A1254
Title: Analysis of disease similarities and bacterial co-occurrences from electronic health records
Authors: Yoshiro Yamamoto - Tokai University (Japan) [presenting]
Hiroto Kubomatsu - Tokai University (Japan)
Sanetoshi Yamada - Tokai University (Japan)
Tadashi Imanishi - Tokai University (Japan)
Abstract: Electronic health record data from 3536 hospitalized patients were analyzed to reduce inappropriate antimicrobial administration when causative bacteria are unknown. Initial empirical treatments can be ineffective, significantly impacting patient mortality. Analysis indicated that inappropriate administration rates were higher in patients with multiple detected bacteria compared to those with a single detected bacterium. To investigate this phenomenon, expected simultaneous detection values and actual simultaneous detection ratios were calculated. Fisher's exact test with Bonferroni correction was applied to identify significant bacterial pairs. Analysis of electronic health record data for pairs of two bacterial types enabled the discovery of co-occurring bacteria, revealing that these co-occurrences differ across medical departments and diseases. Furthermore, calculating similarities between diseases, such as the Jaccard index, based on the frequency of bacterial occurrences and visualizing them using metric multidimensional scaling made it possible to examine their relationships. Incorporating these findings is expected to optimize antimicrobial therapies and reduce inappropriate administration rates.
