When the healthcare industry talks about data, the conversation often centers on interoperability and data standards. These are certainly important topics, but they do not fully address the challenges of making complex forms of clinical data available for exchange and analysis.
Overcoming these challenges is a critical need for organizations aiming to deliver data-driven, high-quality care at the individual and population levels. This is because hospitals, health systems, community clinics and physician offices are increasingly reimbursed and ranked based on patient outcomes.
The most efficient way to connect the dots and get a complete view of the patient is through data normalization. In this process, data from different systems is not only aggregated into a single data warehouse, but also normalized into common terms. Data normalization is not without its challenges, but the right combination of business processes and technologies to capture, store, and normalize data can help organizations realize the clinical, financial, and operational benefits of data normalization.
Why data normalization is more than a technical move
Data sprawl across a typical healthcare organization presents three daunting challenges. First, each clinical data type—diagnostics, procedures, medications, labs, devices, etc.—is stored in its own siloed enterprise application. Second, each data type is encoded differently – integrated circuit for diagnosis, LOINC For lab tests and results, SNOMED For clinical documentation, RxNorm for drugs, and in some cases, there is no code associated with a given dataset. Third, the coding systems overlap so that a diagnosis or drug can be coded in multiple formats.
These data types may be valid in their respective settings, but by themselves they do not provide a complete picture of patient or population health or health system performance. To get a complete picture, each dataset needs to be moved from its separate system into the data warehouse.
However, just a data warehouse is not enough. Data sets may be together, but they are still in their unique format, which can lead to inconsistencies across the data warehouse. Interoperability standards from the Centers for Medicare and Medicaid Services will help in the future, but they are not intended for legacy coding systems or clinical applications.
This is where data normalization comes into the picture. With data normalization, data from disparate systems is normalized into a common set of clinically validated terms when transferred to a data warehouse.
Without standardized datasets, the analytical capabilities of healthcare organizations are limited. They tend to focus on the dataset with the smallest gap. On the clinical side, this is a patient registry, which is actually very limited when looking at patient outcomes. Financially and operationally, they need to put together various documents for compliance, quality or financial reporting.
Normalized datasets enable more sophisticated analysis methods. Organizations have a single source of data in one place that can be combed for actions such as reducing care variability, eliminating waste, managing population health, and introducing predictive analytics at the point of care. This makes data normalization more than a technical initiative – it is an essential tool for value-based care, clinical decision support and data-driven strategic planning.
How to make data normalization easier and faster to connect the dots
Data normalization is not without its challenges. Entering free text and other low-quality datasets into a data warehouse requires the use of an extract, transfer, load (ETL) process because the data must be cleaned before normalization. This requires additional infrastructure and personnel; it also creates a bottleneck that reduces the value of the data, because by the time it ends up in the warehouse, it may be out of date. It is also often redundant because a single data point (eg a patient diagnosed with stage 3 breast cancer) may be represented in multiple datasets (albeit in different ways).
Given these barriers, many healthcare organizations have yet to normalize their data. But this is a critical step in data aggregation, normalization and analysis. Otherwise, the public datasets in the data warehouse may be incomplete at best, or inaccurate at worst. This can have clinical, financial and operational consequences. On an individual level, this can lead to an incorrect diagnosis, prescription or treatment plan by the care team. At the population level, this can lead to population health, quality of care or care management plans moving in the wrong direction.
Start at the point of care
Fortunately, two simple steps can lead organizations down the road to data normalization. First, normalize the data from the start as it is fed into the clinical system at the point of care. This doesn’t need to disrupt the clinical workflow; it just verifies that the correct code is associated with the entered data. A data normalization engine is then used to map each data point from each clinical system to a normalized description and associated code before transferring the data to the data warehouse.
With standardized datasets, any internal or external stakeholder—whether a health system, hospital, insurance company, public health registry, research organization, or health information exchange (HIE)—will have a single version of the truth. This allows organizations to do things they couldn’t do before.
For example, an HIE serving health systems and hospitals in the western United States found that aggregating data from payers, providers and government agencies often led to gaps; this was especially true of Covid-19 laboratory data, which often lacked support for monitoring and quality reporting Valuable LOINC code. Using a data normalization platform, it has normalized more than 1.8 million pieces of information, and its work on normalizing laboratory data has expanded from Covidh-19 test results to include blood bank and microbiology information.
Simple tasks done correctly make complex tasks possible
Today, only about More than a dozen organizations worldwide have reached Stage 7 of the HIMSS Analytical Maturity Model. It’s a far cry from the original Hundreds of HIMSS EHR Maturity Stage 7 Model.
Indeed, prescriptive and predictive analytics are complex tasks. However, they cannot be accomplished without first establishing standardized clinical vocabulary and terminology.
In theory, it should be simple enough to get everyone to use the same term. But this is difficult in healthcare because different clinical systems, let alone different medical disciplines, have historically used different terms to define the same thing.
A data normalization strategy can ensure that data is available in a common language for exchange, use, and analysis, rather than forcing changes across the medical field. Connecting the dots enables organizations to spend less time and resources cleaning data before analysis – and more time using data to improve clinical care and operations.
Photo: Filograph, Getty Images
