Analyzing Patient Complaints in Web-Based Reviews of Private Hospitals in Selangor, Malaysia, Using Large Language Model–Assisted Content Analysis: Mixed Methods Study
Bibliographic record
Abstract
BACKGROUND: Large language model (LLM)-assisted content analysis (LACA) is a modification of traditional content analysis, leveraging the LLM to codevelop codebooks and automatically assign thematic codes to a web-based reviews dataset. OBJECTIVE: This study aims to develop and validate the use of LACA for analyzing hospital web-based reviews and to identify themes of issues from web-based reviews using this method. METHODS: Web-based reviews for 53 private hospitals in Selangor, Malaysia, were acquired. Fake reviews were filtered out using natural language processing and machine learning algorithms trained on yelp.com validated datasets. GPT-4o mini model application programming interface (API) was then applied to filter out reviews without any quality issues. In total, 200 of the remaining reviews were randomly extracted and fed into the GPT-4o mini model API to produce a codebook validated through parallel human-LLM coding to establish interrater reliability. The codebook was then used to code (label) all reviews in the dataset. The thematic codes were then summarized into themes using factor analysis to increase interpretability. RESULTS: A total of 14,938 web-based reviews were acquired, of which 1121 (9.3%) were fake, 1279 (12%) contained negative sentiments, and 9635 (88%) did not contain any negative sentiment. GPT-4o mini model subsequently inducted 41 thematic codes together with their definitions. Average human-GPT interrater reliability is perfect (κ=0.81). Factor analysis identified 6 interpretable latent factors: "Service and Communication Effectiveness," "Clinical Care and Patient Experience," "Facilities and Amenities Quality," "Appointment and Patient Flow," "Financial and Insurance Management," and "Patient Rights and Accessibility." The cumulative explained variance for the six factors is 0.74, and Cronbach α is between 0.88 and 0.97 (good and excellent) for all factors except factor 6 (0.61: questionable). The factors identified follow a global pattern of issues identified from the literature. CONCLUSIONS: A data collection and processing pipeline consisting of Python Selenium, the GPT-4o mini model API, and a factor analysis module can support valid and reliable thematic analysis. Despite the potential for collection and information bias in web-based reviews, LACA of web-based reviews is cost-effective, time-efficient, and can be performed in real time, helping hospital managers develop hypotheses for further investigations promptly.
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 | 0.009 | 0.029 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.001 | 0.001 |
| Bibliometrics | 0.002 | 0.002 |
| Science and technology studies | 0.001 | 0.001 |
| Scholarly communication | 0.001 | 0.001 |
| Open science | 0.001 | 0.001 |
| Research integrity | 0.000 | 0.000 |
| Insufficient payload (model declined to judge) | 0.001 | 0.000 |
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.
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; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
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".