MétaCan
Menu
← Back to cohort
Record W4413802812 · doi:10.2196/76458

Evaluating ChatGPT’s Utility in Biologic Therapy for Systemic Lupus Erythematosus: Comparative Study of ChatGPT and Google Web Search

2025· article· en· W4413802812 on OpenAlexvenueno aff
Kai Li, Luyi Li, Bo Liu

Bibliographic record

VenueJMIR Formative Research · 2025
Typearticle
Languageen
FieldMedicine
TopicArtificial Intelligence in Healthcare and Education
Canadian institutionsnot available
Fundersnot available
KeywordsSystemic therapyMedicineWeb applicationComputer scienceWorld Wide WebInternal medicine

Abstract

fetched live from OpenAlex

Background: Systemic lupus erythematosus (SLE) is a life-threatening, multisystem autoimmune disease. Biologic therapy is a promising treatment for SLE. However, public understanding of this therapy is still insufficient, and the quality of related information on the internet varies, which affects patients' acceptance of this treatment. The effectiveness of artificial intelligence technologies, such as ChatGPT (OpenAI), in knowledge dissemination within the health care field has attracted significant attention. Research on ChatGPT's utility in answering questions regarding biologic therapy for SLE could promote the dissemination of this treatment. Objective: This study aimed to evaluate ChatGPT's utility as a tool for users to obtain health information about biologic therapy for SLE. Methods: This study extracted 20 common questions related to biologic therapy for SLE, their corresponding answers, and the sources of these answers from both Google Web Search and ChatGPT-4o (OpenAI). Then, based on Rothwell's classification, the questions were categorized into 3 main types: fact, policy, and value. The sources of the answers were classified into 5 categories: commercial, academic, medical practice, government, and social media. The accuracy and completeness of the answers were assessed using Likert scales. The readability of the answers was evaluated using the Flesch Reading Ease and Flesch-Kincaid Grade Level (FKGL) scores. Results: The study found that, in terms of question types, ChatGPT-4o had the highest proportion of fact questions (10/20), followed by policy (7/20) and value (3/20). Google Web Search had the highest proportion of fact questions (12/20), followed by value (5/20) and policy (3/20). In terms of website sources, ChatGPT-4o's answers were sourced from 48 sources, with the majority coming from academic sources (29/48). Google Web Search provided answers from 20 sources, with an even distribution across all 5 categories. For accuracy, ChatGPT-4o's mean score of 5.83 (SD 0.49) was higher than that of Google Web Search (mean 4.75, SD 0.94), with a mean difference of 1.08 (95% CI 0.61-1.54). For completeness, ChatGPT-4o's mean score of 2.88 (SD 0.32) was higher than that of Google Web Search (mean 1.68, SD 0.69), with a mean difference of 1.2 (95% CI 0.96-1.44). For readability, the Flesch Reading Ease and Flesch-Kincaid Grade Level scores for ChatGPT-4o and Google Web Search were 11.7 and 14.9, and 16.2 and 20, respectively, indicating that both texts were of high reading difficulty, requiring readers to have a college graduate-level reading proficiency. When asking ChatGPT to respond at a sixth-grade level, the readability of the answers significantly improved. Conclusions: ChatGPT's answers are characterized by accuracy, rigor, comprehensiveness, and professional supporting materials, and demonstrate humanistic care. However, the readability of the provided text is low, requiring users to have a college education background. Given the study's limitations in question scope, comparison dimensions, research perspectives, and language types, further in-depth comparative research is recommended.

Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.

How this classification was reachedexpand

Full frame machine prediction

Teacher imitation

Not calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.

metaresearch head score (Codex)0.008
metaresearch head score (Gemma)0.082
Version: metacan-v3-hybrid-931329e0061cValidation status: machine_predicted_unvalidated
Candidate categoriesnone
Consensus categoriesnone
DomainCandidate signal: none · Consensus signal: none
Study designCandidate signal: Non-randomized trial · Consensus signal: none
GenreCandidate signal: Empirical · Consensus signal: Empirical
Teacher disagreement score0.008
Threshold uncertainty score0.040

Distilled classifier scores by category (both heads)

CategoryCodexGemma
Metaresearch0.0080.082
Meta-epidemiology (narrow)0.0000.000
Meta-epidemiology (broad)0.0010.001
Bibliometrics0.0080.006
Science and technology studies0.0000.001
Scholarly communication0.0020.003
Open science0.0010.002
Research integrity0.0010.001
Insufficient payload (model declined to judge)0.0020.000

Machine scores (provisional)

The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.

Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.

Opus teacher head0.544
GPT teacher head0.616
Teacher spread0.072 · how far apart the two teachers sit on this one work
Validation statusscore_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from it

Classification

machine, unvalidated

Machine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.

The models applied no category: nothing in the taxonomy fit this work.
Study designNon-randomized trial
Domainnot available
GenreEmpirical

How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".

Quick stats

Citations2
Published2025
Admission routes1
Has abstractyes

Explore more

Same venueJMIR Formative Research→Same topicArtificial Intelligence in Healthcare and Education→French-language works237,207→