Enhancing Patient Care and Nursing Practice Through AI-Driven Analysis of Patient Complaints: A Pilot Study

Background/Objectives: The integration of Artificial Intelligence (AI) and Machine Learning (ML) in healthcare has shown promise for analyzing complex data and identifying patterns that can enhance clinical decision-making and patient care. Patient complaints represent a valuable yet underutilized data source for improving healthcare quality and patient safety. This study aimed to investigate whether an AI-driven model could effectively categorize patient complaints and extract actionable insights to support person-centered care and quality improvement in nursing practice. Method: A total of 368 retrospective patient complaint reports collected by a Patients’ Advisory Committee were analyzed using AI-driven clustering and classification techniques. The complaints were pre-processed through text cleaning, tokenization, and vectorization before being analyzed with both an unsupervised clustering algorithm and a supervised classification algorithm. Results: The AI-based classification demonstrated that complaint texts could be classified according to four existing PAC categories (based on 290 complaint reports): (1) Care and Treatment, (2) Communication, (3) Administrative Management, and (4) Availability. These areas encompassed themes such as diagnostic delays, poor communication, fragmented administrative processes, and limited accessibility. The results highlighted that communication failures, insufficient empathy, and lack of continuity were central to patient dissatisfaction. The results indicate that automated analysis can support the systematic identification of recurring issues and patterns in patient complaint reports. Conclusions: This pilot study suggests that an AI-driven model may support the systematic classification and exploration of patient complaint reports. The supervised classification model demonstrated moderate performance in approximating four pre-existing PAC categories, including that the complaint texts contained information relevant to predicting these categories. However, the findings should be interpreted cautiously, as the model was evaluated against existing PAC classifications rather than an independent gold standard. The study did not assess workload, processing time, or cost, and therefore no conclusions can be drawn regarding efficiency gains. However, larger datasets, independent expert validation, and careful consideration of ethical, legal, and governance requirements are needed before the model can be implemented more broadly.