Accuracy of Various Methods for the Surveillance of Hospital-acquired Venous Thromboembolism in a Children’s Hospital

Introduction: Accurate measurement is important in quality improvement. International Classification of Diseases (ICD) codes are often used to measure hospital-acquired venous thromboembolism (HA-VTE) rates, but they have high detection error rates. Our objective was to compare HA-VTE rates of various automated calculation methods and their accuracy. Methods: Single-center retrospective study including children younger than 18 years admitted between 2018 and 2023. We developed 4 automated methods to identify potential HA-VTE cases using (1) ICD-10 codes, (2) vascular extremity ultrasound reports (USG), (3) hematology consult reasons (Heme), and (4) receipt of therapeutic-dosing enoxaparin (Meds). We reviewed charts to determine true HA-VTE, defined as the radiographic presence of a thrombus. Additionally, we reviewed all charts with a vascular extremity ultrasound performed during the study period for true HA-VTE. We computed the accuracy of each method. Results: Across 33,430 admissions, we detected 275 (0.82%) cases by ICD, 508 (1.5%) by USG, 151 (0.45%) by Heme, and 176 (0.53%) by Meds. We manually reviewed 869 charts, of which 180 (0.5%) had confirmed HA-VTE. The false-positive rate was highest for USG (70.7%), followed by ICD (52.4%), Heme (50.9%), and Meds (43.8%). The combined false-positive rate was 59.2%. Sensitivity was highest for USG (82.8%), followed by ICD (72.8%), Heme (41.1%), and Meds (55.0%). Specificity was highest for Meds and Heme (88.8%). Discriminative performance was best for Meds (with an area under the precision–recall curve: 44.3%). Conclusions: Combining multiple methods led to greater identification of HA-VTE without significantly increasing false-positive rates, making it useful for quality improvement surveillance. Detection based on Meds is a more accurate alternative to ICD codes.