As the push to integrate artificial intelligence and increase interoperability evolves, Clinical Architecture sees a dire need for tools that can assess the quality of healthcare data. Poor quality data can lead to incorrect conclusions and wasted resources, hindering the progress of medical research, misguided policy decisions and investments, and ineffective and inequitable care by payers, providers, investors, and the government.
The infrastructure to support interoperability through initiatives such as TEFCA and the number of QHINs continues to grow, but the quality of the data, even structured data, varies widely. There needs to be a way to assess the data flowing through these pipes.
The stakes in healthcare could not be higher. For AI tools to be widely adopted in healthcare, it is paramount that the pools of data required to generate AI-powered clinical decision support software reflect good quality, reliable data that can help improve patient outcomes. If adverse events result from faulty conclusions based on poor quality data, the medical industry’s progress towards widespread adoption of AI tools will falter.
In an interview, Clinical Architecture Founder and CEO Charlie Harp explained how a data assessment framework he developed, the Patient Information Quality Improvement (PIQI) framework, functions as a healthcare data quality taxonomy to address these issues.
With a career steeped in life science and healthcare software, Harp founded the company in 2007 in Indiana because he felt that the healthcare industry needed a company that was focused on data quality, the exchangeability of data, and how to make data current and portable.
The Patient Information Quality Improvement (PIQI) Framework offers a uniform, objective way to assess the quality of data against a selected rubric and highlights the root cause of data quality issues so they can be fixed. Working in collaboration with Leavitt Partners, Clinical Architecture created the PIQI Alliance, which includes work groups to design this open-source framework that can be used throughout the healthcare industry. PIQI is currently going through the balloting process with HL7 to become a standard.
Harp created a rubric for USCDI version 3 that shows what the quality of data should look like.
“We are at a watershed moment for healthcare,” Harp said. “The number of aging patients is increasing and with that, the number of people with multiple comorbidities on multiple medications. We’re doing all these things with genomics and pharmacogenomics and we’re getting to another level of granularity on how humans interact with the world, which gives us better precision medicine. The number of providers we have is declining almost every year. That means the amount of time we have between a provider and a patient decreases. At this moment, we really need to lean on technology to help take care of everybody. But if we don’t improve the quality of the data, it’s not going to work out very well.”
PIQI is about assessing the quality of data an organization sends. Harp likened the PIQI Framework to a standardized test but for data. He explained that PIQI assesses the quality of the data on a granular level and grades it based on data availability, accuracy, conformity and plausibility. Sometimes the source of the problem is inherent in how the data is collected, which is harder to fix.
One challenge is that within a hospital or health system, the data contained within an EMR is often adequate to support the needs of that application. However, in order to share that data outside of that application there is work that needs to be done to make the data interoperable. When this work is not done, or not done well, the data falls short. Physicians tend to write clinical notes in a way that makes sense to them. When assessing the quality of data shared by a provider, semantics and intent behind the data are incredibly important, Harp said. Another challenge is that each EMR has its own dictionary. So even with ICD-10 codes and FHIR standards, there are significant ways the clinical data differs, depending on the EMR that’s used. There are so many data variables that add layers of complexity to transmitting clinical data.
“As the source of most of the patient data we use in healthcare, no market segment is more complex and challenging than the provider space when it comes to data quality,” observed Harp
There are multiple groups relying on data from providers for their own data sets, such as payers who are trying to use clinical data to back up their Healthcare Effectiveness Data and Information Set (HEDIS) measures and their STAR ratings. When government departments such as Social Security or Centers for Disease Control who want to do better surveillance on what’s happening in public health get poor quality, incomplete data, they can’t use that data to get an accurate picture for their needs.
There is a value chain that goes across the spectrum of healthcare. Anybody that wants to receive clinical data should have a way to trust the data, whether that’s for research or claims processing, or to do population health or public health.
“What’s cool about the PIQI Alliance is we’ve got people from across that spectrum. Members include individuals from payers, Social Security, and CMS,” Harp said. “We’re starting to get some providers that are doing this too.”
Clinical Architecture is currently looking for early adopter partners to test and refine the PIQI framework in real-world settings.
By opting to make the PIQI framework open source, Harp wanted to encourage anyone to use the methodology to evaluate and score messages using shared evaluation rubrics. Objectively measuring data using a standard framework will help improve data quality across the healthcare industry.
Clinical Architecture is doing a beta test with several of the health information exchanges (HIEs) Harp walked through a fictionalized example of the PIQI Framework in action. He noted the hospital’s allergy data scored 41% and their condition data scored 52%. Demographics are at 75% and their immunization is 77%. But he noted that users can drill deeper.
Demographics scored only 75% because the birth sex information is not SNOMED for 2,761 messages. The data the hospital provided is not valid, according to the PIQI Framework. It’s possible to review the different pieces of data measured in the rubric. Medication tends to score poorly because 89% of the time the data does not include an indication. USCDI version 3 requires an indication.
If you need data to support value-based care, you might look for patients that are taking Metformin for Type 2 diabetes. Without the indication, the user can’t verify that the reason they’re taking Metformin is because they’re diabetic. Adding the missing information improves the score significantly. So if you are an organization that buys patient data, if you can improve a score of 70% to 95% that’s significant.
Harp views the PIQI Framework and the PIQI Alliance as a wake up call for the healthcare industry, highlighting the need to identify and correct disparate clinical data to improve data quality. When you consider life science data quality, using the Observational Medical Outcomes Partnership (OMOP) standard, the data has been manually curated, touched by a lot of people to get into the OMOP format and normalized. Life science use cases are much more narrow, because they typically are looking for patients in a particular cohort.
“There are so many reasons that we need to drive data quality in healthcare. Ultimately, the economic drivers are what make things happen. If payers like CMS push data quality and require some kind of quality assessment, that will drive adoption.” Harp noted. “If we can drive the adoption of minimum data quality thresholds, I would be very excited to see what comes next.”
Photo: Issarawat Tattong, Getty Images