Assessing the Efficacy of ChatGPT Versus Human Researchers in Identifying Relevant Studies on mHealth Interventions for Improving Medication Adherence in Patients With Ischemic Stroke When Conducting Systematic Reviews: Comparative Analysis
Bibliographic record
Abstract
BACKGROUND: ChatGPT by OpenAI emerged as a potential tool for researchers, aiding in various aspects of research. One such application was the identification of relevant studies in systematic reviews. However, a comprehensive comparison of the efficacy of relevant study identification between human researchers and ChatGPT has not been conducted. OBJECTIVE: This study aims to compare the efficacy of ChatGPT and human researchers in identifying relevant studies on medication adherence improvement using mobile health interventions in patients with ischemic stroke during systematic reviews. METHODS: This study used the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) guidelines. Four electronic databases, including CINAHL Plus with Full Text, Web of Science, PubMed, and MEDLINE, were searched to identify articles published from inception until 2023 using search terms based on MeSH (Medical Subject Headings) terms generated by human researchers versus ChatGPT. The authors independently screened the titles, abstracts, and full text of the studies identified through separate searches conducted by human researchers and ChatGPT. The comparison encompassed several aspects, including the ability to retrieve relevant studies, accuracy, efficiency, limitations, and challenges associated with each method. RESULTS: A total of 6 articles identified through search terms generated by human researchers were included in the final analysis, of which 4 (67%) reported improvements in medication adherence after the intervention. However, 33% (2/6) of the included studies did not clearly state whether medication adherence improved after the intervention. A total of 10 studies were included based on search terms generated by ChatGPT, of which 6 (60%) overlapped with studies identified by human researchers. Regarding the impact of mobile health interventions on medication adherence, most included studies (8/10, 80%) based on search terms generated by ChatGPT reported improvements in medication adherence after the intervention. However, 20% (2/10) of the studies did not clearly state whether medication adherence improved after the intervention. The precision in accurately identifying relevant studies was higher in human researchers (0.86) than in ChatGPT (0.77). This is consistent with the percentage of relevance, where human researchers (9.8%) demonstrated a higher percentage of relevance than ChatGPT (3%). However, when considering the time required for both humans and ChatGPT to identify relevant studies, ChatGPT substantially outperformed human researchers as it took less time to identify relevant studies. CONCLUSIONS: Our comparative analysis highlighted the strengths and limitations of both approaches. Ultimately, the choice between human researchers and ChatGPT depends on the specific requirements and objectives of each review, but the collaborative synergy of both approaches holds the potential to advance evidence-based research and decision-making in the health care field.
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 imitationNot 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.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.633 | 0.787 |
| Meta-epidemiology (narrow) | 0.005 | 0.004 |
| Meta-epidemiology (broad) | 0.028 | 0.033 |
| Bibliometrics | 0.043 | 0.036 |
| Science and technology studies | 0.004 | 0.008 |
| Scholarly communication | 0.015 | 0.017 |
| Open science | 0.006 | 0.013 |
| Research integrity | 0.008 | 0.004 |
| Insufficient payload (model declined to judge) | 0.013 | 0.001 |
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.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; the direct Gemma label and the distilled Codex classifier agree on what is shown here.
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".