A new expert system with diagnostic accuracy for pediatric upper respiratory conditions

• Published in: Healthcare analytics (New York, N.Y.) Vol. 2; p. 100042
• Main Authors: Yazdani, Shahram, Lerner, Carlos, Kulkarni, Deepa, Kamzan, Audrey, Henry, Ronald C.
• Format: Journal Article
• Language: English
• Published: Elsevier Inc 01-11-2022; Elsevier
• Subjects: Computer-assisted diagnosis; Diagnostic analytics; Expert system; Pediatric upper respiratory conditions; Rheumatic heart disease; Variable evoking strength; Variable evoking strength; vES; Expert system; CAD; PMHI; Rheumatic heart disease; Computer-assisted diagnosis; vESCAD; Diagnostic analytics; URI; Pediatric upper respiratory conditions; ES
• ISSN: 2772-4425; 2772-4425
• DOI: 10.1016/j.health.2022.100042

The high prevalence of pediatric upper respiratory conditions combined with poor availability of specialized care for the proper diagnosis of these diseases continues to be a significant cause of morbidity and mortality worldwide. Despite advances in computer-assisted diagnosis, there are no automated triage tools that can effectively provide a first step guide for the parents or cognitive assistance for the care providers to improve the efficacy and accuracy of diagnosis in these children. We propose a new approach to designing expert systems using an integration of logical criteria and variable evoking strength to improve these tools’ diagnostic accuracy. We conducted a retrospective chart review of the electronic health records of children evaluated for several upper respiratory conditions at a large university-based health system in the United States. The diagnoses recorded in the charts were compared to the output from the tool. The accuracy rate of the tool was based on the assumption that human physicians are the gold standard of diagnosis. A total of 138 cases, ranging in age between 6 months to 18 years, were reviewed. Forty-three percent of the participants self-identified as female. The average accuracy of the tool in matching the physician diagnosis of the eight types of upper respiratory conditions that were included in this study was 75% for the first diagnosis and 84% for matching one of the top two differential diagnoses. Integration of logical diagnostic criteria and variable evoking strength into the design of expert systems can significantly improve the diagnostic accuracy of these tools.
[Display omitted]
•Expert systems are alternative to the existing artificial intelligence tools for computer-assisted diagnosis.•Expert systems do not require a large training data set or black-box algorithms prone to bias or user mistrust.•We propose a new approach for designing the inference engine of medical expert systems.•We integrate evidence-based diagnostic criteria and variable-evoking strength values in the inference engine.•A retrospective chart review shows a 75% match between the first-ranked diagnosis by the system and that of the physician.