Jump to comment:

• Why network meta-analysis cannot be used to determine the optimal method of anticoagulation in pregnant women with mechanical heart valves
  Rohan D'Souza, Prakesh S Shah and Claire McLintock
  Published on: 4 March 2020

• Published on: 4 March 2020

  Why network meta-analysis cannot be used to determine the optimal method of anticoagulation in pregnant women with mechanical heart valves

  • Rohan D'Souza, Maternal-Fetal Medicine Physician Mount Sinai Hospital, University of Toronto, Canada
  • Other Contributors:
    • Prakesh S Shah, Neonatologist
    • Claire McLintock, Haematologist

  Dear editor,

  We read with interest, the study protocol by He. et al, comparing anticoagulation strategies in pregnant women with mechanical heart valves (MHVs)[1]. In their protocol, the authors propose the conduct of a network meta-analysis (NMA) to compare various strategies for anticoagulation in this population. The proposed protocol adds to a long list of recently-published systematic reviews on the topic, each of which adopted slightly different methodological approaches, resulting in the drawing of disparate clinical conclusions. The reason for this is the considerable clinical, methodological and statistical heterogeneity within published studies, most of which case series and small cohort studies[2]. Although this protocol has a number of strengths, which includes an exhaustive list of databases, clear definitions of outcomes and the use of GRADE methodology to assess the certainty in evidence and the strength of recommendations, there are a few concerns that must be addressed.

  The choice of anticoagulant in pregnant women with MHVs is often based on resource availability and clinical factors. For example, the use of low molecular weight heparin is cost-prohibitive in low-resource settings and where serial anti-Xa levels cannot be monitored. In settings where a choice of all methods is available, those at highest risk, still receive a vitamin-K-antagonist-based regimen. Further, clinical outcomes are considerably different in lower resource setting...

  Show More

  Dear editor,

  We read with interest, the study protocol by He. et al, comparing anticoagulation strategies in pregnant women with mechanical heart valves (MHVs)[1]. In their protocol, the authors propose the conduct of a network meta-analysis (NMA) to compare various strategies for anticoagulation in this population. The proposed protocol adds to a long list of recently-published systematic reviews on the topic, each of which adopted slightly different methodological approaches, resulting in the drawing of disparate clinical conclusions. The reason for this is the considerable clinical, methodological and statistical heterogeneity within published studies, most of which case series and small cohort studies[2]. Although this protocol has a number of strengths, which includes an exhaustive list of databases, clear definitions of outcomes and the use of GRADE methodology to assess the certainty in evidence and the strength of recommendations, there are a few concerns that must be addressed.

  The choice of anticoagulant in pregnant women with MHVs is often based on resource availability and clinical factors. For example, the use of low molecular weight heparin is cost-prohibitive in low-resource settings and where serial anti-Xa levels cannot be monitored. In settings where a choice of all methods is available, those at highest risk, still receive a vitamin-K-antagonist-based regimen. Further, clinical outcomes are considerably different in lower resource settings, where there is a predominance of rheumatic heart disease, affecting the mitral valve, and the use of more thrombogenic valves, vs. in higher-resource settings where less thrombogenic valves are used to correct congenital valvular disease that is more likely to affect the aortic valve, and access to multi-disciplinary care is less restricted.

  There is therefore, a natural selection bias, which is evident in the published literature. This bias, and the heterogeneity between non-experimental studies means that the assumptions of non-parametric or student T-tests are violated, and this precludes any form of head-to-head comparison. It is for this reason we, and most systematic reviews have presented point estimates for anticoagulant strategy, and have refrained from performing pair-wise meta-analysis. The authors’ decision to perform NMA, by only including studies that have compared two regimens, only one of which is an experimental study, would not only eliminate most case series, which offer vital information, but would also result in erroneous conclusions, from not considering the confounding variables that contribute to the choice of anticoagulant in this population.

  Common assumptions made in a NMA are homogeneity between studies, transitivity and consistency. The homogeneity assumption requires that studies included in the meta-analyses are similar. Transitivity is an assumption that two studies of which one has compared treatment A vs B and other has compared treatment A vs C do not differ with respect to effect modifiers and we can estimate what would be the treatment effect if treatment B and C are compared. This means that the population receiving treatment A is similar, and can therefore be compared with those receiving treatment B and treatment C. Consistency means that direct and indirect effect estimates between two treatments do not differ beyond the expectation by heterogeneity. This would be very crucial in this NMA where likelihood of obtaining direct estimates are very small as majority of studies are case series or small cohorts. In addition, there is no plan for evaluating quality of evidence using CINeMA or other framework which would be crucial in this exercise[3, 4]. Given the reasons outlined earlier, the assumptions of NMA, are difficult to defend for studies on anticoagulation in pregnant women with MHVs.

  The management of anticoagulation in pregnant women with MHVs is a classic example of maternal-fetal conflict, wherein decision-making regarding the choice of anticoagulant depends on individual circumstances and the values and preferences of patients, which is better addressed through clinical decision-analysis[5]. Decision-analysis studies, as well as counselling this complex group of patients requires contemporary data that is best obtained through a prospective international database[6], which will allow for appropriate subgroup analysis, and tailoring the correct approach to each individual. Until a database of this kind is established to further inform clinical practice, systematic reviews should restrict their role to providing pooled estimates of maternal and fetal risks, and refrain from either providing head-to-head comparisons between anticoagulant strategies or attempt to derive indirect estimates.

  1. He S, Zou Y, Li J, Liu J, Zhao L, Yang H, Su Z, Ye H: Anticoagulation regimens during pregnancy in patients with mechanical heart valves: a protocol for a systematic review and network meta-analysis. BMJ open 2020, 10(2):e033917.
  2. D'Souza R, Ostro J, Shah PS, Silversides CK, Malinowski A, Murphy KE, Sermer M, Shehata N: Anticoagulation for pregnant women with mechanical heart valves: a systematic review and meta-analysis. Eur Heart J 2017, 38(19):1509-1516.
  3. Salanti G, Del Giovane C, Chaimani A, Caldwell DM, Higgins JP: Evaluating the quality of evidence from a network meta-analysis. PloS one 2014, 9(7):e99682.
  4. CINeMA: Confidence in Network Meta-Analysis [Software]. In. University of Bern: Institute of Social and Preventive Medicine; 2017.
  5. D'Souza R, Shah PS, Sander B: Clinical decision analysis in perinatology. Acta Obstet Gynecol Scand 2018, 97(4):491-499.
  6. D'Souza R, Silversides CK, McLintock C: Optimal Anticoagulation for Pregnant Women with Mechanical Heart Valves. Semin Thromb Hemost 2016, 42(7):798-804.

  Show Less

  Conflict of Interest:
  None declared.

  • Back to top