AI-Based Diagnostic Platform Capabilities With Lyme Disease as a Use Case: Integrative Exploration

Area of Science:

• Medical informatics
• Infectious diseases
• Artificial intelligence in healthcare

Background:

• Lyme disease (LD) is a common US vector-borne illness, often difficult to diagnose due to overlapping symptoms and limitations in current testing protocols.
• Clinical diagnosis of LD frequently relies on symptoms, exposure history, and clinician judgment, particularly when the characteristic rash is absent.
• Online symptom checkers utilizing AI are increasingly used for diagnostic information, prompting an evaluation of their utility compared to traditional methods.

Purpose of the Study:

• To explore the diagnostic utility (DU) of AI-based online symptom checkers for Lyme disease.
• To compare the performance of AI tools against serological (CDC+) and clinical diagnostic pathways for LD.
• To analyze symptom patterns, severity distributions, and co-occurring conditions related to AI-assisted LD diagnoses.

Main Methods:

• A survey of patients with confirmed or probable LD provided data on diagnostic pathways, symptoms, and time to diagnosis.
• Patient cases were input into three leading AI symptom checker platforms (MediFind, Isabel, WebMD) to assess diagnostic performance.
• Descriptive analytics and logistic regressions were employed to identify patterns in DU and interaction effects.

Main Results:

• Diagnostic utility (DU) of AI platforms varied significantly by platform and symptom severity.
• DU improved with higher symptom severity (≥3) and was slightly higher in clinically diagnosed cohorts versus CDC+ cohorts.
• Clinically diagnosed individuals showed greater sensitivity to symptom severity changes, though platform consistency varied.

Conclusions:

• AI-based symptom checkers show potential as a supplementary tool for early diagnostic reasoning in complex diseases like LD, especially with severe symptoms.
• The study highlights the need for algorithmic refinement and standardized validation frameworks due to observed inconsistencies across AI platforms.
• This preliminary research provides a foundation for larger studies to enhance the reliability of AI diagnostic tools.