Advertisement

The “Risk” of Risk Assessment Models for Venous Thromboembolism in Medical Patients

      To the Editor:
      We read with interest the derivation and validation of a simple risk assessment model (RAM) by Woller et al
      • Woller S.C.
      • Stevens S.M.
      • Jones J.P.
      • et al.
      Derivation and validation of a simple model to identify venous thromboembolism risk in medical patients.
      for identifying at-risk hospitalized medical patients for venous thromboembolism. As the authors point out, previous RAMs for this group have been complex, with incomplete scoring systems and inadequate validation. The authors derived and validated their 4-factor RAM (which included previous venous thromboembolism, bed rest, cancer, and peripherally inserted central catheter lines) from data using an electronic medical record and query of International Classification of Diseases (ICD)-9 codes attributable to hospitalization, with an excellent receiver operating curve of >0.80. However, 2 issues remain.
      First, the ICD-9 codes for venous thromboembolism may lack discrimination and include upper-extremity disease, superficial thrombophlebitis, and chronic deep venous thrombosis, which may explain a ∼3- to 5-fold higher-than-expected 90-day rate of clinical venous thromboembolism in the validation cohort (4.5%). Upper extremity disease, in particular, has a different pathophysiology from lower extremity disease, with uncertain effects of using thromboprophylaxis in reducing that risk.
      • Spyropoulos A.C.
      Upper vs lower extremity deep vein thrombosis: outcome definitions of venous thromboembolism for clinical predictor rules or risk factor analyses in hospitalized patients.
      Hence, the possibility of confounding exists when upper extremity events are included across smaller subgroups of patients when developing risk scores or prediction rules. This especially holds true with weak risk factors, such as advanced age or immobility.
      Second, many of the 86 risk factors used in predicting venous thromboembolism risk were attributable to the hospitalization itself, and not necessarily present at admission, which would be more useful clinically in identifying patients who would benefit from venous thromboembolism prophylaxis begun near the time of admission, enabling the model to be truly predictive rather than associative of venous thromboembolism risk.
      Recently, our group published evidence-derived RAMs from an international 15,156-patient database from the IMPROVE registry of hospitalized medical patients.
      • Spyropoulos A.C.
      • Anderson F.A.
      • FitzGerald G.
      • et al.
      Predictive and associative models to identify hospitalized medical patients at risk for VTE.
      We developed 2 risk models using Cox multiple regression analysis as shown in the Table: a 4-factor predictive model based solely on venous thromboembolism risk factors present at admission, and a 7-factor associative model that included factors present during hospitalization whose timing relative to the venous thromboembolism event could not be completely determined. The final risk scores in both of our models looked very different when we removed upper-extremity disease as an outcome, including elimination of central line catheters as a risk factor.
      TablePredictive and Associative Models to Identify Hospitalized Medical Patients at Risk for VTE
      Hazard Ratio (95% CI)Chi-squaredP ValuePoints
      Predictive model for 3-month VTE and points assigned to each independent risk factor
       VTE risk factor
        Previous VTE5.0 (3.3-7.8)53<.0013
        Known thrombophilia5.2 (1.3-21.5)5.2.023
        Current cancer2.0 (1.3-3.1)11.0011
        Age >60 years1.8 (1.2-2.7)8.5.0041
      Associative
      Previous VTE and age are both known to have occurred before 3-month VTE; the other patient factors are known to have been present at or during hospital admission.
      model for 3-month VTE and points assigned each patient characteristic
       Patient characteristic
        Previous VTE
      Previous VTE and age are both known to have occurred before 3-month VTE; the other patient factors are known to have been present at or during hospital admission.
      4.7 (3.0-7.2)48<.0013
        Known thrombophilia3.5 (1.1-11)5.2.042
        Current lower limb paralysis3.0 (1.6-5.7)11.0012
        Current cancer2.8 (1.9-4.2)27<.0012
        Immobilized ≥7 days
      Days immobile immediately before and during hospital admission.
      1.9 (1.3-2.7)11.0011
        ICU/CCU stay1.8 (1.1-2.9)6.1.011
        Age >60 years1.7 (1.1-2.6)6.3.011
      CCU=critical care unit; CI=confidence interval; ICU=intensive care unit; VTE=venous thromboembolism.
      Associative model: Score 0-1: low VTE risk, Score 2-3: moderate VTE risk, Score 4 or more: high VTE risk.
      low asterisk Previous VTE and age are both known to have occurred before 3-month VTE; the other patient factors are known to have been present at or during hospital admission.
      Days immobile immediately before and during hospital admission.
      Lastly, our model was able to incorporate multiple levels of venous thromboembolism risk, as opposed to a binary-risk vs no-risk approach adapted by the model used by Woller et al. All of the RAMs in this patient group need prospective validation to assess whether venous thromboembolism is reduced by use of the scores.

      References

        • Woller S.C.
        • Stevens S.M.
        • Jones J.P.
        • et al.
        Derivation and validation of a simple model to identify venous thromboembolism risk in medical patients.
        Am J Med. 2011; 124: 947-954
        • Spyropoulos A.C.
        Upper vs lower extremity deep vein thrombosis: outcome definitions of venous thromboembolism for clinical predictor rules or risk factor analyses in hospitalized patients.
        J Thromb Haemost. 2009; 7: 1041-1042
        • Spyropoulos A.C.
        • Anderson F.A.
        • FitzGerald G.
        • et al.
        Predictive and associative models to identify hospitalized medical patients at risk for VTE.
        Chest. 2011; 140: 706-714

      Linked Article

      • Derivation and Validation of a Simple Model to Identify Venous Thromboembolism Risk in Medical Patients
        The American Journal of MedicineVol. 124Issue 10
        • Preview
          Fewer than half of eligible hospitalized medical patients receive appropriate venous thromboembolism (VTE) prophylaxis. One reason for this low rate is the complexity of existing risk assessment models. A simple set of easily identifiable risk factors that are highly predictive of VTE among hospitalized medical patients may enhance appropriate thromboprophylaxis.
        • Full-Text
        • PDF
      • The Reply
        The American Journal of MedicineVol. 125Issue 11
        • Preview
          We congratulate Spyropoulos et al on their recent publication of the International Medical Prevention Registry on Venous Thromboembolism (IMPROVE) registry findings1 and are grateful for their thoughtful comments. We agree that our research using International Classification of Diseases, Ninth Revision (ICD-9) codes for outcomes is inherently limited by the accuracy of ICD-9 coding and the retrospective nature of data extraction and analysis. In our case, coding limitations may have led to a portion of patients that experienced upper-extremity deep venous thrombosis, superficial thrombophlebitis, or chronic deep venous thrombosis being included among the group classified as having deep venous thrombosis, increasing our observed rate of venous thromboembolism.
        • Full-Text
        • PDF