Assessment of Algorithms to Identify Patients with Thrombophilia Following Venous Thromboembolism

BACKGROUND: Thrombophilia testing after venous thromboembolism (VTE) is controversial, and studies to-date have relied on medical record abstraction of lab results to assess patterns of testing and outcomes. Administrative databases are increasingly used for VTE outcomes research; however, the ability to identify thrombophilic patients in large datasets has not been evaluated. Our study examined how algorithms using ICD-9 codes and/or pharmacy data accurately reflect lab-based thrombophilia diagnoses.

METHODS: This was a retrospective cohort analysis. We included patients ≥21 years old with their first VTE (based on the presence of an ICD-9 code for deep vein thrombosis [DVT] or pulmonary embolism [PE]) between 1/2004 and 12/2010 at Kaiser Permanente Colorado (KPCO). All VTE events were validated using manual chart review. We excluded patients with atrial fibrillation, prior VTE, warfarin prescription <3 years prior to VTE, <1 month of continuous health plan enrollment and drug benefits after VTE, recurrent VTE during follow-up, or provoked VTE (surgery <1 month or pregnancy <1 year prior to VTE, or active cancer). Patients were followed for 1 year after VTE. Thrombophilia lab test results were extracted from KPCO’s electronic lab database and confirmed with manual chart review as necessary. Tests included factor V Leiden/activated protein C resistance, prothrombin gene mutation, antithrombin activity, protein C activity, protein S activity, lupus anticoagulant (hexagonal phase and Russell’s viper venom time), cardiolipin immunoglobulin (Ig)G, and β-2 glycoprotein IgG. Patients were considered positive (+) for lab-confirmed thrombophilia if ≥1 test was positive, negative (-) if no tests were positive, and unknown if testing was not performed. “ICD-9 only” (+: ≥1 ICD-9 code for primary [289.81] or secondary hypercoagulable state [289.82], - otherwise) and “pharmacy only” criteria (+: extended anticoagulation [AC] >6 months after VTE, - otherwise) were applied individually and in combination (“combined ICD-9/pharmacy”) to identify possible thrombophilia cases. Using lab-confirmed thrombophilia as the gold standard, sensitivities, specificities, positive predictive values (PPVs), and negative predictive values (NPVs) of the 3 thrombophilia identification strategies were calculated along with 95% confidence intervals (CIs). Baseline characteristics were analyzed overall and by thrombophilia test result status.

RESULTS: Out of 1314 patients with VTE, 636 had an unprovoked clot and met our criteria for this study. The analytic cohort had a mean age of 62.7 years, 49.5% of the study patients were male, and 55.2% and 54.4% had a lower extremity DVT and PE, respectively. Thrombophilia testing was performed in 32.4% of the patients, and 7.6% of the analytic cohort had ≥1 positive test result. After applying the various algorithms, 6.5% had ≥1 ICD-9 thrombophilia code and 47.6% received extended AC following VTE. There were differences in the mean age (55.1 vs. 63.3 years), positive family history of VTE (18.8% vs. 6.1%), and presence of ≥1 ICD-9 thrombophilia code (35.4% vs. 4.1%) between the thrombophilia test + (n=48) and thrombophilia test -/unknown (n=588) groups, respectively (all p<0.05). There was no difference in the proportion of + or -/unknown thrombophilia patients who received extended AC following VTE. Sensitivity, PPV, specificity, and NPV results of each thrombophilia identification strategy are presented below.

CONCLUSIONS: ICD-9 codes for thrombophilia are highly specific for lab-confirmed cases. However, many patients with positive lab tests for thrombophilia do not get coded as such. Although utilizing extended AC as an identifier for patients with thrombophilia has a higher sensitivity than using ICD-9 codes alone or in combination, none of the 3 algorithms had a value >50%. Our results suggest that abstraction of lab results is still required to accurately capture all cases of thrombophilia. Further development of methods to identify thrombophilia status in large datasets is warranted.