US Conference, page-4

Search Results Print Page
Share Page
Session 224 - CPHM12 - Molecular Diagnostic Microbiology: Pre-Analytical Considerations
SATURDAY - 349 / SATURDAY - 349 - Discrepant Result Analysis for Emerging Technologies in Microbiology
Itinerary
June 9, 2018, 11:00 AM - 1:00 PM  Exhibit and Poster Hall, Building B, Halls B2-B5
Authors
T. Stanford1, L. Brenton2, S. Giglio1; 1LBT Innovations, Adelaide, Australia, 2St. Vincent's Hosp. Melbourne, Melbourne, Australia
Disclosures
T. Stanford: D. Employee; Self; LBT Innovations. L. Brenton: None. S. Giglio: D. Employee; Self; LBT Innovations.
Abstract
Background: To evaluate the effectiveness of a newly introduced test or system, the truth status of a sample result needs to be established, typically by the use of a gold standard. This is problematic where the gold standard is imperfect, resulting in difficulties to understand the true performance of the test or system under investigation. This is especially relevant now in microbiology where a number of plate imaging systems, some with the ability to interpret microbial growth using artificial intelligence, are evolving. The gold standard in this case is the microbiologist assessment (or panel of microbiologists). Methods: Discrepant analysis traditionally involves investigating only those results where the test and the imperfect truth (microbiologist assessment) do not agree. These samples have their truth status re-assessed with the expectation the new truth status and the test’s classification agree. This method ignores the possibility of samples previously in agreement having the truth status re-assessed in disagreement with the test’s classification, biases interpretation and statistical analysis, and therefore alternative methods are required. Using data from an evaluation of the APAS Independence at St. Vincent’s Hospital Melbourne screening urine samples, we present three unbiased methods to undertake discrepant analysis: a novel simulation method using re-sampling, the composite reference standard method (Alonzo & Pepe, 1999), and latent class analysis (Walter & Irwig, 1988). The APAS Independence is a stand-alone plate reading and interpretation system that utilizes artificial intelligence algorithms to provide a detailed screening result for the sample. Results: In addition to the St. Vincent’s case study data, simulated data are used to illustrate characteristics of the modelling processes and detail the limitations of the proposed methods. We also demonstrate that the novel re-sampling simulation method produced an estimate of significant growth sensitivity close to that of discrepant resolution while, importantly, using defensible methodology. Conclusions: This presentation is timely given the advances in microbiology imaging automation and will provide information to laboratories to objectively resolve samples that are discrepant.