Mining Patient Feedback and Complaints for Healthcare Insights
Leveraging existing patient comments and AI to uncover hidden healthcare themes and improve service quality safely.
Stock photo for illustration only, not from the actual event
- Australian health services sit on years of unread patient comments containing valuable insights.
- Theme-mining projects require no new data collection as existing corpuses are ready to use.
- The HCAT framework provides a standardized and comparable approach for classifying complaints.
- Human oversight and strict privacy compliance are essential when deploying AI in healthcare settings.
Most health services in Australia are sitting on several years of patient comments that nobody has read from start to finish. Likert scores are typically tabulated and placed into quarterly reports, while free text is merely skimmed by whoever compiles the pack, leaving the rest archived. This post forms part of the Practical AI in Health series, highlighting a low-cost, board-visible AI project that stays well away from direct clinical decision-making.
For those running Australian hospitals, data collection via AHPEQS—the ACSQHC's 12-item patient experience question set including a free-text field—is likely already underway. One implementation study analyzed 86,180 surveys across 36 private hospitals over 18 months, proving that a theme-mining project requires no new data collection; the corpus already exists.
Understanding data volumes is crucial before scoping. In a provincial health system's inpatient survey analysis, 43.4% of adult patients and 46.9% of paediatric caregivers left free-text comments. Topic modelling generated 86 adult and 35 paediatric topics, uncovering care elements that standard survey questions never asked about. Closed items can only measure what creators thought to ask in advance.
Stock photo for illustration only, not from the actual event
Applying AI to analyze free-text patient comments offers an effective entry point because it focuses on management and sentiment rather than direct clinical diagnosis. This low-risk setting allows hospitals to catch systemic issues—such as communication breakdowns or wait times—long before they escalate into formal complaints.
Complaints represent a separate and more sensitive data stream. Ahpra received 13,327 notifications regarding practitioners in 2024/25, marking a 19% increase from the previous year. Clinical care accounted for 35% of all notifications, while communication issues ranked second with 4,313 cases. Communication problems serve as early signals that appear in free-text comments months before surfacing formally.
Rather than jumping straight to unsupervised clustering, a structured approach is recommended. Starting with the Healthcare Complaints Analysis Tool (Gillespie and Reader, BMJ Qual Saf 2016)—derived from a systematic review of 59 studies and 88,069 complaints—distils 729 codes into three domains and seven problem categories: quality, safety, environment, institutional processes, listening, communication, and respect and patient rights.
Source: Dev.to
Found something wrong in this article? Report an issue with this article
Comments
Leave a Comment