Boosting healthcare payment integrity with predictive analytics
Improper payments cost Medicare and Medicaid an estimated $100 billion in 2023 alone—a staggering amount that underscores the challenges of fraud, waste, and abuse in these critical programs. While the Centers for Medicare & Medicaid Services (CMS) has made strides in reducing improper payment rates, the fight to safeguard taxpayer dollars is far from over.
One of the biggest hurdles lies in addressing fraud effectively. The Center for Program Integrity (CPI) at CMS receives far more leads about potential fraud each year than its staff can realistically investigate, making it essential to prioritize the most egregious cases.
The scale of this work demands innovative approaches to protect resources and restore the public’s trust. Technology can help by transforming large CMS datasets into actionable insights that can enhance fraud detection, ensure payment integrity, and promote the efficient use of taxpayer dollars.
How technology can enhance CPI investigations
1. Speeding up medical reviews
Currently, reviewers (e.g., nurses or other clinicians) manually sift through reams of clinical documentation to ensure a paid claim was substantiated. They also must determine whether the documentation provided is legitimate—a growing challenge in the era of artificial intelligence. Techniques like machine learning can help detect irregularities in these files, such as billing spikes, and natural language processing can flag inconsistencies in documentation. By streamlining the manual efforts currently employed by reviewers, clinical staff can focus their time on medical review decision-making and process a larger volume of reviews given the same labor pool.
2. Identifying priority leads
Tips about potential Medicare fraud come to CPI from a variety of sources—internal investigations, hotlines for patients and caregivers, and referrals from other agencies, contractors, and lawmakers. The data in these leads can be complicated, particularly if patient harm is involved. Employing tools like predictive modeling can help identify fraudulent patterns in claims submissions and attributes of the providers who are billing them, as well as the resources needed to audit or investigate them. This makes it easier for investigators to prioritize those leads.
3. Spotting serial offenders
A provider who was involved in a previous fraud scheme is a bigger concern for the CPI than one who’s new on the radar. Yet it’s easy for providers to get a new tax ID, billing credentials, or change their business’s name to keep from being discovered. Advanced analytic techniques can break down silos among claims, provider, and beneficiary datasets to help investigators discover those links quicker and more efficiently than current manual methods.
4. Simplifying investigation summaries and referrals
CPI investigators serve many different stakeholders, all of whom have different reporting requirements. Staff spend a lot of time typing up reports for the same case in multiple formats to answer the diverse questions that stakeholders will ask of a referral. Generative AI can help with this time-consuming work, enabling users to develop many different reports based on the same pool of data. A human investigator can then review each output to ensure accuracy and appropriateness.
Tech as supplement, not a replacement
For all the fanfare, AI cannot—and will not—make the ultimate decision to take action against a provider or facility for alleged Medicare fraud, waste, or abuse.
For all the fanfare, AI cannot—and will not—make the ultimate decision to take action against a provider or facility for alleged Medicare fraud, waste, or abuse. If a regulator or defense attorney asks a CPI investigator, “Why are you looking into Dr. Smith?” that investigator still must show a chain of custody, and AI cannot do that work on its own.
Instead, AI and similar technologies can be a force multiplier. These tools can increase the speed and accuracy of CPI investigations by handling the “grunt work.” They can comb through unstructured patient files faster, freeing up time for CPI and support staff to pursue more cases without sacrificing the integrity of their investigations.
Increasing the productivity of CPI is a worthy goal. In 2022, CPI’s activities produced more than $8 for every $1 spent. Incorporating technology and advanced analytic techniques into CPI’s processes can drive that ROI even higher.
A peek at what’s possible
What might an AI-enabled investigative tool for CPI look like?
Imagine an engine like ChatGPT that:
- Connects to CMS’ public data on billing and claims data.
- Has built-in knowledge of CMS and industry medical codes.
- Offers an easy-to-use interface that allows non-technical users to query the database.
- Answers both basic and complex questions and prompts.
- Can build on previous questions and prompts using a “Chat with History” function.
- Validates and documents its work.
An executive user may prompt the engine just like they would ChatGPT, asking questions such as, “How has the billing of office visits changed from 2019 through 2022?” The engine would return a graphical plot of the expenditures related to the correct set of procedure codes selected by the model and describe the trends in text using key statistics. The user could then submit follow-up inquiries of the tool. For example: “Based on these trends, what can we expect the billing of these codes to look like in the coming two years?”
Likewise, an investigator engaged in a provider review may submit inquires such as “When did this provider’s referrals to the fraudulent lab start? Did they take off quickly at a certain point in time?” And while auditing that same provider’s professional billing, the reviewer may wish to answer questions like “What services is the provider billing for the patients they referred to the fraudulent lab?” In a similar capacity, a medical reviewer who is making claims determinations may ask questions such as “Did this patient ever see the referring provider before the provider made the referral for expensive DME? If so, when, and what services did the provider bill for this patient?”
A tool like this would drastically reduce the amount of time CPI investigators, analysts, and leadership spend manually gathering data points, and instead allow them to focus their energies on making expert decisions that require human insight. This functionality will be essential as CPI strives to match the speed with which healthcare fraud is occurring.
ICF is currently developing a prototype of this tool within the framework and policies of the CMS AI Playbook. It’s a product of ICF’s deep expertise in large-scale development and deployment, as well as our broad experience working with federal agencies on digital transformations. Partnering with ICF can help CPI modernize its processes and address more claims—and, in the process, help ensure the American health care system is fairer, more efficient, and more trusted by the public.
To learn more about how ICF can help agencies uncover or prevent Medicare fraud, waste, and abuse, please contact us today.
