Data normalization is finding an expanding role in a range of healthcare settings.
Health information is scattered across multiple organizations and myriad IT systems throughout the healthcare community. Those systems often use disparate terminologies designed to support their specific needs. With the growing need to aggregate data for reporting and analytics purposes, many organizations are now working through a number of interoperability challenges both at the syntactic and semantic layers.
To see how this may play out in practice, let’s follow a hypothetical example. A care management program may want to connect the dots between the order and the lab result in order to create and manage a cohort of high-risk patients. In order for the care nurses to reach out to the right patients, the program’s quality metrics need to establish relationships between claims (encoded in CPT®) and lab results (encoded in LOINC®) utilizing high quality maps. Without mapping the claims and clinical data to support this specific program, the care nurses run the risk of calling the wrong patients or missing the high risk ones completely.
Terminology mapping or normalizing data, provides a mechanism to translate data to support a variety of use cases. To simplify things, here are two common areas that we’ve found applicable to Data Normalization:
Making Your Data within Data Warehouse Useful
A data warehouse or clinical data repository, pulls together patient data from a variety of clinical and IT systems. There are many reasons why providers and payers may establish a data warehouse. The most common reason is organizations may want to centralize their quality programs and reporting projects, leverage data to create predictive models, or use clinical and claims data to monitor the caregiving process. The contents of the repository may also vary, but in general, it’s not uncommon to find claims data consisting of ICD-9 and CPT codes as well as clinical data like lab results, medications, patient vitals, etc.
So where would terminology mapping be applicable in a data warehouse? Let’s say a hospital provider wants to trend A1C hemoglobin levels for a population of diabetic patients. Unfortunately, it’s very common that laboratory systems have their own proprietary lab codes in place and aren’t standardized on LOINC. That hospital’s data warehouse or analytics platform may contain lab results from Lab A encoded as 1234/Hgb A1c blood. In the same data repository, the provider may have lab results from LAB B already standardized on LOINC as code 17855-8.. Both of these represent lab results of A1C, but the data must be normalized to a standard such as LOINC in order for the healthcare system to accurately represent and analyze the results from both labs.
Representing Your Patients within Your Quality Measures
Another scenario that combines data normalization with another terminology challenge - managing code groups, or value sets - is Quality Measures. As you know, Quality Measures, are a mechanism for assessing observations, treatment, processes, experience, and/or outcomes of patient care to support a number of different initiatives like attesting for Meaningful Use if you’re a provider and supporting the HEDIS program if you’re a health plan.
To understand how managing code groups and normalizing data come together, let’s look at a Clinical Quality Measure that applies to all patients with heart failure and a low ejection fraction who don’t have any contraindication to taking an ARB or ACE inhibitor medication. Within the denominator of the measure, the heart failure value set has to be represented using ICD-9 diagnoses codes. If the provider wants to represent all their patients that have problem list entries within their EHR, they would need to normalize their free text entries or SNOMED CT codes to ICD-9 in order to include them within the Heart Failure value set. In addition, if they want to represent ICD-10 diagnoses codes within the same value set, the provider might need to map the ICD-10 diagnoses codes back to ICD-9 in order to ensure a seamless transition post October 2015.
What normalization use cases are you encountering in labs, drugs, problems and diagnoses and other domains? Leave your comments below.