Cardiovascular risk factors are major determinants of thrombotic risk in patients with the lupus anticoagulant

The Vienna Lupus Anticoagulant and Thrombosis Study (LATS) is an ongoing, single-center, biobank-based, prospective observational cohort study enrolling adult patients who repeatedly test positive for LA (two positive tests at least 12 weeks apart) with or without a history of thrombosis or pregnancy complications [5]. The primary endpoint of the study is a composite of symptomatic, objectively confirmed arterial and/or venous thrombosis. Comprehensive details about the design of this study have been reported previously [5] and can be found in Additional file 1: paragraph 1.

Blood sample preparation procedures are reported in Additional file 2: paragraph 2. LA was diagnosed according to Scientific and Standardization Committee (SSC)/International Society on Thrombosis and Hemostasis (ISTH) recommendations [17, 18]. A lupus-sensitive activated partial thromboplastin time (PTT-LA, Diagnostica Stago, Asniere-sur-Seine, France) and a diluted Russell’s viper venom time (dRVVT) were used as screening tests. For screening during therapy with vitamin K antagonists (VKAs), only the aPTT was used. Confirmatory tests were performed following the methodology of Wenzel et al. in the case of prolongation of one or both screening tests [19]. As confirmatory assays the StaClot LA (Diagnostica Stago, Asniere-sur-Seine, France) and the dRVVT-LA Confirm (Life Diagnostics, Clarkston, GA, USA) were used. In case of a not definitely positive confirmatory test during the follow-up period, LA was still regarded as positive if the Rosner Index, calculated as 100 × (clotting times of the 1:1 mixture - normal plasma)/patient’s plasma, was higher than 15 [20]. Commercially available indirect solid-phase enzyme immunoassays were used to determine IgG and IgM antibodies against cardiolipin (aCL) and β2-GPI. Between 2001 and September 2005, the Varelisa Cardiolipin test (Pharmacia (Phadia AB), Uppsala, Sweden) was performed semi-automatically with a Tecan Genesis liquid handling system (Tecan Group Ltd., Maennedorf, Switzerland). From October 2005 the Orgentec Cardiolipin and, from October 2006, the Orgentec β2-GPI tests (both from Orgentec, Mainz, Germany) were performed on a fully automated BEP2000 Advance System (Siemens Healthcare Diagnostics, Marburg, Germany). All assays were used following the manufacturers’ instructions. Positivity for aCL IgG and aCL IgM was defined as results >40 IgG phospholipid units (GPL)/IgM phospholipid units (MPL) U/mL for both the Varelisa Cardiolipin and the Orgentec Cardiolipin tests according to the revised Sapporo criteria [4]. Two further aCL cut-offs were analyzed as a sensitivity analysis (see the legends of Tables 1 and 3). For aβ2-GPI IgG and IgM (Orgentec assays), results >8U/mL were regarded as positive (corresponding to the 99th percentile of healthy controls). IgM- and IgG-isotype antibodies against prothrombin and protein Z were measured using commercially available enzyme-linked immunosorbent assay kits from the Zymutest product line (Hyphen Biomed, Neuville-sur-Oise, France). The annexin A5 resistance ratio (A5R) was measured as previously described, expressed as the ratio of in vitro coagulation times with and without annexin A5 [21]. IgG-isotype autoantibodies against domain I of β2-GPI were measured with a chemiluminescent immunoassay (QUANTA Flash/Bioflash, Inova Diagnostics, San Diego, CA, USA) [9].

