response rates (P = .012) with a borderline improvement in failure-free survival (P = .054) but did not reach statistical significance for overall survival (P = .098).40 A subsequent venous thrombosis prevention study, in which patients with metastatic breast cancer were randomized to receive either warfarin or placebo, demonstrated a significant reduction of venous thromboembolism without a substantial improvement in the secondary survival outcome.41 A follow-up warfarin study in patients with SCLC produced mixed results42; however, both subsequent studies with UFH and LMWH reported significant improvements in overall survival.43, 44

Available meta-analyses investigating the effect of anticoagulation on mortality in patients without a diagnosis of venous thrombosis either included noncancer patients, analyzed nonrandomized cohort reports and post-hoc subgroup analyses of cancer patients from trials that were performed in general medical patients, or included studies investigating a combination of anticoagulants.45–47 In addition, these meta-analyses were largely not based on a comprehensive, systematic review; researched only 1 type of anticoagulant; provided little or no bleeding information; and preceded recent trials.48–52 The objective of the current analysis was to perform a comprehensive, systematic review and meta-analysis of the evidence from RCTs on the efficacy and safety of anticoagulants, including LMWH, UFH, and vitamin K antagonists, as treatment in cancer patients without venous thromboembolism by assessing the effect of these drugs on survival and bleeding complications.

MATERIALS AND METHODS

Literature Search

An exhaustive, systematic literature review of RCTs that examined the influence of anticoagulant therapy on survival and bleeding complications in cancer patients was performed. The comprehensive search included the following electronic databases through May 2006 with subsequent weekly updates to the end of 2006: Medline (including Cancerlit), EMBASE, the Cochrane Database of Systematic Reviews, the Cochrane Central Register of Controlled Trials, the Database of Abstracts of Reviews of Effect, and the National Guidelines Clearing House and Conference Proceedings (American Society of Clinical Oncology, American Society of Hematology, and International Society of Thrombosis and Hemostasis). References from included articles, relevant excluded reports, and guidelines were searched by hand. In addition, content experts from the Venous Thrombosis Guideline Committee of the American Society of Clinical Oncology were asked to review the list of identified articles for completeness. Subject headings and keywords used in the search process included four major categories including medical subject headings and text words: 1) survival, including death and mortality; 2) all types of malignancies, including solid tumors and hematologic malignancies; 3) anticoagulation, including vitamin K antagonists, UFH and LMWH; and 4) RCTs. The recommended search strategy from the Cochrane Collaboration was used.53, 54 These 4 major search categories were combined by using the Boolean “AND.” The terms that were used within these major search categories were combined by using the Boolean “OR.” Because some trials were performed that used survival as a secondary outcome and the prevention of venous thromboses development as primary outcome, this search was also performed by replacing the survival search terms with venous thromboembolism terms, including deep vein thrombosis and pulmonary embolism. The literature search was performed without any language restrictions.55

Inclusion and Exclusion Criteria

The studies included had to be RCTs of adult cancer patients without a concurrent diagnosis of venous thromboembolism comparing anticoagulation drug therapy and an appropriate control group. Anticoagulation had to consist of LMWH, UFH, or an oral vitamin K antagonist. To ensure an adequate amount of therapeutic time, anticoagulation had to be given on a continuous basis for >4 weeks without interruption unless clinically indicated. Studies were required to have overall mortality as a planned primary or secondary outcome. To avoid publication bias, reports that were published only in abstract form were included if they had been presented at a major international conference in the last 3 years.

To avoid biased results, studies were excluded if they were nonrandomized reports or post-hoc subgroup analyses or if they included noncancer patients. Given the significantly different clinical setting, studies related to indwelling catheters, surgery, or intraportal heparin infusion were not included. To ensure an adequate assessment of the drug efficacy, trials were not allowed to examine a combination of anticoagulants or to have any other differences between study arms apart from the anticoagulation therapy under investigation. Among duplicate publications, the most recently updated information was included.

