By continuing to browse this site you agree to us using cookies as described in About Cookies
Notice: Wiley Online Library will be unavailable on Saturday 7th Oct from 03.00 EDT / 08:00 BST / 12:30 IST / 15.00 SGT to 08.00 EDT / 13.00 BST / 17:30 IST / 20.00 SGT and Sunday 8th Oct from 03.00 EDT / 08:00 BST / 12:30 IST / 15.00 SGT to 06.00 EDT / 11.00 BST / 15:30 IST / 18.00 SGT for essential maintenance. Apologies for the inconvenience.
Willem M. Lijfering, Department of Clinical Epidemiology, Leiden University Medical Centre. PO Box 9600, 2300 RC Leiden, The Netherlands. E-mail: firstname.lastname@example.org
Epidemiological research throughout the last 50 years has provided the long list of risk factors for venous thrombosis that are known today. Although this has advanced our current understanding about the aetiology of thrombosis, it does not give us all the answers: many people have several of these risk factors but never develop thrombosis; others suffer from thrombosis but have none. In this review, we discuss how risk factors for venous thrombosis can be interpreted with use of several epidemiological models. We comment on how to explain why risk factors for first venous thrombosis differ from recurrent venous thrombosis, and use a causal model to better understand risk of first and recurrent venous thrombosis.
Venous thrombosis is a disease of blood coagulation that occurs in the veins, most often in the calf veins first, from where it may extend and cause deep vein thrombosis or pulmonary embolism (Cogo et al, 1993). The first described case of venous thrombosis that we know of dates back to the 13th century, when deep vein thrombosis was reported in the right leg of a 20-year-old man (Mannucci, 2002). Several hypotheses have been suggested through the centuries for understanding venous thrombotic disease, such as the idea of the ‘milk leg’ (Rosendaal, 2009). This hypothesis postulated that thrombosis was caused by milk accumulating in the leg. In the late 1700s this led to the first public-health advice to breast feed during puerperium as a prevention of milk leg (Deslandes, 1759). It was only in 1856 that the modern pathogenesis of venous thrombosis was proposed by Virchow. According to this triad (Virchow, 1856), thrombosis is a result of changes in the vessel wall, changes in the blood flow and changes in the blood composition. It is Virchow’s triad that provided the basis and inspiration for his successors to seek the causal factors for thrombosis that still hold today. In the 1940s and 1950s, important environmental risk factors for venous thrombosis, such as surgery and trauma, were identified by mere observation (Malone & Agutter, 2008). With increasing refinement of mathematical and laboratory methods, the detection of risk factors has increased exponentially, resulting in the long list of risk factors known today (Table I). The effect of these factors on the risk of venous thrombosis varies from about 20-fold (antithrombin deficiency) to only a little above unity [single nucleotide polymorphisms (SNPs)]. Although the existence of such a list may seem reassuring, it does not come close to giving us all the answers: many people have several of these risk factors but never develop thrombosis; others suffer from thrombosis but have none. Therefore, the challenge that we are facing is not to just add more risk factors to this list but rather to integrate them all in a causal model that allows us to understand how and when thrombotic disease develops. In this review we discuss the causal meaning of several known risk factors; we describe concepts of integration and finally apply a causal model to better understand recurrence risk.
Table I. Risk factors for first venous thrombosis.
References with risk estimates†
Discussed in the present review
In some cases, these were such early papers with unstable estimates, that we also included references to more recent papers with precise estimates.
SNPs, single nucleotide polymorphisms; TAFI, thrombin activatable fibrinolysis inhibitor; HIV, human immunodeficiency virus; CRP, C-reative protein; APC, activated protein C.
*Risk for first venous thrombosis compared to the general population.
†We have attempted to give a reference to the first paper from which a risk estimate could be inferred.
