Denials are a substantial and persistent problem all healthcare providers face.
They can cost healthcare organizations between 3 to 5 percent of their net revenue annually, according to some estimates, while others project the financial implications are even higher. With tens of millions of dollars at stake, hospitals and health systems are constantly searching for ways to improve their denials management systems.
Many hospitals and health systems have devised best practices for the revenue cycle, created lean processes for coding and billing and introduced new analytics systems. Regardless, denials management remains problematic.
"The reason the problem is persistent is that there is no longer any 'low-hanging fruit,'" Alison Gilmore, PhD, a data scientist at Ayasdi, said during a Sept. 15 webinar. "The solutions that are left are complex, not obvious."
Dr. Gilmore says the underlying challenge preventing hospitals from reducing the number of denied claims is the complexity of the data.
"There are at least 100 features associated with every claim. That's a lot of data, and it's very complex, intensely multivariate data," said Dr. Gilmore. "Denials are not related to any single one of those features that you know about any particular claim, they're related to combinations of those features, and that makes for a challenging analytics problem."
A revenue cycle manager might decide to take a macro analysis approach to determine the factors that contribute to denials, and then determine an intervention to remedy them. But where should one start? The manager could look at the payer, provider, facility type and denial type. Unfortunately, this approach presents only basic information and lacks the specific descriptions necessary to create an actionable solution.
An alternative is to take a narrow micro analysis approach, according to Dr. Gilmore. Someone can review individual claims and look for patterns. Perhaps 10 claims were miscoded in the same way. Maybe one facility doesn't follow proper precertification procedures, or physicians aren't following physician documentation procedures. Unfortunately, anecdotal patterns such as these are local and it is impossible to tell if they will generalize, according to Dr. Gilmore.
"The macro level analysis and the narrow micro level analysis both have their place, but they're both very limited," she said.
The ideal type of denials pattern is specific and rich in description. This allows revenue cycle teams to act on them with a nuanced and targeted intervention. That is where most of the opportunity for addressing denials lies, according to Dr. Gilmore.
How can hospitals' and health systems' revenue cycle management departments find these patterns? By using machine intelligence.
"Traditional analytics have hit the wall because they all start with people asking questions," said Dr. Gilmore, noting that the vast amount of data related to each claim opens the door to an exponential amount of insights and combinations of factors leading to denials. "Smart people asking smart questions still cannot overcome this fundamental complexity problem to produce rich solutions."
Instead, with machine intelligence, you start with the data, let software "do what it's good at" (automated discovery), identify relevant patterns and then present the patterns to the human decision-makers to interpret. The incorporation of software solutions and automated discovery alleviates the burden of manually searching for and analyzing patterns, which is extremely time consuming.
How it works
Machine intelligence can reduce denials in three ways.
- It identifies drivers of rejections and denials for groups of claims
- It prioritizes and streamlines the work queue for claims resubmission
- It informs upstream process changes to prevent future denials
First, the automated discovery software applies algorithms to a data set. The data sets include all relevant data recorded on UB-04 hospital billing data, additional account data from a provider's financial databases, associated 835 remit transactions (such as denials, rejections, underpayments and payments), primary payer data and personal or family guarantors. The algorithms used to analyze all of this data include statistical, geometric and machine learning algorithms. The software then constructs a multifaceted notion of similarity, and clusters similar data points on similarity maps that reveal mid-level patterns.
The visual map shows groups of similar data points, represented by "nodes" or dots, with "edges" or lines connecting similar nodes. Similar claims will tend to be close together in the map, while very different claims will tend to be far apart. Nodes are also colored according to a scale from low to high concentration of denials, giving the human decision-maker interpreting the map to discover which distinct groups of similar claims have higher concentrations of denials.
"Hot spots," or areas of similar claims where there is a high concentration of denials, are investigated to produce an output denials hot spot report, which includes the most relevant factors leading to denials among that group. RCM staff can then identify the drivers of rejections, drive process modifications and reduce denials upstream.
Challenges ahead: ICD-10
With the transition to ICD-10 and multitudes of new codes effective Oct. 1, there will be even more data to sort through at an even more granular level. CMS estimates the transition to ICD-10 could double the rate of denials, or worse.
In terms of denials analytics, this transition is going to take a hard problem and make it even more complex — it will make traditional methods of denials analysis hit the wall even faster, according to Dr. Gilmore.
The first thing ICD-10 will do is increase the amount of codes substantially, which will increase the granularity of data captured about each claim. This surge of data will not do well with macro or micro types of analyses, Dr. Gilmore explained.
"You really need the data-first approach — you won't be able to rely on the same heuristics. [Healthcare organizations] need a solution that that can infer new rules as they arise, and that can be automated and adapt as new patterns emerge," said Dr. Gilmore. "Finally the transition is going to make obsolete any kinds of methods or heuristics that rely on any understanding of the ICD-9 hierarchy, or heuristics on the existing code sets. You need a code-agnostic method that doesn't care particularly about the ICD-9 hierarchy, but uses the general analytics approach to understand the codes as features of claims."
Click here to view the webinar on YouTube.
Click here to download the webinar as a PDF.