How algorithmic value sets enhance clinical decision-making

A partnership to enhance utilization management

Last year, Wolters Kluwer, an IT software and services provider, partnered with MCG Health, a developer of evidence-based decision support tools that help providers with utilization management. MCG turned to Wolters Kluwer's Health Language division for help managing value sets, or lists of codes, as well as keeping them up to date and converting them into structured rules. These codes define clinical concepts related to laboratory observations, diagnoses and medications.

Nurses and clinicians use MCG's Indicia Synapse platform to make care decisions on hospitalizations.

"MCG writes the rules, but we power those rules by helping them build those clinical definitions," Phansalkar said.

Wolters Kluwer's Health Language Data Quality Workbench, a value-set management tool, helps MCG ensure that codes are up to date and that clinicians are collecting the most relevant data for utilization management decisions.

MCG's software generates codes for vitals, such as respiratory rate and oxygen saturation. For example, LOINC codes help providers match oxygen saturation and respiratory rate to care guidelines. However, Phansalkar noted that some LOINC codes for conditions like oxygen saturation could expire.

By using Data Quality Workbench, along with its Indicia Synapse, to update algorithmically defined value sets automatically, MCG can help save thousands of manual hours in code set creation and cut clinical review time for care decisions by providers, payers and government agencies. Converting the value sets into algorithms allows clinicians to automate routine medical necessity decisions in utilization management, so they can focus on more complex decisions, according to a case study published on Feb. 5.

A reasoning engine and natural language processing capabilities within Indicia Synapse help providers match data from EHRs with evidenced-based criteria that help determine medical necessity.

This matching capability of Indicia Synapse eliminates the need to navigate multiple tabs in a patient chart to make care decisions.

In a Dec. 4 AMIA webinar, Jason Gillman, MD, senior director of clinical informatics at MCG Health, discussed how the matching process works.

"We're basically retrieving patient data from the EHR, and we're attempting to match it back to those medical necessity criteria for inpatient level of care or observation level of care," he said.

Gillman noted that the ability to match patient data in EHRs with evidence-based guidelines criteria also affects payers' decisions because inpatient and observation levels of care have different rates of reimbursement, which are key revenue drivers for hospitals.

"We want to ensure that hospitals have a sustainable operation, and that patients are admitted to the appropriate level of care," Gillman explained in the webinar.

Intensional value sets enhance data quality

Within two weeks, the partnership allowed MCG Health to modernize more than 2,500 value sets in separate spreadsheets, converting them from extensional value sets, that is, a flat list of codes, to intensional value sets, which are structured intelligent algorithms. Intensional value sets can update automatically as terminologies evolve, according to the case study. The "flat list of codes" were rebuilt as algorithms, Gillman said in the webinar.

Tania Elliott, MD, a dual board-certified allergist and internist as well as a clinical instructor at NYU Langone Health, noted that converting extensional to intensional value sets is important for data quality because they allow healthcare organizations to process structured data.

"You want to work with as much structured data as possible to reduce variability and subjectivity," Elliott told Healthtech Analytics.

Phansalkar further explained that the "messy" combination of data in structured and unstructured forms and the fact that every EHR is different make it difficult to know where in the EHR to extract data.

Providers use FHIR APIs to retrieve EHR data, which includes diagnoses, procedures, lab results, medications, diagnostic reports and physician notes, Gillman said.

"We both surface the clinical data and then we match it back to the medical necessity indications using informaticist-developed reasoning," he explained in the webinar.

He added that Health Language's API was important for exporting value sets in a proprietary, machine-readable format to integrate them with the MCG software.

Automation improves decision-making around hospital admissions

Good quality and complete data impact clinical decisions on whether a patient is a candidate for observation or inpatient care, according to Phansalkar.

Health systems are looking to reduce the volume of observation care, which is the middle point between emergency and inpatient care, she said. In the webinar, Gillman mentioned "appropriately dosed IV diuretics" as an example of observation care.

Gillman also discussed how evidence-based medical criteria can be combined with automation and AI to aid care decisions.

"Using evidence-based medical criteria along with automation and machine learning and AI, you're able to admit patients to inpatient level of care who are appropriate for that level of care and for whom you can justify that decision," he explained.

For example, by using Indicia Synapse, a large 13-hospital health system reduced observation stays in the hospital and more easily identified which patients satisfy inpatient criteria, according to the case study.

The scale of the AI tools' operations can also improve decisions on potentially unnecessary admissions.

"It can look at data at massive scale beyond what an individual can do with manual chart review," Elliott said.

"We become less reliant on manual input and human error, which can be impacted by tardiness in putting in entries, missed entries, and individual variation in subjective interpretation or technique, as well as the amount of time it would take for a human to individually search charts and records and unstructured data to find patterns," she continued.

Going forward, MCG will further develop Indicia Synapse to process unstructured data using LLMs. The platform aims to make utilization management decisions faster and more accurate by combining GenAI with algorithmic rules-based matching of EHR data and evidence-based guidelines, the case study said. 

"This gives us the potential to leverage that strength in data foundation to standardize and maybe even help codify some unstructured data to leverage it better with our new GenAI solutions," Gillman said in the webinar.

By using automation, healthcare providers can avoid a cumbersome "needle-in-a-haystack search" of reading through unstructured text and decrease provider burnout for utilization nurses.

"Needle in a haystack is a very time-consuming activity," Gillman said. "You're going through medical records in a complicated EMR, navigating through multiple tabs. You're probably reading through lots of free text, which is full of abbreviations."

Gillman said that through the ongoing partnership, Indicia Synapse can reduce utilization review time from around 20 minutes to less than three minutes.

The collaboration between Wolters Kluwer and MCG, which includes the automated conversion of value sets and making the evidence-based guidelines actionable with computable data, means better decision-making on medical necessity, not just for providers but also payers.

 "All of that decision-making is driven not just by the guideline but by the availability of the right data at the point of care," Phansalkar said.

Brian T. Horowitz started covering health IT news in 2010 and the tech beat overall in 1996.