Design and validation of algorithms to identify venous thromboembolism in the French National Healthcare Database.

Abstract

This paper aims to introduce an algorithm designed to identify Venous Thromboembolism (VTE) in the French National Healthcare Database (SNDS) and to estimate its positive predictive value. A case-identifying algorithm was designed using SNDS inpatient and outpatient encounters, including hospital stays with discharge diagnoses, imaging procedures and drugs dispensed, of French patients aged at least 18 years old to whom baricitinib or Tumor Necrosis Factor Inhibitors (TNFi) were dispensed between September 1, 2017, and December 31, 2018. An intra-database validation study was then conducted, drawing 150 cases identified as VTE by the algorithm and requesting four vascular specialists to assess them. Patient profiles used to conduct the case adjudication were reconstituted from de-identified pooled and formatted SNDS data (i.e., reconstituted electronic health records-rEHR) with a 6-month look-back period prior to the supposed VTE onset and a 12-month follow-up period after. The positive predictive value (PPV) with its 95% confidence interval (95% CI) was calculated as the number of expert-confirmed VTE divided by the number of algorithm-identified VTE. The PPV and its 95% CI were then recomputed among the same patient set initially drawn, once the VTE-identifying algorithm was updated based on expert recommendation. For the 150 patients identified with the first VTE-identifying algorithm, the adjudication committee confirmed 92 cases, resulting in a PPV of 61% (95% CI = [54-69]). The final VTE-identifying algorithm including expert suggestions showed a PPV of 92% (95% CI = [86-98]) with a total of 87 algorithm-identified cases, including 80 retrieved from the 92 confirmed by experts. The identification of VTE in the SNDS is possible with a good PPV.