Anything that affects the incidence of disease occurrence is called a risk factor. This includes genetic, acquired, environmental risk factors or a mixture of such origins. The term ‘risk’ is generally used as a synonym for a probability or rate. In this sense, a risk factor is a characteristic in the presence of which the probability of developing disease is higher than in its absence. While risk factors are those antecedents that are considered to have a causal role in the development of disease occurrence, risk markers are biological traits that indicate a non-causal association with developing the disease. Although risk markers cannot explain pathophysiology, this does not mean that risk markers cannot be used for clinical purposes. For example, high C-reactive protein (CRP) levels are associated with an increased risk of vascular disease (Doggen et al, 2000). Although this association appears not to be causal, as individuals who have a genetic predisposition of high CRP are not at increased risk of vascular disease (Lawlor et al, 2008), they may be considered candidates for preventative measures because they are predicted to develop this disease more often than individuals with normal CRP levels.
Aetiological research, i.e. the science that deals with the causes or origin of disease, has by definition not a particular interest in risk markers. As the focus in this review is the understanding of venous thrombotic disease from an aetiological point of view, we will mainly discuss risk factors for venous thrombosis and not risk markers. When interpreting a risk factor, several aspects should be considered: its effect on absolute and relative risk, its strength, its prevalence and its causal nature.
Interpretation of a risk factor: absolute versus relative risk
There are two types of risk: absolute and relative risks. An individual’s absolute risk of venous thrombosis refers to the actual likelihood of getting the disease, and is not compared to any other risk. They may be expressed as rates (incidence rate) or percentages (cumulative incidence). For example, large population studies have shown that the incidence rate of venous thrombosis is around 1·5 per 1000 person-years and that an individual’s absolute lifetime risk of venous thrombosis is approximately one in nine or 11% (Heit et al, 2001; Naess et al, 2007). As shown in Fig 1, the absolute risk of venous thrombosis varies throughout life. On average, the incidence rate in childhood is 1 per 100 000 per year which increases exponentially to nearly 1 per 100 person-years in old age (Fig 1) (Naess et al, 2007). Reports that state that a factor increases or reduces the risk of venous thrombosis refer to rate differences or rate ratios, i.e. relative risk, as a comparison between effects of different exposures. (For convenience we use the term relative risk throughout this paper, irrespective of the relative risk type, such as odds ratio, rate ratio or hazard ratio. Explanation of how these different relative risk estimates are calculated and in what way these estimates may differ from one another can be looked up in epidemiological textbooks, e.g. Rothman, 2002.) A relative risk can tell us, for example, whether women have a higher risk of venous thrombosis than men. From Fig 1 it can be calculated that the risk of venous thrombosis in women compared to men is increased 1·2-fold, or that the relative risk is 1·2. In other words, the risk of venous thrombosis in women is 20% higher than in men. Relative risks are used particularly for understanding aetiology of disease, where the size of the relative risk reflects the strength of the association. The absolute risk usually determines the clinical implications. Consider air travel and venous thrombosis. Long haul flights are associated with a threefold increased risk of symptomatic venous thrombosis in the subsequent 8 weeks (Chandra et al, 2009), which indicates a moderate to strong association. However, as the overall absolute risk is 1 in 4600 over the time frame shortly after the flight (Kuipers et al, 2007a), a clinical implication, such as the use of temporary thromboprophylaxis, is probably not justified as the number of individuals needed to treat to prevent one case of venous thrombosis is too high, relative to the risk of serious side effects, such as major bleeding, induced by anticoagulants (Rosendaal, 2006). When we consider only severe pulmonary embolism directly after air travel this example becomes even more extreme: the relative risk of severe pulmonary embolism in passengers who flew more than 10 000 km compared to travellers who flew <5000 km was very high in a French study: a more than 400-fold increase (Lapostolle et al, 2001). However, the absolute risk of severe pulmonary embolism immediately after air travel was only 4·8 per million passenger arrivals in flights longer than 10 000 km (Lapostolle et al, 2001). Extrapolating these results to the Dutch population, we calculated that only one Dutch person per year travelling further than 10 000 km is seen with severe pulmonary embolism at an emergency department after disembarkment (Lijfering & van der Meer, 2009). Thus, although air travel and severe pulmonary embolism are probably causally related (as demonstrated by the high relative risk), the clinical implication of this finding seems limited due to the low absolute risk. For certain a-priori high risk groups, such as patients with previous venous thrombosis, this may be different. If such individuals have a high absolute risk after long haul flights compared to the overall absolute risk of venous thrombosis in long haul air travellers, they may benefit from temporary thromboprophylaxis. Information on absolute risk estimates for venous thrombosis after air travel in a-priori high risk groups is unfortunately limited (Kuipers et al, 2007b), so definite conclusions on this issue cannot be drawn yet (Geerts et al, 2008). Nevertheless, the prevalence of air travel is high [two billion passengers flew in the year 2005 (Annual Review of Civil Aviation, 2005)]. Therefore, even a small increase in risk will have a major impact on the number of events. Consider that long distance air travel (more than 4 h) is associated with an absolute risk of venous thrombosis of 1 in 4600 passengers, and that 32% of all flights are long distance flights (Kuipers et al, 2007a). From these numbers it can be calculated that more than 140 000 air travellers per year experience venous thrombosis that is related to long distance air travel. Of note, this latter number merely serves as an example that when interpreting a risk, not only the strength of association (measured by a relative risk) or clinical impact (measured with an absolute risk), but also the prevalence of a certain exposure (in this case air travel) can have an impact on health status, in this case venous thrombotic disease, even when relative or absolute risks are low.
