Table of Contents

The healthcare industry deals with vast amounts of data daily. This data comprises clinical research, doctor’s notes, medical imaging, etc. With such diverse information, knowledge graphs (KGs) can analyze clinical data to extract unseen patterns and relationships. 

A healthcare knowledge graph can help doctors understand patient diagnoses and treatments, discover new medicines, and systematically advance clinical research. It can help build a robust clinical decision support system when combined with modern technologies like artificial intelligence and machine learning. 

A KG-powered clinical decision support system can help doctors relate to previously unforeseen diseases, such as the recent COVID-19 pandemic, to enable easy data retrieval and information aggregation and, eventually, derive better treatment. 

We’ll talk more about all of this but first, let’s learn more about different types of healthcare data.

What Type of Medical Data Fuels a Healthcare Knowledge Graph?

The modern healthcare system collects structured, unstructured, and semi-structured data from multiple touchpoints. These touchpoints carry all the information to describe a patient's medical journey and are vital to building knowledge graphs. Some common touchpoints include

  • EMR Interface: Data in an Electronic Medical Record (EMR) system is collected when a patient comes in for a check-up. At this point, the attendant collects the patient's personal information, such as name, age, and address. Additionally, the doctor records vitals (weight, height, blood pressure, etc.) and diagnoses (symptoms, previous medication, and any available medical history).
  • Lab Results: Results from various lab tests are sent back to healthcare providers and are stored with the rest of the patient’s information. Lab tests can include image data from X-rays or CT scans and vital readings from diagnostic tests (e.g., blood glucose levels).
  • Smart Gadgets: Although not in mainstream use, smart bands and watches track your vitals, including heart rate, SpO2, and blood glucose levels. This information can prove critical in an emergency, daily self-monitoring, and disease and medication management. 
  • Medical Research: The medical research community works tirelessly to bring new information to light. They help derive biological connections between diseases, symptoms, and the human body. 99 medical publications were registered in 2021 in the Archives of Medical Research alone.

All this information is scattered and meaningless. Doctors manually connect the dots for each patient and derive decisions based on limited knowledge. Healthcare knowledge graphs create connections between this distributed data and provide the basis for clinical decision support. 

Now, let’s discuss how a biomedical knowledge graph system is built.

Building a Biomedical Knowledge Graph

A biomedical knowledge graph amalgamates all the information available in various databases and creates a traversable graph. This graph consists of nodes representing different entities, such as disease, symptom, and patient information. These nodes are linked together via relation mappings that define how two entities relate to each other. The figure below represents what a healthcare knowledge graph looks like.

healthcare-knowledge-graph

A general healthcare knowledge graph

Building such a graph requires data and statistical models to learn and create relations. The statistical models build probabilities for relationships between entities, e.g., all possible symptoms of “Type 2 diabetes”. The final links are determined based on these likelihood estimations.

KGs can be used to create detailed patient profiles with robust disease mapping. With an appropriate graph traversal, healthcare providers can use these graphs to relate patients to known diseases via their symptoms and propose optimal treatment options with minimum side effects. The figure above only represents a small portion, and in reality, a KG is an ever-growing knowledge space with new information added every second. 

Some popular healthcare knowledge graphs and resources are:

  • Google Knowledge Graph (GKG): It contains information from multiple fields. As of 2020, the knowledge base grew to 500 billion facts and 5 billion entities. Any healthcare knowledge is verified by in-house doctors at Google.
  • Clinical Knowledge Graph: Contains more than 16 million nodes and 220 million relationships made up of numerous clinical data sources.
  • BioPortal: Contains over 1000 ontologies and about 80 million mappings of healthcare-related entities.
  • DBpedia: Constructed from structured data within the Wikipedia database. The 201604 release contained data about 301,000 species and 5000 diseases. As of June 2021, it contains 850 million RDF triples

Now, let's move on to how knowledge graphs in healthcare aim to transform this industry.

Assisting Clinical Decisions with Knowledge Graphs

A clinical decision support system (CDSS) assists healthcare providers in their day-to-day clinical practice. These include reminders for scheduled checkups and help them identify any unusual information from the recorded medical data. 

Data scientists have been working tirelessly to create intelligent AI models for healthcare. These models allow a CDSS to make critical decisions with high precision. In the future, researchers hope that these CDSS will work without human supervision. However, human characteristics are unique, so it is difficult to define optimal medical outcomes for each individual.

In clinical medicine, it is not uncommon for doctors to come across a situation where it is difficult to determine a diagnosis. Perhaps the patient accompanies several symptoms, making it hard to reach a conclusion.

Hence, researchers are actively experimenting with the possibility of utilizing complex knowledge graphs for clinical decision support systems. Presently, there is a lot of work being done for drug discovery using knowledge graphs. The information within the KG nodes helps test hypotheses related to possible treatments for certain diseases. These also help healthcare professionals better understand their patients. 

A clinical decision support system backed by a dense knowledge base helps doctors visualize all possibilities. Since all the nodes are linked, the knowledge graph can link the present symptoms to common diseases. This narrows the doctor's search and encourages their confidence in making an informed medical decision.

Knowledge graphs also answer the age-old question that has haunted AI, How do AI models work? 

One major complication holding back AI in healthcare is the inexplicable nature of the models. AI models are mostly black-box systems with no understandable human reasoning attached to their outputs.  KGs are naturally built to be explainable. Combined with AI models, they offer reasoning about the system's working. The outputs from AI can be tracked using a KG to form explanations. For instance, if AI diagnoses a patient with diabetes, doctors can use a knowledge graph to link the inputs, such as high blood glucose levels and family history of diabetes, to make a more personalized and understandable decision. 

An example of a network that semantically injects a KG in layers of an AI model to form an Explainable AI system (XAI) is illustrated below.

A sample illustration of an XAI-enabled clinical decision support system using knowledge graphs

Biomedical Knowledge Graph With Wisecube

About 80% of biomedical data is unstructured. Wisecube offers an easy and efficient way to construct a knowledge graph from a multitude of biomedical data sources. Our Knowledge Graph Engine uses state-of-the-art NLP and knowledge graph techniques to integrate and analyze disparate data, extract relationships between entities, and leverage predictive analytics based on the extracted information. 

For instance, using NLP and graphs, we collaborated with our friends at Roche, to develop a biomedical knowledge graph containing 10k+ relationship types, information about 500k clinical trials, 85M entities, and extracted 5B+ semantic facts. 

Wisecube is built on top of the open-source library graphster. We provide users with our products and access to all graphster libraries. This means you can inject personalized datasets and produce KGs that best suit your needs.

Schedule a call with our consultants today to accelerate biomedical innovation using knowledge graphs.