Transforming Patient Feedback Into Actionable Insights Through Natural Language Processing: Knowledge Discovery and Action Research Study
Bibliographic record
Abstract
Background: Patient feedback has emerged as a critical measure of health care quality and a key driver of organizational performance. Traditional manual analysis of unstructured patient feedback presents significant challenges as data volumes grow, making it difficult to extract meaningful patterns and actionable insights efficiently. Objective: We attempted to develop and evaluate a comprehensive methodology for analyzing patient feedback data using natural language processing and Knowledge Discovery in Databases (KDD) approaches, aiming to identify key patterns, themes, and variations in patient experiences across different demographic groups and care settings, and to translate these insights into actionable improvements in health care delivery. Methods: This study applied an integrated KDD-action research framework to analyze 126,134 patient feedback entries collected at Alexandra Hospital, Singapore, in 2023. A comprehensive suite of text mining techniques, including sentiment analysis, topic modeling, emotion detection, and aspect-based sentiment analysis, was employed to uncover patterns in patient-reported experiences. The dataset included 92,578 (73.4%) entries containing free-text comments, comprising 1,568,932 tokens with an average comment length of 16.9 words. Multiple analytical techniques were used to ensure the validity and reliability of the findings. Stakeholder engagement throughout the research process facilitated the translation of analytical insights into practical improvements. The study was granted an exemption from ethical review by the National Healthcare Group Domain Specific Review Board (number: NUS-IRB-2025-087E), with a waiver of informed consent granted for this retrospective analysis of deidentified patient feedback data. Results: Text mining analysis revealed a moderately positive overall sentiment across the feedback corpus (average polarity score: 0.42), with 68.8% (63,685/92,578) of comments classified as positive, 25.4% (23,515/92,578) as neutral, and 5.8% (5378/92,578) as negative. Topic modeling identified 10 distinct topics, including staff attitude and service (9443/92,578, 10.2%), health care staff professionalism (9350/92,578, 10.1%), hospital environment (9258/92,578, 10.0%), and waiting time (9258/92,578, 10.0%). Aspect-based sentiment analysis highlighted nurse attitude (sentiment score: 0.65), staff helpfulness (0.61), and doctor expertise (0.58) as the most positive aspects, while waiting time (-0.42) and billing transparency (-0.28) emerged as the most negative. Demographic segmentation revealed significant variations in patient priorities, with younger patients (<35 years) expressing 37% more concerns about digital accessibility and efficiency than older patients who valued face-to-face interactions 42% more highly. Implementation of targeted interventions based on these findings resulted in measurable improvements, including an 18% increase in waiting time satisfaction, a 15% improvement in doctor-patient communication ratings, and a 23% reduction in billing-related complaints. Conclusions: The integration of natural language processing techniques with KDD and action research principles provides a powerful framework for transforming unstructured patient feedback into actionable insights for health care improvement. This approach enables health care organizations to understand the complex patterns and drivers of patient experience, identify targeted improvement opportunities, and implement evidence-based initiatives that enhance care quality and patient-centeredness.
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 | 0.053 | 0.097 |
| Meta-epidemiology (narrow) | 0.001 | 0.000 |
| Meta-epidemiology (broad) | 0.001 | 0.001 |
| Bibliometrics | 0.003 | 0.003 |
| Science and technology studies | 0.002 | 0.004 |
| Scholarly communication | 0.005 | 0.004 |
| Open science | 0.002 | 0.004 |
| Research integrity | 0.001 | 0.002 |
| Insufficient payload (model declined to judge) | 0.002 | 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".