Evaluation of the Association Between Hereditary Thrombophilias and Recurrent Pregnancy Loss: A Meta-analysis

Background  Recurrent pregnancy loss (RPL) is a significant clinical problem. Recently, thrombophilias have been implicated as a possible cause. Factor V Leiden (FVL) and prothrombin gene (G20210A) mutations are the most common types of hereditary thrombophilias, but are usually undiagnosed because most carriers are asymptomatic. The relationship between FVL, G20210A, and RPL has been investigated with conflicting results. This study analyzed existing data to determine whether an association exists.

Methods  A systematic review of the literature was performed. Only case-control studies that defined RPL as 2 or more pregnancy losses in the first or second trimester and that confirmed mutations by DNA analysis were included. Sixteen studies were selected for the FVL meta-analysis and 7 for the G20210A analysis. Stratified and multivariate logistic regression analyses were performed with the use of aggregate data. Results were confirmed by means of fixed- and random-effects meta-analyses models.

Results  The combined odds ratios for the association between RPL and FVL and between RPL and G20210A were 2.0 (95% confidence interval, 1.5-2.7; P<.001) and 2.0 (95% confidence interval, 1.0-4.0; P = .03), respectively. Similar results were produced by the logistic regression and both fixed- and random-effects meta-analysis models.

Conclusions  Carriers of FVL or prothrombin gene mutations have double the risk of experiencing 2 or more miscarriages compared with women without thrombophilias. Hereditary thrombophilias may be an unrecognized cause of RPL. We recommend testing for these mutations in women with RPL.

Recurrent pregnancy loss (RPL), defined as 2 or more spontaneous abortions, affects approximately 5% of women of reproductive age.¹ Although several causes of RPL have been established, more than 50% of cases remain unexplained. This is a challenging dilemma for both patients and physicians.¹ Recently, thrombophilias have been suggested as a possible cause of RPL.^2-8 Hereditary thrombophilias are a group of genetic disorders of blood coagulation resulting in a hypercoagulable state, which in turn can result in abnormal placentation. Early in pregnancy this may manifest as spontaneous loss.^9,10 In later pregnancy, thrombophilias have been associated with complications such as preeclampsia, intrauterine growth restriction, placental abruption, and stillbirth.^11-15

Most carriers of the mutation will not develop any clinical signs and remain undiagnosed because these conditions result in a small absolute risk of clinically significant thrombosis. However, when carriers are exposed to additional risk factors, such as pregnancy or possibly oral contraceptives, the risk of life-threatening thrombotic events is significantly increased and may become clinically evident.^16,17

The 2 most common causes of hereditary thrombophilias are factor V Leiden (FVL) and prothrombin gene (G20210A) mutations.¹⁸ Factor V Leiden is a point mutation (G1691A) that results in an altered factor V, which is resistant to inactivation by protein C. This results in a hypercoagulable state with a 5- to 10-fold risk of thrombosis in heterozygotes and an 80-fold risk in homozygotes.¹⁷ Factor V Leiden is responsible for 20% to 40% of isolated thrombotic events and 40% to 45% of familial thrombophilias. Its prevalence in the United States is estimated to be between 3% and 7%, with the highest frequency in whites.^18-20 Many studies have investigated the relationship between FVL and RPL, and the majority found an association, with odds ratios ranging from 0.5 to 18.^21-43

The G20210A mutation affects 1% to 4% of the US population. Its prevalence is highest among whites of Southern European decent. This mutation is associated with a 20% to 50% rise in prothrombin plasma levels.^44,45 Affected women have a 3-fold increased risk of venous thrombosis, and a higher chance of myocardial infarction and cerebral thrombosis than noncarriers.^16,46,47 Studies of G20210A have also shown conflicting associations with RPL.^{32,33,35,38,39,41,48,49}

Since the association between hereditary thrombophilias and RPL has not been conclusively established, most women with RPL are not tested for these mutations unless they have a personal or family history of thrombosis. Because most carriers are otherwise asymptomatic, the diagnosis would usually be missed. Preliminary studies of thrombophylaxis during pregnancy in carriers suggest that treatment may significantly improve pregnancy outcome.^9,50,51 Thus, by identifying heritable thrombophilias in young women, we might potentially prevent miscarriages, as well as serious maternal and neonatal complications.

