Twitter Facebook LinkedIn
Center for Global Research Data

Individualized prediction of recurrence risk reduction and risk of bleeding with extended anticoagulation in patients with venous thromboembolism

Lead Investigator: Mathilde Nijkeuter, UMC Utrecht
Title of Research Proposal: Individualized prediction of recurrence risk reduction and risk of bleeding with extended anticoagulation in patients with venous thromboembolism
Vivli Data Request: 5951, 4543
Funding Source: Employment Contracts – UMC Utrecht
Potential Conflicts of Interest: None

Summary of the Proposed Research:

Venous thromboembolism (VTE) comprises deep vein thrombosis (DVT) and pulmonary embolism (PE). In DVT, a blood clot blocks a vein in the leg, whereas in PE, a vessel in the lungs is blocked. VTE is the third most frequent cardiovascular disease, occurring in 1-2 per 1000 people annually. Anticoagulation for a minimum of 3 months is the mainstay of therapy. After the initial 3 months, a decision should be made to stop anticoagulation or to continue to prevent recurrent VTE, with consideration of the patient’s preferences. As prior duration of treatment after the first 3 months does not influence risk of recurrence, it is advised to either stop or continue indefinitely[4]. Indefinite anticoagulant treatment, which is advocated in the absence of a high risk of bleeding, minimizes the risk of recurrent VTE, but requires lifelong compliance and gives an annual 1-2% risk of major bleeding. These major bleedings comprise, among others, intracranial bleeds and bleeding of an internal organ.

In daily clinical practice, it is difficult to decide whether to continue or stop treatment, because both risk of recurrence and risk of bleeding differ between individuals. Several risk scores for bleeding and recurrent VTE have been developed. With a moderate predictive accuracy and methodological limitations, guidelines and recent reviews do not recommend using these scores. Thus, how to weigh risk of bleeding and risk of recurrent VTE for individual patients remains an important clinical dilemma worldwide. Inadequate risk assessment can lead to unfavorable treatment choices, increasing risk of recurrent VTE and bleeding. Moreover, personalized information on risks and benefits of treatment is required to enable patient involvement and shared decision-making, which will increase patient satisfaction with care and treatment adherence while reducing costs.

In the proposed project, we will combine data from existing international data from large cohorts and trials on VTE. Using this data, we will develop models for the prediction of recurrent VTE within 5 years after stopping anticoagulation, and for the prediction of bleeding during 5 years of anticoagulation, after the initial >3 months. With these models, the effect of extended anticoagulant treatment with different types of anticoagulants can be predicted for individual patients. Our aim is to develop a well performing prediction model that is simple to use in clinical decision making with individual patients. The statistical methods used will be similar to those used in previous analyses of our group. Our models will be integrated in a web-based, online calculator on https://U-Prevent.com, which is developed by group members Frank Visseren and Jannick Dorresteijn. It already includes 11 interactive calculators of cardiovascular prediction models, and is increasingly used worldwide by physicians and their patients.

The proposed model will have the potential to assess an individual’s risk of recurrent VTE as well as the risk of major and clinically relevant bleeding. These individual risks can be used to decide on treatment duration by weighing individual thresholds for treatment via shared decision-making. This will improve patient satisfaction and adherence to medication. Moreover, overtreatment as well as undertreatment can be prevented and reduce the worldwide burden of VTE-related and bleeding-related morbidity and mortality.

Statistical Analysis Plan

Study population

As specified under Study design – 1, the intended study population consists of patients with VTE after initial treatment of >3 months anticoagulation whom participated in cohort studies or trials. Trials comparing direct oral anticoagulants (DOACs) vitamin K antagonists (VKA) and/or placebo and, prospective cohorts with a study follow-up time of at least 12 months and data available on candidate predictors, treatment duration and relevant outcomes are eligible. For model derivation and validation, Hokusai-VTE, RE-SONATE, RE-MEDY, PREFER in VTE registry and the Bleeding Risk Study will be used via the Vivli platform. Other eligible studies will be used for external validation (e.g. EINSTEIN-CHOICE, GARFIELD in VTE, ETNA-VTE, Tromso study, Danish National Patient Registry, MEGA study).

