AI-Supported Electrocardiogram Interpretation: The Effect of Support Presentation on Diagnostic Accuracy, Psychological Need Satisfaction, and Diagnosis Time

Paper Title

AI-Supported Electrocardiogram Interpretation: The Effect of Support Presentation on Diagnostic Accuracy, Psychological Need Satisfaction, and Diagnosis Time

Publication Info

• Topic area: Interaction design for AI-supported clinical decision systems.
• Keywords: Explainable AI, ECG interpretation, clinical decision support, user experience, psychological need satisfaction, autonomy, competence, security, two-stage support, diagnosis accuracy.

Background and Problem

• Problem / challenge: Despite the availability of diagnostic support algorithms for ECG interpretation, the optimal presentation format and timing of AI support remain underexplored. Existing systems often fail to balance diagnostic accuracy with user experience (UX), particularly psychological need satisfaction.
• Significance: Accurate ECG interpretation is critical for patient safety, but it is a cognitively demanding and error-prone task. Enhancing the interaction between clinicians and AI systems could improve both diagnostic outcomes and clinician satisfaction.
• Motivation and related work: Prior studies have shown that AI support can improve diagnostic accuracy but often neglect UX factors like autonomy and competence. Research has also highlighted the potential for explainable AI to enhance understanding, but the impact of visual explanations and two-stage protocols on UX and efficiency is unclear.

Solution

• Proposed approach: A preregistered experimental study comparing five ECG diagnostic support conditions, including no support, immediate support (text or text + visual marking), and two-stage support (text or text + marking provided after initial independent review).
• Novelty:
  1. Introduction of a two-stage protocol without requiring a preliminary written diagnosis.
  2. Integration of psychological need satisfaction (autonomy, competence, security) into the evaluation of AI-supported clinical decision-making.
  3. Quantitative evidence on the impact of visual explainable AI and interaction design on diagnostic accuracy, UX, and diagnosis time.
• Procedure and key techniques:
  • Participants (n=58 clinicians) diagnosed ECGs under five conditions.
  • Dependent variables: diagnosis accuracy, psychological need satisfaction/frustration (autonomy, competence, security), diagnosis time, and condition preference.
  • Randomized within-subjects design with preselected ECGs and simulated AI support (always correct).
  • Statistical analysis included repeated-measures ANOVA and exploratory post hoc tests.

Results

• Concrete findings:
  • Diagnosis accuracy improved significantly with AI support (75%-78% vs. 57% for no support).
  • Two-stage support conditions resulted in the shortest diagnosis times (57-65s vs. 85-101s for other conditions).
  • Two-stage conditions yielded higher competence (4.83 vs. 4.39-4.47), autonomy (5.12-4.97 vs. 4.58-4.89), and security satisfaction (4.60-4.69 vs. 3.88-4.39) compared to immediate support or no support.
• Advantage over baselines:
  • Visual marking improved accuracy over text-only support (76%-78% vs. 70%-75%).
  • Two-stage support mitigated autonomy and competence frustration seen in immediate support conditions.
• Experiments / evaluation:
  • Participants: 58 clinicians (29 female, 29 male; mean age 32.3 years; mean work experience 2.2 years).
  • Materials: 15 ECGs per condition, randomized presentation.
  • Metrics: Accuracy, need satisfaction/frustration (7-point Likert scale), diagnosis time, and preference.
• Limitations and future work:
  • AI support was always correct, which may not reflect real-world scenarios.
  • Limited generalizability due to the use of American-standard ECGs with German clinicians and absence of patient history.
  • Future work should explore imperfect AI support, team decision-making, and domain-specific UX scales.

Summary

This study demonstrated that AI-supported ECG interpretation improves diagnostic accuracy and clinician experience when designed thoughtfully. Two-stage support protocols, which allow clinicians to form independent opinions before receiving AI assistance, enhanced competence, autonomy, and security satisfaction while reducing diagnosis time. Visual explainable AI (e.g., ECG segment highlighting) further boosted accuracy. These findings highlight the importance of integrating UX considerations into clinical decision support design to balance performance and user satisfaction, ultimately contributing to safer and more effective healthcare.