#41 Manual Validation of an Automated Tool to Extract Blood Culture and Susceptibility Data from the Electronic Health Record for Children with Acute Myeloid Leukemia

Abstract Background Children with acute myeloid leukemia (AML) receive high-intensity chemotherapy to achieve durable remission. AML chemotherapy causes bone marrow suppression resulting in vulnerability to infection, most frequently bloodstream infections (BSI). While the most commonly described organisms include Gram-negative pathogens and streptococci, the epidemiology and resistant profile of these pathogens can change. It is important to monitor the epidemiology of these infections over time to inform clinical care. Previously capture of these microbiology data required laborious manual chart reviews that are often done intermittently and with variable accuracy, limiting the impact of the results. We sought to develop and validate an automated process for extracting blood culture results, including antimicrobial susceptibility profiles for positive results from the electronic health record (EHR). Method An automated tool to extract blood culture results from the EHR (Epic Systems, Verona WI) was developed using SQL (Structured Query Language). The tool was applied to the EHR of all children with newly diagnosed AML treated at the Children’s Hospital of Philadelphia from January 1, 2011 to December 31, 2020, regardless of subsequent relapse and treatment. Data from all blood cultures (including standard, fungal, mycobacterial, and subacute bacterial endocarditis blood cultures) were captured. Manual chart review was performed by an Infectious Diseases physician to abstract the same blood culture results to determine accuracy of the automated extraction tool. The manual abstraction was considered the gold standard. The BSI epidemiology of AML patients during this time period were described to illustrate the utility of this tool. Results There were 91 children with newly diagnosed AML who received chemotherapy during the study period. Of the collected 3,150 cultures obtained and resulted, 206 (6.5%) were positive. There were 37 distinct pathogens identified (Table 1). Of positive cultures, 114 (55.3%) were reflexed to antimicrobial susceptibility testing (AST) per institutional standards. In total, 1,427 AST results were captured in the automated tool. Manual validation confirmed that concordance between the automated abstraction and chart review was 100% for the organisms grown on culture and accurately identified all AST results. The majority of organisms were Gram-positive. The most frequently identified species was Streptococcus mitis/oralis, found in 40/206 (19.4%) of positive cultures (Table 1). Of S. mitis/oralis that underwent AST, only 8/21 (38.1%) were susceptible to penicillin. Escherichia coli was the most common Gram-negative pathogen. Candida spp. accounted for 9.3% of detected pathogens, with C. krusei identified most commonly. Conclusion We developed and automated tool to extract blood culture results from the EHR of children and adolescents with AML at a single center. This tool was manually validated and found to be 100% accurate. This tool has the ability to efficiently capture the BSI epidemiology of a specific patient population at high risk for BSI. Further work is needed to confirm the accuracy of this tool at more centers. Implementation at other sites will allow this tool to be employed for various purposes including large, multicenter epidemiology research studies or quality improvement projects.