Duke Health is a world-class academic healthcare system with three hospitals serving the Raleigh-Durham, NC region. Duke Health is the home of Duke Cancer Institute (DCI), a National Cancer Institute-Designated Comprehensive Cancer Center that leverages its network of oncology scientists and physicians to provide cutting-edge, evidence-based cancer treatments to over 66,000 cancer patients each year from across the country.
Approximately 20-30% of breast cancer patients will develop metastatic breast cancer (MBC), which is often fatal. However, thanks to more effective treatments survival rates among MBC patients are on the rise. Duke Cancer Institute has long been interested in tailoring treatments to these individuals, but the cohort is far from monolithic. Indeed, some therapies may be better suited to specific patients based on certain clinical characteristics, or phenotypes. Yet to effectively study these differences, patients must be stratified into subgroups based on phenotype, which can be difficult to achieve.
Finding patients who fit a particular phenotype requires highly specific clinical data, much of which is obtained through time-consuming manual chart review. To streamline this process, Duke Cancer Institute wanted to leverage SNOMED CT® – a widely accepted standard used to codify clinical data – to build a computational phenotype that they could then deploy to drive automatic querying of the EHR. The ultimate goal was to create a more efficient way of finding MBC patients without sacrificing the high level of accuracy seen with manual review.
However, this posed a challenge as not all data in the EHR is coded in SNOMED CT. In fact, much of it is coded using ICD-10-CM billing codes, which are typically not granular or precise enough to find specific MBC patients. Additionally, data that is coded in SNOMED CT often lacks the secondary and tertiary codes that are needed to appropriately capture the highly detailed MBC diagnoses required for precise cohort identification.
In order to accurately and reliably build and test their SNOMED-based queries, Duke Cancer Institute turned to IMO’s industry-leading terminology, which comprehensively maps over five million diagnosis terms to 36 reference code systems, including SNOMED CT. Duke was already successfully using IMO’s ICD-10-CM maps for patient cohort identification and welcomed the opportunity to further unlock the potential of IMO’s extensive code maps to construct their SNOMED-based phenotypes. The results were significant.
“IMO gives us the ability to easily query the EHR to identify very specific patient cohorts,” said co-investigator and lead author, Ben Neely, MS, a senior statistician at Duke Health. “IMO provides comprehensive SNOMED mapping, an important consideration for any institution that aims to leverage this method of cohort identification.”
Duke Health’s IMO-enabled computational phenotype produced an accuracy rate of 95% and was found to be more sensitive than other more commonly used methods of identifying MBC patients, such as pulling patients from tumor registries. Importantly, this increased sensitivity came without sacrificing precision. In other words, they were able to correctly identify more patients with specific types of MBC without increasing the rate of false positives.
These findings illustrate the value of IMO’s robust and accurate mapping to facilitate effective patient cohort
identification. More importantly, by leveraging IMO’s SNOMED CT maps, Duke Health was able to successfully create a highly efficient and accurate computational phenotype. In addition, by linking diagnoses to standardized codes their proposed method can now be easily implemented by other institutions.
“If adopted, these phenotypes could be used to construct more comprehensive patient registries, which include information on treatments and outcomes,” said principal investigator, Jennifer Plichta, MD, MS, associate professor of Surgery and Population Health Sciences at Duke University. “Having this type of registry available could help accelerate the development of more tailored treatment guidelines and more accurate prognostic estimates for MBC patients.”
The work conducted by Duke Health and IMO highlights the benefit of using IMO’s precise problem terminology and comprehensive code mapping to create accurate and specific patient cohorts. This, in turn, will facilitate research and care management not only for the treatment of cancer, but for many other diseases as well.
Moving forward, the Duke Health research team plans to take advantage of IMO’s robust terminology to:
Optimize query performance by including more precise exclusion and inclusion criteria
Validate their computational phenotypes at other institutions that also use IMO
Share executable cohort definitions based on computational phenotypes incorporating IMO-encoded components
For more on the challenges of precise cohort identification, download our white paper, Accurate value sets: The key to effective clinical initiatives.
SNOMED and SNOMED CT are registered trademarks of SNOMED International.