Restarting Oral Anticoagulant Therapy after Major Bleeding in Atrial Fibrillation: A Systematic Review and Meta-Analysis
Supplementary Materials
S1Methods
Search Strategy
((“atrial fibrillation”MeSH Terms) OR (“atrial”All Fields AND “fibrillation” All Fields) OR (“atrial fibrillation”All Fields)) AND ((“major bleeding”All Fields OR “major bleeding”MeSH Terms OR (“major” All Fields AND “bleeding”All Fields) OR (“gastrointestinal bleeding”All Fields OR “gastrointestinal bleeding”MeSH Terms OR (“gastrointestinal” All Fields AND “bleeding”All Fields) OR (“gastrointestinal hemorrhage”All Fields OR “gastrointestinal hemorrhage”MeSH Terms OR (“gastrointestinal” All Fields AND “hemorrhage”All Fields) OR (“intracranial hemorrhage”All Fields OR “intracranial hemorrhage”MeSH Terms OR (“intracranial” All Fields AND “hemorrhage”All Fields) OR (“brain hemorrhage”All Fields OR “brain hemorrhage”MeSH Terms OR (“brain” All Fields AND “hemorrhage”All Fields) OR (“hematemesis”[All Fields] OR “hematemesis”[MeSH Terms]) OR (“melena”[All Fields] OR “melena”[MeSH Terms]).
Data Synthesis and Analyses
All statistical analyses were undertaken using Review Manager (RevMan) version 5.3 (The Cochrane Collaboration 2014, Nordic Cochrane Centre Copenhagen, Denmark) and R version 3.4.0 (R development core team, Vienna, Austria). Risk of events according to OAC restarting was reported as odds ratio (OR) and 95% confidence interval (CI), relative risk reduction (RRR) or absolute risk reduction. The I2-statistic was quantified to describe the percentage of variation across all included trials. Given the low number of studies, assessment of heterogeneity was not reliable and hence as per guidelines(1) we performed a meta-analysis via a heterogeneous Bayesian model with informative priors. Log transforms of the OR were assumed to be normally distributed. Each study sample was assumed to arise from a Gaussian centred on a study-specific effect and with variance corresponding to the square of its estimated standard error, inflated by 25% in order to guarantee conservative statements(1). The study-specific summary was assumed to be Gaussian, centred on an unknown pooled measure, which was the main object of interest. Summaries were then back-transformed appropriately. As per guidelines with limited number of studies involved (1), informative priors were used. For the variance of the pooled measures we assumed an inverse Gamma centred on an estimator obtained with a moment-based approach (inflated by 25%, once again in order to guarantee a conservative statement). Similar approaches were previously used(2).
To assess the clinical benefits and risks of restarting OAC after a major bleeding event, we performed a NCB analysis. Given the characteristics of the cohort considered, we developed specific weights on the basis of a previous Markov-model base decision analysis whether or not to restart OAC in patients reporting an ICH(3). Three outcomes were considered in the NCB: i) any stroke/any thromboembolic event (those two were taken together due to lack of data for some of the studies considered); ii) recurrent major bleeding; and iii) all-cause death.
In order to assess the effectiveness and safety of OAC in patients restarting therapy, two distinct NCB models were constructed using IRs, one including and another excluding death (4,5). Weights were assessed by the swing weighting method (4). The first model balanced incidence of any stroke/any thromboembolic event with incidence of recurrent major bleeding (relative weight: 0.66), as follows:
NCB= (IR Any Stroke/Any thromboembolic eventOAC restarters - IR Any Stroke/Any thromboembolic eventOAC non-restarters) + 0.66 (IR Recurrent Major BleedingOAC restarters - IR Recurrent Major BleedingOAC non-restarters).
In the second model, all-cause death was balanced with any stroke/any thromboembolic event (relative weight: 0.20) and recurrent major bleeding (relative weight: 0.13) occurrence, as follows:
NCB= (IR All-Cause DeathOAC restarters - IR All-Cause DeathOAC non-restarters) + 0.20 (IR Any Stroke/Any thromboembolic eventOAC restarters - IR Any Stroke/Any thromboembolic eventOAC non-restarters) + 0.13 (IR Recurrent Major BleedingOAC restarters - IR Recurrent Major BleedingOAC non-restarters).
In order to account for weights variability, a sensitivity analysis was performed. In the first NCB model, the NCB was computed for 1000 different weight values equally spaced between 0.2 and 1.5; in the second model, all combinations for 100 values equally spaced in the interval [0.1,1] for each of the two weights were used (total number of scenarios: 10000). We then evaluated sign agreement between NCB values computed with optimal weights and each value computed at sensitivity analysis. After being evaluated for every study, all NCBs were pooled together through a meta-analysis (described below).
