2023medRxivOpen access

Diagnosis in Bytes: Comparing the Diagnostic Accuracy of Google and ChatGPT 3.5 as Diagnostic Support Tools

Guilherme Rodrigues Guimarães, Caroline Santos Silva, Jean Carlos Zambrano Contreras, Ricardo G. Figueiredo, Uros - Grupo de Pesquisa, Ricardo Brianezi Tiraboschi, Cristiano Mendes Gomes, José de Bessa

Open full text 0 citations

Abstract

ABSTRACT Objective Adopting digital technologies as diagnostic support tools in medicine is unquestionable. However, the accuracy in suggesting diagnoses remains controversial and underexplored. We aimed to evaluate and compare the diagnostic accuracy of two primary and accessible internet search tools: Google and ChatGPT 3.5. Method We used 60 clinical cases related to urological pathologies to evaluate both platforms. These cases were divided into two groups: one with common conditions (constructed from the most frequent symptoms, following EAU and UpToDate guidelines) and another with rare disorders - based on case reports published between 2022 and 2023 in Urology Case Reports. Each case was inputted into Google Search and ChatGPT 3.5, and the results were categorized as "correct diagnosis," "likely differential diagnosis," or "incorrect diagnosis." A team of researchers evaluated the responses blindly and randomly. Results In typical cases, Google achieved 53.3% accuracy, offering a likely differential diagnosis in 23.3% and errors in the rest. ChatGPT 3.5 exhibited superior performance, with 86.6% accuracy, and suggested a reasonable differential diagnosis in 13.3%, without mistakes. In rare cases, Google did not provide correct diagnoses but offered a likely differential diagnosis in 20%. ChatGPT 3.5 achieved 16.6% accuracy, with 50% differential diagnoses. Conclusion ChatGPT 3.5 demonstrated higher diagnostic accuracy than Google in both contexts. The platform showed acceptable accuracy in common cases; however, limitations in rare cases remained evident.

Open-access reader

About this research paper

What this paper is about

ABSTRACT Objective Adopting digital technologies as diagnostic support tools in medicine is unquestionable. However, the accuracy in suggesting diagnoses remains controversial and underexplored. We aimed to evaluate and compare the diagnostic accuracy of two primary and accessible internet search tools: Google and ChatGPT 3.5. Method We used 60 clinical cases related to urological pathologies to evaluate both platforms. These cases were divided into two groups: one with common conditions (constructed from the most frequent symptoms, following EAU and UpToDate guidelines) and another with rare disorders - based on case reports published between 2022 and 2023 in Urology Case Reports. Each case was inputted into Google Search and ChatGPT 3.5, and the results were categorized as "correct diagnosis," "likely differential diagnosis," or "incorrect diagnosis." A team of researchers evaluated the responses blindly and randomly. Results In typical cases, Google achieved 53.3% accuracy, offering a likely differential diagnosis in 23.3% and errors in the rest. ChatGPT 3.5 exhibited superior performance, with 86.6% accuracy, and suggested a reasonable differential diagnosis in 13.3%, without mistakes. In rare cases, Google did not provide correct diagnoses but offered a likely differential diagnosis in 20%. ChatGPT 3.5 achieved 16.6% accuracy, with 50% differential diagnoses. Conclusion ChatGPT 3.5 demonstrated higher diagnostic accuracy than Google in both contexts. The platform showed acceptable accuracy in common cases; however, limitations in rare cases remained evident.

Why it matters

A significance statement is not available in the OpenAlex record.

Key contribution

A contribution statement is not available in the OpenAlex record.

Method / approach

Method details are not available in the OpenAlex metadata.

Main findings

Findings are not separately available in the OpenAlex metadata.

Limitations

Limitations are not available in the OpenAlex metadata.

Applications

Application details are not available in the OpenAlex metadata.

Available abstract

ABSTRACT Objective Adopting digital technologies as diagnostic support tools in medicine is unquestionable. However, the accuracy in suggesting diagnoses remains controversial and underexplored. We aimed to evaluate and compare the diagnostic accuracy of two primary and accessible internet search tools: Google and ChatGPT 3.5. Method We used 60 clinical cases related to urological pathologies to evaluate both platforms. These cases were divided into two groups: one with common conditions (constructed from the most frequent symptoms, following EAU and UpToDate guidelines) and another with rare disorders - based on case reports published between 2022 and 2023 in Urology Case Reports. Each case was inputted into Google Search and ChatGPT 3.5, and the results were categorized as "correct diagnosis," "likely differential diagnosis," or "incorrect diagnosis." A team of researchers evaluated the responses blindly and randomly. Results In typical cases, Google achieved 53.3% accuracy, offering a likely differential diagnosis in 23.3% and errors in the rest. ChatGPT 3.5 exhibited superior performance, with 86.6% accuracy, and suggested a reasonable differential diagnosis in 13.3%, without mistakes. In rare cases, Google did not provide correct diagnoses but offered a likely differential diagnosis in 20%. ChatGPT 3.5 achieved 16.6% accuracy, with 50% differential diagnoses. Conclusion ChatGPT 3.5 demonstrated higher diagnostic accuracy than Google in both contexts. The platform showed acceptable accuracy in common cases; however, limitations in rare cases remained evident.

Key concepts: Medical diagnosis, Differential diagnosis, Diagnostic accuracy, Byte, Medicine, Differential (mechanical device), Computer science, Data mining

Related papers

Back to paper searchBrowse research topicsOriginal source
Diagnosis in Bytes: Comparing the Diagnostic Accuracy of Google and ChatGPT 3.5 as Diagnostic Support Tools — Research Paper | ScholarLens