Outcomes

1. Recurrent VTE, defined as objectively confirmed DVT or PE, a death in which PE was contributing or could not be ruled out, or a condition treated as VTE;
2. Bleeding, defined as a composite of major bleeding and CRNMB according to the International Society on Thrombosis and Haemostasis (ISTH);
3. Non VTE- or bleeding-related mortality.

Sample size calculation

The ideal number of events per variable (EPV) varies among studies. It was calculated that a model for the prediction of recurrent VTE requires a dataset with 23 EPV. As the same set of predictors will be used for bleeding and recurrent VTE, our aim is to build a model with at least 23 EPV in the total dataset. To produce accurate and precise estimates of model performance, external validation of prognostic models requires a minimum of at least 100, but ideally 200 events. Aforementioned eligible studies comprise >1800 recurrent VTE events and >2000 bleeding events. As some cohorts are still ongoing, the exact number of events is not yet known.

Data preparation

A variable dictionary has been developed for the purpose of this study. Each dataset will be modified so that variable names, data types and values are coded accordingly. Subsequently, all datasets for model derivation will be merged into one single dataset. Multilevel multiple imputation will be used for sporadically and systematically missing variables.
Studies may differ in design, outcome measures and enrollment criteria. This is of limited influence with the proposed methodology. All datasets, both from trials and cohorts, will be analyzed as if it were prospective cohort data. Due to differences in enrollment criteria, variables may be missing in datasets. Besides missing variables, we do not expect differences in enrollment criteria to be of influence. The same outcome definitions will be used across studies. As all of the eligible trials use ISTH definitions or comparable definitions for bleeding and require objective confirmation of recurrent PE or DVT, no differences are expected. We will apply the same definitions to cohort data.

Descriptive statistics

Baseline characteristics will be presented as proportions and percentages, mean value and standard deviation, or median and interquartile range, depending on underlying distribution, for derivation and validation datasets separately. Kaplan-Meier curves will be presented to show cumulative event rates for bleeding and recurrent VTE.

Model development

The statistical methods used will be similar to those used in previous analyses of our group.
Two complementary Fine & Gray Subdistribution Hazard models will be derived: one for prediction of 5-year recurrent VTE risk after an initial treatment of ≥3 months anticoagulation, using non-VTE related mortality and bleeding as competing endpoints, and one for prediction of 5-year bleeding risk after an initial treatment of ≥3 months anticoagulation, using non-bleeding related mortality and recurrent VTE as competing endpoints. Candidate predictors will be based on existing risk scores, expert opinion and availability in clinical practice. Because both models will eventually be part of one calculator, the same set of predictors will be used for both models. Penalized regression techniques will be used for shrinkage and predictor selection.
Different types of antithrombotic treatment will be modelled using dummy-variables. By doing this, data of both treated and untreated patients can be used for model development. The prognostic model will then produce probabilities of recurrent VTE and bleeding if left untreated. In the final model, these dummy-variables will be replaced by unbiased hazard ratios for extended anticoagulation with full-dose or low-dose direct oral anticoagulants, vitamin K antagonists or antiplatelet therapy, derived from trials and meta-analyses.
Proportional hazards assumptions will be assessed based on Schoenfeld residuals. If violated, an interaction term of the covariate with time will be added.
Model coefficients and subdistribution hazards (sHR) with 95% confidence intervals (CI) will be shown for all included predictors for recurrent VTE and bleeding separately.
Sensitivity analyses will be conducted to assess whether the same model can be used in patients with provoked and unprovoked VTE, and for male and female patients.

Internal validation

Discrimination (Harrell’s C-statistic) will be assessed within the derivation dataset. Goodness-of-fit will be assessed using a calibration plot.

External validation

Both the VTE- and bleeding part of the prediction model will be validated in combined multiple external datasets by plotting predicted versus observed incidences in a calibration plot (goodness-of-fit) and calculation of Harrell’s c-statistic (discrimination). The validation dataset will be divided into pooled cohorts based on geographic regions of study participants to enable recalibration for different geographic regions.

