29 Data-driven identification of suspicious test patterns based on serial values from laboratory and point of care testing. (23rd February 2023)

• Record Type: Journal Article
• Authors: Thomson, Atticus; Bowyer, Stuart A; Pissaridou, Eleni; Booth, John; Briggs, Lydia; Bryant, William A; Key, Daniel; Shah, Mohsin; Spiridou, Anastassia; Sebire, Neil J
• Is Part Of: Archives of disease in childhood. Volume 108(2023)Supplement 1
• Page Start: A11
• Page End: A11
• Publication Date: 2023-02-23
• Subjects: Children -- Diseases -- Periodicals; Infants -- Diseases -- Periodicals; 618.920005
• DOI: 10.1136/archdischild-2023-gosh.29
• Languages: English
• ISSNs: 0003-9888

Abstract

Abstract : Introduction: Digital health data forms the foundation for most research and many patient treatment decisions. Whether developing a new clinical decision-making tool or carrying out data-driven research, it is essential that the data itself can be trusted and relied upon. Laboratory tests are a common data source, which generate results based on patient samples, with minimal user input. Point of Care Tests (POCT) rely more heavily on user-entered values, increasing the potential for introducing human error, systematic biases, and approximation. We analysed a statistical method to detect tests that show the most non-random variation, indicative that the data may be less credible and require further investigation. Methods: Laboratory data for 2020-2021 at the Great Ormond Street Hospital were extracted from the electronic patient record system. These data were explored with the programming language R using the secure Aridhia Digital Research Environment (DRE). Using runs analysis on a per-test per-day basis, the Anhøj criteria were used to quantitatively generate a 'suspicion index'. The average suspicion index per test over the year was used for comparison between tests. A high suspicion index indicated that the values observed were less likely due to random variation and therefore potentially less credible for further analyses. Results: Based on 4, 749, 278 test results across 2, 371 distinct test types, when ranked in order of credibility, the 23 least credible tests were all POCT. The suspicion indices ranged from 0.8 to 15.6. There was a clear separation of POCT from the laboratory tests, with the latter being generally more credible. Impact: Low credibility tests could adversely affect clinical tools designed to utilise them. By calculating the credibility of tests, low scoring tests could be identified and targeted for improvement or excluded from, or modified for, clinical tool development.