Data Extraction and Study Outcomes

Two reviewers independently extracted the data (N.M.K. and A.A.K.) on basic study design, patient characteristics, study outcomes, and measures of study quality. Any disagreements were resolved by consensus and third-party adjudication (G.H.L.). Data for analysis were abstracted systematically from the published reports and included 1) authors and citation; 2) type and stage of malignancy and other demographic patient characteristics; 3) drugs, doses, and schedule of anticoagulation therapy; 4) study design parameters; 5) the number of patients randomized; 6) the number of patients evaluable; and 7) the cumulative proportion of patients that experienced specific outcomes. Overall study quality was evaluated by using the method of Jadad et al., a validated quality-assessment instrument.56, 57 The instrument evaluates study quality based on the methods of randomization and blinding and the description of study withdrawals. The possible scoring range is from 0 to 5, with good quality represented by a score ≥3. The primary study outcomes were 1-year overall mortality and all bleeding complications. Major bleeding, as defined by each study, as well as fatal bleeding events were secondary study outcomes.

Meta-analysis and Statistical Testing

The overall summary risk estimates represent the weighted sum of the individual estimates using the reciprocalof the variance as weights. Forest plots are presented that display point estimates and 95% confidence limits for each study as well as the weighted summary estimates. When such data were not available directly, the cumulative proportion surviving at 1 year was extracted from the published survival curves.58 When no significant heterogeneity was observed, a fixed-effects model was employed using the method of Mantel and Haenszel59; whereas a random-effects model based on the method of DerSimonian and Laird was used when significant heterogeneity was encountered, employing a conservative P value < .10.60 Potential heterogeneity was evaluated based on the Cochran Q statistic, which represents the weighted sum of squared differences between the pooled effect estimates across studies and the individual study effects.61 The power of the Q statistic for identifying heterogeneity may be too low or too high with a small or large number of studies, respectively. Therefore, it is used best in conjunction with the inconsistency index (I²),62 which is relatively independent of the number of studies.63 Effect measures also were analyzed for a priori defined subgroup and sensitivity analyses, including type of control group, study quality, type of anticoagulant, length of anticoagulation, concurrent chemotherapy, type of cancer, and stage of disease. The potential influence of these study and patient characteristics on outcomes also was explored in meta-regression analysis by regressing the estimated treatment effect on prespecified subgroups.

Estimates of standard error and 95% confidence intervals (95% CIs) for all individual studies and for the weighted summary effect estimate were based on a z statistic.64 Interaction was evaluated between the type of anticoagulation and a priori specified subgroups. Statistically significant differences between subgroups were studied by comparing the ratio of the difference in the natural logarithm of the subgroup relative risk (RR) estimates and the standard error of the difference with the standard normal distribution.65, 66 The presence of potential publication or selection bias was evaluated qualitatively using funnel plots that displayed the treatment effect as the logarithm of the risk ratio versus the inverse standard error of the logarithm of the treatment effect. Potential publication or selection bias was evaluated quantitatively by using the rank correlation method of Begg and Mazumdar based on the Kendall τ67 and also was evaluated by using Egger et al.'s regression intercept method, which employs the actual values of the effect sizes and their variance to regress the standardized effect on precision or inverse of the standard error.68 Significance was based on 2-tailed testing for all analyses.

RESULTS

Literature Search

Among 12,279 potential articles that were screened, 985 abstracts were obtained for further analysis. From these, 90 publications were retrieved in full for a more careful evaluation (Fig. 1). Most of these studies were nonrandomized. Thirteen studies included noncancer patients, or did not make the information regarding cancer patients available, and/or constituted a post-hoc subgroup analysis. Three studies were cost analyses, and 3 studies were duplicate publications. Two studies did not describe the methods or results adequately, making them impossible to evaluate. Overall, 11 separate RCTs, which were reported in 12 publications, were eligible for the currentanalysis (Table 1).38, 40–44, 48–52, 69 One trial reported mortality data and bleeding results separately in 2 different publications.39, 69

Figure 1. This quorum flow diagram of the literature search details the reasons for article exclusions. Overall, 12 eligible publications were identified that reported on 11 randomized clinical trials.

