Venous thrombosis: understanding the paradoxes of recurrence


  • S. C. Cannegieter,

    Corresponding author
    1. Department of Clinical Epidemiology, Leiden University Medical Center, Leiden, the Netherlands
    • Correspondence: Suzanne C. Cannegieter, Department of Clinical Epidemiology, Leiden University Medical Center, PO Box 9600, 2300 RC Leiden, the Netherlands.

      Tel.: +31 71 526 4037; fax: +31 71 526 6994.


    Search for more papers by this author
  • A. van Hylckama Vlieg

    1. Department of Clinical Epidemiology, Leiden University Medical Center, Leiden, the Netherlands
    Search for more papers by this author


The decision to continue anticoagulant treatment in patients with a first venous thrombosis after the initial treatment period has strong, life-long implications. Both the risk of recurrence when treatment is stopped and the risk of bleeding when it is continued are high and will persist over a patient's lifetime. For clinicians, rational strategies to stratify their patients into levels of risk of recurrence are limited. To support in the decision to continue or not, it is of the utmost importance to understand why some people develop a second event and others do not and how these people can be identified. This is not easy as, contrary to intuition, the risk profile of a recurrent event is entirely different from that of a first: Some genetic factors that have a major effect on first thrombosis only marginally predict recurrence, while, for instance, the opposite is true for male sex. These paradoxes can be explained when we understand etiology of a first event, how rates for first and second event cannot be directly compared, and how fixed risk factors cannot be predictors, while factors that are not causes can yet be predictors. Integrating all knowledge and combining the best predicting variables will ultimately lead to ways to estimate an individual's recurrence risk and hence to decide on optimal further treatment.


Patients with a first deep vein thrombosis or pulmonary embolism are treated with anticoagulants for a limited period of time, generally 3–6 months, which serves as treatment of the acute episode, that is, prevention of extension or (further) embolization of the thrombus [1]. After this initial period, continuation of the anticoagulant treatment serves as prevention of a second event. At this point lies the most challenging decision for clinicians. On the one hand, the risk of recurrence is high: In the 3 years after a first event, more than 15% of the surviving patients will experience a recurrent episode of venous thrombosis (Fig. 1) [2-4]. On the other hand, continued anticoagulant treatment causes bleeding, with an annual risk of major bleeding of 3%, and a case-fatality rate of 13% [5, 6]. Considering that the risks and benefits of continued treatment persist over a patient's lifetime, the cumulative likelihood of either thrombosis or bleeding will be very high. Hence, the decision to continue treatment or not has strong, life-long implications.

Figure 1.

Cumulative incidence of recurrent venous thrombosis in 1626 patients. From Ref. [2].

Currently, clinicians have limited rational strategies to identify and treat those with high risk of recurrence [1]. To support in the decision to continue or not, it is of the utmost importance to understand why some people develop a second event and others do not and how these people can be identified. The aim of this review is to offer a conceptual framework for better understanding of the data currently available, and how this knowledge should be integrated to identify patients at different levels of risk of recurrence.

Recurrence paradoxes

Contrary to intuition, studies into risk factors for recurrence do not simply show the same results as for first events (this is the case for several diseases, not only thrombosis). In some cases, the results of recurrence research even show the opposite of what was found for a first event, which was called paradoxes of recurrence research in a recent paper by Dahabreh and Kent [7]. Some examples they give include how in a database of trials into thrombolysis for acute myocardial infarction it was found that prior aspirin use was associated with an increased risk of recurrent myocardial infarction [8]. In a study of patients who had undergone percutaneous coronary intervention (PCI), patients with normal body mass index (BMI) had higher rates of cardiac death than overweight or obese patients [9]. Lastly, the so-called ‘smokers-paradox’ describes the better prognosis of smokers compared with non-smokers among patients with myocardial infarction [10].

Another situation that is less paradoxical, but seen many times, is that a risk factor is found to be much weaker for a second event than for a first. This is the case, for example, for hypertension, which increases the risk of first-time stroke about 4-fold, but is a much weaker risk factor for stroke recurrence, with relative risks ranging from 0.9 to 1.6 [11]. For venous thrombosis, a similar situation has been observed: Thrombophilia, although a strong risk factor for a first event, increases risk of recurrence only about 1.5-fold, the so-called thrombophilia paradox [3, 12]. Age, the strongest risk factor for a first event, even has no effect on recurrence risk whatsoever [13, 14]. Understanding these phenomena, which obviously do not represent truly causal relations, requires insights into risks in general, etiology of a first event, and into the differences between causes and predictors of recurrence, which will all be addressed below.

