Data Science

Clinical Data Usability: A Critical Step in the Shift to Value-Based Care

Doctor looking at computer smiling

Value-based payment models (VBPMs) depend on differentiating and rewarding entities that demonstrate higher quality, more equitable care delivery at a lower cost when compared to similar entities. For VBPMs to succeed, quality and utilization measures assess and reflect the results of meaningful, actionable aspects of care, and the variation in performance can be linked to variation in outcomes. Attribution of these results needs to be sufficiently accurate for the intended purpose; and benchmarks need to be based on robust, representative and reliable analyses to generate valid, reliable and reproducible comparisons. A prerequisite for all these elements is accessible, usable, high-quality and (reasonably) complete data.  

And lacking these comprehensive data presents a major problem for U.S. healthcare. 

“Dirty Data” are Pervasive 

Industries outside of healthcare have been struggling with data quality and usability for many years. When there is a stupendous flop of a highly visible, anticipated product—for example, Apple Maps in 2012—the results of inaccurate, incomplete, and unusable information are available for everyone to see. However, transparency about the presence and effects of poor-quality data does not currently exist in healthcare. The implications are considerable—not just for performance under VBPMs, but for how we care for people. 

Health plan executives know there are “dirty data” in the system. A recent survey by Sage Growth Partners of 100 healthcare executives showed that only 20% of organizations fully trust their data; more than 50% indicate poor data quality has serious consequences; and 84% say creating and sharing high-quality data is a top strategic priority for the next 36 months. Yet, 63% are still using tools like Excel, and only 17% aggregate social determinants of health (SDOH) data. Deficiencies in data can affect clinical, operational and financial analyses which, in turn, influence how, where and whether care is provided. Gaps in the quality of social determinants data make it challenging to identify and address gaps in health equity due to factors including misclassification and non-random, missing data. Systematic data issues can lead to biased analyses and misdirection of resources. A lack of confidence in the underlying data can lead clinical teams to question decision-support algorithms designed to nudge clinical teams towards guideline-concordant, cost-effective best practices. Inaccurate and incomplete data challenge most health plans’ ability to optimize performance on VBPM accountability measures, including Healthcare Effectiveness and Data Information Set (HEDIS®) scores, The Centers for Medicare & Medicaid Services (CMS) Medicare Star Ratings, and state-based incentive programs.  

But Wait … There’s More (Data)  

The 21st Century Cures Act will facilitate secure access, exchange and use of electronic health information through Health Level Seven (HL7®) Fast Healthcare Interoperable Resources (FHIR®) application programming interfaces (APIs) and reduce information blocking. But the usefulness of that data will only be as good as its quality. The Office of the National Coordinator must establish the policies and procedures that streamline the arduous burden of each entity attempting to connect and interoperate with the various organizations necessary to deliver care and evolve into the learning health system we envision. The first Common Agreement (CA) framework was published Jan 2022, and the Recognized Coordinating Entity (RCE) – The Sequoia Project – has been holding question and answer sessions for those organizations looking to apply for Qualified Health Information Network (QHIN) status since April 2022. As QHINs become more common, the expectation is that the liquidity of clinical data will become more ubiquitous and less arduous to acquire.   

But More Data Will Not Fix Poor Data  

The need for more and better clinical data will continue to increase. The CMS National Quality Strategy (NQS) includes a commitment to transition to digital quality measurement and National Committee for Quality Assurance (NCQA) has also signaled its intent to digitalize HEDIS measures 

The impetus for the shift to digital measures is primarily driven by the overly burdensome, expensive, underperforming and under differentiating performance measurement ecosystem. New digital quality measures will generate actionable insights for improving care by leveraging clinical data beyond what most health plans and healthcare organizations currently can access. A good portion of the measures used in VBPMs are based on administrative data, yet health plans often attempt to acquire clinical data via a sampling methodology, by using chart abstraction to improve performance rates (i.e., supplemental data). The cost of obtaining these supplemental data can be upwards of ~$250K per HEDIS submission for a process that covers only ~20% of the reportable measures. As the demand for the use of clinical data increases, so too will these costs and associated burden until the ecosystem shifts to FHIR-APIs and provisions of the 21st Century Cures Act are fully implemented.  

Further, the CMS NQS aims to ensure that, “… individuals have access to their information to support informed decision-making as well as referrals and community planning … and give people access to their health data when and where it is needed.” October 6, 2022, was a major step on that journey. The so-called “data liberation day” opened the door for patients to get their health data and share it with whomever they choose. Many people will eventually, and routinely, avail themselves of that level of access.  

Health plans, accountable care organizations, health systems and other data-dependent entities need to focus on clinical data quality and usability now in anticipation of the need for more and better clinical quality data. This is the only way to ensure that the newly required transparency does not result in the equivalent of the Apple Maps debacle in healthcare. Achieving enhanced data quality will make it possible to consume greater quantities of clinical data from multiple sources, share those data with patients, and improve performance under VPBMs. The ROI from growing investments in EHR optimization and systems to support population health initiatives will be hampered if data quality is not addressed. 