Table 1. Summary and General Characteristics of Included Randomized Controlled Clinical Trials

References	No. of patients	Tumor type [tumor stage]	Drug type	Anticoag. dosing/timing (start of anticoag.)	Length of anticoag.	Length of follow-up	Jadad score	Study design parameters
Anticoag. indicates anticoagulation; SCLC, small cell lung cancer; LMWH, low-molecular-weight heparin; C, chemotherapy; IU, international units; N, no; additional Tx, type of additional cancer therapy in both groups; +, with; ±, with or without; R, radiotherapy; descr., description; random, randomization, NA, not available; ITT, intention-to-treat analysis; Y, yes; TOPIC, Thrombosis Prophylaxis in Oncologic Patients with Certoparin studies; NSCLC, nonsmall cell lung cancer; BID, twice daily; UFH: unfractionated heparin; PTT, partial prothrombin time; PT, prothrombin time; INR, international normalized ratio (coagulation laboratory values). * Report published in abstract form only. † The results of this study are reported in 6 specified stratifications by cancer type (1, advanced NSCLC; 2, SCLC; 3, advanced colorectal cancer; 4, prostate cancer; 5, head and neck cancer; and 6, early NSCLC and early colorectal cancer). ‡ This study reports the data on bleeding complications for the 1984 study by Zacharski et al.39
Altinbas et al., 200444	79	SCLC [all stages: 57% limited, 43% extensive]	LMWH	Dalteparin 5000 IU daily (with C)	18 wk (during C)	Median, 10 mo	1	Placebo control, N; additional Tx, +C, ±R; blinding descr., NA; random. descr., N; withdrawal descr., N; ITT, modified
Haas et al., 200550 (TOPIC 1)*	362	Breast cancer [advanced stage]	LMWH	Certoparin 3000 IU daily (unknown)	6 mo	Unknown	Unknown	Placebo control, Y; additional Tx, unknown; blinding descr., unknown; random. descr., unknown; withdrawal descr., unknown; ITT, unknown
Haas et al., 200551 (TOPIC 2)*	546	NSCLC [stage III, 47%; stage IV, 53%]	LMWH	Certoparin 3000 IU daily (unknown)	6 mo	Unknown	Unknown	Placebo control, Y; additional Tx, unknown; blinding descr., unknown; random. descr., unknown; withdrawal descr., unknown; ITT, unknown
Kakkar et al., 200448	374	Mixed solid tumors [stages III or IV]	LMWH	Dalteparin 5000 IU daily (unknown)	12 mo	Median, 10.2 mo	4	Placebo control, Y; additional Tx, ±C, ±R; blinding descr., Y; random. descr., Y; withdrawal descr., N; ITT, modified
Klerk et al., 200549	302	Mixed solid tumors [advanced stage: mainly metastatic disease]	LMWH	Nadroparin (weight-based) BID×2 wk then daily (unknown)	6 wk	Mean, 12 mo	5	Placebo control, Y; additional Tx, ±C, ±R; blinding descr., Y; random. descr., Y; withdrawal descr., Y; ITT, strict
Sideras et al., 200652	138	Mixed solid tumors [advanced stages]	LMWH	Dalteparin 5000 IU daily (unknown)	2 y	Planned, 2 y	4	Placebo control, Y/N; additional Tx, ±C, ±R; blinding descr., Y; random. descr., Y; withdrawal descr., N; ITT, modified
Lebeau et al., 199443	277	SCLC [all stages: limited, 44%, extensive, 56%]	UFH	Heparin 500 IU/kg/d 2–3 × daily; PTT × 2–3 control (1 d before C)	5 wk	Unknown	3	Placebo control, N; additional Tx, +C, ±R; blinding descr., NA; random. descr., Y; withdrawal descr., Y; ITT, strict
Chahinian et al., 198940	189	SCLC [extensive disease]	Warfarin	Warfarin; PT, 1.5–2.0 (D 1 of C)	Median, 162 d	Median, 36 mo	2	Placebo control, N; additional Tx, +C, ±R; blinding descr., NA; random. descr., Y; withdrawal descr., N; ITT, modified
Levine et al., 199441	311	Breast cancer [stage IV]	Warfarin	Warfarin 1 mg daily × 6wk; then INR, 1.3–1.9 (within 1 mo of C)	Mean, 181 d (until 1 wk post-C)	Mean, 199 d	5	Placebo control, Y; additional Tx, +C; blinding descr., Y; random. descr., Y; withdrawal descr., Y; ITT, modified
Maurer et al., 199742	347	SCLC [limited stages]	Warfarin	Warfarin daily (PT, 1.4–1.6 × normal; D 1 of C)	Range, 113–197 d (3 wk after C stopped)	Median, 69 mo	1	Placebo control, N; additional Tx, +C, ±R; blinding descr., NA; random. descr., N; withdrawal descr., N; ITT, modified
Zacharski et al., 198439†; Zacharski et al., 198138‡	418	Mixed solid tumors [all stages]	Warfarin	Warfarin (“therapeutic range”; 1 wk before C)	Mean, 26 wk (lifelong)	≥1 yr	3	Placebo control, N; additional Tx, +C; blinding descr., NA; random. descr., Y; withdrawal descr., N; ITT, unknown