Interpretation of a risk factor: influence of confounding
A prerequisite that should be taken into account when interpreting an association is its causal nature. A confounder is a variable that is associated with the exposure, affects the outcome, but acts not as an intermediate link in the chain between exposure and outcome. The result of the presence of a confounder is that the effect of the confounder will be incorrectly attributed to the risk factor of interest. Therefore, when interpreting a risk factor, it should always be ascertained whether sufficient adjustment for confounders has taken place. The difference between confounding and bias is that risks that are due to confounding are a result of indirect associations, and are therefore real but not causal, while biased results are spurious associations. It is essential to know something about pathophysiology of disease when one deals with confounding. Otherwise adjustments for ‘confounders’ that actually are not confounders can lead to incorrect interpretation of data. One approach that can help in determining whether a certain exposure is a potential confounder is by presenting the data graphically, such as a directed acyclic graph (DAG) (Greenland et al, 1999). Confounding is often easier to understand from examples than from definitions. Therefore, we have given three examples that are related to confounding by using DAGs (Fig 2).
Suppose, for example, that one would like to investigate whether subjects with high levels of clotting factor VIII are at increased risk of venous thrombosis. As can be appreciated from Fig 2A, the effect of exposure (high factor VIII) on occurrence of disease (venous thrombosis) can be confounded by age (third variable), as increasing age is both associated with venous thrombosis (Naess et al, 2007) and increase in factor VIII levels (Bank et al, 2007). Indeed, one study showed that the relative risk of recurrent venous thrombosis in persons with high factor VIII levels was 3·4 before adjustment for age and 2·9 after adjustment for age compared to persons with normal factor VIII levels (Kyrle et al, 2000). Thus, age partly explained the effect of high factor VIII on venous thrombotic risk in that study.
An investigator may wonder if a genetic variant, like for example hereditary antithrombin deficiency, is a confounder for the association between high factor VIII levels and risk of venous thrombosis, As can be inferred from Fig 2B, hereditary antithrombin deficiency is related to venous thrombosis (Lijfering et al, 2009a) but is not related with the exposure, levels of factor VIII (Mahmoodi et al, 2008a). Therefore it is not a confounder but an additional risk factor. Adjustment would be redundant but not change estimates.
If an investigator would like to estimate the relative risk of venous thrombosis for women who use oral contraceptives as compared to non-users, adjusting for a variable like high factor VIII would lead to incorrect results. High factor VIII levels are an intermediate in the chain of causation between exposure (oral contraceptive use), resulting in an increase of procoagulant factors (like factor VIII), that finally results in venous thrombosis (Fig 2C) (van Hylckama Vlieg & Rosendaal, 2003). Adjustment here would lead to an attenuation of the effect measure and thus to an underestimation of the true relationship.
