Replication Data for: A multi-analyte machine learning model to detect wrong blood in tube errors

Misidentification of blood specimens is an important pre-analytical risk that can lead to patient harm. We developed several machine learning models to detect this problem using Complete Blood Count (CBC) data in a large pediatric inpatient population. We achieved accuracy of >97% using CBC with differential cell counts. We then utilized a validation set designed to mimic real world prevalence, achieving a positive predictive value of 20%. Datasets are tabular data at the test level containing CBC with Diff and CBC no Diff analyte deltas (absolute deltas: current value - previous value and percent deltas: current value / previous value) for patients at the Children's Hospital of Philadelphia (CHOP) meeting certain inclusion criteria. Also included are patient sex, age and hours between CBCs. The analysis was conducted for CBC with Diff and CBC no Diff tests in parallel but separately. Therefore, each test has its own notebook and corresponding train/validation/test datasets.
118,314 total tests: 8,253 Complete Blood Count (CBC) no Differential (Diff) 110,061 CBC with Diff tests
The raw clinical data consisting of Complete Blood Count (CBC) test results was extracted from the Children's Hospital of Philadelphia (CHOP) Clinical Data Warehouse (CDW). The data was analyzed with a novel machine learning model.