Artificial Intelligence - A key to better risk-adjusted payments

Use of value-based contracts, which hold organizations accountable for healthcare outcomes, continues to be strong among payers and providers. The adoption of payment for value is causing seismic shifts in the healthcare ecosystem and is the basis for much recent healthcare M&A activity. Indeed, some innovative models have been reported to reduce costly emergency department use and inpatient stays.

The concept of risk sharing has bipartisan support. The Trump administration has embraced a pay-for-value concept. Health and Human Services Secretary Alex Azar stated that there is a “need for value-based transformation” because it “can improve quality, decrease cost and make our programs more sustainable.”

Value-based arrangements have also been put forward by pharmaceuticals and device manufacturers such as GE and Medtronic. In February 2018, Medtronic announced a five-year partnership with Lehigh Valley Health Network (LVHN), an eight-hospital network, with the goal of reducing cost of care by $100 million and improving quality of care for 500,000 patients.

Underlying shifts and challenges in healthcare

The rising cost of care is not leading to better care

The cost of care in the U.S. is rising at a rate greater than that of overall inflation. Healthcare costs were $3.3 trillion in 2016 alone—that’s an annual healthcare cost of $10,348 per person compared to $146 per person in 1960. The Centers for Medicare and Medicaid Services (CMS) projects that in 2017 national healthcare spending grew to $3.5 trillion. The average rate of annual healthcare inflation was approximately 3 percent each year for the past 20 years. Not only do we spend more and more money each year, it is estimated that the U.S. wastes more than 25 cents of each dollar on over-treatment, unnecessary visits and procedures, and lack of coordination.

With the acceleration of medical advances and genetic discoveries, we can effectively diagnose and treat a wider variety of diseases and reduce the burden of illness better than ever before. However, these life-saving treatments come at a very high cost. Given the big price tag, it is important that the right diagnostics and therapies are selected. And with an aging population and a greater prevalence of lifestyle-induced chronic diseases, there is a larger pool of potential recipients of these costly interventions. In this setting, wrong choices translate into unnecessary costs.

Under the prevailing payment structure, costs are often passed from the provider to the payer to the purchaser – either the tax payer or the employer – which is the primary driver for healthcare inflation each year. Over the last decade, there has been a shift towards payment for value to arrest the growth in costs and improve outcomes.

Rather than paying for each service or procedure, value-based payment is constructed to be outcomes- oriented. There are many types of value payment in healthcare from pay for performance (e.g., bonuses for higher quality physicians) to payment to cover an episode of care (e.g., bundled payments for elective knee replacement surgery) to payment to cover all costs of care over 12 months (e.g., Medicare Advantage [MA] plans). In the two latter examples, the providers have an incentive to offer care that results in the best, less costly outcomes. MA plans provide Medicare beneficiaries a better healthcare experience compared with traditional Medicare – better physician coordination, proactive care, and improved benefits – which makes it very popular with seniors.

For most of these arrangements, it is important to adjust the payment based upon the underlying health of the individual – a process called risk adjustment. Without this adjustment, the payment made would either be too much or too little for the same procedure, and physicians and hospitals would have little incentive to participate in value-based payment arrangements.

Why risk adjustment compliance done right is key for value-based care

Risk adjustment employs sophisticated models that group patients into one or more categories associated with predicted costs. The data input to these models come from medical records, billing data and other individual data. Before the widespread adoption of electronic medical records, these models (or groupers) relied upon data from administrative claims in the form of diagnosis, procedure and prescription drug codes. Now that 96 percent of U.S. hospitals have electronic medical records, a much richer source of information about patients and their healthcare is available.

Risk adjustment has been used on a wide scale for programs such as MA plans – to which CMS provides a lump sum payment for each enrolled member based upon predicted costs of care over a 12-month period. For most of this period, these models were informed by feeding codes from encounter claims or codes abstracted by certified medical coders painstakingly reviewing millions of medical records into risk scoring models. CMS mandates that risk adjustment must be performed each year for each enrollee. There is no “carryover” of diseases from last year to the present year – providers need to supply documentation that supports the active treatment of a relevant (risk-adjusting) disease. Errors are introduced by mistakes of over and under-coding made both at the physician’s office and by vendors working on behalf of the health plan.

CMS currently pays about $200 billion to MA plans caring for over 20 million members. The Office of the Inspector General of the U.S. Department of Health and Human Services (OIG) released a report that revealed that in FY 2016 approximately 10 percent of MA reimbursement is not justified. Health plans are under increased scrutiny to submit well-supported codes. The cost of non-compliance can be massive. A judge has ruled that the U.S. can move forward with its False Claims Act lawsuit suing UnitedHealth Group Inc. for the $1 billion it retained from the government healthcare program Medicare. The Department of Justice (DOJ) dropped much of the suit, but it has set a precedent that the DOJ is ready to go to court for overpayment claims.

A very small number of charts are audited each year to determine whether the codes submitted by health plans to CMS to determine the risk score and payment have the required supporting evidence in the medical records. The effort of reviewing charts to audit reported codes takes a lot time, effort and money. It takes at least twice as long to audit a chart than to review a chart for codes missed – about two provider charts per hour. At that rate, it would need 15 million hours of coder time to review every chart. If each coder could perform audits for 1,800 hours per work year on a full-time basis, we would need about 8,500 coders. There are approximately 4,000 self-identified Hierarchical Condition Category (HCC) coders on LinkedIn. Even if the actual number is double that amount, we still do not have the coders required to perform basic risk adjustment to meet current and future needs.

How can AI help with risk adjustment?

Many new types of computer algorithms can be trained to read and decipher text. These learning-based algorithms, which use artificial intelligence, have been effectively used to review hospital and clinic charts to find well-documented HCCs. Cloud-based systems running these algorithms, when applied to images of documents such as faxes and scanned records or to text from electronic records, can help coders perform audits faster and better.

In fact, a study that we conducted compared manual and technology-assisted coding applications for 75,000 records across 10 states showed coders are three times more productive with a technology-assisting coding application. The same study showed a 24 percent more supported diagnoses that similarly trained and skilled coders working unassisted. In a follow-up study, we evaluated 286,000 records and found similar results with respect to accuracy – 27 percent more codes were found with machine assistance –and productivity.

Together these two sets of findings represent the largest side-by-side studies of AI technology for risk adjustment. We have been able to demonstrate that learning algorithms can work effectively across a variety of practices, specialties, document types and templates. These algorithms can also be used to conduct coding audits, in addition to identifying missed diagnoses.

Many workflows are key to making a value-based healthcare system work. All of these rely on humans who have very little assistance. Healthcare payers and providers now have an opportunity to use technology-assisted automation of laborious processes to provide a more accurate view of patient care and its cost. Such automation can enable people to focus their attention on healthcare delivery rather than reporting, billing and data gathering. As the healthcare ecosystem embraces value-based contracts, understanding how AI platforms assist with risk adjustment can open new doors to better patient care and a more efficient system for all.

© Copyright ASC COMMUNICATIONS 2018. Interested in LINKING to or REPRINTING this content? View our policies by clicking here.

 

Top 40 Articles from the Past 6 Months