Lexigram: The Future of Healthcare Data
Americans visit their doctors over 1.2 Billion times annually. Such a large number of interactions and transactions has led to the generation of a tremendous amount of data in healthcare. Furthermore, as technology permeates the sector, newer events and metrics are being tracked, leading to further exponential growth. Besides structural concerns around storage, security and management, a key issue is how to extract value from this vast amount of data and translate the information into better outcomes. Unfortunately, lack of interoperability, easy access and standards have long plagued the industry.
Lexigram breaks down these information silos and organizes data from a variety of sources into a single platform. More importantly, it provides an intuitive and powerful interface to access, understand and act upon that data. At Storm, we are excited to invest in Lexigram and support the team on their journey to realizing the promise of Big Data in Healthcare.
Structured Vs Unstructured
The U.S. healthcare system has expended a vast amount of energy and resources to make our medical data available electronically. The Health Information Technology for Economic and Clinical Health Act (HITECH), for example, mandated the use of electronic health records. Today, each patient interaction, diagnosis, prescription, treatment and follow up is a collection of data points that is tracked and analyzed. Ensuring that this data is available in an easy to access and understandable form is vital in providing care, especially as the industry transitions to a value based model.
Unfortunately, the data is spread across disparate systems such as wearables, apps, EHR systems, medical devices, labs, claims databases, pharmacy systems and various ancillary services. Much of the data stays dark. Therefore, the first challenge is interoperability. Secondly, while some of the data like is well organized and structured, a majority is unstructured. Doctor’s notes, for example, can contain everything from family history to treatment information. These notes are in free text form and EHRs often don’t even have the right fields to enter this information. Therefore, practitioners often need to combine both forms to get the full picture.
A significant portion of the data is also still in paper form. This makes accessing and understanding the data time consuming and error prone.
Another challenge that healthcare faces is the use of a unique ontology. The industry has developed a specialized terminology and lexicon to communicate healthcare information. Understanding and utilizing this ontology is not easy, but will be vital for any platform or technology to support data interoperability and automation.
As the amount of data grows, it is increasingly difficult to retrieve relevant information. As discussed above, information may be buried in difficult to search, unstructured, or free-text narratives, which is estimated to be the majority of the content. EHR search tools today are inefficient, simplistic, and unable to parse healthcare ontology to rank the relevance of information for a particular problem or complaint. This is especially true in scenarios where providers have limited, if any, prior relationship with a patient and often have to make quick decisions around life-threatening conditions. Therefore, timely identification of relevant clinical information is critical for medical decision making and the right solution will play a huge role in driving better patient outcomes.
Lexigram
This is precisely the problem that Lexigram solves. Lexigram has developed a queryable platform that can efficiently combine unstructured and structured data using popular healthcare ontologies. The platform has three main components.
- NLP Engine and Feature Engineering: The Lexigram platform uses natural language processing and machine learning to analyze unstructured data such as doctor’s notes and faxes. It then combines this information with structured data such as EHR to create a knowledge graph, which ultimately holds all relevant information in an organized and accessible form. Domain specific medical knowledge is also used to extract a rich set of features from the data and enrich the graph further.
- Ontology Layer: As discussed above, ontologies are key to establishing interoperability since they help define a common language through which all the disparate systems can communicate. Lexigram uses a database of over 100 healthcare ontologies to establish relationships between the data silos and form a usable knowledge graph. This graph, for example, allows the platform to understand that both hydrocodone and morphine are opiates and that diabetes is a disease with with complications that can manifest in the foot or eye. The system can then analyze additional knowledge automatically and update the relevant connections. The uniqueness of healthcare ontology is also why a general purpose AI platform cannot just use compute power and data to address this challenge. Doctors don’t write in plain english and a specialized interpretative layer is critical to uncover the rich connections in the data.
- Query Engine: The final piece is an interface and an API that patients, providers, payers or third parties can use to quickly access the underlying knowledge graph and retrieve insights from the patient record.
A key differentiator for Lexigram is the strength of the founding team, which came out of a Stanford project called Bioportal. The project gathered a database of biomedical ontologies (over 500) from all over the world and built the tools to work with them. Lexigram’s founders were the lead developers and helped scale the project to support over 100m requests a month. This background is one of the reasons the team is uniquely suited to address the challenge of structuring the world’s healthcare data and providing access to its knowledge.