Risks of a first and recurrent event are not directly comparable

An incidence rate has number of events in the numerator and number of person-years in the denominator. Incidence rates of a first event and recurrence rates are not directly comparable because the denominator consists of person-years of different people: in the first case of ‘everyone’ in the population that was studied, and in the second case of people who had a first event. This is a situation that is by definition inevitable, but needs to be realized when the effect of a risk factor is compared. If we take, for example, a general population with a baseline risk of venous thrombosis of 1 in 1000 people per year and we find a relative risk of factor (F)V Leiden of 7, this means that the risk in the exposed is seven per 1000 people per year, or in other words that factor (F)V Leiden leads to six extra cases per year. If we subsequently take a population of subjects who had a venous thrombosis, we will find a baseline risk of recurrence of about 30 per 1000 per year. Suppose factor (F)V Leiden again leads to six extra cases per year, we would find a recurrence risk in these people of 36 per 1000 per year or a relative risk of 1.2. This is quite in line with what is found in several studies [15], and it therefore does not immediately mean that the effect of factor (F)V Leiden on a recurrent event is less than on a first, but it means that we are comparing at different scales.

Etiology of a first thrombosis and its relation to risk of a second event

It is well established that venous thrombosis is a multicausal disease, with genetic and acquired factors playing a role in its etiology [16]. When sufficient causes have been accumulated in a patient, the ‘thrombosis threshold’ is crossed and an event occurs (Fig. 2). After this first event, three things can happen to the ‘thrombotic potential’ (i.e. the combination of risk factors that was present in a person at the time of the thrombosis): It can decrease, it can stay the same, or it can increase. To understand this, an important distinction needs to be made between transient causes, for example, surgery, pregnancy and plaster cast, and persistent factors, which again can be further distinguished into fixed and varying ones. Examples of fixed persistent factors are age (or rather its underlying processes) and genetic thrombophilia, while factors as obesity and chronic comorbidity are persistent after the first event but may vary over time. Removal of transient factors will immediately lower the ‘thrombotic potential’ of a person to an extent that depends on their strength. This explains, for example, why recurrence risk is very low (<1%) in patients who developed their first event after surgery. This model also explains why recurrence risk remains high in patients in whom nothing changes, that is, in whom all (or most) of the causes of the first event were persistent. Lastly, in persons in whom the strength of a risk factor increases, such as for example in patients with cancer, where metastasis and chemotherapy may play additional roles over time, the ‘thrombotic potential’ is expected to increase (Table 1).

Table 1. categorization of risk factors for recurrence risk, as a tool for estimating their effect on recurrence risk
Type of risk factorNature of risk factorExpected effect on recurrence riskExamples of risk factorResults from studies: relative risk compared to unprovoked first VTReference
  1. a

    OC, oral contraceptive.

Transient causesStrongMuch reducedSurgery0.09 (95 CI, 0.03–0.29)17
IntermediateSomewhat reducedPregnancy0.6 (95 CI, 0.4–0.9)19
OC usea~0.713,18
Persistent causesFixedNone to marginalAge~1.013,14,31
Hereditary thrombophilia~1.515,21–23
Varying over timeIncreased or reduced, depending on direction of variation Cancer~2–434–38
Inflammatory bowel disease2.5 (95 CI, 1.4–4.2)42
Chronic kidney diseaseunknown 
Obesity~1.530, 43–45
New transient causesOccurring after first eventIncreasedOC usea2.8, (95 CI, 1.7 to 4.7)47
Surgery~25 13
PredictorsMarkers of thrombogenic processDistinguish low from high riskNegative D-dimer~ 0.448–50
Thrombin generation1.651
Residual VT2.4 (95 CI, 1.3–4.4)55
Prox vs distal DVT2 4
Mechanism unclearIncreasedMale sex~231,62
High levels of procoagulant factors6.7 (95 CI, 3.0–14.8)64
Figure 2.

Models of thrombosis risk. In each panel, the figure shows the thrombosis (black) potential of each risk factor present during an individual's life and the resultant thrombosis potential (red). From Ref. [16].

