Computerised antibiotic susceptibility prediction: accounting for prior knowledge and cross-resistance
Abstract number: P819
Andreassen S., Zalounina A., Kariv G., Leibovici L., Paul M.
Objectives: To develop an automated method for estimation of a broad range of antibiotic susceptibilities given a restricted antibiogram.
Methods: Amended susceptibility for a non-tested antibiotic was determined using the probability of it being sensitive (S) to this antibiotic a priori; available susceptibility results for all antibiotics in the antibiogram and cross-resistance, expressed as the conditional probability of resistance for non-tested antibiotics given the tested antibiotics. The method was derived using a bacteraemia database of 3347 clinically significant isolates (CSI) collected between 20022004 and validated on a bacteraemia database of 4848 CSI between 20062008, both at Rabin Medical Center, Israel. The Brier score (BS) was used to measure the accuracy of the predicted amended susceptibilities. This score measures the distance between the predicted susceptibility and the actual outcome (0% indicates perfect agreement and 100% complete deviation).
Results: The table exemplifies how the susceptibility of Acinetobacter to imipenem is amended. The knowledge that Acinetobacter is sensitive to amikacin and meropenem increases the odds for imipenem=S by odds ratios (OR) of 5.8 and 21.5, respectively. Resistance (R) to minocycline reduces the odds by an OR of 0.47. The remaining antibiotics carry insignificant ORs (p0.10). In the proposed method the biggest OR (21.5) is then multiplied by the smallest OR (0.47) and the a priori odds for imipenem=S (56/44 = 1.27) is multiplied by that product. The resulting amended probability for imipenem=S is 93%. Across all pathogen/antibiotic combinations in the derivation DB, the BS for a priori susceptibilities was 39%, reduced to 25% for amended susceptibilities. For the validation DB, the BS for a priori susceptibilities was 41%, reduced to 29% for amended susceptibilities, indicating that there is a significant advantage to the amendment (p < 10-99).
Conclusion: Amended susceptibilities can be used to guide the prescription of antibiotics not included in the antibiogram. This is necessary to allow for a restricted antibiotic panel testing in the laboratory, to assist clinicians and to improve the management of polymicrobial infections. The computerised algorithm will be incorporated into the TREAT computerised decision support system for antibiotic treatment.
Susceptibility of Acinetobacter spp.
|Session name:||19th European Congress of Clinical Microbiology and Infectious Diseases|
|Location:||Helsinki, Finland, 16 - 19 May 2009|
|Back to top|