Introduction
Venous Thromboembolism (VT) is a common cardiovascular disease with an annual incidence of ~ 1 to 3 per 1,000 in the general population which increases with age [
1]. This pathology can manifest as either deep vein thrombosis (DVT) or pulmonary embolism (PE) with a mortality rate within a month of diagnosis at 6% and 12%, respectively [
2].
After a first VT, the recurrence rate is approximately 30% within 10 years [
3]. VT recurrence could be prevented by a continued anticoagulant treatment but this therapy leads to a substantial risk of bleeding and a significant cost to society [
4]. Understanding the pathophysiological mechanisms of VT recurrence may facilitate the identification of groups of patients at lower risk of recurrence who do not require extended treatment. While more than 100 loci are now well established to be associated with the genetic susceptibility to VT [
5‐
7], less is known about the genetic susceptibility to VT recurrence which possibly differs from that of first VT [
8]. Among VT disease loci, the
ABO locus, coding for the
ABO blood groups, is one of the most important genetic risk factor due to the magnitude of the genetic effects associated with the A1 and B at-risk blood groups (Odds Ratio ~ 1.5) and their prevalence in the general population (18% and 8%, respectively) [
9‐
12]. The attributable risk associated with
ABO blood groups can be as high as 30% in some specific groups of individuals [
13]. To date, few studies have explored the effect of
ABO blood types on VT recurrence risk [
14‐
17]. These analyses have generally been conducted in studies of moderate size with few recurrent events and have often relied on serological measurement of blood groups. Recently, our group showed that molecularly defined blood groups are more reliable than serological measurements [
12]. In this work, we wish to investigate the effect of molecularly defined
ABO blood groups on the risk of recurrence in two large VT cohorts, the MARTHA and MEGA studies [
18,
19].
In MEGA, participants were included at the time of their first VT which represents the beginning of the at risk period for the recurrence. To study the risk of first recurrence a standard time-to-event analysis among which the Cox model is the most popular one [
20] can be used. The MARTHA study has a different design since it included all subjects who visited a Thrombophilia centre in Marseille (France) between 1994 and 2012 and had a history of VT (possibly many years before inclusion). Information on recurrence post-inclusion was collected at a follow-up visit several years later but many participants had already experienced a VT recurrence at the time of inclusion.
In the literature, several choices have been proposed to deal with recurrent events occurring before the time of inclusion: i) analysing only first recurrent events that occurred post-inclusion, while discarding that experienced the event of interest (VT recurrence in our case) before their inclusion [
21,
22], ii) analysing only recurrent events that occurred post-inclusion while stratifying according to the number of events before inclusion [
23], iii) analysing all recurrent events that occurred post-inclusion without distinguishing between patients who did or did not experience a recurrence prior to inclusion [
24].
Most of these approaches have been proposed to assess risk factors that are time-dependent variables such as biological measurements explaining why they only include the events occurring after the collection of the studied variables as the exposure must be measured before the event occurrence in order to avoid bias due to reverse causality. However, when the explanatory variables do not change over time, as genetic factors, this bias is avoided. When the dates of the events that occurred before the inclusion in the study are known, considering this information in the analysis could greatly increase the statistical power. However, specific data analysis procedures should be considered to avoid selection bias by death.
Therefore we here propose an original weighted survival analysis that enables the joint analysis of patients with or without recurrence prior inclusion in the study. This approach can be applied in any studies about recurrent events when information is collected about events that occurred before the inclusion and when the explanatory variables are time-independent, such as genetic factors. The proposed methodology is then employed to efficiently analyse the impact of ABO blood groups on first VT recurrence in MARTHA.
Discussion
The motivation of this work was to investigate the risk of VT recurrence associated with ABO blood groups in two large cohorts of middle-aged VT patients, MARTHA and MEGA, the former being built upon an ambispective design.
In order to maximize the power of the MARTHA study where about 70% of first VT recurrences occurred in patients before their inclusion in the study, we developed a weighted approach to analyse non-time dependent risk factors (such as genetic polymorphisms) of an event which could have occurred in patients before their inclusion in the study. Such recurrent events are generally discarded in standard approaches that focus only on recurrent events occurring post-inclusion [
21‐
24]. Our proposed modelling relies on a weighted Cox model where the use of weights allows to limit the selection bias associated with the use of pre-inclusion events and thus to gain statistical power by jointly analysing pre- and post-inclusion recurrent events. This method differs from the weighting approach for repeated events proposed to deal with event-dependent sampling [
38]. Indeed, the inclusion in MARTHA depends on an event, the first VT, which is not the outcome of interest; the first VT defines the beginning of the period at risk for the recurrence. Our weighting approach handles potential bias due to mortality until the time of data collection for the recurrence.
Our proposed weighted estimation approach is unbiased if the weights are well-specified which means that the Cox model for death is correct. In this work, the death model has two main limitations. First, as the information on VT recurrence was often missing for subjects who died, it was not possible to include VT recurrence (and other possible unknown variables) as a risk factor in our model for death risk. Second, we assumed the proportionality of the risk of death and did not account for the calendar time that could modify either the baseline risk of death or the association with risk factors for death. Moreover, as the number of death during the follow-up in MARTHA is quite small, a Monte Carlo analysis was performed to evaluate the sensitivity of the results to the uncertainty on the weights (description is available in the Supplement). Despite some slight variability in HRs’ estimates (especially for O2 group), the overall results remain unchanged.