Study Characteristics

All 11 clinical trials (12 reports) were performed in patients who had solid tumor patients, with some studies utilizing a mixture of different tumor types (Table 1). Six studies (55%) were placebo- controlled,41, 48–52 and 1 placebo-controlled study switched to a nonplacebo-controlled protocol after it was identified as a major obstacle for adequate accrual.52 Six studies (55%) compared LMWH with a control group, and 4 studies (36%) investigated warfarin; whereas only 1 study was performed with unfractionated heparin.43 Haas et al. have not published their findings from the Thrombosis Prophylaxis in Oncologic Patients with Certoparin (TOPIC) I and TOPIC II studies in full, and more detailed information is not yet available.50, 51 Therefore, the current results are based on an abstract publication (Table 1). Three studies (27%) identified that had Jadad scores of ≤2, and only 2 studies (18%) performed a strict intention-to-treat analysis.

One-Year Overall Mortality

Across all eligible studies, according to the fixed-effects model, anticoagulation significantly decreased 1-year overall mortality with an RR of 0.905 (95% CI, 0.847–0.967; P = .003) (Fig. 2). The RR of dying at 1 year for patients who received LMWH was 0.877 (95% CI, 0.789–0.975; P = .015), and, for patients who received warfarin, the RR of dying at 1 year was 0.942 (95% CI, 0.854–1.040; P = .239). The RR estimates remained virtually unchanged when a random-effects model was employed (all studies: RR, 0.915; 95% CI, 0.858–0.976; P = .007; LMWH: RR, 0.875; 95% CI, 0.788–0.972; P = .013; warfarin: RR, 0.958; 95% CI, 0.876–1.047; P = .345) (data not shown). This trend in improved survival was observed among all types of anticoagulants but reached statistical significance only with LMWH. However, in a comparison of LMWH with warfarin, no significant difference regarding 1-year overall mortality between these 2 types of anticoagulants was observed (P = .418). The absolute risk difference (ARD) in mortality was 8.0% (95% CI, 1.6–14.3%) among LMWH studies and 3% (95% CI, −1.9–7.9%) among warfarin trials.

Figure 2. Forest plot for 1-year overall mortality by type of anticoagulation treatment. Rate 1 indicates anticoagulation group; Rate 2, control group; RR, relative risk; SCLC, small cell lung cancer; Mixed, mixture of solid tumors; LMWH, low-molecular-weight heparin; UFH, unfractionated heparin; NSCLC, nonsmall cell lung cancer; SCLC, small cell lung cancer; CRC, colorectal cancer; NH, head and neck cancer; I², inconsistency index.