A sensitivity analysis according to risk of bias was performed. A subgroup analysis on recurrent index bleeding event was also performed. When necessary, the numbers of events were calculated using event rates, sample size and follow-up duration or by using subgroups sample size, hazard ratios (HR) and 95% CI. A p-value <0.05 was considered statistically significant.
1
Study / SelectionBias / Performance Bias / AttritionBias / DetectionBias / Reporting Bias / OverallBiasHernandez, 2017 / LowRisk / LowRisk / LowRisk / Medium Risk / LowRisk / LowRisk
Kuramatsu, 2015 / LowRisk / LowRisk / LowRisk / Medium Risk / LowRisk / LowRisk
Nielsen, 2017 / Medium Risk / LowRisk / LowRisk / LowRisk / LowRisk / LowRisk
Witt, 2012 / High Risk / LowRisk / LowRisk / LowRisk / High Risk / High Risk
Qureshi, 2014 / Medium Risk / LowRisk / LowRisk / LowRisk / LowRisk / LowRisk
Sengupta, 2015 / High Risk / LowRisk / LowRisk / LowRisk / High Risk / High Risk
Staerk, 2015 / Medium Risk / LowRisk / LowRisk / LowRisk / LowRisk / LowRisk
Table S1: Risk of Bias Assessment
Table S2: Patients Cohorts and Incidence Rates* of Major Adverse Events Used for Net Clinical Benefit Analysis
Study / N / Stroke/TE / Recurrent Major Bleeding / All-Cause DeathrOAC / nrOAC / rOAC / nrOAC / rOAC / nrOAC / rOAC / nrOAC
Hernandez et al. / 696 / 843 / 14.8 / 10.7 / 11.9 / 6.8 / 4.6 / 13.2
Kuramatsu et al. / 110 / 456 / 10.0 / 26.1 / 7.3 / 5.7 / 8.2 / 37.5
Nielsen et al. / 405 / 778 / 38.6 / 46.2 / 20.6 / 10.7 / 69.4 / 88.4
Qureshi et al. / 653 / 676 / 13.8 / 19.4 / 37.8 / 17.4 / 14.3 / 20.4
Staerk et al. / 725 / 924 / 4.0 / 9.7 / 7.3 / 5.4 / 12.2 / 31.4
Legend: *incidence rates are expressed as per 100 patient-years; nrOAC= non-restarting oral anticoagulant; rOAC= restarting oral anticoagulant; TE= thromboembolic events.
Figure S1: Flow-Chart of Studies’ Selection
Figure S2: Risk of Recurrent Intracranial Hemorrhage
Legend: CI= confidence interval.
Figure S3: Risk of Recurrent Gastrointestinal Bleeding
Legend: CI= confidence interval.
Figure S4:Bias Stratified Analysis for Any Stroke Occurrence
Legend: CI= confidence interval.
Figure S5:Bias Stratified Analysis for Any Thromboembolic Event Occurrence
Legend: CI= confidence interval.
References
1. Higgins JPT, Thompson SG, Spiegelhalter DJ. A re-evaluation of random-effects meta-analysis. J R Stat Soc Ser A (Statistics Soc [Internet]. 2009 Jan [cited 2017 Jun 7];172(1):137–59.
2. Iacovelli R, Pietrantonio F, Farcomeni A, Maggi C, Palazzo A, Ricchini F, et al. Chemotherapy or Targeted Therapy as Second-Line Treatment of Advanced Gastric Cancer. A Systematic Review and Meta-Analysis of Published Studies. Wei Q-Y, editor. PLoS One [Internet]. 2014 Sep 30 [cited 2017 Jun 26];9(9):e108940.
3. Eckman MH, Rosand J, Knudsen KA, Singer DE, Greenberg SM. Can Patients Be Anticoagulated After Intracerebral Hemorrhage?: A Decision Analysis. Stroke [Internet]. 2003 Jul 1 [cited 2017 Jun 7];34(7):1710–6.
4. Ma H, Jiang Q, Chuang-Stein C, Evans SR, He W, Quartey G, et al. Considerations on Endpoint Selection, Weighting Determination, and Uncertainty Evaluation in the Benefit-Risk Assessment of Medical Product. Stat Biopharm Res [Internet]. 2016 Oct 21 [cited 2017 Jun 26];8(4):417–25.
5. Towse A. Net clinical benefit: the art and science of jointly estimating benefits and risks of medical treatment. Value Health [Internet]. 2010 Jun [cited 2017 Jun 26];13 Suppl 1:S30-2.