As more revenue is tied to care quality under value-based contracts, measuring performance against established measure benchmarks is now a critical business activity. Plans, providers, and measure set developers are still in the process of defining which measures are most valuable for benchmarking quality and sustainable from a reporting perspective.
Different types of quality measures come with different sets of challenges. For structural quality measures—those focused on provider capacity, systems, and processes—the measurement process relies on standardized, labeled data that’s relatively straightforward to capture and analyze. Some types of procedure measures—those that evaluate provider activities to maintain or improve patient health per clinical practice guidelines—are simple to measure from claims submissions, but others require searching through clinical charts for specific pieces of evidence. Outcome measures—those that capture the impact of care activities on the health status of patients—are the most difficult to quantify at scale, but also the most aligned with clinical activities.
While each of these quality measure types has value, procedure and outcome measures generally paint a richer picture of the effectiveness, safety, and impact of care provided to patients. To date, however, quality frameworks have limited the number of required procedure and outcome measures due to the burden they place on healthcare organizations to perform related measurement and reporting activities. But artificial intelligence (AI) has the power to transform quality measurement, allowing providers and payers to evaluate care using increasingly nuanced, clinically-focused measures.
Current Obstacles to Procedure & Outcome Measure Adoption
Although many healthcare leaders want to adopt procedure and outcome measures that more accurately reflect care quality, doing so is a challenge for a number of reasons. Firstly, gathering information on procedures that aren’t captured on claims is complicated. Most of the data in patient notes, imaging, and labs is unstructured, meaning it isn’t tagged to be easily readable by computers. The only way to extract information from this kind of data is to hire abstractors to manually review charts, or implement advanced technology to mine for trends. (More on this later.) Secondly, developing more sophisticated measures is complex. Clinical perspectives on chronic condition management, screenings, and appropriate procedures change frequently, which means organizations performing quality measurement have to change their processes to accommodate. Lastly, outcome measure data can be difficult to capture and analyze at scale. This is why most measure set developers have focused their efforts on structural and procedural measures that are easy to glean from existing structured data sources such as claims, which capture patient, provider, and procedure information in a standardized format.
AI: Paving the Way for More Meaningful Measures
In order to move away from simplistic measures and adopt more meaningful, sophisticated measures to benchmark care quality, we need a way to extract insights from large, diverse, unstructured datasets. Artificial intelligence (AI) techniques such as natural language processing, machine learning, and machine vision can help streamline data analysis and make it possible to extrapolate trends from a variety of different kinds of records.
Unlike traditional computational techniques, which require explicit programming to perform discrete functions or look for specific trends, AI models learn and adapt over time, improving their accuracy and performance. They also don’t require labeled data fields—they can look at a mass of unstructured data and, with the right training, locate specific pieces of information relevant to care quality.
Because of this flexibility, AI opens the door to new kinds of quality measures that rely on medical imaging, patient feedback during in-person encounters, or even health history information captured as free-form text. These quality measures, based on data captured during patient encounters, could potentially serve as a more reliable gauge of patient health, treatment outcomes, and care experience. Additionally, AI could give quality measure developers a way to better gauge clinical best practices and results within populations to inform new measures and compliance thresholds.
The Impact of AI on Quality Measure Adoption
The reality is that, without the support of technology, most healthcare organizations won’t be able to adopt more robust quality measures. The burden of reporting on measures that require multiple data points over time (such as medication adherence, disease management, addiction counseling, etc.), or require looking at large sets of patient records (for example, flagging cases of unnecessary imaging or preventable readmissions), isn’t scalable with manual reviews. But AI algorithms can do the heavy lifting to find relevant values for certain measures, pinpoint specific evidence in patient charts and imaging, and even determine whether data meets measure compliance criteria.
When healthcare organizations adopt AI to augment human-driven processes, quality measurement will improve across the board. Measurement accuracy will increase. Operational lift for measurement activities will decrease. Procedure and outcome measurement adoption will rise while less valuable structural measures will fall by the wayside. In turn, providers and plans will have more useful clinical intelligence to inform quality improvement activities. When acted upon, these insights will improve care quality, patient experience, and physician workloads, improving healthcare for everyone.
Curious to learn more about how AI can improve your quality measurement program? Request a demo of Quality Identifier, Apixio’s AI-powered abstraction solution.