An algorithm for the identification of heparin-induced thrombocytopenia using a medical information database

Using effective algorithms for extracting relevant drug and patient information from electronic medical information data systems is likely to form an increasingly important aspect of pharmacovigilance. To this end, we aimed to develop and validate a novel algorithm for detecting heparin-induced thrombocytopenia (HIT) using a medical information database (MID) and for identifying possible risk factors for HIT.

We developed a new algorithm for detecting HIT based on platelet count at distinct time-points and diagnostic information from patients treated with unfractionated heparin (UFH) from April 2008 through March 2012 at Hospital of Hamamatsu University School of Medicine, Japan. Definitive diagnoses of HIT were made through reviews of the medical records by a skilled haematologist. The performance of the algorithm was assessed using the positive predictive value (PPV). Multivariate logistic regression analysis was used to identify possible risk factors for HIT.

This algorithm detected 47 patients with suspected HIT in the source population (n = 2875). Of these, 41 were identified as patients with definitive HIT after review of the medical records. The PPV for the algorithm was 87·2%, and the frequency of definitive HIT was 1·4%. Longer-term treatment (≥4 days) with UFH was identified as a risk factor for HIT, with an odds ratio of 5·38 (95% CI: 2·35-12·32) for definitive HIT.

We developed a novel, high-PPV detection algorithm for HIT and identified longer-term treatment with UFH as a risk factor for HIT. Our results support the utility of MIDs for improving pharmacovigilance.