Comparison with existing scores

To assess the clinical value of our newly derived prediction models, their predictive accuracy in terms of discrimination and calibration will be compared to the following existing models and scores using the validation dataset:
• Recurrent VTE: DASH, Vienna, HERDOO2, DAMOVES, L-TRRiP
• Bleeding: EINSTEIN, VTE-BLEED, Hokusai, ACCP VTE

Individual predictions

The individualized absolute risk reduction (ARR) that could be achieved by treatment with a specific anticoagulant can be estimated by subtracting an individual’s recurrence risk with treatment from their risk without treatment. Individualized number needed to treat (iNNT) can be calculated by dividing 1 by ARR. Similarly, the absolute risk increase (ARI) and associated individualized number needed to harm (iNNH) for bleeding can be calculated. These effects can be weighed against one another and compared between different anticoagulants.
Net treatment effect will be estimated for equal severity of bleeding and recurrent VTE, and other scenarios, to be discussed with the advisory board.

Concrete revenues

The prediction models for the individualized absolute treatment effect of extended anticoagulation will be made available for use in a clinical setting as a web-based calculator. Depending on a physician’s preferences, the calculator will show 1-year, 2-year or 5-year risks and treatment effects. Individual treatment effect estimates could improve long-term health outcomes of treatment and lower costs. From a patient’s perspective, knowing and discussing their individual treatment effect can result in improved understanding of VTE and perception of risks and benefits. This can result in optimal shared decision-making and increased patient satisfaction with care. Moreover, this can stimulate treatment adherence in those who are found to benefit from extended treatment and thus further improve results on a group level.

Software

All analyses will be performed in R-Statistic Programming. The following add-on packages will be used: glmnet, mvtnorm, Hmisc, editrules, VIM, lme4, mice, foreign, survival, survAUC, pec, rms, mstate, ggplot2, logistf, ipw, mnormt, dplyr, riskRegression, cmprsk, crrstep, timereg, crskdiag, crrp, survminer, xlsx, car, nephron, data.table, tableone, nephro, crrSc, micemd, metafor, jomo, mitml, corrplot, etm

Requested Studies:

A Phase 3, Randomized, Parallel-Group, Multi-Center, Multi-National Study for the Evaluation of Efficacy and Safety of (LMW) Heparin/Edoxaban Versus (LMW) Heparin/Warfarin in Subjects With Symptomatic Deep-Vein Thrombosis (DVT) and or Pulmonary Embolism (PE).
Sponsor: Daiichi Sankyo, Inc.
Study ID: NCT00986154
Sponsor ID: DU176b-D-U305

A Phase III, Randomised, Multicenter, Double-blind, Parallel-group, Active Controlled Study to Evaluate the Efficacy and Safety of Oral Dabigatran Etexilate (150 mg Bid) Compared to Warfarin (INR 2.0-3.0) for the Secondary Prevention of Venous Thromboembolism.
Sponsor: Boehringer Ingelheim
Study ID: NCT00329238
Sponsor ID: 1160.47

Twice-daily Oral Direct Thrombin Inhibitor Dabigatran Etexilate in the Long-term Prevention of Recurrent Symptomatic Proximal Venous Thromboembolism in Patients With Symptomatic Deep-vein Thrombosis or Pulmonary Embolism.
Sponsor: Boehringer Ingelheim
Study ID: NCT00558259
Sponsor ID: 1160.63

(Note: Additional study added as part of data request 5951)

Prevention of Thromboembolic Events – European Registry in Venous Thromboembolism (PREFER in VTE)
Sponsor: Daiichi Sankyo, Inc.
Study ID: DRKS00004795
Sponsor ID: DSE-VTE-01-12

Public disclosure:

de Winter MA, Dorresteijn JAN, Carrier M, Cohen AT, Hansen J-, Kaasjager HAH, Middeldorp S, Raskob GE,Sørensen HT, Visseren FLJ, Wells PS, Büller HR, Nijkeuter M. Individual Benefits and Harms of Extended Anticoagulation in Patients with Venous Thromboembolism: The VTE-PREDICT Model [abstract]. Res PractThromb Haemost. 2021; 5 (Suppl 1). https://abstracts.isth.org/abstract/individual-benefi ts-and-harms-of-extended-anticoagulation-in-patients-with-venous-thromboembolism-the-vte-predict-model/. Accessed July 23, 2021.