The statistical analysis is described in detail in Additional file 3: paragraph 3. Briefly, median follow-up time was estimated with the reverse Kaplan-Meier estimator [22]. Patients who became LA negative during follow-up (n = 11) were censored at the date of the first negative LA test. The cumulative incidence of the primary endpoint was calculated with cumulative incidence estimators according to Marubini and Valsecchi, treating death from any cause as a competing risk [23]. Differences in thrombosis incidence functions between two or more groups were investigated using Gray’s test [24]. The association between potential risk factors and the cumulative incidence of thrombosis was modeled with uni- and multivariable proportional subdistribution hazards models according to Fine and Gray [25]. To increase external generalizability, modeling was also performed using an aPTT ratio, which was defined as the ratio of the lupus-sensitive aPTT of a patient divided by the mean of the lupus-sensitive aPTT in healthy controls at our department (mean = 34.09 s, SD = 0.476). A backward selection algorithm (p for exclusion = 0.10) including all four univariable predictors of thrombotic risk with p < 0.10 (lupus-sensitive aPTT ratio (dichotomized into a binary variable at the 75th percentile of the distribution (117.5 s)), diabetes, smoking, and aCL IgM antibodies) was applied to construct a multivariable model for the prediction of thrombotic risk [26]. The algorithm selected the three variables diabetes, smoking, and a prolonged lupus-sensitive aPTT, and we constructed an empirical risk stratification rule by assigning 2 points for diabetes, and 1 point for each of the risk factors smoking and a prolonged lupus-sensitive aPTT. These points were chosen because they were consistent with an additive effect of the underlying predictor variables on the log hazard scale (further details are reported in Additional file 3: paragraph 3) [27]. Discrimination of the proposed stratification rule was assessed using Harrell’s C statistic, and calibration was explored by comparing the observed and predicted 5- and 10-year cumulative incidences of thrombosis [28]. Finally, in a sensitivity analysis, we assessed the separate association between the three risk stratification variables and the prospective risk of arterial and venous thrombosis (see Additional file 4: Table S1).

One hundred and fifty patients were included in the analysis. Patients were predominantly female, and 74.2% had an established diagnosis of APS (Table 1). All patients were positive for LA, and 67 (44.7%), 105 (71.0%), and 64 (43.2%) patients also had above-cut-off antibody levels against cardiolipin (aCL), β2-GPI (aβ2-GPI), or both (“triple positivity”). IgM- and IgG-isotype aCL and aβ2-GPI antibodies were moderately strongly correlated with each other. Some correlations were also observed between elevated levels of these antibodies and (1) higher levels of antibodies against prothrombin and protein Z, (2) a lower annexin A5 anticoagulant ratio, and (3) higher levels of IgG-isotype antibodies against domain I of β2-GPI (Additional file 5: Table S2). A long lupus-sensitive aPTT was significantly correlated with a higher level of IgG-isotype antibodies against domain I of β2-GPI (rho = 0.40, p < 0.0001) and a lower prothrombin time (given as percent of normal, rho = –0.27, p = 0.0007). The average levels of the lupus-sensitive aPTT and fibrinogen were slightly but non-statistically significantly elevated in VKA users (Additional file 6: Table S3).

Ninety-eight patients (65.3%) had a history of thrombotic events (TEs) before study inclusion (arterial: n =21, venous: n = 84, both: n = 7). Patients with a history of TE were significantly younger than patients without a history of TE and had a much higher probability of being on oral anticoagulation with VKA (odds ratio (OR) = 56.7, 95% CI: 12.9–248.2, p < 0.0001, Table 1). The median levels of IgG-isotype antibodies against aCL and β2-GPI (Table 1) and the odds of being “triple positive” were also significantly higher in patients with prior TE (OR = 2.5, 95% CI: 1.2–5.1, p = 0.01). In patients with a prior history of TE, we observed a significantly higher average antibody level against domain I of β2-GPI and a significantly lower annexin A5 anticoagulant ratio. LA-related antibodies against prothrombin and/or protein Z did not appear to consistently differ according to anamnestic thrombosis status. Forty (42.6%) of the 94 female patients who had at least one documented pregnancy had at least one pregnancy complication according to Sapporo criteria. These 40 women had significantly higher levels of IgG-isotype aCL, aβ2-GPI, and domain 1-β2-GPI antibodies and were also more likely to be “triple positive” (Additional file 7: Table S4). Other parameters did not appear to differ between women with and without pregnancy complications.

