Statin therapy and recurrent venous thromboembolism in the elderly: a prospective cohort study

Previous studies reported lower rates of recurrent venous thromboembolism (rVTE) among statin users, but this association could be influenced by concurrent anticoagulation and confounding by statin indication. This study aimed to confirm the beneficial association between statins and rVTE, stratified according to periods with and without anticoagulation, and additionally employ propensity score weighted approach to reduce risk of confounding by indication. The setting was a prospective multicentre cohort study and the outcome was time to first rVTE in statin vs. non-statin users. 980 participants with acute VTE were enrolled (mean age 75.0 years, 47% women), with median follow-up of 2.5 years. Of 241 (24.3%) statin users, 21 (8.7%) suffered rVTE vs. 99 (13.4%) among 739 non-users. The overall adjusted sub-hazard ratio (aSHR) for rVTE comparing statin users to non-users was 0.72 (95%CI 0.44 to 1.19, p = 0.20). This association was only apparent during periods without anticoagulation (aSHR 0.50, 95%CI 0.27 to 0.92, p = 0.03; vs. with anticoagulation: aSHR 1.34, 95%CI 0.54 to 3.35, p = 0.53). Using propensity scores, the rVTE risk during periods without anticoagulation fell further (aSHR 0.20, 95%CI 0.08 to 0.49, p < 0.001). In conclusion, statin use is associated with a more pronounced risk reduction for rVTE than previously estimated, but only during periods without anticoagulation.

A recent systematic review and meta-analysis of randomised clinical trials (RCT) and observational studies demonstrated a 15-25% risk reduction in a first episode of venous thromboembolism (VTE) with statin use 1 . The underlying mechanism of action for this benefit could lie in a reduction of markers of inflammation 2 , resulting in a decreased coagulant state and ultimately a reduced risk for VTE 3,4 . For persons who have already suffered a VTE, an RCT with statins is lacking. According to a meta-analysis of eight observational studies, the risk of recurrent VTE (rVTE) was 27% lower under statin therapy, and this association was homogenous across the included studies (I 2 = 0%), but results were not stratified according to whether or not patients took anticoagulants 5 .
Statin users are more likely to be older (>60 years) and suffer from cardiovascular comorbidities which predispose for VTE, compared to non-users [6][7][8] . In this instance, a traditional multivariable statistical analysis approach may fall short of balancing comorbidities between statin users and non-users, and the results could be confounded by statin indication 9 . Thus, in the absence of an RCT, it is advised to analyse observational data with a statistical approach that improves balance of baseline comorbidities, and this could be achieved with a propensity score weighted approach 9 .
We aimed to first assess whether statin use was associated with a lower rate of rVTE in a prospective cohort study with standardised assessment of rVTE, second to examine if the magnitude of the association between statin use and rVTE using a traditional multivariate statistical analysis approach was comparable with a propensity score (PS) weighted approach, and third to explore whether markers of inflammation (ultrasensitive C-Reactive Protein [uCRP], fibrinogen) explained the effect of statins.