Low recurrence risk in presence of a transient risk factor for a first event

As described above, removal of a transient risk factor clearly lowers the thrombotic potential of an individual, thereby immediately lowering this person's risk of a second event. This has been described several times for most transient risk factors. Iorio et al. summarized these results in a comprehensive review of studies of recurrence in patients with a first episode of symptomatic venous thrombosis, provoked by a transient risk factor [17]. Using data from 11 studies on 2268 patients, they found an overall risk of recurrence of 3.3 per 100 patient-years. This risk was lowest in the subgroup of patients who had their first venous thrombosis provoked by surgery, that is, 1 per 100 patient-years (95% confidence interval (95CI) 0–2.3 per 100 patient-years) in the first 12 months after stopping anticoagulant therapy. In the first 2 years after stopping, this risk was 0.7 per 100 patient-years (95CI: 0%- 1.5 per 100 patient-years). In patients who had a transient non-surgical risk factor shortly before their first event, this risk was somewhat higher, at 5.8 per 100 patient-years (95CI: 3.2–8.3 per 100 patient-years) for the first year and 4.2 per 100 patient-years (95CI: 2.8–5.6 per 100 patient-years) for the first 2 years. The risk factors considered in this analysis consisted of trauma, travel, plaster cast, bedridden, pregnancy, post partum, hormone use, and non-surgical illness. The number of patients per risk factor in the studies included in this review (three in total) was too small to determine estimates for the risk factors separately.

On the effect of discontinuation of the oral contraceptive pill, some data are available. In the LETS study, the risk of recurrence was lower in women who had been using the pill at the time of their first event and stopped, than in those women who had not used the pill: 1.0 per 100 patient-years (95CI: 0.4–2.2 per 100 patient-years) vs. 1.6 per 100 patient-years (95CI: 0.9–3.0 per 100 patient-years) [13]. Similar results were found in the REVERSE study, in which 314 women with a first VT were followed for 2 years. At the time of the index event, 67 women were on estrogen therapy, none of whom continued after their thrombotic event. The annual recurrence risk in users was 3.8 per 100 patient-years (95CI: 0.8–6.8 per 100 patient-years), vs. 5.0 per 100 patient-years (95CI: 3.0–7.0 per 100 patient-years) in non-users [18].

In a study using administrative data from the State of California, the incidence of recurrent venous thrombosis was compared between 1085 women first diagnosed with acute venous thrombosis during pregnancy or puerperium and 7625 women who had a first-time venous thrombosis that was unrelated to pregnancy. The cumulative risk of recurrent venous thrombosis in the pregnancy-associated group was 5.8% after 2 years, as compared to a risk of 10.4% in the unprovoked group (adjusted Hazard Ratio 0.6; 95CI: 0.4–0.9) [19].

The results of these studies consistently show that the recurrence risk is reduced in subjects in whom a transient risk factor was present at the time of their first event, where the strength of the reduction in risk is related to the strength of the risk factor.

High recurrence risk in the absence of a transient risk factor

As follows from the above, the thrombotic potential remains high in subjects in whom the causes of the first event persist. Indeed, in their review, Iorio et al. also calculated a summarized risk in patients with unprovoked venous thrombosis who were enrolled in the studies they included and found this to be much higher, at a rate of 7.9 per 100 patient-years (95CI: 4.9–10.9 per 100 patient-years) in the first year, and a rate of 7.4 per 100 patient-years (95CI: 6.5–8.2 per 100 patient-years) in the first 2 years (as compared to 0.7 per 100 patient-years in the group of patients with a surgical factor and 4.2 per 100 patient-years in the group with a nonsurgical risk factor [17]). This group of patients with an unprovoked first event clearly forms the most challenging clinical problem, but also most can be gained here. For this, we need to distinguish between fixed persistent causes, varying persistent causes, new transient causes, and predictors of recurrence (Table 1).

Persistent causes

To distinguish causes from predictors, we need to define the concept of cause. One definition that is given by Rothman is that a cause of a disease is an event, condition, or characteristic that preceded the disease onset and that, had the event, condition, or characteristic been different in a specified way, the disease either would not have occurred at all or would not have occurred until some later time [20]. Several transient causes have been listed above. The group of persistent causes of a first event contains risk factors such as age, height, and genetic mutations that lead to a prothombotic state of the coagulation system, non-genetic thrombophilic defects such as antiphospholipid syndrome, chronic comorbidity such as cancer, inflammatory conditions or kidney disease, and obesity. In all people who had a first thrombosis in the absence of a transient risk factor, several of these persistent risk factors have to be present, as well as, most likely, other risk factors that we have not yet identified as risk factors for a first thrombosis. When we contrast these people to each other, no strong differences are expected to be found, as everyone will have a high risk, irrespective of the exact composition of the individual risk factors. This is indeed what we see happening in studies. For hereditary thrombophilia, for example, four systematic reviews combined the literature on risk of recurrence in patients with factor (F)V Leiden or the prothrombin G20210A mutation and found only marginally increased relative risks of about 1.5 [15, 21-23]. The risk of recurrence in patients with deficiencies of the natural coagulation inhibitors, such as antithrombin, protein C, or protein S, is mainly derived from family studies [24-27]. In one of the few population-based studies, the LETS study, recurrence risk related to coagulation inhibitor deficiency was found to be strongest of all thrombophilic defects, but this effect was still modest at a RR of 1.8 (95CI: 0.9–3.8) [3]. Similar results were found in the Cambridge study [28].