During a median follow-up of 9.5 years (range: 12 days–13.6 years) and 1076 patient years, 32 patients developed TE (arterial: n = 16, venous: n = 16). The most frequent type of events were lower extremity deep vein thrombosis (n = 6) and pulmonary embolism (n = 6) in the venous vasculature, and cerebrovascular incidents (n = 9) and myocardial infarction (n = 5) in the arterial vasculature. Twenty-one of the 32 events occurred in patients with a prior history of thrombosis (“recurrent thrombosis”), and 11 events occurred in LA-positive patients without a prior history of thrombosis. Data on antithrombotic therapy at the time of thrombosis were available for 31 out of 32 patients (Table 2). Twenty-three (74.2%) of these 31 events occurred while patients were receiving antithrombotic therapy (Table 2). In detail, 14 (45.2%), 4 (12.9%), and 7 (22.6%) of these patients were receiving VKA, low molecular weight heparin, and/or low dose aspirin at the time of thrombosis, respectively. Among the 14 patients receiving VKA, the international normalized ratio (INR) was insufficient (i.e., <2) in 6 patients, within therapeutic range in 5 patients, and unknown in 3 patients. The cumulative 1-, 5-, 10-, and 15-year incidences of TE accounting for competing mortality were 4.0% (95% CI: 1.7–8.1), 13.3% (95% CI: 8.3–19.4), 24.3% (95% CI: 17.0–32.5), and 27.6% (95% CI: 19.3–36.6), respectively (Additional file 8: Figure S1). With 12 patients having died during follow-up without developing TE, death was clearly present as a competing risk in this population. Of the 32 patients who developed thrombosis during follow-up, 2 patients developed a further TE (1x venous thrombotic event (VTE) after VTE, 1x myocardial infarction after cerebrovascular insult).

In a univariable competing risk analysis, diabetes (subdistribution hazard ratio (SHR) = 5.18, 95% CI: 1.87–14.31, p = 0.002), active smoking (SHR = 2.11, 95% CI: 1.06–4.20, p = 0.034), and a prolonged lupus-sensitive aPTT (SHR per 10 seconds increase = 1.10, 1.00–1.21, p = 0.044) were univariably associated with a higher risk of TE (Table 3). In detail, the 10-year cumulative risk of TE was 60.0% in patients who were diabetic at baseline, as compared to 21.6% in non-diabetic patients (Gray’s test p = 0.002). The 10-year thrombotic risk was estimated at 36.5% in active smokers, as compared to 18.8% in ex- or never-smokers (p = 0.022). In patients with a lupus-sensitive aPTT > or ≤ the 75th percentile of its distribution (cut-off at 117.5 s (or for the aPTT ratio at 3.4 multiples of the median in healthy individuals)), we observed 10-year thrombotic risks of 43.7% and 17.6%, respectively (p = 0.004, Additional file 9: Figure S2A–C). A borderline significant association was observed between an increased baseline IgM-isotype aCL antibody level and a higher risk of TE risk (SHR = 1.30, 95% CI: 0.98–1.74, p = 0.068). Risk of TE was comparable between patients with established APS and LA-positive-only patients (SHR = 0.75, 95% CI: 0.36–1.58, p = 0.448), as well as between patients with or without a prior history of thrombosis (SHR = 0.94, 95% CI: 0.45–1.95, p = 0.865). Oral anticoagulation with a VKA at baseline was not associated with prospective thrombotic risk (SHR = 0.93, 95% CI: 0.47–1.86, p = 0.839, Additional file 10: Figure S3A). Antibodies against domain I of β2-GPI also did not emerge to be associated with prospective risk of thrombosis in the univariable analysis (SHR per 1000 CU increase = 0.93, 95% CI: 0.65–1.34, p = 0.711), and this result prevailed when analyzing the subgroups of patients (1) with and without a prior history of thrombosis (p for interaction = 0.323) and (2) younger or older than 50 years at study entry (p for interaction = 0.514). In a multivariable analysis, we adjusted the results for diabetes, smoking, and a prolonged lupus-sensitive aPTT ratio (Table 3, i.e., the variables that were selected below). The joint multivariable association between diabetes, smoking, the prolonged lupus-sensitive aPTT ratio and a higher risk of thrombosis prevailed upon inclusion of all reported variables. This also held true when adjusting for oral anticoagulation at baseline (Additional file 11: Table S5). None of the studied variables was significantly associated with thrombotic risk after adjusting for these three variables. However, a weak multivariable association between exposure to statins and a higher risk of thrombosis was observed. In a sensitivity analysis by event type, diabetes and smoking appeared to contribute prognostic information towards arterial events, and the prolonged lupus-sensitive aPTT towards venous events (Additional file 4: Table S1). Further sensitivity analyses by event type did not identify signals for associations between other studied variables and the risk of arterial and or venous events (not shown).