All Bleeding Complications

All bleeding complications were increased with anticoagulation compared with controls (RR, 2.309; 95% CI, 1.928–2.764; P < .0001) (Fig. 3). A significant increase in all bleeding was observed for both LMWH and warfarin with estimated RRs of 2.029 (95% CI, 1.205–3.417; P = .008) and 2.366 (95% CI, 1.954–2.866; P < .0001), respectively (Fig. 3). Bleeding risk was not increased significantly in the single UFH study. In the random-effects model, the RR for bleeding was 2.110 overall (95% CI, 1.645–2.705; P < .0001), 1.900 for LMWH (95% CI, 0.948–3.806; P = .070), and 2.171 for warfarin (95% CI, 1.632–2.889; P < .0001) (data not shown). Patients who received warfarin experienced a significant increase in ARD (22.5%; 95% CI, 18.2–26.9%) for all bleeding events compared with an ARD of 2.4% (95% CI, 0.7–4.1%) in patients who received LMWH (P < .0001).

Figure 3. Forest plot for all bleeding complications by type of anticoagulation treatment. Rate 1 indicates anticoagulation group; Rate 2, control group; RR, relative risk; SCLC, small cell lung cancer; NSCLC, nonsmall cell lung cancer; Mixed, mixture of solid tumors; LMWH, low-molecular-weight heparin; UFH, unfractionated heparin; NH, head and neck cancer; CRC, colorectal cancer; I², inconsistency index.

Major Bleeding Complications

Across all studies, major bleeding events increased with anticoagulation (RR, 2.598; 95% CI, 1.936–3.488; P < .0001) (Fig. 4). Among the 3 anticoagulants, only warfarin reached a statistically significant increased risk of major bleeding events (RR, 2.979; 95% CI, 2.134–4.157; P < .0001) compared with controls, whereas patients who received LMWH experienced a nonsignificant increased RR of 1.678 (95% CI, 0.861–3.269; P = .128) (Fig. 4). These results remained unchanged in a random-effects model (overall: RR, 2.374; 95% CI, 1.679–3.357; P < .0001; LMWH: RR, 1.571; 95% CI, 0.743–3.321; P = .237; warfarin: RR, 2.682; 95% CI, 1.783–4.036; P < .0001). Patients who received warfarin experienced substantially more major bleeding events (ARD, 11.5%; 95% CI, 8.5–14.5%) compared with patients who received LMWH (ARD, 1.0%; 95% CI, 0.3–2.3%; P < .0001).

Figure 4. Forest plot for major bleeding events by type of anticoagulation treatment. Rate 1 indicates anticoagulation group; Rate 2, control group; RR, relative risk; SCLC, small cell lung cancer; NSCLC, nonsmall cell lung cancer; Mixed, mixture of solid tumors; LMWH, low-molecular-weight heparin; UFH, unfractionated heparin; NH, head and neck cancer; CRC, colorectal cancer; I², inconsistency index.

Fatal Events

Six of 1226 patients who received anticoagulation treatment (0.49%), compared with 2 of 1209 control patients (0.17%), experienced a fatal bleeding complication in the 9 studies that reported this outcome (RR, 1.422; 95% CI, 0.459–4.412; P = .542; data not shown). No patient with a fatal pulmonary embolism was reported in any of the studies.

Sensitivity and Subgroup Analyses and Meta-regression

All subgroup analyses should be considered as hypothesis-generating only. Because of limited statistical power, they also are prone to produce false- negativeresults. Nevertheless, results from the a priori defined sensitivity and subgroup analyses indicated that the reduction in overall mortality and the increase in bleeding complications associated with anticoagulation were observed independent of study quality, use of placebo controls, length of anticoagulation treatment, and the intended use of chemotherapy. There were no significant differences in estimated treatment effects for all 3 outcome measures based on these study or patient characteristics (Table 2). The available trials did not permit subgroup analyses of the influence of cancer type and stage or anticoagulation timing and dosing. The potential influence of these patient and study characteristics on mortality risk reduction also was explored in meta-regression analysis by regressing the estimated treatment effect on these prespecified subgroups. In concurrence with the findings described above, none of these characteristics, except for tumor type (P = .005), significantly influenced the mortality risk reduction caused by anticoagulation.

