Explaining Call Recommendations in Nursing Homes: A User-Centered Design Approach for Interacting with Knowledge-Based Health Decision Support Systems

Title of the Paper

Explaining Call Recommendations in Nursing Homes: A User-Centered Design Approach for Interacting with Knowledge-Based Health Decision Support Systems

Bibliographic Information

• Subject Area: Healthcare recommendation systems, user-centered design, interaction design in high-risk scenarios
• Keywords: Health recommendation systems, decision support systems, user-centered design, interactive recommendation systems, priority recommendations, user feedback, transparency, trust, user interface

Research Background and Issues

• Identified Problems or Challenges:

  1. Health recommendation systems are still in their early stages of development, with trust and reliability needing improvement.
  2. Call systems in nursing homes lack priority recommendations, transparency, and user feedback features, potentially leading to delayed response times and reduced satisfaction with care processes.
  3. Common challenges faced by recommendation systems include data sparsity, cold start problems, and the "black box" phenomenon.

• Why It Matters:

  1. Call systems are critical communication tools between nursing home residents and caregivers, directly impacting residents' safety and well-being.
  2. Rapid response and clear prioritization are crucial for improving care efficiency and resident satisfaction.

• Research Motivation and Related Work:

  1. A literature review indicates that recommendation systems in high-risk scenarios need to consider user interaction, transparency, and user control.
  2. Existing healthcare decision support systems are mostly rule-based or machine learning-based, with limited research on user interaction methods.

Proposed Solution

• Proposed Method or Solution: The authors propose a mobile application that integrates a user-centered design process with a health decision support system. Starting from the needs of residents and caregivers, the user interface was designed and iteratively optimized to support call recommendations and their explanations in nursing homes.

• Innovative Aspects of the Solution:

  1. This is the first application of a user-centered design approach to a priority recommendation system for nursing home calls.
  2. A hybrid recommendation method (rule-based and machine learning) was adopted, combining user feedback and historical data.
  3. Explored explainable recommendation techniques to enhance transparency and user trust.

• Implementation Steps and Key Technologies:

  1. Exploratory Qualitative User Research: Conducted observations and semi-structured interviews with caregivers and residents to understand the strengths and weaknesses of existing call systems.
  2. Prototype Design and Iteration:
     • Low-fidelity prototype: Designed five priority recommendation labels.
     • High-frequency prototype iteration: Optimized interface transparency, added historical record functionality, and introduced a quick feedback mechanism.
  3. Final Prototype Development:
     • Added a suspension list feature to reduce caregivers' workload.
     • Provided detailed explanations and past call records to help caregivers better understand residents' needs.
  4. Recommendation Engine:
     • Used a random forest model as the machine learning algorithm, combining sensor data and user feedback.
     • During the cold start phase, a rule-based engine provided priority predictions.

Research Outcomes

• Specific Outcomes:

  1. Successfully deployed a health decision support system for priority call recommendations in a nursing home, operational for four months.
  2. The system significantly increased the frequency of caregiver feedback, with an intuitive and convenient feedback process.
  3. Surveys showed high levels of user-perceived transparency and interaction satisfaction.

• Advantages Compared to Existing Solutions:

  1. The system provided detailed and dynamic recommendation explanations, significantly improving users' understanding of the algorithm and recommendation rationale.
  2. Priority recommendations for call notifications demonstrated higher levels of user control and trust.
  3. The model improved through user feedback, mitigating the cold start problem.

• Experimental or Evaluation Results:

  1. The system processed a total of 2,610 resident call records, with 29.5% of the records receiving caregiver feedback.
  2. Accuracy evaluation showed good performance in priority predictions, with the highest classification accuracy reaching 71%.

• Limitations and Future Directions:

  1. Limitations:
     • Some sensor data were not fully integrated due to conditional constraints.
     • The user group size was relatively small, potentially limiting the generalizability of the results.
     • The system's performance in predicting medium-priority calls was suboptimal and requires further improvement.
  2. Future Directions:
     • Optimize the recommendation engine algorithm, including feature selection, hyperparameter tuning, and class sampling.
     • Further simplify the user interface to enhance usability.
     • Expand the user base for larger-scale testing and feedback collection.

Conclusion

This paper presents a user-centered design-driven solution for a priority recommendation system for nursing home calls, offering a new research direction to improve the quality and efficiency of care for residents. The study demonstrates that integrating transparency, user control, and dynamic feedback into interaction design can significantly enhance the user experience and trust in healthcare decision support systems.