Our objective was to evaluate the relationship between RPL and these 2 common thrombophilias. To do so, we conducted a comprehensive meta-analysis of the existing data. We believe that this was the best approach to address this controversial issue because collecting prospective data would be exceedingly difficult, given the relatively low prevalence of both the risk factor (thrombophilia) and the outcome (RPL). Furthermore, many of the previous studies lacked power to detect a small, but clinically important, association.

We conducted a systematic review of the literature to identify studies exploring the relationship between FVL or G20210A and RPL. An English-language MEDLINE search using Ovid and PubMed (1966-2002) was performed with the use of various combinations of the following terms: thrombophilia, factor V Leiden, prothrombin, G20210A, miscarriage, abortion, and pregnancy loss. Pertinent studies were also identified from article bibliographies.

Inclusion criteria required that RPL be defined as 2 or more losses in the first 2 trimesters of pregnancy, and that mutations be identified by DNA analysis (polymerase chain reaction). Studies that did not provide control subjects were excluded. If a study as a whole did not meet inclusion criteria, but a subset of the subjects qualified, only the subset was included. The search yielded 34 relevant studies. Among articles addressing FVL, 17 met inclusion criteria. Of these, 16 were case-control studies and 1 was a retrospective cohort. Because of the major difference in design, the cohort study was excluded from analysis. Seven articles investigating G20210A met inclusion criteria; all were case-control studies. All studies excluded women with a history of thrombosis and those previously diagnosed as having a hereditary thrombophilia. Two of us (G.K. and C.R.G.) performed the literature search, reviewed the articles, and abstracted the data independently. All differences were resolved by consensus.

The data were abstracted by means of a standardized spreadsheet. The information extracted from each study consisted of the number of cases and controls with and without each mutation, mean age of cases and controls, minimum number of losses (2 or 3), race (whites only or mixed), and whether subjects were excluded on the basis of an evaluation for other causes. When possible, data were recorded separately for women with pregnancy losses exclusively in the first trimester and for those with losses in both the first and second trimesters.

All analyses were performed with Stata (version 6.0; Stata Corp, College Station, Tex). Rather than assigning arbitrary quality scores to this set of studies with similar designs, we analyzed the associations between outcome and relevant variables that we were able to extract.⁵² Data were entered as the aggregate number of cases or controls with or without the mutation in each study. Independent analyses were carried out for the 2 mutations. The association between presence of mutation and RPL was evaluated by means of multiple techniques: crude bivariate analysis, analysis stratified by study, logistic regression, and meta-analysis of weighted averages. The multivariate logistic regression model was constructed by means of indicator variables for the individual studies. The meta-analysis was performed with both fixed-effects and random-effects models. This approach involved computing weighted averages of the study-specific log odds ratios.

The degree of among-study heterogeneity was investigated and possible sources were examined. First, the analysis was stratified by study and an overall test of among-study heterogeneity was performed. Then, interactions between the presence of mutation and aggregate-level covariates (ie, mean age, race, minimum number of losses, and exclusion based on other causes) were evaluated by likelihood-ratio tests. To estimate the influence of various factors on the association of interest, we also performed analyses stratified by race, trimester of RPL, minimum number of losses, and whether other causes of RPL were excluded. Finally, to further assess among-study heterogeneity, weighted random-effects meta-regression models of the log odds ratios were fitted by means of the "metareg" procedure in Stata. Covariates were added to the model individually and in all possible combinations.

As one large study contributed approximately half of the cases to the FVL analysis, we decided a priori to perform the complete analysis with and without this study. Publication bias was assessed with both Begg and Egger tests and funnel plots.⁵³

The 16 individual studies included in the analysis are summarized in Table 1, and the odds ratios (ORs) are depicted in Figure 1. Table 2 summarizes the ORs produced by each method. The initial bivariate analysis of the association between the FVL mutation and RPL produced a crude OR of 2.1 (95% confidence interval [CI], 1.6-2.7; P<.001). However, significant among-study heterogeneity was observed when the analysis was stratified by study (Breslow and Day test of homogeneity, P = .03). Nevertheless, this result was confirmed by the logistic regression model, while attempting to control for among-study variation. Furthermore, both fixed-effects and random-effects meta-analyses also reproduced this estimate (fixed effects: OR, 2.0; 95% CI, 1.5-2.6; P<.001; random effects: OR, 2.2; 95% CI, 1.4-3.3; P<.001).