Table 2. Summary Measures for A Priori Defined Study Subgroups: Relative Risks and 95% Confidence Intervals for 1-Year Overall Mortality, All Bleeding Complications, and Major Bleeding Complications

Subgroup (No. of studies)	1-year overall mortality		All bleeding complications		Major bleeding complications
	RR	95% CI	RR	95% CI	RR	95% CI
RR indicates relative risk, 95% CI, 95% confidence interval. * Studies with Jadad scores ≥3 (study quality could not be assessed adequately in 2 studies). † Studies in which the addition of chemotherapy was intended per protocol in all patients in both the anticoagulation treatment arm and the control treatment arm (the protocol could not be assessed adequately in 2 studies).
Placebo-controlled studies
Yes (6)	0.89	0.80–0.98	2.32	1.38–3.88	2.11	1.00–4.45
No (5)	0.92	0.83–1.02	2.40	1.98–2.91	3.03	2.17–4.24
Good-quality studies*
Yes (6)	0.93	0.87–0.99	1.88	1.51–2.33	2.48	1.80–4.43
No (3)	0.86	0.75–0.99	3.39	2.41–4.78	4.62	1.58–13.50
Length of therapy ≥6 mo
Yes (8)	0.95	0.88–1.02	2.24	1.87–2.69	2.60	1.92–3.50
No (3)	0.83	0.74–0.93	5.56	1.46– 21.17	2.68	0.63–11.38
Chemotherapy in all patients†
Yes (6)	0.90	0.80–1.00	2.37	1.96–2.86	2.90	2.09–4.03
No (3)	0.91	0.84–0.99	1.92	0.94–3.93	1.34	0.49–3.68

Publication or Selection Bias

Neither major nor minor study outcomes were subject to any publication or selection bias based on funnel plot analysis. The absence of publication bias was confirmed quantitatively for all outcomes on the basis of nonsignificant P values both for the rank-correlation method of Begg and Mazumdar and for Egger et al.'s regression-intercept method.

DISCUSSION

The current report represents a comprehensive, systematic review regarding the efficacy and safety of anticoagulants as antineoplastic therapy. The meta-analysis demonstrated a significant reduction in overall mortality with anticoagulant therapy in cancer patients without venous thromboembolism. These findings were most pronounced for LMWH, which produced an RR reduction in mortality of 13.3% compared with a nonsignificant reduction with warfarin of 5.8%. Both major and all bleeding complications were increased significantly in anticoagulant-treated patients compared with controls and in patients who received warfarin compared with patients who received LMWH. LMWH-treated patients experienced a 2.4% absolute risk increase in any bleeding episode and a 1% absolute increase in major bleeding events. This is in contrast to 22.5% of warfarin-treated patients developing any type of bleeding and 11.5% experiencing a major bleeding event. No difference between anticoagulants was observed among fatal bleeding episodes, but this may have been due to the small number of events. The single study that used UFH did not permit any firm conclusions to be drawn.

To ensure the reporting of the most valid effect estimates and to avoid potential publication or selection bias, trials that reported subgroup analyses of cancer patients were not included in the current analyses, and no language restrictions were imposed.55 The improved survival observed with anticoagulants does not appear to be the result of decreased thromboembolic events. None of the trials reported a case of fatal venous thromboembolism. The diagnosis of fatal venous and arterial thromboembolic events may be underestimated, because diagnosis can be difficult, and none of the studies performed routine autopsies. This constitutes a limitation of the available data. Furthermore, screening for venous thromboembolic events was based solely on clinical evaluation in all published trials. Nevertheless, fatal thromboembolic events are unlikely to account for the magnitude of the observed survival benefit from anticoagulation. This observed treatment effect translates into absolute risk reductions for mortality of 8% with LMWH and a nonsignificant 3% reduction for warfarin. However, we observed no statistical difference in mortality reduction between LMWH and warfarin. It remains unclear whether there is a true difference in the impact on mortality between the 2 anticoagulants or whether this reflects a lack of statistical power. A meta-analysis examining differences in efficacy between LMWH and oral vitamin K antagonists in cancer patients with a concurrent diagnosis of venous thromboembolism did not report a difference in mortality.70 Our study also did not permit any firm conclusions about potential differences between LMWH and UFH. Several studies that are investigating cancer subgroups have suggested a survival improvement for LMWH as initial therapy for venous thrombosis compared with UFH in cancer patients.71–75