The same applies for increasing age which is most likely a composite of several states that increase thrombosis risk. If we consider age as a marker of other processes accumulating over time, this explains why age no longer has an effect on recurrence. These processes are currently not exactly known, but examples could be degeneration of the venous valves, increased immobility, increased incidence of comorbidities, or decline of muscle strength in the calves [29]. At any rate, it is likely that most of these processes will be persistent in nature, and therefore, the risk in these patients will not change, and age has no effect on recurrence risk. Although some studies found a slightly increased risk of recurrent venous thrombosis with increasing age [2, 30], most gave negative results [13, 14, 31] or even showed a lower risk in older patients [32, 33].

Persistent causes that vary over time

Some causes of a first event are persistent but vary over time in strength. This is the case for most comorbidity, such as cancer, which may be cured after some time or may relapse and need treatment that in itself is also thrombogenic. Also, a high BMI, although generally quite persistent over time, may change in either direction. Inflammatory diseases, such as inflammatory bowel disease, will also vary in intensity. Other comorbidity, such as kidney disease, may progress gradually. In our conceptual framework of the thrombosis potential, it may well be that the thrombotic potential in patients with progressive disease increases after their first event. It can therefore theoretically be expected that their recurrence risk is higher than in those in whom it stays the same or in whom it goes down. This is indeed the case for cancer, for which several studies describe a 2–4 fold increased risk of recurrence compared with non-cancer patients with a first venous thrombosis, both during and after anticoagulant therapy [34-38]. The antiphospholipid syndrome is also associated with a high risk of recurrence, even in patients on anticoagulation therapy [32, 39-41]. In a study in 116 patients with inflammatory bowel disease (IBD) with a history of first venous thrombosis, the probability of recurrence 5 years after discontinuation of anticoagulation therapy was higher among patients with IBD than in patients with a first venous thrombosis without IBD (33.4%; 95CI: 21.8–45.0 vs. 21.7%; 95CI: 18.8–24.6). After adjustment for potential confounders, the hazard ratio was 2.5; 95CI: 1.4–4.2) [42]. For other progressing comorbidity (e.g. chronic kidney disease), the risk of recurrence is not well described. Results of studies on the relation between obesity and recurrence risk are not all in agreement, but most studies show a slightly increased risk (RR~1.5) [30, 43-45].

New transient causes

In all patients with a first venous thrombosis, new situations can occur that will temporarily increase risk of recurrence, such as surgery, plaster cast, pregnancy, long-haul travel, or oral contraceptive use. Not many data are available on these risks; in the LETS study (474 patients with a first venous thrombosis of whom 90 had a recurrence), three subjects had a recurrence after surgery and three during or shortly after pregnancy, which led to a 25-fold increased recurrence risk for surgery compared with no surgery and an 8-fold increased risk of pregnancy compared with women who were not pregnant [13]. In a summary of the available literature, the authors from the most recent ACCP guidelines estimate that women with prior venous thrombosis have a risk of recurrent venous thrombosis of 2–8 per 100 pregnancies [46]. Also in the LETS study, 53 women continued hormonal contraception therapy, of whom 11 developed a recurrence, leading to a RR of about four compared with women not using hormonal contraception [13]. In a large follow-up study of 700 premenopausal women with a first venous thrombosis followed for a median of 7 years (MEGA study), we recently found that 210 women used hormonal contraceptives, mainly orally administered, at some point during follow-up. These women experienced 21 recurrent thrombotic events resulting in a rate of 3.9 per 100 woman-years, which was triple the rate of non-use during follow-up (HR 2.8, 95%CI: 1.7–4.7) [47].

Even though all of these results are limited due to small numbers, it appears clear that exposure to transient risk situations should be avoided as much as possible in patients who suffered from a venous thrombosis, which is relatively easy for oral contraceptive use, where good alternatives are available. During other, unavoidable situations like surgery, strict thromboprophylaxis is indicated.

Predictors of recurrence

Predictors are factors that are not necessarily a cause of disease, but nevertheless associated with an increased risk. For example, gray hair is a predictor of mortality, but not a cause. For recurrent venous thrombosis, we have seen that true causes actually predict rather poorly. The absence of a cause (presence of a transient factor at first event) in fact predicts best, as well as new or progressive risk factors (e.g. new oral contraceptive use, or progressive malignancy). Even though age and thrombophilia must be causes of a second event, they hardly predict its occurrence. Therefore, we need to look for other factors that are not really causes of thrombosis, but rather markers of a thrombogenic process or factors that do predict through a mechanism that we do not yet completely understand. Markers are factors like D-dimer levels, thrombin generation levels, extent of a residual thrombosis, and whether the first event was proximal or distal. Predictors where the mechanism is as yet unknown are factors like male sex and high levels of procoagulant factors, such as factor VIII.

