Interaction Techniques for Providing Sensitive Location Data of Interpersonal Violence with User-Defined Privacy Preservation

Research Background and Issues

• Problems or Challenges Identified by the Authors:

  1. Violence is a significant public health issue, and understanding the locations of violent incidents is crucial for prevention efforts. However, existing data suffer from incompleteness and privacy concerns.
  2. In hospital-collected data, many patients are unwilling or unable to provide exact addresses of violent incidents, making it difficult to identify potential high-violence areas.
  3. The Cardiff Model Tool (CMST), currently used for collecting incident locations, has issues with data quality and map accuracy, making it difficult to meet community-specific needs in terms of location detail.

• Importance of the Research:

  1. Violence incident maps can guide Community Safety Partnerships (CSPs) in allocating resources more effectively and reducing violence.
  2. Privacy concerns prevent some of the most vulnerable populations from sharing location information, potentially leading to a permanent lack of violence prevention resources in these areas.
  3. Improving data collection technologies can address the gaps in crime data, enhancing public health and safety research and decision-making capabilities.

• Motivation and Related Work:

  • The Cardiff Model combines hospital and police data to provide preliminary violence hotspot maps, but its implementation in the U.S. faces challenges due to data insufficiency and patient privacy concerns.
  • Existing privacy protection technologies (e.g., differential privacy and obfuscation techniques) may not be fully interpretable or applicable in medical or violence prevention contexts.

Solution

• Proposed Methods or Solutions:
  1. Interactive Violence Incident Reporting Interface (VIIPR):
     • A patient-driven interactive tool that allows selective sharing of geographic locations based on privacy needs.
     • Utilizes spatial subdivision algorithms to help users protect privacy through a "progressive zoom-in selection" approach.
  2. Exploration of Three Subdivision Algorithms:
     • Uniform Grid (UG): Divides areas into rectangular sections, offering simplicity and intuitiveness.
     • Voronoi Polygon (VP): Generates polygonal regions based on random points, providing more even segmentation.
     • Political Area Units (PAU): Uses existing administrative units (e.g., census tracts), enabling compatibility with socioeconomic or crime data.
• Key Technologies and Implementation Steps:
  • Algorithm Development:
    • Recursive multi-level subdivision (based on tree structures) to progressively divide large areas into smaller, clickable regions.
    • Based on Voronoi methods or existing political units like communities or census tracts.
  • Interactive Interface Design:
    • Map-based interactive tablet interface: Patients can provide location input via clicking or dragging.
    • Supports address search, manual location selection, and fuzzy area drawing features.
  • Data Storage and Analysis:
    • Interaction with secure cloud servers to store GeoJSON data.
    • Heatmap aggregation analysis based on confidence scores (generating heatmaps from the size and location of patient-input areas).
• Innovations:
  1. Offers an expandable "fuzzy" location reporting method, allowing users to stop sharing more detailed data based on their comfort level.
  2. Provides a map segmentation system compatible with existing health systems, police, and community data.

Research Outcomes

• Specific Outcomes:
  1. Improved location data collection rates:
     • The web-based VIIPR increased mappable locations from 13% using the CMST tool to 98.3%.
     • The average location information resolution provided by participants reached six-click precision.
  2. Provided three spatial subdivision algorithms, with experiments identifying the PAU method as the most suitable for practical needs while supporting multiple map layers.
  3. Enhanced secure storage mechanisms to ensure patient data privacy protection.
• Advantages Compared to Existing Solutions:
  • Compared to traditional CMST tools, VIIPR significantly improves location provision rates while respecting user privacy. Its results support finer-grained community violence data analysis.
  • The fuzzy data representation inherits the core concept of differential privacy and is easier for patients to understand.
• Experimental or Evaluation Results:
  1. Field Trial:
     • Among 60 participants, VIIPR collected 59 mappable locations.
     • Results were validated for significance using standard statistical methods (e.g., McNemar’s test, p<0.00001).
  2. Simulation Testing:
     • Evaluated the impact of different privacy levels (fewer or more clicks) on geographic heatmap analysis using real police and violence incident data.
     • Heatmaps from users with fewer clicks retained major trends but showed reduced fine-grained hotspots.
  3. Feedback Sessions:
     • Feedback from Cardiff and its partners (hospitals, public health, and police systems) indicated that users preferred heatmap formats over direct patient shapes.
     • Suggested adding filtering features by age, time, and other dimensions to meet practical needs.
• Limitations and Future Directions:
  1. Data Bias: The study sample is concentrated in Atlanta community hospitals, and regional and demographic characteristics (e.g., high proportions of non-white and low-income populations) may limit generalizability.
  2. Domestic Violence: The current tool cannot map domestic violence incidents, requiring further development of methods.
  3. International Challenges: Broader experiments and adjustments are needed to adapt the technology to different national contexts.
  4. Clinical Integration: Explore ways to reduce technical/personnel barriers for deployment, such as developing standalone self-service or home-based versions.