None of the patient characteristics, trial quality, or other design parameters appear to influence the mortality outcome other than the type of cancer, as suggested previously.38, 39 Although this finding requires further validation, it suggests the importance of studying the association between anticoagulants and cancer survival by specific tumor sites. The currently available trials did not allow us to analyze the influence of cancer stage, anticoagulation timing, or dosing. In post-hoc subgroup analyses, previous trials have suggested the absence of a beneficial effect of anticoagulation on mortality in patients who have advanced-stage cancer, compared with patients who have either early-stage cancer or a more favorable prognosis.48, 76

Our current findings are consistent with abundant in vitro and in vivo laboratory data, which suggest that anticoagulants, particularly LMWH, exert an antineoplastic effect.15–30, 50, 77 It has been reported that heparins inhibit primary tumor growth, reduce tumor cell motility, inhibit tumor cell migration, reduce the formation of metastases, enhance the antitumor effects of chemotherapy, and prolong the survival of laboratory animals after tumor cell inoculation.18, 24, 27, 31, 78–91 The specific mechanism of this effect has not been determined clearly, but it most likely involves several pathways, including the inhibition of angiogenesis26, 92–96 and a change in tumor biology.19–22, 25, 30, 79, 97, 98 In addition, fibrin deposits surrounding many viable tumors and tumor cells not only protect against an immunologic attack99–101 but also enable metastasis formation19, 20, 97, 102–107 and stimulate angiogenesis.108 Furthermore, it is believed that thrombin acts as an autocrine growth factor that can increase the expression of c-myc proto-oncogene and tissue factor.109–113

Not surprisingly, anticoagulation significantly increased bleeding complications. However, life-threatening bleeding events were extremely rare. LMWH appears to have a more favorable bleeding profile. Patients who received warfarin experienced substantially more major bleeding events (11.5%) compared with patients who received LMWH (1%). Although this difference in bleeding risk between LMWH and warfarin may have been caused in part by differences in the definition of bleeding events and in trial patient populations, these results correspond to previously reported findings in patients with and without cancer.76, 114–121 Despite the administration of anticoagulants in conjunction with chemotherapy in many of the trials, the prevalence and impact of chemotherapy- and heparin-induced thrombocytopenia remain unknown.

Limitations of the current study include the unavailability of mortality data from the recently reported TOPIC I and II studies.50, 51 However, no evidence for publication or selection bias was identified for any study outcome. Similar to the majority of meta-analyses, individual patient data were not made available for this study122; however, in the absence of confounding, it has been reported that meta-analyses based on aggregate data produce very similar results compared with individual patient data meta-analyses.123, 124 The limited quality of some of the identified studies did not appear to alter summary effect measures. Although they increased the clinical relevance of the current meta-analysis, all subgroup analyses need to be interpreted as hypothesis-generating only. Given the small numbers in subgroups, negative results may not be caused by the absence of a treatment effect but may be secondary to a lack of statistical power to confirm treatment efficacy. All identified trials were performed in patients with solid tumors and, thus, did not provide any information regarding patients with hematologic malignancies.

Important questions that remain with regard to the therapeutic value of anticoagulation in cancer patients include the optimal cancer types; the influence of disease stage on treatment effect; duration, and timing of anticoagulation; and the extent of patient monitoring to ensure safety. Given the limitations of available studies and the potential for serious complications from anticoagulation therapy, these findings need to be confirmed in future, well-designed, adequately powered, RCTs before the clinical use of anticoagulants as cancer therapy can be recommended. Nevertheless, our findings are reassuring with regard to the overall efficacy and safety of especially LMWH in the treatment of cancer patients and should encourage the further study of these agents as antineoplastic therapy.

Acknowledgements

We thank the expert panel of the Venous Thrombosis Guideline Committee of the American Society of Clinical Oncology for their valuable comments on the preliminary analysis.