The question of whether a factor is the cause or the consequence of disease can be difficult to answer, yet it is pivotal, as only treatment or prevention of causal factors will prevent the outcome of disease. A way to be more certain that a putative risk factor is not a result of the disease is in study designs where the exposure was determined before the disease occurred. This is the case in follow-up (or cohort) studies, where the incidence of disease is compared in groups according to the presence or absence of an exposure assessed before the beginning of follow-up. It may also happen in case-control studies (in so-called ‘nested’ designs), where cases and controls are taken from a cohort in which questionnaires and blood sampling took place at baseline in the cohort, or in case-control studies where exposure status had been recorded routinely for another purpose. However, in most conventional case-controls studies the data are collected in the cases after disease onset. One may argue that genetic markers predispose individuals to exposure before the event. However, abnormal laboratory markers may be difficult to interpret. As an example of a case-control studies where temporality was difficult to infer: in the Leiden Thrombophilia Study (LETS) an increased risk of venous thrombosis was reported for persons with high levels of the inflammatory marker interleukin 8 (van Aken et al, 2002). As samples were collected after the event, this finding could have been either a cause or a result of the thrombosis (e.g. by post-thrombotic syndrome and inflammation) (Roumen-Klappe et al, 2009). A nested case-control study, in which blood samples were collected prior to the onset of disease, on the relationship between venous thrombosis and an altered inflammatory profile yielded negative findings on increased interleukin 8 levels on venous thrombotic risk (Christiansen et al, 2006), which suggests that they are more likely to be a result of venous thrombotic disease than a cause.
Bias is a misrepresentation of data in a study that arises from the way that the data are collected. Several taxonomies exist for classification of bias; Sackett’s landmark compilation, for example, included 35 different biases (Sackett, 1979). Most of these biases can be termed under the denominator ‘selection bias’ (particularly in case-control studies, when the control group is not randomly sampled) or ‘information bias’ (in all types of study design when data on exposure are not collected independently from the outcome, or vice versa). Almost all studies have built-in bias. The challenge for investigators, reviewers and readers is to assess the validity of results. As an example of selection bias: when an investigator would like to perform a case-control study on the risk of venous thrombosis associated with air pollution, two groups of persons are compared: those with venous thrombotic disease (cases) and a sample of the general population who do not have venous thrombosis (controls). Selecting neighbours of cases as controls will obviously result in a biased null finding as both cases and controls are similarly exposed to air pollution. In this example, selection bias was a result of matching controls on a factor closely related to the exposure (air pollution). The simplest way to overcome this problem is not to match in case-control studies, and take a random sample of the source population (the population from which the cases originated) as a control group. However, if one would like to study the effect of a genetic defect on venous thrombotic disease, selecting neighbours of cases as controls is appropriate as genes are not related to living in a certain neighbourhood. Likewise, in a study on ABO blood group and thrombosis in men, newborn female babies would be good controls.
Integration of several risk factors into one causal model
Why certain exposures yield higher risks for disease occurrence than others is an intriguing question. The thrombosis potential model, first described in 1999, may serve well to illuminate such findings (Rosendaal, 1999). The idea is that an individual is at risk for venous thrombosis throughout life, which is reflected in the ‘thrombosis potential’ and that each risk factor contributes to the potential. Only when the combination of thrombosis risk factors reach a certain potential, will venous thrombosis occur (crossing of the thrombosis threshold). A simple worked out example is given in Fig 3. The black lines in the figure represent the thrombosis potential of each separate risk factor for a particular individual, and the grey line the thrombosis threshold. The horizontal axis shows time (i.e. the age of the individual). As the thrombosis potential is age-dependent, based on the observation that the incidence of thrombosis increases with age (Naess et al, 2007), the black line increases through time. Consider a person who has F5 R506Q (factor V Leiden; person A) and a person with blood group non-O (person B). F5 R506Q carriers have a 5–7 fold increased risk of venous thrombosis (Rosendaal et al, 1995), while blood group non-O is associated with a 1·5-fold increased risk (Jick et al, 1969; Koster et al, 1995). Thus, person A has a larger inherited thrombosis potential in the example than person B. They are of similar age, i.e. 55 years old. At this age, person A encounters a short period of immobilization due to a strained ankle. In this example, this short period of a high thrombosis potential exceeds the thrombosis threshold and consequently leads to deep vein thrombosis. For person B, however, a similar acquired risk factor has not led to a thrombosis potential exceeding the thrombosis threshold. Therefore, this person does not develop the disease at this age. In this example, risks are considered constant over categories of other risk factors (i.e. the lines ‘add up’), which is a simplification, since the actual joint effect of two risk factors may differ from the expected joint effect, which is known as effect modification or interaction. Another simplification is that ‘increasing age’ is used as a container concept here, that is a mix of unknown and known risk factors that either become stronger with age or become more prevalent with age.
