Enabling Analytics Innovation
Exploring Time, Space, Medicine, and Beyond
What’s the challenge?
Artificial intelligence, machine learning, precision medicine, data analytics, and predictive modeling hold great promise to advance healthcare—possibly as dramatically as the introduction of scientific research methodology to medicine in the past century. While the ‘big data’ healthcare analytics field swells, we still lack an indispensable element for analytics and natural language processing (NLP)—interpretation and quantification of free text temporal statements from unstructured data.
Precision medicine that lacks clear timelines is, simply put, not all that precise. Predictive modeling can only determine a pattern when temporal relationships capture intervals and sequencing. With the estimated 80% of data in a patient’s medical record entered as free text, our inability to tap information contained in that narrative leaves a substantial gap in the data we can utilize for analytics.
Why is this a big deal?
Incorporation of temporality into analytics and modeling fills a critical gap in the interpretation of the data (precursors, outcomes, related events). Not only can it augment phenotypic associations for patients, necessary for life sciences initiatives and other health studies, but it will be key for deriving meaningful links between medical treatment and health outcomes and for constructing advanced decision support systems.
​
Beyond healthcare, temporality and spatiality play a central role in other arenas, like insurance claims, business models, scientific research, and investigatory analyses, which often rely on the capture and interpretation of free text narratives for key tasks and construction of interlaced timelines.
What’s the opportunity?
In medicine, two initial business cases stand out—first, inclusion of temporality found in unstructured text from a single patient’s record to build a clearer picture of the patient; and second, enablement of investigation across all medical records in a system for events (temporality associated with elements like findings, problems, procedures, orders, observables, etc.).
For precision medicine and life sciences research, temporal incorporation supports natural language processing and can be used to evaluate data for building a patient’s longitudinal electronic medical record (LEMR), providing temporal relationships (e.g., age at event, length of event, sequence of events, time between events, etc.) from records across multiple sources of data. For population health research, by running queries through a system that has incorporated temporal relationships into the patient data (aligning patient records and being able to consider these relationships within large cohorts), we can provide a fundamental characteristic for artificial intelligence systems and machine learning to elucidate context—a building block which currently does not exist in production.
An apparatus for incorporating temporality will support health information exchanges, accountable care organizations (ACOs), data warehouses, disease registries and future “wide area network” data sharing, thus enabling patient phenotypic matching, and population health studies. Moreover, inclusion of temporality into patient records will enable the leap beyond inferred relationships to clear-cut associations between events.
The origin of Exploring Time, Space, Medicine, and Beyond
Over the past 20 years I have advocated for the incorporation of de-identified patient records into our corpus of accessible research data. The Health Record Banking Imperative: A Conceptual Model promoted a business case and route to access anonymized patient data for specific research questions. Medical Researcher, Meet Mrs. Roseburg built on this theme further, proposing that real-world experience and the search for doppelgangers across a vast database of patients could help us tailor treatment and establish the most accurate predictive models.
​
The missing item, of course, was the capture of events from free text, specifically linking temporality and spatiality to elements and being able to plot these accurately. This would enable us to index a patient's entire medical record. For the past 15 years, I have worked on defining temporal and spatial aspects of medicine, initially reviewing syntax, but also exploring how to assign mathematical formulae to pre-coordinated phrases. After considering how we might assign a time for a single event recorded in a single account, it became clear that there were often multiple notes regarding that same event. How could we know that this is the same event? What should happen when, for the same event, there were inconsistencies as to when the event occurred?
One goal has been to provide a dynamic index to a patient’s health record. Could we construct a health timeline for a single patient by analyzing the structured text and the multiple unstructured notes from multiple sources about a patient’s multiple events? Beyond this, might it be possible to search for the elusive doppelgangers by comparing all patient indices and establishing rules pertaining to the degree of concordance necessary (and discordance prohibitory) for use in building a highly specific predictive model for that individual?
​
Clearly, to get there, it all must start with capturing temporality and spatiality.
I'm always looking for new and exciting opportunities to incorporate temporospatial relationships into the medical record. Let's connect.