Methods
Material and methods of this long-term cohort study has been described in detail in earlier publications [
4‐
6]. In brief, we used a cohort of 4337 young women (18–55 years) in Bavaria (Germany) who had at least one follow-up.
We collected data from demographics, reproductive life, lifestyle pattern, conditions/diseases, and particularly potential risk factors for VTE through a questionnaire for self-administration. Whenever possible, time-related information was documented. Using this method we were able to set up the common starting point for the cohort as 1993. Telephone enquiries were made to supplement, clarify and verify the data in the questionnaires.
The primary source for the data on VTE was the follow-up questionnaire (self-reported VTE or symptoms potentially compatible with VTE). This information was completed by telephone interviews with the woman and with the treating physician. All available information about diagnostic and therapeutic measures taken was recorded. Clinical data and/or invasive or non-invasive diagnostic procedures were assigned to one of the following categories of likelihood of a VTE:
Definite VTE
Unequivocal positive finding in at least one imaging test, e.g., phlebography or duplex sonography for deep venous thrombosis (DVT); pulmonary angiogram, VQ scan, spiral computed tomography (spiral CT) for pulmonary embolism (PE).
Probable VTE
Typical clinical symptoms for VTE without unequivocal imaging test but positive findings in other diagnostic tests (e.g. Doppler US or plethysmography) and subsequent specific therapy over a longer period (low-dose heparin or other anticoagulants).
Possible VTE
Typical clinical symptoms for VTE and unknown or equivocal result on imaging, only suspicion of VTE suggested by a non-imaging diagnostic tests (such as Doppler US, plethysmography, ECG, blood gas analysis or others for PE) and no subsequent specific therapy (e.g., only short-term low-dose heparin and bandage).
Potential VTE
Typical clinical symptoms for VTE without further diagnostic tests or negative results or diagnostics unknown. Unspecific therapy – but nonetheless the treating physician maintained the diagnosis VTE based on clinical findings.
All possible and potential "VTE cases" were excluded from the analyses in this paper because of diagnostic uncertainty, i.e. lacking information whether they should be better classified as VTE cases or non-cases.
Women with a history of cancer or with known antiphospolipid syndrome were not in our follow-up study.
Laboratory methods
After having given informed consent the women included in this study gave a blood sample at one time point during the observational period (1996/97). An independent ethics committee approved all study related activities.
Whole blood samples were obtained from resting subjects. Blood was put into tubes with trisodium citrate. Plasma was prepared soon after venipuncture by centrifugation for 15 minutes with 3000 to 4000 / min at room temperature and stored at -20°C.
Protein C and antithrombin activities in plasma were measured by chromogenic substrate assays (Dade Behring, Marburg, Germany). For "antithrombin" the activity against factor IIa was determined, Protein C activity was measured after activation of the proenzyme by snake venom. Plasma activities are given as percentage (units/dl) of pooled human normal plasma.
Genomic DNA was isolated by mean of QIAmp
® DNA Blood Kit (Qiagen) according to the manufacturer's instructions. The genetic polymorphisms Factor V R506Q (G1691A), the prothrombin promoter G2010A and the 5-, 10-methylenetetrahydrofolate reductase (MTHFR) A223V (C677T) were determined using a multiplex PCR with allele-specific primers slightly modifying a previously described method [
7].
All blood tests were performed blinded, i.e. the investigators had no clinical information, and had no access to the clinical database.
Method of data analysis
Due to the importance of the temporal relationship the database was structured to accommodate both concurrent as well as time-dependent variables. Concurrent variables are variables, which describe the woman's status at the time of questionnaire response, whereas the outcome variable (VTE) is time-dependent. While concurrent variables were held in a fixed data set, a periodic data set containing information on lifetime exposures and the occurrence of VTE events along a time axis was created for the time-dependent variables of each participant, using months as a unit of measurement. The exposures of interest in this publication, such as VTE risk factors including genetic markers, refer to the baseline time point.
Some of the variables in the database (age, BMI, Protein C, AT) were continuous. These variables were dichotomized in order to define a categorical exposure status (exposed – non-exposed) for the analyses based on incidence or logistic regression. We arbitrarily separated the continuum in two roughly equal intervals such as age under/over 30 or BMI under/over 25 in order to have sufficient case numbers for analyses with further stratification. For protein C and AT we used the 5th percentile (lower 5% of the distribution in non-cases) as cut off point. This limit was considered as usual definition for deficiency and clinically relevant [
8].
All analyses concerning the occurrence of VTE events over time were performed by adding up individual observation time (1993 until the last contact) for different exposure-cohorts and in total. The incidence rate of VTE was calculated per 10,000 women years of observation (WY).