Methods
We based our study report on the guidelines provided in the Strengthening the Reporting of Observational studies in Epidemiology (STROBE) statement 10 . Study design, setting, and participants. This study was conducted between September 2 nd , 2009 and December 6 th , 2013. It was part of the Swiss Cohort of Elderly Patients with VTE (SWITCO65+) prospective multicentre cohort study, which assessed long-term medical outcomes in elderly patients with acute VTE. The study population was recruited from all five university (Basel, Bern, Geneva, Lausanne and Zurich) and four high-volume non-university hospitals (cantonal hospitals of Baden, Frauenfeld, Luzern and St. Gallen) in Switzerland. The Ethics Committee at each of the nine participating centres approved the study. All research in this observational study was performed in accordance with relevant ethics and research guidelines/regulations. Informed consent to participate in the SWITCO65+ study was obtained for all enrolled adults. Participants aged 65 years or older, with an acute, objectively confirmed symptomatic deep vein thrombosis (DVT) or symptomatic pulmonary embolism (PE), were prospectively recruited in the inpatient and outpatient settings of all study sites, and followed-up until December 6th, 2013 (participants were censored earlier in case of death or rVTE) 11 . Symptomatic PE was defined as a positive spiral computed tomography or pulmonary angiography, a high probability ventilation-perfusion scan, or proximal DVT confirmed by compression ultrasonography or contrast venography, in participants with acute chest pain, new or worsening dyspnoea, haemoptysis, or syncope. Those with catheter-related thrombosis, insufficient German-speaking or French-speaking ability, no follow-up possibility (e.g. terminal illness), an inability to provide informed consent (e.g. severe dementia), or previous enrolment in the cohort, were excluded. The SWITCO65+ study was funded by grants from the Swiss National Science Foundation (grants 33CSCO-122659/139470).
Definition of exposures, outcomes, and potential confounders. The main exposure was statin use at study entry (statin user vs. non-user). The main outcome was the time to the first recurrence of a symptomatic, objectively confirmed VTE (rVTE) during the follow-up period, defined as PE or DVT (proximal and/or distal) based on previously published criteria [11][12][13] . Follow-up started directly after diagnosis of index VTE. The first follow-up took place after three months. During follow-up, study nurses interviewed participants (or respectively general practitioners in case of death, or relatives in case of no information [if participant died at home]) about the date and type of clinical events (rVTE; death). These interviews were implemented as one telephone call and two surveillance face-to-face evaluations during the first year of study participation. Following this, semi-annual contact was initiated which alternated between face-to-face evaluations (clinic visits or home visits in housebound participants), telephone calls, and periodic review of the participant's hospital chart. If a clinical event did occur, the participant's medical chart was reviewed, and interviews were held with the participant's primary care physicians and family members.
In addition, information about the following variables that were relevant for the statistical analyses was collected: age (analysed as a continuous variable), smoking history (never, past, current), polypharmacy (<5 drugs vs. ≥5 drugs), prior VTE (i.e. a VTE that occurred before the index VTE), transient provoked VTE (surgery during the last 3 months, immobilisation during the last 3 months, oestrogen therapy during the last 3 months), cardiovascular disease (coronary heart disease, peripheral artery disease, cerebrovascular disease [stroke, TIA]), arterial hypertension, diabetes mellitus, body-mass index (BMI) ≥ 25 [yes vs. no], fibrinogen and uCRP (uCRP was measured at University Hospital Zurich in serum aliquots using a latex-enhanced immunoturbidimetric assay on a cobas c502 autoanalyser [Roche Diagnostics, Mannheim, Germany] with assay characteristics as reported by the manufacturer). A committee of three blinded clinical experts adjudicated all outcomes, and final classifications were made based on the full consensus of this committee 11 . Statistical analyses. We compared baseline characteristics among statin users and non-users with the chi-squared test and the non-parametric Wilcoxon rank-sum test as appropriate. rVTE rates were displayed for statin users and non-users, as well as stratified into periods with and without anticoagulation. We examined the overall association between statin use and the time to a first rVTE using competing risk regression, accounting for non-PE-related death as a competing event. The method yields subhazard ratios (SHR) with corresponding 95%CIs and p-values for the failure event of primary interest. In the traditional statistical analysis approach, we adjusted for established risk factors for VTE: age, sex, symptomatic PE, prior VTE, transient provoked VTE, cardiovascular disease (i.e. coronary heart, peripheral arterial or cerebrovascular [stroke, TIA] disease), active cancer, and periods of anticoagulation as a time-varying covariate. In a sensitivity analysis, we additionally adjusted for Aspirin, other antiplatelet therapy, hypercholesterolaemia and diabetes after observing that these characteristics were not evenly distributed among statin users and non-users. In another sensitivity analysis, we stratified the results into periods with and without anticoagulation. Moreover, since the study population included both provoked and unprovoked index VTE, we also performed subgroup analysis to examine whether the association between statin use and rVTE rates was similar between these two groups. Furthermore, we assessed fibrinogen and uCRP as potential inflammatory mediators, additionally correcting models for their log-transformed values. Beforehand, we performed multiple imputation for missing values of fibrinogen and uCRP.
In the PS weighted analysis approach, we calculated the propensity to receive a statin in a logistic model with statin use as dependent and all the variables that were used for adjustment in the main analysis as independent variables. The advantage of the PS approach was the ability to not only better balance the two groups (statin users and non-users) in terms of baseline characteristics, but to also account for more potential confounders if rVTE events were few. Therefore, in the PS approach we additionally adjusted for hypertension, polypharmacy, smoking status, and BMI. Missing values for BMI and smoking status were assumed to be normal (i.e. BMI as <25 [n = 4], (2019) 9:14804 | https://doi.org/10.1038/s41598-019-51374-8 www.nature.com/scientificreports www.nature.com/scientificreports/ smoking status as never [n = 3]). For completion, we compared the results from the PS approach with a sensitivity analysis from the traditional statistical approach, which adjusted for all the same variables as in the PS analysis.
We trimmed the PS values using the 2.5 th percentile of the PS in the statin users as the lower cut-off and the 97.5 th percentile of the PS in the non-users as the upper cut-off. Observations in both groups that had PS values beyond these two cut-offs were dropped 14 . We then calculated the inverse probability of treatment weights (IPTW) for the remaining participants. The competing risk model was weighted according to the IPTW and additionally adjusted for periods of anticoagulation as a time-varying covariate. Again, we stratified the results into periods with and without anticoagulation in sensitivity analyses, and also assessed fibrinogen and uCRP as potential inflammatory mediators, additionally correcting models for their log-transformed values. We used Stata 15 for all analyses (Stata Corporation, College Station, Texas).

essentials.
• Statin users have lower rates of recurrent venous thromboembolism (rVTE), but the magnitude of this association is uncertain and could be influenced by concurrent anticoagulation. • While statin therapy was not associated with reduced risk of rVTE during periods of anticoagulation, we observed a 50% reduced risk of rVTE with statin therapy during periods without anticoagulation. • Using a propensity score weighted approach, statins were associated with an 80% reduced risk of rVTE during periods without anticoagulation.