These models also confirmed the presence of significant among-study heterogeneity (P = .02). When we attempted to identify a likely source for the heterogeneity, we found that the only covariate that showed a significant interaction with the presence of FVL mutation was race (likelihood ratio test, P = .005). However, race was not shown to be a significant source of heterogeneity by the weighted meta-regression model (P = .14). In fact, none of the covariates was found to be a significant source of heterogeneity by this method.

Subgroup analyses confirmed that race produced statistically significant variation in the strength of the main association (Table 3). We also found that limiting the analysis to women with exclusively first-trimester losses weakened the association (OR, 1.6; 95% CI, 1.2-2.2; P = .002). Despite these stratifications, the association between FVL and RPL remained significant in all of the subgroups. We also found that greater age was associated with higher risk. Finally, tests for publication bias did not detect any significant bias (Begg test, P = .78; Egger test, P = .47).

The complete analysis was repeated with exclusion of the study by Rai et al.³⁴ Stratification by study could no longer reject the null hypothesis of homogeneity (Breslow and Day test, P = .26). The adjusted OR increased to 2.5 (95% CI, 1.8-3.4; P<.001). This result was corroborated by the logistic regression model (OR, 2.6; 95% CI, 1.9-3.5; P<.001) and both meta-analysis models (fixed effects: OR, 2.5; 95% CI, 1.8-3.4; P<.001; random effects: OR, 2.5; 95% CI, 1.7-3.7; P<.001). Neither the interaction tests nor the meta-regression analyses yielded any significant findings. Thus, exclusion of the study by Rai et al eliminated significant among-study heterogeneity.

The 7 studies included in the analysis are summarized in Table 1, and ORs are given in Table 2 and Figure 2. The initial analysis of the association between the G20210A mutation and RPL produced a crude OR of 2.5 (95% CI, 1.3-4.7; P = .004), and the analysis stratified by study resulted in a Mantel-Haenszel adjusted OR of 2.0 (95% CI, 1.0-4.0; P = .03). No significant among-study heterogeneity was observed (Breslow and Day test of homogeneity, P = .51). The association between G20210A and RPL in the logistic regression model, while adjusting for the influence of among-study variation, was not substantially different from the adjusted OR. Both fixed-effects and random-effects meta-analyses achieved a similar result; however, the statistical significance became borderline. These models also confirmed the absence of among-study heterogeneity (P = .44). Consistent with the absence of significant heterogeneity, both the interaction analyses and the meta-regression models found that no factors significantly influenced the association between G20210A and RPL.

The subgroup analyses demonstrated that stratification by each of the variables also yielded no statistically significant variation in the strength of the association (Table 3). The association between G20210A and RPL remained positive in all of the subgroups, although statistical significance was not achieved in all strata. Specifically, a significant association was not found in studies where RPL was defined as 3 or more losses, races other than white were included, and only unexplained losses were included. Finally, tests for publication bias did not detect any significant bias (Begg test, P = .89; Egger test, P = .71).

Recurrent pregnancy loss is a devastating condition, both medically and emotionally. While most cases remain unexplained, inherited thrombophilias have recently been implicated as a potential cause. The FVL and G20210A mutations, the two most common inherited thrombophilias, have been investigated and have shown inconsistent results. By performing a meta-analysis, we established an association between FVL mutation and RPL, with a cumulative OR of 2. In other words, carriers of the FVL mutation have double the risk of RPL when compared with noncarriers. This association was confirmed by several methods of analysis.

Carriers of the G20210A mutation also were found to have double the risk of RPL when compared with noncarriers. However, while the magnitude of the association between G20210A and RPL remained similar in all methods of analysis, statistical significance was dependent on the technique of analysis. This may be due to a smaller sample size in the G20210A group; only 7 studies met the inclusion criteria for this analysis.