The predictive model was developed using the discriminant function analysis technique [
9]. This technique permits to determine which of the clinical and laboratory data discriminate best between two groups: future VTE cases vs. non-cases. In other words, this multivariate analysis determines which variables at baseline are the best predictors of a future event.
We multivariately ranked the predictive power of the variables with suspected effect on occurrence of VTE (i.e., possible or established risk factors). Other available information in the database or information not known at baseline (e.g. later occurring conditions like surgery or longhaul flights; see discussion) were not included into the model since they cannot contribute to a predictive model (i.e. the setting for the application of the results of this study is women consulted by a physician independent of an acute VTE event). Technically, we used stepwise discriminant analysis with forward inclusion or backward elimination of variables. The p-value for entry into the model was 0.49 and for removal 0.50.
The p-value of the parameter provided by the discriminant analysis at the last step determines its rank. Variables with the smallest p-value get the highest rank. This permits the comparison of predictive power of potential risk factors – documented at baseline (1993) – for later occurring VTEs.
This technique permits to classify persons by the discriminant function value (separated by the case status). The true incidence of new VTE cases was determined in strata of persons with different pattern of risk profile to characterize groups with lower or higher absolute VTE risk according their risk profile at entry. We used for this analysis those variables that depicted the highest 5 ranks in the stepwise discriminant analysis.
All analyses were performed with the statistical packages SPSS 10.2, SAS 8.2 or STATA 8.2.
Discussion
To our knowledge, long-term, community-based cohort studies with the aim to evaluate or compare the predictive power of clinical as well as genetic risk markers for VTE are lacking. Most studies related to VTE risk factors were restricted to clinically available markers such as age, BMI, previous VTE, family history, or acute factors (immobilization, surgery, accidents, pregnancy, and also hormone use) and usually based on clinical observations or case-control studies or studies in administrative databases, and also a few cohort studies (overview about incidence and risk factor studies in [
10,
11]). One recent publication of a historic cohort [
12] assessed carefully the impact of clinical risk factors for the prevalence of VTE and determined relative VTE risk estimates. Cohort studies in the population rarely included/reported genetic markers for thrombophilia and acquired risk factors, except the Physicians Health Study for example – however only for males over 40 years of age [
13]. An important recently published cohort study in Denmark analyzed specifically the incidence of VTE and compared carriers of FVL compared with non-carriers [
14].
Other studies with focus on markers for hereditary thrombophilia were performed in patients (e.g. in anticoagulation clinics), in relatives of carriers of genetic mutations but not in the "normal female population" [
15‐
18]. In addition, the evaluation of the importance of genetic markers for VTE risk does rarely consider the impact of clinically available risk factors and the design is often restricted to case-control studies. Overall different study designs and restricted views may come to similar conclusions in an ideal world, but not necessarily.
Our BATER cohort study covers more than 4,000 cohort members with a fairly long observation period (1993 – 2003), translating into over 32,000 WYs. Thirty-four VTE cases, classified as definite or probable, occurred within this period. This is equivalent to an incidence of about 10 per 10,000 WYs. It is important to realize that we put great effort on the detection of potential cases and – even more important – we included all definite and probable cases, whereas most reported incidence rates refer only to definite and so-called "idiopathic VTE", i.e. most reported rates excluded all cases that occurred in temporal relationship to possible "acute" causes such as pregnancy/delivery, surgery, and immobilization. Idiopathic VTEs, however, reflect only a part of all confirmed VTE cases [
17]. We found in our cohort study roughly 50% so-called "idiopathic" VTE cases, and the other 50% of cases had a previous VTE in the past, pregnancy, delivery, surgery, accident, or immobilization/long bed-rest shortly prior to the VTE event. Thus, the incidence of "idiopathic VTE" observed in this study would be 5 per 10,000 WYs and thereby likely to be in the same range as other reported incidence rates in the general population. The incidence estimates for definite VTE ranges between 1 to 6 per 10
4 WYs in OC non-users and 2 to 10 per 10
4 WYs in OC users [
10]. Older studies depicted almost always-higher incidence rates than more recently performed studies (see overview in [
10]). A recent systematic review [
11] came to a pooled incidence of definite VTE for the general population of 5 per 10,000 person years, similar in males and females, and found that around 40% of VTE cases were "idiopathic". A large cohort study found a similar incidence rate in males aged 40–49 years: 4.7/10
4 person-years [
13].
The objective of this study was to provide a simple algorithm for medical practice to predict the future VTE risk with a simple scheme based on usually available information, i.e. to discriminate women with virtually no VTE risk in the foreseeable future from those at a high absolute risk to suffer from VTE. Incidence rates associated with different clinical and genetic factors will be published separately [
6].