It is important to note that the thrombosis potential model is an explanatory model. The importance of the model is mainly as a tool to better understand how life time exposure to several risk factors for thrombosis may lead to occurrence of the disease in an individual. This is of great benefit as it translates epidemiological research to the clinic. Because epidemiology is based on two principles: group comparison and population thinking (Morabia, 2004), this approach seems to hamper usefulness for medical doctors, as population thinking appears to conflict with a fundamental principle of medicine where every patient is unique. The use of the thrombosis potential model may be a pragmatic solution to this tension, as this epidemiological model is able to illustrate how results of group comparisons apply to an individual’s risk. For example, with the thrombosis potential model in mind, it becomes obvious why strong genetic risk factors will more often lead to disease than weaker genetic risk factors, and also why acquired risk factors may more often lead to disease in carriers of strong genetic risk factors: the thrombosis potential of the complete genetic make-up lies closer to the thrombosis threshold level when a person has a strong inherited risk factor compared to a person who has a weaker inherited risk factor, and additional, acquired factors will often push the potential past the threshold. This theoretical model works well in practice. Consider the issue of genetic variation a bit further: in a previous study it was shown that individuals with inherited antithrombin, protein C or protein S deficiency had a high risk of venous thrombosis (Lijfering et al, 2009a). They were much younger when they encountered venous thrombotic disease (median age 29 years), compared to individuals with weaker risk factors like F5 R506Q or F2 G20210A (prothrombin G20210A; median age 40 years), while the median age of venous thrombosis in the community was 62 years (Heit et al, 2001). Apparently, the thrombosis potential for the complete genetic make-up lies progressively closer to the threshold for those with stronger genetic risk factors, so an increase in age in the presence of antithrombin, protein C or protein S deficiency sooner leads to a crossing of the thrombosis threshold than in the presence of F5 R506Q or F2 G20210A, or the absence of all of these. In the same study, individuals with genetic defects encountered a venous thrombotic event at a younger age when environmental risk factors, like oral contraceptive use, immobilization or trauma, were present while older persons usually encountered unprovoked venous thrombotic events. For the latter group increasing age was the likely trigger for venous thrombotic onset (the necessary part to cross the thrombosis threshold) (Lijfering et al, 2009a). This concept of multicausality is illustrated by studies that describe an increasing risk with an increasing number of risk factors present (Rosendaal, 1999; Vossen et al, 2005; Brouwer et al, 2006; van Vlijmen et al, 2007; Kuipers et al, 2009; Lijfering et al, 2009a). For example, in a study of thrombophilic families, the incidence rate of first venous thrombosis in relatives without a coagulation abnormality was 0·05 per 100 person-years. In relatives with a single F2 G20210A mutation, F5 R506Q or elevated level of factor VIII, incidence rate ranged from 0·2 to 0·3 per 100 person-years and increased to 0·6–1·5 per 100 person-years when it was combined with two or more coagulation abnormalities (Lijfering et al, 2009a). A similar effect of multicausality was found in the Multiple Environmental and Genetic Assessment (MEGA) study, where risk of venous thrombosis in travellers was obtained (Kuipers et al, 2009). When one coagulation abnormality was present, the risk of venous thrombosis was fourfold increased compared with travellers without a coagulation abnormality. This risk was 10-fold increased in travellers who had two or more clotting abnormalities.
