The task of achieving semantic interoperability ranks among the toughest problems in health IT.
The rapid adoption of IT systems has digitized masses of clinical data, but at a cost. Different health IT (HIT) systems tend to represent data in different ways, which creates numerous interoperability issues. Accordingly, the healthcare community faces significant hurdles when it comes to efficiently sharing and using data across clinical care processes, business functions and systems.
Data normalization aims to bridge the semantic gap between different reference terminologies and classification systems. Normalization solutions map local content to terminology standards and also translate between standards. It’s a difficult task, made easier by automation. Here are some of the most challenging terminology domains:
Prescription drug nomenclature over the years has been embedded in myriad vocabularies and coding systems. The lack of standards has made it extremely difficult to aggregate and share medication-related data.
Indeed, healthcare organizations grapple with sorting, categorizing, and labeling medication data in light of the many disparate medication terminologies. That list includes The National Library of Medicine’s RxNorm, the Food and Drug Administration’s National Drug Code (NDC), the Veterans Health Administration’s National Drug File Reference Terminology (NDF-RT) and Wolters Kluwer Health’s Medi-Span.
Another issue: drug terminologies often lack hierarchies and therapeutic classes. As a consequence, grouping medications becomes a challenge. For example, a health system may want to group medications into drug classes to create a clinically friendly view of patient data. This grouping might involve, for instance, placing Captopril and Benazapril within an ACE inhibitor list. The health system's clinical data repository, however, receives data from two systems. System A feeds a medication list that includes Benazepril 10mg, which has an NDC code of 54868235002, while System B feeds a medication list that includes Captopril 25mg, which has an RxNorm code of 197436.
In this situation, the normalization task involves multiple mappings between NDC and RxNorm as well as RxNorm and NDF-RT.
Labs may be the worst offenders when it comes to having clean, structured data. Many lab orders and results are cryptic to read or make use of proprietary coding paradigms. And even when labs use terminology standards such as LOINC, they often implement the codes with subtle differences (e.g., they may use different systems, also known as specimen sources). The coding confusion makes it difficult to trend lab results over time, to deduplicate lab lists so that they are easier to view, or to use lab information for downstream analytics such as quality-of-care measurement.
Here’s one example of the problem: A patient may have two lab results for a sodium test. Both are mapped to LOINC, but one has a system of serum, the other of blood. Hence, there are two different LOINC codes. A data normalization solution must be able to make use of LOINC hierarchies to group these two lab items next to each other in a clinical system.
In another case, a healthcare system may seek to trend A1C hemoglobin levels for diabetes. The organization may receive a local code from Lab A (1234/Hgb A1c blood) and a standard code from Lab B (LOINC 17855-8). Both of those codes represent lab results of A1C, but they must be normalized to one standard for trending to be accurate.
Problems and Diagnoses Terminology
Many health organizations are wrestling with the terminology breakdown between problems and diagnoses. The federal government’s Meaningful Use program has designated SNOMED CT as the standard for problems and ICD-9/ICD-10 for diagnoses. A healthcare provider might want to deduplicate problem lists coming from ICD-9- or SNOMED-encoded data, or walk between ICD-9/ICD-10 and SNOMED for disease and diagnoses stratification efforts. Data normalization can play a role in both of those use cases.
What’s Your Top Challenge
Incompatible terminologies present a number of interoperability problems in healthcare. What are your main challenges with conflicting terminologies? Are you considering a data normalization solution to address them? Leave your comments below.