The Human Connection

Healthcare can often feel fragmented and impersonal. Records are scattered, communication is difficult, and patients can feel lost in the system. Through our work with Helix, we're tackling this challenge head-on with innovative graph-based technology solutions.

Traditionally, healthcare data is stored in relational databases, where information is divided across numerous locations. While effective for transactional operations, this model often struggles to represent the complex, interconnected nature of healthcare relationships. A patient's journey usually involves multiple doctors, clinics, and family members. Representing these intricate connections in a relational model can lead to complex queries, performance bottlenecks, and a fragmented view of the patient's health record.

We have developed a new approach: building a person-centered healthcare network, a graph of interconnected data points where the person is the central node. This is a fundamental shift in perspective. Instead of treating patients, doctors, and organizations as separate records in different tables, we model them as nodes in a graph, connected by meaningful relationships.

This graph-based approach leverages concepts from network theory. In our healthcare context, we can traverse relationships to discover connections: a patient's primary care physician connects to specialists, who connect to other care providers, creating a rich web of healthcare relationships that can be explored algorithmically.

To build a truly person-centric system, we needed a data model that mirrors the real world of healthcare. In a graph, we don't think in terms of tables and rows, but in terms of entities and their relationships.

Our graph architecture contains interconnected node types like:

• Persons: The individuals at the center of healthcare, whether they are acting as patients or healthcare professionals.

• Organizations: The hospitals, clinics, and other institutions that provide care.

• Roles and Responsibilities: The specific roles people have within an organization, like a doctor, nurse, or administrator.

• Resources: Things like medical equipment or rooms that are part of the care process.

• Treatments and Care Plans: The links between a patient and the specific care they are receiving, from individual appointments to long-term treatment plans.

Instead of using complex database joins to connect these concepts, we create direct relationships (edges) between them. This enables powerful graph traversal algorithms that can explore multi-hop relationships efficiently. For example, when a patient is discharged, we can automatically identify all relevant care team members who need to be notified and coordinate follow-up appointments. The graph structure naturally supports these computational approaches, making complex healthcare workflows both automated and performant.

Healthcare data is sensitive, and controlling who can see what is paramount. Traditional access control systems can be rigid and difficult to manage. With our graph, we've implemented a more intuitive and powerful approach called Relationship-Based Access Control (RelBAC).

With RelBAC, permissions are not just a set of roles and policies; they are part of the data model itself. We can ask questions directly of the graph through sophisticated traversal queries, such as:

• "Is this person a doctor in the same department as this patient?"
• "Does this nurse have the correct relationships to view these lab results?"
• "Can this specialist access patient data through their referral network?"

The graph database can answer these questions by executing path-finding algorithms that follow the relationships we've defined. For instance, to determine if a nurse can access a patient's lab results, the system traces the relationships: nurse → works in → cardiology department → treats → patient → has → lab results. If this path exists and meets our security criteria, access is granted. Our system can efficiently compute these authorization paths, ensuring that data is only accessible to those with the right relationship chain to it, providing a flexible and robust security model that adapts as relationships evolve.

To bring our healthcare network to life, we selected Neo4j, a leading graph database technology. Our backend services communicate with the database through Cypher queries that can efficiently traverse complex relationship patterns. However, one of our key architectural decisions was to implement a hybrid approach that separates relationship storage from property storage.

In our Neo4j graph, we store the structural relationships between healthcare entities - who is connected to whom and how they're connected. However, we keep the graph itself "thin" and anonymous by storing minimal data directly in the graph nodes. Each node contains primarily an ID that serves as a key to corresponding document(s) stored in DynamoDB.

This hybrid architecture provides several critical advantages:

• Privacy and Security: The graph structure remains anonymous, containing only relationship patterns without sensitive personal or medical information.

• Performance: Graph traversals are extremely fast since nodes are lightweight, while property lookups are optimized by using the same keys for both the graph and document store.

• Scalability: We can scale relationship queries (Neo4j) and property storage (DynamoDB) independently based on different usage patterns.

• Flexibility: DynamoDB's document model allows us to store varying schemas for different entity types without rigid table structures.

When executing queries, we first traverse the graph to identify the relevant nodes and relationships, then perform key-based lookups in DynamoDB to retrieve the detailed properties. This pattern allows us to leverage the strengths of both technologies.

By placing the person at the center of our data model and leveraging the computational power of graph databases, we are building a more intelligent, secure, and connected healthcare platform. Our person-centric healthcare network is not just a technical implementation detail; it's a paradigm shift that allows us to apply network science principles to healthcare data. This approach has enabled us to implement sophisticated features like RelBAC, and it provides a flexible foundation for building the next generation of healthcare applications powered by graph theory and network analysis.