Clean Dirty Data with A Purpose 

Context and prioritization matter. The basic definition of data quality can be summed up as “fit for the intended purpose,” though several definitions and frameworks have been proposed. For Apple Maps, the critical need was data to guide users to the right roads and connections to generate accurate directions. In healthcare, there are many different types and sources of data, and a wide range of potential uses. The intended use of data should drive the data quality (usability) objectives, independent of the format in which the data are transported (e.g., ADT, C-CDA, FHIR). Some obvious uses of data include billing, care delivery and performance measurement. Incorrect or incomplete standardized codes [e.g., Current Procedural Terminology (CPT), International Classification of Diseases, Tenth Edition (ICD-10)] could significantly undermine usability of data for billing and claims-based performance measures. The quality of clinical data becomes a much higher priority in the setting of care delivery, for example, to produce time- and context-sensitive clinical decision support. Next generation digital quality measures would need reliable, credible clinical data to generate performance rates and achieve their full potential to drive improvements in care. However, significant variability in completeness, correctness and timeliness of data can undermine its usability. For example, suboptimal design and implementation of electronic health records can produce records with the right data in the wrong place, wrong data in the right place, missing or incomplete codes, and bias because of templates that systematically fail to promote the necessary data inputs.  

Take Important Steps Now 

There are some important steps healthcare organizations and health plans should take to support better clinical data usability. 

  • Determine the current and future need for clinical data within your organization. 
  • Inventory current clinical data sources and whether existing data can support those needs with respect to internal and external quality reporting, VBPMs, health plan and CMS Medicare Part C and D ratings. 
  • Assess the usability of data sources, for example the percentage improvement (or numerator hits) that a particular data source provides for specific clinical quality measures. What is missing? What can be improved at the source and in the collection/transport process and to what level of refinement? For example, data used to help guide nurse care coordinators does not necessarily need to be refined as highly as that which would be needed to inform clinical decision support. 
  • Engage with the regional health information exchange (HIE) and encourage network providers not already connected to participate.  
  • Advocate for care delivery organizations to submit clinical data to the HIE via HL7 2.x – 4.x formats, which will facilitate more efficient data sharing.  
  • Prepare your organization for FHIR API implementation, including an assessment of internal operations to consume and use clinical data. 
  • Encourage clinical data sources to achieve FHIR-readiness (HL7 4.x transport protocol) and monitor progress. 
  • Begin the shift to FHIR API data feeds and prioritize connections to sources with sufficient data quality to support the intended use.  

… And Leverage Existing Tools/Data Validation Programs 

Despite best efforts, in most situations, a significant data usability gap will remain. Industry sources estimate that less than 40% of clinical data are usable for most intended purposes without further processing. In addition to internal efforts to address data quality and usability, forward-thinking organizations should consider engaging an external entity to support data quality enhancement and understand the benefit of attaining or requiring National Committee for Quality Assurance (NCQA Data Aggregator Validation. 

Several companies in the market focus on clinical data usability. They leverage approaches like machine learning, natural language processing and stepwise algorithms designed to curate clinical data. In some cases, these products are functions within a larger offering. Other companies provide focused solutions that can be integrated into existing data streams and workflows to support data quality enrichment for specific use cases. Determining the value proposition for engaging such support can be a challenge. Evaluation criteria should include whether the entity can: 

  • Assess data sources for compliance with industry standards (e.g., HL-7 2.x-4.x). 
  • Evaluate data for completeness, validity, etc., against a standard such as the United States Core Data for Interoperability (USCDI). 
  • Rate the usability of data by source and stratify its value for specific use cases (e.g., quality performance measures such as HEDIS/CMS Stars). 
  • Estimate the potential performance measure improvement the entity can generate through data quality enhancement. 
  • Model the opportunity cost of not improving data quality for specific measures versus the likely ROI of better data. 
  • Reflect all steps in the process through a dashboard and provide an audit trail for all transformations across every step of the process. 
  • The results of such evaluations could enable providers and users of these data streams to prioritize and assess the appropriate value of different data sources based on the potential for that data to, for example, improve CMS Medicare Advantage and Part D Star ratings 

The NCQA Data Aggregator Validation (DAV) program evaluates data streams flowing into and out of clinical data aggregation points [e.g., HIEs, Qualified Health Information Networks (QHINs)]. Validated entities can provide assurances to organizations that report HEDIS® quality measures that there are no data processing errors—an important component of data quality. Another benefit is that validated data streams are considered standard supplemental data in HEDIS reporting, and therefore, primary source verification for HEDIS audits is not required. However, the DAV program does not consider the context of the data, and therefore, could be considered necessary but not sufficient.  

Avoid Drowning in a Sea of Dirty Data 

The potential to improve quality and reduce health inequities through digital quality measures, Artificial Intelligence/Machine Learning (AI/ML)-based algorithms, and value-based payment models, means that the demand for more and better data will accelerate as the implementation of the 21st Century Cures Act unfolds. The Release of Information (ROI) under VBPMs from improved data completeness may be easier to generate than through efforts to change practice patterns or implementation of complex new technology solutions or processes. Those organizations that do not have an informed and forward-looking clinical data quality strategy could incur avoidable financial and reputational risk as public reporting of quality, health equity and utilization become more heavily dependent on access to robust, high-quality and usable clinical data.