Human–AI Collaboration for Explainable Discovery of Atypical Heart Failure Profiles Through Unsupervised Anomaly Detection and Clinician-Guided Decision Support
Main Article Content
Abstract
Artificial intelligence (AI) has demonstrated promising capabilities for discovering complex patterns in clinical data; however, meaningful clinical use requires more than autonomous model output. Clinicians must be able to inspect, contextualize, and challenge AI-generated findings before those findings influence patient assessment. This study therefore reframes explainable unsupervised anomaly detection as a human–AI collaboration framework for identifying and interpreting atypical clinical profiles among patients with heart failure. The AI component uses Isolation Forest to discover unusual multidimensional patient profiles without incorporating mortality outcomes during model development, while SHAP-based explanations expose the clinical variables contributing to global and patient-level anomaly patterns. The framework was evaluated using the Heart Failure Clinical Records dataset consisting of 299 patients and 12 clinical variables. The model identified 45 patients (15.1%) as anomalous profiles and 254 patients (84.9%) as non-anomalous profiles. The anomalous subgroup demonstrated a mortality rate of 60.0%, compared with 27.2% among non-anomalous patients, and showed a mortality enrichment ratio of approximately 1.87 relative to the overall cohort. Clinical profile analysis revealed substantially higher creatine phosphokinase and serum creatinine levels among anomalous patients, with fold changes of 2.99 and 1.76, respectively. SHAP analysis identified high blood pressure, sex, creatine phosphokinase, smoking status, ejection fraction, platelet count, age, and serum creatinine as important contributors to anomaly formation. A local explanation further demonstrated an atypical case characterized by preserved ejection fraction together with severe biochemical abnormalities. Within the proposed collaborative workflow, these AI outputs are treated as interpretable evidence for clinician review rather than autonomous diagnoses: the AI highlights unusual patterns and explains them, while the clinician contributes patient context, medical knowledge, and final judgment. Because this retrospective study did not include a clinician user study, the results demonstrate the technical readiness of the framework for human–AI collaboration rather than improved team performance. Future prospective evaluation should examine how clinicians use, verify, and override explainable anomaly findings in real clinical workflows.
Article Details

This work is licensed under a Creative Commons Attribution-ShareAlike 4.0 International License.
Authors who publish with International Journal for Applied Information Management agree to the following terms: Authors retain copyright and grant the International Journal for Applied Information Management right of first publication with the work simultaneously licensed under a Creative Commons Attribution License (CC BY-SA 4.0) that allows others to share (copy and redistribute the material in any medium or format) and adapt (remix, transform, and build upon the material) the work for any purpose, even commercially with an acknowledgement of the work's authorship and initial publication in International Journal for Applied Information Management. Authors are able to enter into separate, additional contractual arrangements for the non-exclusive distribution of the journal's published version of the work (e.g., post it to an institutional repository or publish it in a book), with an acknowledgement of its initial publication in International Journal for Applied Information Management. Authors are permitted and encouraged to post their work online (e.g., in institutional repositories or on their website) prior to and during the submission process, as it can lead to productive exchanges, as well as earlier and greater citation of published work (See The Effect of Open Access).