Introducing AI for clinical text interpretation in the 3terra platform
(For a brief overview of clinical text AI and how it will change hospital analytics, please first watch this introductory video.)
Over the past three years, the advancement of natural language processing (NLP) within healthcare has been remarkable to watch. In that time, a new service model has emerged where the top technology giants (Microsoft, Amazon, IBM, and Google) have released world class medical NLP Application Programming Interfaces (APIs) that enable new opportunities to use unstructured medical text to solve a wide variety of analytical and research problems.
We recently completed the integration of Microsoft’s Cognitive Services into our platform to assist in the translation of clinical text to identify medical concepts, entities, relationships, and syntactical context. This brings world class medical AI to all of our clients with significant and immediate implications to their medical coding, research, and hospital data analytics practices. We have been preparing our platform in anticipation of the general availability of Microsoft’s service, which was finally launched in June 2021. It was a long wait.
NLP for medical text has been a topic of substantial research over the past decade with significant improvements coming out of academia in the past few years alone. While there are many straightforward approaches to implementing NLP natively within our own engine, it would not be feasible to match the features, accuracy, and flexibility of what the formidable data science divisions at Microsoft, Google, IBM, and Amazon provide.
Microsoft has invested billions of dollars in their Azure Cognitive Services platform and they now provide a cloud infrastructure that is transforming healthcare. Coupled with the AI and analytics that 3terra has developed over the past 6 years to address Canadian specific issues, our hospital clients now have access to a suite of tools that was simply unimaginable a short time ago.
How this affects our platform
The initial use of this functionality for most of our clients will be to supplement our 3terra DataHub and DQA platforms as NLP is a natural fit within those tools. The introduction of NLP affects our current platform in two ways:
3terra DataHub: In addition to all of the current functionality, DataHub now allows users to view clinical notes pertaining to any patient encounter along with all clinical concept labels, relationships, and pertinent contextual modifiers. Like other coding assistance tools, we employ selective highlighting and contextual popups to assist with choosing the proper medical codes. There are also workflow tools such as annotations, bookmarks, and search. The DataHub database also maintains an archive of all processed records for other purposes (see the next section).
Data Quality Assist: High probability data quality errors detected from the clinical notes have been included into the auditing engine, similar to any other cross-reference data source. Like our other data quality engine rules, there are several options to tune the engine to reduce both false positives and false negatives.
Unlike other medical NLP platforms that have emerged over the past decade, there is no need to train the AI (therefore no investment of professional coding resources) to get immediate value from the system. The performance of the Natural Language Processing engine without any data preprocessing or adaptive training is astonishing. However, we do have supplementary layers within our engine to improve results based on idiosyncrasies that may exist in your clinical notes. Similar to other rules in the system, coders have the ability to mark invalid classifications to improve future results.
With the introduction of this NLP technology, both the efficiency and data quality of your hospital’s medical coding process will be greatly enhanced.
Limitless possibilities
Utilizing NLP for medical coding support is just the beginning of how this technology will benefit our Canadian clients. Taking unstructured clinical notes and decomposing them into semi-structured labeled medical concepts and relationships has many practical uses.
By storing and indexing these processed medical records, hospitals can now extract and utilize this data for a wide variety of analytical, statistical, and machine learning applications to drive quality improvement initiatives. Gathering certain types of research data will no longer be a laborious process of text pattern matching and manual data cleansing. Medical concepts are automatically extracted into unified medical terminology that is easily queried by researchers and analysts.
The NLP engine handles semantical complexities such as conditional terms and negation. For instance, “patient refused intubation” will correctly detect that intubation was not performed. It can use context to disambiguate concepts and breakdown entities into their constituent parts (e.g. name, dosage, dosage form, frequency for drug prescriptions).
While it cannot be used as a substitute for professional medical expertise, it can assist in many tasks. When faced with a wall of clinical notes, clinicians can now use visualization aids (e.g. highlighting) to improve comprehension and reduce cognitive fatigue by focusing attention on key areas of interest and by eliminating certain types of textual noise.
The healthcare AI landscape has reached a significant milestone. Clinical text interpretation has effectively been refined and commercialized as a SaaS offering and there is much more to come. Our commitment, as always, is to adapt the leading data analytics tools, technological services, and best practices to serve the specific needs of Canadian hospitals.
A brief update on 3terra
We’ve been making significant investments in our platform to maximize the opportunities that recently released technology enables for healthcare. Earlier this year, we released a major update of DQA and DataHub that introduced a completely new user interface and greatly enhanced functionality. Since early 2020, we’ve rearchitected a large portion of our platform in preparation for new analytical capabilities enabled by Microsoft Power BI along with the integration of an NLP engine. Behind the scenes, we’ve gradually transitioned much of our data engine from relational database models to document-based NoSQL approaches that open up significant new analytical opportunities. This will be a new way of using data for many hospitals, but we can help guide you through the learning curve.
Over the past 6 months, we have piloted portions of our new analytical platform (Aperture DS) within several hospitals. The feedback has been enormously positive and we are currently rolling out initial modules as they have immediate applicability. The first full version of the Aperture DS platform is scheduled to launch in 2022, and is the most exciting project that I’ve personally worked on. The confluence of technologies over the past couple of years have opened up a staggering amount of new opportunities in the healthcare analytics space.
This is a time of growth for 3terra as we are well positioned to be an early leader in this new field of applied analytics for clinical metadata. Our company is focused on using this technology to serve the specific needs of Canadian hospitals and we are establishing partnerships to bring our platform’s unique capabilities to international markets.
I’d like to thank our hospital and technology partners that have helped us redefine our platform over the past 18 months. For our clients, please contact us if you’d like to setup the coding support that our new NLP module provides.
