Many organizations have taken on the quest—and the enormous business opportunity—of measuring the quality of clinical care. For instance, health care registries collect data about procedures and outcomes for particular disciplines, such as oncology or heart failure. These registries don’t try to serve as a gigantic, general-purpose data repository like the failed Health Information Exchange (HIE) model. Instead, a limited set of data elements are collected so that hospitals can compare themselves with others in their discipline.
Hospitals try to collect correct and consistent data sets, both to participate in registries and to conduct quality reviews internally (usually across a large regional system where one chain owns many hospitals). Registries tend to share insights based on three-month-old data, whereas a hospital chain that treats data as a critical resource can derive insights from its internal data in near real time.
I talked to Brian Foy, chief product officer of Q-Centrix, about their approach to clinical data and to collecting high-quality data for hospitals. This article considers two problems in particular: important information is buried in unstructured data (according to Foy, only 20% of data in the patient’s medical record is structured), and it’s hard to code diagnoses consistently using systems such as ICD-10, which notoriously offers 68,000 codes to choose from.
The bane of most doctors is “structured coding,” the division of patient information into numerous separate fields and menus in the electronic health record (EHR). Making providers enter this data directly puts too much onus on the provider and sucks up precious time as they search the interface. Moreover, different providers tend to code the same problem differently. And the data schemas used by registries might include 500 potential elements to measure. Foy said that a traditional clinician could take several hours to code a single patient.
Q-Centrix offloads quality data management from the hospitals. To achieve this at scale, the company employs clinicians (usually nurses) and gives them extra training. Q-Centrix now serves more than 300 therapeutic areas and some 1,200 hospitals. For security reasons, coding is not outsourced to other countries.
The company is also partnering with companies who are expert in natural language processing (NLP) and other artificial intelligence (AI) techniques, including a recent partnership with Realyze Intelligence. Foy points out that for billing and other parts of the revenue cycle, NLP has been in use for some time under the name “computer assisted coding.” Q-Centrix is breaking new ground by applying these tools to the much more complex space of clinical data, particularly in oncology, cardiology, and surgery.
Foy says, “It is early in this process, but we feel that because of the number of hospitals we work with and our large clinical staff, we are best positioned to unlock some big efficiencies using automation.” In principle, coding and other data classification activities should be perfect for artificial intelligence, which thrives on tasks that have a huge number of input parameters and decisions to make. While creating accurate models will probably be difficult, the effort will pay off because you’re making highly paid people more efficient.
Published in Healthcare IT Today. Read the article here.