Inclusion and exclusion criteria are a critical part of a meta-analysis and can substantially affect results. We chose to be conservative in our study selection, which likely underestimated the discovered association. A number of articles investigating FVL from the Israeli collaborators were excluded because they included third-trimester miscarriages in their definition of RPL, and the necessary information for the subset of women who met our criteria could not be extracted.^4,5,12,24 The ORs in these publications ranged from 4 to 18. Similarly, the cohort study, which was excluded on the basis of design differences, found an OR of 2.5.²⁶ Thus, inclusion of these studies in our analysis would have only strengthened the association between FVL and RPL.

Notably, the disproportionately large study by Rai et al³⁴ showed no association (OR, 0.8). This study contributed 45% of the FVL-positive RPL cases to our analysis and was the cause of significant among-study heterogeneity. Nevertheless, we found a significant association between FVL and RPL both with and without inclusion of this study. One possible reason for the null effect in this study is that women with RPL of known causes were included in their analysis. In fact, 20.7% of the cases in this study were diagnosed as having antiphospholipid antibodies. When we repeated the analysis without this study, a stronger association was observed (OR, 2.5; 95% CI, 1.8-3.4).

We stratified our analysis by using several descriptive variables in an attempt to estimate their influence on the main association (Table 3). We found that race had a significant influence on the association between FVL and RPL; studies recruiting only white women demonstrated a weaker association than those with mixed populations. This effect was reversed in the G20210A analysis, but the difference was not statistically significant. Also in the FVL analysis, limiting the data to women with first-trimester RPL appeared to weaken the association, while such an effect was not found in the G20210A analysis. It would not be surprising to find that the association is weakened in the first trimester because abnormal fetal karyotype has been consistently shown to be responsible for most of these losses,¹ thus resulting in a lower prevalence of losses caused by abnormal placentation. On the other hand, we are not aware of any reason why race may have influenced the association. It is possible that selection bias was present in studies that did not recruit the cases and controls from populations with equal prevalence of the mutation. However, in interpreting these results, it is important to remember that the analysis was performed on data composed of aggregate values from studies, and not on data from individual subjects. Such analyses are vulnerable to ecological bias, and further investigation of these factors is needed. Most importantly, the association between FVL and RPL remained significant in all subgroups, while the association between G20210A and RPL remained significant in most and borderline in the rest.

Our study focused on women with RPL, but it is likely that hereditary thrombophilias are also a cause of isolated pregnancy losses. Carriers of these mutations exhibit highly variable phenotypic expression, with most having no clinically discernible abnormalities. With respect to pregnancy losses, only a fraction of carriers will develop RPL, while most will experience no losses or possibly a single loss with an otherwise normal reproductive history. Since women who have a single early pregnancy loss are not commonly tested, and hereditary thrombophilias are usually clinically silent, the true cause of the miscarriage thus remains undiagnosed. Clearly, the effect of thrombophilias on isolated pregnancy loss deserves further research.

In addition to pregnancy loss, there is evidence that hereditary thrombophilias are associated with severe complications in late pregnancy and with increased risk of venous thrombosis during pregnancy, post partum, and in users of oral contraceptives.^11-15 The role of thrombophilias in all aspects of reproduction should be closely examined. As the clinical implications of inherited thrombophilias on pregnancy and women's health become better understood, a comprehensive cost-effectiveness analysis will be essential to assess the benefits of screening all women of reproductive age.

For testing to be truly beneficial, effective treatment of carriers is needed. Thrombophylaxis has been shown to be valuable in preventing recurrent thrombosis in carriers, and preliminary evidence from case series suggests that it also may help to prevent miscarriages and improve neonatal and maternal outcomes in pregnancy.^50,51 However, clinical trials are necessary to establish the safety and efficacy of treatment before widespread application is instituted.

In summary, our meta-analyses have demonstrated a strong association between RPL and the FVL mutation and shown a probable association between RPL and the G20210A mutation. Given the clinical implications and the relatively high prevalence of these mutations, we recommend testing women with RPL for their presence. Until the efficacy of thrombophylaxis for RPL is proved, anticoagulation of carriers should be considered on an individual basis.

Corresponding author: George Kovalevsky, MD, CONRAD Clinical Research Center, 601 Colley Ave, Norfolk, VA 23507.

Accepted for publication April 28, 2003.

This study was presented in part at the meeting of the American Society of Reproductive Medicine; October 15, 2002; Seattle, Wash.