Differences in risk of first versus recurrent venous thrombosis explained with the thrombosis potential model
Incidences and risk profiles for a first venous thrombosis are very different from those for recurrent venous thrombosis (Zhu et al, 2009). The absolute risk for recurrence is much higher (2–5% per year) than for first venous thrombosis (0·1–0·2% per year) (Hansson et al, 2000; Baglin et al, 2003; Christiansen et al, 2005; Naess et al, 2007; Prandoni et al, 2007). But obviously, the populations at risk differ, and the risk profile is also quite different: Age, an important risk factor for first venous thrombosis, seems to play little role in the risk of recurrence (Eischer et al, 2009). Male sex, which was weakly associated with first venous thrombosis in some studies, (Andreou et al, 2008) but not in others (Naess et al, 2007), appears to strongly increase the risk of recurrence (nearly fourfold higher than in women) (Kyrle et al, 2004; Christiansen et al, 2005; McRae et al, 2006). Results from several cohort studies suggest that the role of the known thrombophilias is limited as the recurrence risk is hardly increased in patients with coagulation abnormalities as compared to those without, with a possible exception for rare deficiencies such as heritable natural anticoagulant deficiencies or homozygous F5 R506Q (Baglin et al, 2003; Christiansen et al, 2005; Middeldorp & van Hylckama Vlieg, 2008; Lijfering et al, 2009a). Environmental risk factors seem to play a more important role in predicting a recurrent event than thrombophilia, as several reports showed higher recurrence rates after a first unprovoked event than after a first provoked event with relative risks of 2–3 (Baglin et al, 2003; Christiansen et al, 2005; Prandoni et al, 2007). Much of these differences can be explained by the thrombosis potential model. For example, the effect of thrombophilia: every individual who suffered from a first thrombosis has been demonstrated to have a sufficiently thrombosis-prone genetic makeup to develop it. So, when we compare, in these people, risks for recurrence among those who do and those who do not carry thrombophilic defects, it should not be surprising that relative risks are close to unity: these people have the same genetic thrombosis potential, although in some it is made up of as yet unknown genetic variants. The same applies for age: every patient with thrombosis has reached an age at which, given the potential of all other individual risk factors thrombosis can develop. In other words, they developed thrombosis at the age which was sufficient to reach the thrombosis threshold, and afterwards, age will no longer have an effect on recurrence.
Consider the difference in recurrence risk after provoked or unprovoked first events: nearly 50% of first venous thrombotic events are associated with an environmental risk factor and are classified as ‘provoked’ (Naess et al, 2007). These environmental risk factors have led to the thrombosis potential exceeding the thrombosis threshold for the first event. They are often transient and will subsequently cease to exist or be removed (e.g. cessation of oral contraceptive use, surgery, travel, plaster cast). The thrombosis potential will drop because of this and will not readily exceed the thrombosis threshold level again. In contrast, a person with an unprovoked first venous thrombosis surpassed a thrombosis threshold level for first venous thrombosis unrelated to any known environmental risk factor. As we do not know them, they can not be removed, and many of them may be non-transient environmental, or unknown genetic variants. Therefore, the thrombosis potential will not drop sharply after the event and this individual will readily cross the thrombosis threshold again. It follows from this reasoning that research should be focussed on identifying new ‘preventable’ risk factors, as the most efficient way to reduce the considerable risk of a recurrent event.
According to the thrombosis potential model, differences in risk of recurrence between men and women are more likely to be based on differences in transient environmental risk factors than in genetic risk factors. If sex-specific genetic variation explained the difference, men would also have a higher risk of first events. Obvious transient environmental sex-specific risk factors are pregnancy and oral contraceptive use. As oral contraceptives are discouraged after first venous thrombosis and thromboprophylaxis is often recommended in women during pregnancy or puerperium after a first event (De Stefano et al, 2006; Bates et al, 2008), women may have a lower thrombosis potential level for recurrence than men where environmental risk factors are less often at play on first venous thrombotic risk (Lijfering et al, 2009b). Some studies indeed suggested that the lower risk of recurrent venous thrombosis in women could be explained by a reduced rate of recurrence after venous thrombosis associated with oral contraceptive use or pregnancy (Cushman et al, 2006; Pengo & Prandoni, 2006; Lijfering et al, 2009b). However, a meta-analysis of most of these studies concluded that the lower risk of recurrent venous thrombosis in women could not to be fully accounted for by a reduced rate of recurrence after venous thrombosis associated with oral contraceptive use or pregnancy (McRae et al, 2006). Whether male sex in itself is or is not a potential factor in the development for recurrent venous thrombosis is therefore still unclear.
In this review, we discussed how risk factors for venous thrombosis could be interpreted with use of several epidemiological models. We commented on how to explain why risk factors for first venous thrombosis differ from recurrent venous thrombosis. A great benefit of the thrombosis potential model is that it integrates a plurality of risk factors into one causal model. This accords well with the fundamental principles of medicine where every individual is unique.