It is a limitation of this long-term cohort study, however, that the number of confirmed (definitive and probable), incident VTE cases was still too small in absolute numbers (n = 34). In other words, some sub-cohorts with certain combinations of risk factors did not contain one single new VTE case. The consequence was that the number of subgroups at risk was minimized to the extent possible to make it a feasible tool for the practice. In so far, the results and conclusions should be considered as rough but the best we can possibly do at this stage, i.e. future analyses will benefit from an improved point of departure (more cases, longer observation). Another limitation is that we did not have the chance yet to test the validity of the model in another, independent cohort. This is a task for the future. Therefore we focused this paper on a simple scheme with a rough classification of the future VTE risk.
The interested (or worried) women and her treating physician might like to know (or to get confirmation) whether the future VTE risk is higher than "normal" (no/low risk). This information could have an impact on further medical surveillance, especial counseling, proposals as how to reduce of changeable risk factors or on suggestions for preventive measures under certain circumstances and – of course- with respect to the compliance regarding preventive measures.
Using stepwise discriminant analysis the rank order of 12 (11) clinical or laboratory data at baseline (1993) was multivariately determined concerning the power to predict future VTEs. This was the information needed to select a minimal set of parameter combinations to build a "VTE prediction model". Finally we ended up with a model covering the five variables with highest ranking (impact) regarding predictive importance for future VTE only: history of previous VTE, family history of VTE, higher age, higher body mass index, and carrier of FVL or PTM. The decision to form the composite genetic marker "FVL or PTM" was guided by the small numbers of cases who were carrier of this two mutations and the low predictive power of all other lab parameters we had in the data set (see table
2).
Four levels of future VTE risk were arbitrarily defined-based on a steeply increasing absolute VTE risk:
No/low risk (4 per 10
4 WY),
moderate (12/10
4 WY),
high risk (47/10
4 WY), or
very high (171/10
4 WY). The low-risk group was chosen to reflect an assumed VTE risk of the normal population (see above), the group with "moderate risk" because VTE risk over 10/10
4 is indicative for an increased risk, and the "high and very high risk" groups are clearly out of the normal range. One should also consider in this context, that these cut-off points reflect an average risk with an assumed variation within these groups – as can be seen in the scheme of a decision tree (figure
1).
In accordance with clinical experience the overwhelming majority (61%) depicts a low risk of a future VTE. Only a minority of 6% and 0.9% is facing a high or very high VTE risk. The women who fall into the two highest risk categories have a previous own history of VTE or a positive VTE family history, have a higher BMI or a genetic mutation (FVL or PTM). Even though, the contribution of genetic appears to be limited. Using an analysis based only on clinically available data, i.e., without use of the information about lab parameters, we arrived at very similar three risk categories with almost identical absolute VTE risk (data not shown separately but are part of Figure
1).
Another point for discussion is the impression suggested by figure
1 that persons with previous VTE do not require genetic testing because they are in the "high risk" category without any further considerations. The risk might well be different for persons with/without inherited risk (family history), younger/higher age, or overweight. This however we cannot further disentangle because we are lacking new VTE events particularly in this high-risk group of our study. Thus, the conclusions are rather crude as discussed before and require clinical experience for the interpretation of individual cases. The need for genetic testing depends on the judgment in a specific clinical situation and the usefulness of this additional information for the physician and/or the patient (family).
Decisions based on clinical variables about preventive measures will be made in any case – even if no genetic information is available. The possible approaches are a matter of a current controversy in the literature [
15,
20,
21]. Clinical reports point often towards a high VTE recurrence rate in patients with previous VTE [
22], but despite being the "best" single predictor we found this phenomenon only in 4 of our 34 incident VTE cases.
The predictor variables used in our model seem to be plausible and consistent with the clinical experience: History of previous VTE, age and obesity are indeed important clinical information for the VTE risk assessment, and also genetic marker were discussed as predictors of a future VTE. These are also risk parameter that are commonly used when recommending preventive measures in situations like long-haul flights, immobilization (such as accidents, surgery) and are also labelled as risk factor in drugs containing sexual hormones (e.g., oral contraceptives or hormone therapy).
We assume that physicians will appreciate these results as a possibility to double-check if their own decision are supported by evidence coming from this large cohort study or may even alter their decision. At least the proposed model contains a reassuring element. It should be underlined that these algorithms do not obviate the need of weighing individual risks and benefits.
We conclude from our observation that the prediction of future VTEs can well be done on clinical data alone – at least until better genetic markers are established. In other words, well-established genetic parameters alone are relatively weak long-term risk factors, the occurrence of VTE requires interaction of both inherited and acquired risk factors [
23]. Results of several recent studies support arguments against the possibility that testing for thrombophilia could help to better predict future VTEs [
15,
20,
21]. Nonetheless one can argue that genetic testing in families with significant family history of VTE or previous experience of a VTE might well give additional information for clinical decisions and may increase efforts to comply with preventive measures. The limitation of our study is that we cannot further divide the risk spectrum in these sub-groups due to small numbers of new events or too short total WY of observation. In any case, when a genetic test is recommended, the physician should know how a positive test would influence his/her clinical judgment [
15]. These early results of our cohort study contribute to this interpretation or decision-making, respectively.
