External validation of the Leiden Thrombosis Recurrence Risk Prediction models (L-TRRiP) for the prediction of recurrence after a first venous thrombosis in the Heart and Vascular Health study

Abstract

Background: Long-term outcome after a first venous thromboembolism (VTE) might be optimized by tailoring anticoagulant treatment duration on individual risks of recurrence and major bleeding. The L-TRRiP models (A-D) were previously developed in data from the Dutch Multiple Environment and Genetic Assessment of Risk Factors for Venous thrombosis study to predict VTE recurrence.

Objectives: We aimed to externally validate models C and D using data from the United States Heart and Vascular Health (HVH) study.

Methods: Data from participants with a first VTE who discontinued initial anticoagulant therapy were used to determine model performance. Missing data were imputed, and results were pooled according to Rubin's rules. To determine discrimination, Harrell's C-statistic was calculated. To assess calibration, the observed/expected (O/E) ratio was estimated, and calibration plots were created, in which we accounted for the competing risk of death. A stratified analysis based on age <70 or >70 years was performed.

Results: Of 1430 participants from the HVH study, 187 experienced an unprovoked VTE recurrence during follow-up. The C-statistics of L-TRRIP models C and D were 0.62 (95% CI, 0.56-0.67) and 0.61 (95% CI, 0.55-0.67), respectively. The O/E ratio (1.00; 95% CI, 0.84-1.17 and 1.09; 95% CI, 0.91-1.27, respectively) and calibration plots indicated good calibration. The discrimination was similar between participants <70 or >70 years, whereas overall calibration was lower in participants <70 years.

Conclusion: The L-TRRiP models showed moderate discrimination and good calibration in a different population and can be used to guide clinical decision making. To assess the added value in daily clinical practice, a management study is needed.

Keywords: anticoagulants; clinical decision rules; prognosis; validation study; venous thromboembolism.

Figure 1

Flow-chart of included patients. DVT, deep venous thrombosis; VTE, venous thromboembolism.

Figure 2

Calibration plot model C. Calibration plot showing estimated risks of recurrent VTE at 2 years according to L-TRRiP model C against observed proportions of recurrent VTE in the HVH study data. The curve including confidence intervals was estimated using pseudo-observations and LOESS smoothing. The gray line indicates perfect calibration. The histogram at the x-axis indicates the distribution of risk estimates. HVH, Heart and Vascular Health; LOESS, locally estimated scatterplot smoothing; VTE, venous thromboembolism.

Figure 3

Cumulative incidence of unprovoked recurrent VTE stratified by predicted risk category. Cumulative incidence curves for unprovoked VTE recurrence during follow-up stratified by risk group; a low, intermediate, and high recurrence risk was defined as a predicted recurrence risk according to L-TRRiP model C of <6%, 6% to 14% or >14% within 2 years, respectively. VTE, venous thromboembolism.

Figure 4

Distribution of predicted risks for patients with a provoked and unprovoked first VTE. Histogram of predicted risks of VTE recurrence at 2 years according to model C for patients with a provoked (top) or unprovoked (bottom) first VTE. VTE, venous thromboembolism.