A backward selection algorithm included diabetes, smoking, and a prolonged lupus-sensitive aPTT ratio (binary specification) into a model for thrombotic risk stratification for patients with LA (Model 1, Table 4). According to the relative contribution of these variables to the model, 2 points were assigned for diabetes, and 1 point each for smoking and a prolonged lupus-sensitive aPTT (Models 2 and 3, further sensitivity analyses for point assignment are reported in Additional file 12: paragraph 4). The three-category point-based model (Model 3) could stratify patients into subgroups with a very low and very high risk of thrombosis (10-year risk of TE in patients with 0 (n = 77), 1 (n = 51), or ≥2 (n = 22) points: 9.7%, 30.9%, and 56.8%, respectively; Fig. 1). Internal validation procedures showed a strong discrimination according to this point-based rule (Harrell’s C: 0.72), and calibration was excellent for 10-year thrombotic risk and moderate for 5-year thrombotic risk (Additional file 13: Figure S4).

Although this study represents one of the very few prospectively executed studies with stringent inclusion criteria in the field of APS research, several limitations have to be mentioned. First, this study includes patients who tested repeatedly positive for LA with and without a history of thrombosis and included subgroups of patients with other potentially relevant factors, such as concomitant autoimmune rheumatic diseases. Consequently, our results are not directly generalizable to other patient groups included in the wide and heterogenic spectrum of APS, such as patients with SLE [34, 38] or patients who are positive for aCL or aβ2-GPI but not LA [39]. Second, our present analysis cannot provide a valid estimate for the potential benefit of anticoagulation in LA-positive patients. In this observational study, patients who were anticoagulated at baseline had a similar risk of prospective thrombosis as patients who were not anticoagulated. Of course, this must not be interpreted as an absence of efficacy, because reverse causality has likely confounded this observational result. Indeed, patients with a history of thrombosis had a substantially higher probability of being on oral anticoagulation, so anticoagulation may rather reflect the decision of individual physicians to anticoagulate those patients with the highest perceived risk of thrombosis. Other prospective investigators in the field have also faced this problem [32]. Third, because prospective data in LA-positive patients are scarce, we could not yet validate our empirical risk stratification rule in an external cohort. Fourth, our suggested risk stratification rule has some loss of information as compared to the full multivariable model, which is attributable to “rounding” of regression coefficients to a point-based system. In detail, the relative contribution of diabetes towards thrombotic risk, as compared to smoking status and the lupus-sensitive aPTT, would have led to an uneven point score of 1.6 for diabetes, which was rounded to 2. Nevertheless, we would like to mention that our cohort features several strengths, including a stringent prospective design with a long follow-up, a small drop-out rate, and the time-dependent censoring of patients who became LA negative over time.