Forecast of VTE risk cannot be based on genetic characteristics alone but only in combination with important clinical data (acquired risk information). Genetic markers play obviously a limited role in the long-term prediction of VTE – at least in the age group under 50. Genetic markers together with these "personal characteristics" constitute the disposition. Family history of cardiovascular events, specifically venous events have to be taken into account. As described before and confirmed by our data the probability whether the disposition translates into an event is obviously more influenced by "personal characteristics" such as higher age, or higher BMI. However, there are obviously other important, more acutely affecting environmental factors such as immobilization, surgery, accidents, and treatment with drugs that influence coagulation. The latter factors can be used to reduce the risk as estimated by the model (or own clinical experience).
Another issue for discussion is the validity of our calculated incidence rates: The lowest VTE incidence level observed in this model was 4 per 10,000 WYs. Due to an active search for findings compatible with the diagnosis of VTE, the inclusion of definite and probable diagnosis as well as of so-called "non-idiopathic VTEs", the incidence figures were expected to be higher than in normal "medical statistics" or administrative databases as discussed above. The equivalent incidence rate for only definite and idiopathic VTE could be expected to be approximately 2 per 10
4 per year and therefore very low for women who were using oral contraceptives as the average population. We like to stress that our study was rigorous in documenting the VTE diagnosis. We conclude that the data can be generalized for the female population in the fertile age range. In analogy, the incidence might be compared with results of a prospective, community-based cohort study [
13] that found in males aged 40–49 years a VTE incidence rate of 2.7 primary VTEs cases per 10
4 person-years (equal to idiopathic: no previous VTE history, no cancer, no surgery or trauma).
The influence of other, more acutely acting risk modifiers – such as immobilization, surgery, long-haul flights, and use of drugs (e.g. OCs and other hormones) was intentionally excluded from this model. Only parameters that were available at baseline and likely to affect the long-term development were eligible for this prognostic model. We saw no possibility to introduce parameters in the model that may or may not operate later, shorter or longer during the observational period. In other words, only long-term characteristics at baseline (both clinical and genetic variables) were initially included into the modeling. Other influential risk factors or preventive measures have to be considered when discussing activities to reduce a predicted increased risk in the medical practice. It was not the aim of the study and data are neither available to test the effect of preventive measures nor the effect of additional risk factors in the immediate period before the event occurred. This would require another study design and a separate study with sufficient power for such questions.
The variables selected for the model fit the expectations of the skilled clinician. The model was developed to assist physicians- we hope for feedback from medical practice.
This "prognostic model" seems to be worthwhile to be tested in clinical practice. There is a minority of women that would need additional genetic testing, intense counseling, suggestions for risk reduction (if possible), and efforts to prevent avoidable risk situations (e.g. treatment with OCs or hormones) or to take other appropriate preventive measures in situation of an acute risk (immobilization, surgery, long-haul flights and others), e.g. compression stockings/ heparin. Several other options to reduce the VTE risk profile are principally available, but sometimes not easy to achieve (e.g. reduction of BMI). Prevention of VTE in medical practice can be improved if the main risk factors are known (importance to document medical history and established risk factors), but also knowledge of their relative importance in the risk-network as well as of interactions with environmental factors. The predictive models discussed in this paper may assist doctors to pay particular attention to labeling prior to prescription (e.g. oral contraceptives or hormones) and to use convincing evidence-based data when counseling women of a predicted higher risk.
We abstained – at the current stage – from the temptation to use a complex equation to calculate an apparently exact risk for the individual person (e.g., using a "risk calculator") because it suggests inadequate accuracy and we rather prefer to provide a very simple scheme that can be handled during routine work. Moreover, we are planning a validation of this model in an independent cohort study, as the first step the part related to clinical risk factors.
Competing interests
The five authors from research institutes (LAJH, TDM, AA, WS, MS) and the two authors from industry (RS, JH) see no conflict of interest.
Authors' contributions
LAJH: designed together with WS the cohort study and both are the principal investigators, LAJH planned all analyses, wrote a first draft of the manuscript. DMT: developed and maintained the database, performed the majority of analysis, and contributed to the manuscript. AA: responsible for running all field work, performing quality control and designing the validation of diagnoses, contributed to the manuscript. WS: one of the PIs, contributed to the manuscript. RS: responsible together with JH for interpretation of the findings, major contributions to the manuscript. JH see RS. MS: responsible for the haemostasiological lab work during the entire study period, contributed to the manuscript.