Health plan SIUs find deep learning models for fraud detection provide advantages over rules-based tools.
The onset of the novel coronavirus pandemic highlighted a glaring weakness in rules-based FWA systems: the need to develop rules well ahead of their actual use in order to attempt to catch coronavirus-related FWA in a timely manner. At best, these rules are educated guesses at schemes that health plans are only beginning to pursue. But, as we’ve mentioned before, a fraud detection solution powered by deep learning eliminates the need for specific rules to be created. Even during an unprecedented pandemic, an A.I.-based solution natively surfaces aberrances that indicate fraud, waste and abuse. While the benefits of that change are significant, it’s also a huge mindset shift. How does fraud detection without rules even work?
But first, let’s discuss why rules-based fraud detection is no longer the ideal tool for the job, especially in the current environment of stressed providers and increasingly complex novel fraud schemes.
Why Not Rules?
If your health plan has been working to mitigate fraud, waste and abuse for any length of time, you likely have invested a great deal of resources in a rules-based system. But the novel coronavirus pandemic has exacerbated some concerns around relying solely on this traditional approach to fraud detection.
From the beginning of the pandemic-created healthcare crisis, there have been concerns that the rapid adoption of telehealth would usher in a wave of fraud designed to exploit the relaxed rules that made its expansion possible. While that outcome remains to be seen, virtual care encounters certainly go against the conventions of in-person provider visits that fraud rules have been structured for. Legitimate providers conducting majority telehealth visits could easily trigger rules designed to catch outliers in place of service, impossible days and numbers of visits.
False positives risk provider abrasion
How are you distinguishing legitimate providers whose billing patterns have changed for good reason over the last few months from a few fraudulent opportunists? With the fallout from COVID-19 battering providers, health plans have been understandably reluctant to overly burden their providers with aggressive pursuit of overpayments.
Unfortunately, traditional rules-based systems – even those updated with coronavirus-specific queries – are prone to creating false positives that overwork investigators and risk provider abrasion. Providers have already raised the alarm that COVID-19 diagnosis codes don’t accurately reflect some of the more severe complications of the diagnosis, which could worsen the false positive rate of rules-based tools that can’t evaluate multiple approaches to the data simultaneously and fare poorly with unstructured clinical information.
Novel fraud schemes
As we presented in our short on-demand webinar on the topic, a unique healthcare crisis like this one can absolutely give rise to bad actors gaming the system for personal gain. But traditional rules-based fraud detection tools are primarily designed to look for fraud schemes you already know about. As a result, the SIU could easily miss novel fraud schemes featuring more subtle aberrances that nonetheless add up over time. The success of these more sophisticated approaches depends on the assumption that health plans have limited their fraud strategy to the same tools in use for decades.
Rules vs. Deep Learning Models
If rules-based tools are increasingly outmatched by the growing complexity of healthcare data and fraud schemes, what kind of solution features the flexibility and advanced technology to help health plans meet their FWA goals? In a previous article, we covered just how deep learning is uniquely suited to overcoming many of the common challenges faced by the SIU. But, how does fraud detection without rules operate?
To help illustrate how fraud detection differs in a rules-based environment and with a deep learning model, let’s explore a typical COVID-19 claim you might encounter in the SIU.
A 55-year old patient calls his Internal Medicine provider with complaints of chest congestion and a sore throat. The patient has underlying bronchial asthma and sees his Allergist for this condition once per quarter. The Internal Medicine provider schedules a telehealth visit and, based on his findings from the visit, prescribes a full lab workup for COVID-19. The payer is billed the following:
- Dx: COVID-19
- Office visit 99214
- Lab codes 87635, 86328, 86769
How does this new claim make it through the SIU to be evaluated for potential fraud?
COVID-19 scenario: typical rules-based environment
If the SIU works primarily with a rules-based system, it’s possible the tool has been updated with a query for Level 4 office visits with a lab panel based on a diagnosis of COVID-19. Otherwise, the rules will need to be updated accordingly with the help of the IT or analytics team. Then, the SIU would query other Internal Medicine providers billing the same combination and manually compare rates and paid amounts looking for clusters of outlier claims.
These findings may lead the SIU to additional queries to increase the likelihood of understanding if there is fraud or not:
- Query patient for other visits?
- Query Internal Medicine provider for past 6 months on COVID-19?
- Is this a trend for this provider or a one-off?
- Is the provider’s concentration of COVID-19 diagnoses higher because of his geography?
COVID-19 scenario: deep learning model
On the other hand, if the SIU is working with a fraud detection solution powered by deep learning models, much of the work related to historical claims and peer-to-peer provider comparison information has already been accounted for in the system. In this scenario, the patient’s history of asthma, peer provider comparison based on block of dates, spikes vs. trends, geography, patient acuity and more are present in the models, which continuously update as new information comes in.
From there, the deep learning models in the system push likely provider fraud leads to the SIU. With Pareo Fraud: Detection, the billing provider could be highlighted with a high Super ScoreTM based on high model scoring for Unusual Active Days, Ratio of Px/Codes and High Claim Volume. The investigator could then one-click drill down to the provider for more supporting detail on these scores and optionally conduct additional peer-to-peer comparison.
This workflow supports the investigator’s expertise while greatly reducing the manual effort and eliminating the need to “pull” unprioritized query results from the system.
Benefits of Deep Learning Models for Fraud Detection
Deep learning models for fraud detection allow the SIU to go beyond the Boolean, if-then logic of rules-based tools. Freed from these restrictions, investigators are better able to discern the true potential fraud offenders and realize several differentiators that offer unique value:
- Eliminates need for multiple queries and manually layering those results
- Continuously learns and evolves so that the more investigators work with it, the more precise it becomes
- Adds specificity to results to reduce false positives and bring focus to providers with high likelihood of fraudulent or abusive billing practices – including the amount of dollars at stake
- Digs deeper and wider into the data in multiple dimensions to find the subtle changes in provider behavior that may indicate FWA
- Extends capabilities of identifying relationships in the data to find hidden, previously unknown schemes
Pareo Supports the SIU in Transitioning to Fraud Detection Powered by Deep Learning
The benefits provided by deep learning models go a long way towards your compliance goals, including FWA prevention. But making a move like this, along with its associated mindset shift, can also feel like a gamble. The perceived risks associated with upgrading your FWA solution are largely centered around the discomfort of change for users and concerns about wasted resources. But conserving resources – time, effort and expertise – is exactly what a deep learning-powered fraud detection solution helps accomplish.
And, for those SIU teams that have invested heavily in a rules-based tool and value its reliability and familiarity, that effort is not lost. As an integrative platform, Pareo allows you to easily connect this resource as another useful data input that Pareo Fraud: Detection can leverage to improve the precision of results more quickly.
Finally, as the only solution for the SIU that seamlessly integrates overall payment integrity efforts, Pareo Fraud can help you quickly covert detection results to leads or cases, communicate relevant information back-and-forth with the audit function, and refer honest billing mistakes for provider education. This ability to maximize recoveries and avoidance at the most optimized cost end-to-end is unique to Pareo and ensures this solution proves valuable for all stakeholders – both internal resources and network providers.
THE FUTURE OF FWA IS NOW
Learn more about the ClarisHealth 360-degree solution for total payment integrity and FWA, Pareo Fraud: Case Management and Detection.