In this work, we were interested in the association of
ABO blood groups with the risk of first VT recurrence and not with the risk of multiple VT recurrences. Indeed, at inclusion in MARTHA, only detailed information on first VT and possibly first recurrence was collected, preventing us from investigating the association with multiple recurrences. Besides, the analysis of such multiple events would require more complex modeling that would take into account the correlation of repeated events [
39].
The analysis of these two studies, totaling 2,752 VT patients including 993 recurrences, revealed that the A1 and A2 blood groups were both associated with a moderate increased risk of VT recurrence, HR ~ 1.20 for both, compared with O1. Note that, likely because of the modest frequency of the A2 blood group (~ 5%), the association was only marginally significant (
p = 0.06). Some studies have already investigated the association of
ABO blood groups with the risk of VT recurrence [
14‐
17], but often with a moderate sample size or using serological
ABO phenotypes whereas we here used genetically defined
ABO blood groups which has been shown to be more efficient to capture the effect of
ABO on VT risk [
12]. Our results are consistent with those showing a higher risk of recurrence in non-OO patients [
15‐
17]. However, they are discordant with the study of
Baudouy et al. who found a higher risk of VT recurrence in B blood group patients [
14], while no association (HR = 1.01,
p = 0.90) was observed in our work. This lack of association in our work is unlikely due to a power issue as the B blood group was more frequent than A2 which was significantly associated here with VT recurrence. Beyond its rather modest size (
N = 100) and the analysis of serological
ABO phenotypes, the work of
Baudouy et al. focused on patients with PE as first VT. A stratified analysis of
ABO blood groups with recurrence according to the type of the first event (DVT or PE) did not reveal in our work any evidence for specific sub-group
ABO effects (Supplementary Table S
2).
MARTHA and MEGA are composed of middle-aged VT patients, with average age at first VT event ~ 45 yrs. While the association of
ABO blood groups with VT recurrence was consistent between patients with age at first event lower and higher than 45yrs (Supplementary Table S
3), our study is not well-suited to assess whether the observed
ABO association also holds in older ages. Our results cannot then be generalizable to older populations and further studies are mandatory to investigate this issue.
As the proposed
ambispective modelling is only valid for analysing non-time dependent variables, we could not adjust the
ABO blood group‘s effect on biological variables that have only been measured at the time of inclusion, such as von Willebrand Factor (vWF). Adjusting for vWF plasma levels could have allowed us to assess whether
ABO blood groups impact on VT recurrence independently of vWF, at least partially. This is plausible as the observed pattern of association of
ABO blood groups with recurrence does not match the known associations between
ABO blood groups and vWF plasma levels [
12]. The observed pattern of association with recurrence does not match either the known associations with first VT risk. Indeed, the B blood group did not show any trend for association with VT recurrence while it is associated with an Odds Ratio of ~ 1.5 for first VT [
12]. Interestingly, the observed pattern matches with the one observed between
ABO blood groups and plasma levels of Intercellular Adhesion Molecule 1 (ICAM1) where both A1 and A2 groups associate with ICAM1 levels, but not B [
12]. These observations suggest that the biological factors involved in the association of
ABO blood groups with VT recurrence differ from those involved in their relation with incident VT. More than 50 plasma proteins have been shown to be under the genetic influence of the
ABO locus [
40]. Determining which of them are associated with the risk of incident and/or recurrent VT merit further deep investigations. Finally, we could not adjust our analysis for the familial history of VT as the available information in MARTHA refers to the presence of a history at the time of inclusion and many recurrences arose earlier.
Nevertheless, our modelling enabled us to assess the impact on the risk of VT recurrence of several clinical variables that are fixed after the first VT event such as age at first VT and the type of first VT (DVT vs PE; provoked vs unprovoked). Consistent in MARTHA and in MEGA were the associations of male sex and DVT as first VT with an increased risk of first VT recurrence, confirming previous observations [
8,
41]. However, we did not observe consistent results with respect to age at first VT nor with the provoked status of the first VT. For the effect of the provoked character on VT recurrence, the different trends observed can be due to the different design and sample selection between MARTHA and MEGA [
42,
43]. Indeed, participants in MARTHA were included for at least one previous VT which may have occurred more than fifty years before their inclusion (Mean = 6 years; Standard Error = 10 years) whereas in MEGA, patients were recruited at the time of their first VT. Besides, the definition of the provoked character slightly differs between MARTHA and MEGA (Supplementary Table S
5). We also observed some differences between MARTHA
prospective and
ambispective that might be due to the calendar time which has not been taken into account in our work. We are aware that the differences in the management, prevention and identification of VT events may have masked the association between the provoked character of the first event and VT recurrence in the
ambispective analysis. Indeed, among the 25% of MARTHA patients that had their first VT before the start of the study (1994), for 80% of them the VT was provoked. Whereas in the remaining sample of 75% of MARTHA patients, the first VT was provoked in only 62% of cases. Of note, when we restrict the analysis to recurrent events that occur within less than 2 years after the first event, the provoked status of first VT was protective against recurrence, consistently in MARTHA and MEGA (Supplementary Table S
4). These results are in line with previous findings from a 2-years follow-up study [
42]. Furthermore, the association between A1 and A2 blood groups with VT recurrence remain unchanged when focusing on patients whose first VT occurred after the start of the MARTHA study. Altogether, we feel that such differences may have modest impact when one is interested in genetic factors as illustrated here with the consistent patterns observed for
ABO blood groups in both MARTHA
prospective and
ambispective samples as well as in MEGA (Supplementary Table S
6).
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.