Discussion
In this population-based prospective cohort study, we found a lower rate of rVTE among statin users compared to non-users, but this beneficial association was only apparent during periods without anticoagulation, and stronger in the PS weighted approach than in the traditional statistical analysis approach. Furthermore, the potentially beneficial effect of statins was not explained by markers of inflammation such as fibrinogen or uCRP. www.nature.com/scientificreports www.nature.com/scientificreports/ Our results from the traditional statistical analysis approach were very similar to a recent meta-analysis of eight observational studies 5 , indicating an almost 30% risk reduction in rVTE with statin therapy. However, our results suggest that statins may only reduce the risk for rVTE in the absence of anticoagulation. Therefore, they might be a second best option to prevent rVTE when anticoagulation is contraindicated. Moreover, under the PS weighted approach, the risk reduction became even more substantial (i.e. 80% during periods without anticoagulation). There are arguments supporting the hypothesis that the traditional statistical analysis approach underestimates the true effect of statins in observational studies: statin users have more comorbidities and are sicker than non-users 6,7 , which is also observed in our own study population's baseline characteristics (Table 1), and multimorbid persons have a higher risk of VTE 8 . Thus, the potentially beneficial effect of statins could be partly concealed by comparing multimorbid statin users with healthier non-users in observational studies using a traditional multivariable statistical analysis approach. In contrast, in the PS weighted approach, participants with a PS value below the 2.5 th percentile and above the 97.5 th percentile were discarded, as at these extremes of range there was no comparable participant in the other group with similar comorbidities/risk profile, thus making the two groups (statin users vs. non-users) as similar as possible 14 .
Our study has several limitations. First, the study population was recruited consecutively from most large-volume hospitals in Switzerland. As a result, it is uncertain if the results also apply to patients with potentially less severe (r)VTE who were treated in smaller hospitals or by general practitioners. Second, almost all study participants were white, so our results are only generalisable to a Caucasian population. Third, we tested if the potential beneficial effect of statins was explained by a reduction in inflammation -we observed no such association. Yet, we only considered fibrinogen and uCRP. Both markers have been suggested by previous studies to be affected by statin therapy 17,18 . It is possible that statins reduce the risk of (r)VTE via other inflammatory markers for which we had no information available or via direct impact on endothelial function 17,19 . In fact, newly published data suggests that the protective effect of statins in decreasing rVTE rates may not be down to a reduction in inflammation, but rather due to a reduction in coagulation factors 20 . Fourth, our study did not have sufficient power to analyse different statin agents separately, and it is possible that the risk reduction seen in (r) VTE is statin-specific and therefore, cannot be attributed to a class-effect 21,22 . Fifth, all participants were 65 years or older. Therefore, it is uncertain whether our results also apply to a younger population. Sixth, whether or not a participant used statins was only assessed at baseline. Therefore, our exposure assessment (i.e. statin use) might be inaccurate in cases where participants started or stopped statin therapy during follow-up. However, as this inaccuracy is not related to the outcome of interest (rVTE), results may at worst be biased towards the null (i.e. underestimate the association between statin use and rVTE).
Our study has several strengths. First, it highlights that statistical analysis approaches accounting for confounding by indication (i.e. the PS weighted approach) can produce different results from traditional multivariable analyses of observational data. Analysing observational data with both a traditional multivariable analysis and a PS weighted analysis, comparing the findings and attempting to explain differences, can result in new insights. Furthermore, it makes optimal use of already existing data. Second, our study included elderly patients with VTE (mean age 75 years), a population that is often underrepresented or excluded from clinical trials. Third, the cohort had a small dropout rate of 4.8% during follow-up 11 , and the data collection was near complete with less than 5% missing values. Fourth, rVTE was adjudicated by a blinded expert committee using clearly defined previously published objective criteria [11][12][13] , reducing the risk of detection bias. Finally, both inpatients and outpatients with an index VTE were included in a nationwide multicentre setting.

conclusion
Statin use was associated with a more pronounced risk reduction for rVTE than previously estimated, but only without concurrent anticoagulation. An RCT would be needed to substantiate the potential benefits of statins for prevention of rVTE during periods when anticoagulation is not indicated or feasible.

Data Availability
All necessary data that support the findings of this study are included within this published article. Any further supplemental data is available from the authors upon request.