A negative D-dimer test after stopping of anticoagulation is actually quite a good predictor of low risk of recurrence, as was shown in several studies [48-50]. Patients with a first unprovoked venous thrombosis and D-dimer concentrations of less than 250 ng mL−1 had a 60% lower recurrence rate than those with concentrations of 250 ng mL−1 or more [50]. Their cumulative incidence of recurrence was only 3.7% (95CI: 0.9–6.5%) 2 years after stopping anticoagulant therapy. Also for thrombin generation assays, several studies showed good distinctions between low- and high-risk patients. However, the tests are not yet standardized with respect to method and reagent, so further refinement of this method in prediction is still necessary [51-54]. The presence of residual thrombus after discontinuation of anticoagulant treatment also appears to distinguish low- and high-risk patients in several studies [55-59]. Illustrating that such presence is indeed a marker of a prothombotic state rather than a trigger of a new event is that in patients with residual thrombosis, the recurrences also occurred at sites remote from the original location [59]. Furthermore, also probably as a marker of a more prothrombotic state, it appears that the more proximal the initial clot, the higher the recurrence risk. Patients with an iliofemoral deep vein thrombosis have a higher risk of recurrence than those with femoral or popliteal deep vein thrombosis [60, 61], and patients with a proximal deep vein thrombosis are at a 2-fold higher risk than those with a distal deep vein thrombosis [4].

Male sex has been associated in several studies with a 2-fold increased risk of recurrence [31, 62]. This finding has not been explained so far and is the more puzzling considering that the risk of a first event is equal for both sexes [63]. This discrepancy can be explained in case a similar risk difference exists for a first event, but is masked through exposure to hormonal causes in women of fertile ages, in whom the risk is therefore ‘artificially’ high. In this situation, the difference in risk between men and women may be intrinsic, that is, due to genetic causes. When this is not the case, the discrepancy has to be explained through environmental factors that are different between men and women after a first event. Neither explanation has been studied so far. At any rate, male sex is a good predictor for recurrence.

High levels of coagulation factors FVIII and FIX have also been found to predict recurrence in some studies [64-66] while others did not [3]. These differences may be due to different cut-off levels between the studies. If such high levels do predict recurrence, the question is whether this is because these levels are a marker of some underlying prothombotic state, such as for example in cancer or other comorbidity, or because these levels are intrinsically high and therefore cause thrombosis directly.

Prediction models

Considering the above, useful predictors are the presence of a transient risk factor at the first event (low recurrence risk), male sex, presence of a progressing disease, and markers of a prothrombotic state. Furthermore, it may be that information on many other factors, which are in themselves not very predictive, may lead to more precise risk estimates when they are considered jointly. Three prediction models have so far been developed, that is, the HERDOO2 Rule, the Vienna prediction model, and the DASH Score [67-69]. Variables that are included in these models in various combinations are sex, D-dimer, BMI, age, post-thrombotic syndrome, location of first event, and hormone use during first venous thrombosis. As we saw above, not all of these variables are truly very predictive. Furthermore, in the important decision, with life-long implications, of deciding to continue or to stop anticoagulation, it is worthwhile to make the risk prediction as accurate as possible, that is, to include as many variables as necessary for the greatest discriminatory power. Therefore, further studies have to be aimed at development of extensive, integrated models that will accurately classify a person's risk, not only into low and high, but to a much more refined extent. Such models will need to be further validated and developed in several databases before they can be tested and used in practise. Until that time, theoretical considerations about the risk factors present in a certain individual can help in determining that person's individual recurrence risk.


Contrary to intuition, the risk profile of a recurrent event is entirely different from that of a first. Age, the strongest risk factor for a first venous thrombosis, is not associated with recurrence risk, just like hereditary thrombophilia. Male sex, while not associated with risk of a first event, doubles recurrence risk. These paradoxes can be explained when we understand etiology of a first event, how rates for first and second event cannot be directly compared, and how fixed risk factors cannot be predictors, while factors that are not causes can yet be predictors. Integrating all knowledge and combining the best predicting variables will ultimately lead to ways to predict an individual's recurrence risk and hence to decide on optimal further treatment.

Disclosure of Conflict of Interest

The authors state that they have no conflict of interest.