BUILD EFFICIENCY, EMBRACE OPPORTUNITY
THE OVERPAYMENT PROBLEM
The complexity of the health care delivery and payment system means health plans frequently overpay for care, in some cases losing millions of dollars that would be far better spent on members who need it. That translates into higher costs for insurers, employers and consumers, who must absorb those wasted dollars. For some Americans, these higher costs could put health insurance and the care it covers out of reach.
The industry has made progress in some areas. For example, the rate of improper payments under Medicare Advantage dropped from 14.1% in 2010 to 9.5% in 2015, according to CMS data.1 However, data problems, contracting issues, opportunistic bad actors and other challenges mean overpayments continue today. Generally, health care providers that are overpaid must return those funds within 60 days,2 but in most cases it falls to health plans to make providers aware of the issue.
Health plans that identify and recoup overpayments improve their business standing, and members may be the ultimate beneficiaries through lower premiums and cost sharing. Although it would be more efficient to avoid payment errors to begin with, there are systemic barriers, increasing expectations among providers and strict regulations that require prompt payment of claims. Plans that fail to meet prompt payment standards are potentially subject to penalties, and in some cases payment disputes have escalated to lawsuits. It behooves plans to pay quickly, even if correction may be required later. The need for overpayment identification is unlikely to diminish anytime soon.
To ensure payment integrity, health plans have a variety of levers to pull, including network design and contracting, provider behavior modification programs, pre-authorization, and pre- and post- payment clinical review audits. One area that can be overlooked, however, is data mining.
Health plans that update their data mining programs with the latest techniques are likely to realize significant gains without the provider abrasion that can be associated with other tactics. It is worth the time and investment to modernize these data mining programs to reduce false positives, identify and address root causes of overpayment, and maximize recovery.
THE DATA MINING OPPORTUNITY
Health plans need payment integrity programs with the capability to detect overpayments. But any time the provider is brought into the process, such as through medical record reviews, provider abrasion is a concern. Plans can minimize abrasion and maximize recovery by instead optimizing their data mining programs. The right program allows plans to uncover issues and often make fixes internally, minimizing the involvement of the provider.
Inefficient data mining programs can carry their own headaches, including high false-positive rates that compromise return on investment and an inability to help plans address root causes of overpayment. The most effective programs enable health plans to identify and act on overpayments based on known patterns while also flagging unknown patterns that might indicate a new source of overpayments. Programs should be calibrated to zero in on only actionable overpayments, and they should take a smarter approach to the data-mining process itself.
Health plans that update their data mining programs with the latest techniques are likely to realize significant gains without the provider abrasion that can be associated with other tactics.
LONGITUDINAL DATA MINING
Longitudinal data mining is a higher-value process that replaces the point-in-time, individual claims analysis most plans use. Taking a wider view across
claims trends enables identification of patterns that are missed with only one-by-one snapshots. This approach unlocks the ability to capture repeated errors by the same provider, mistakes at the same point in the adjudication process, or some other pattern that becomes apparent over time. Not only can plans act to recover overpayments, they can also use what they learn to address root causes and reduce the overpayment problem itself. One key to longitudinal data mining is a unified data strategy that enables the ability to connect massive amounts of data from disparate data stores and stratify it at higher levels such as by system, specialty or facility type. Additional patterns emerge at this level, and reliance on a partner that can bring together multiple large data sets also allows for benchmarking, so plans understand where they fall among their peers.
CARE TRANSITION ANALYSIS
One particularly useful application for longitudinal data mining is known as care transition analysis, which uncovers overpayments that can occur as members move between sites of care. Scrutiny of these transitions can spotlight inappropriate referrals, of particular interest in the age of provider consolidation. The ability to link and track claims from site to site also allows health plans to uncover questionable coding and billing patterns. Care transition analysis flags anomalies such as when hospitals and skilled nursing facilities both bill for the same rehabilitative care. Such an error is costly but likely unintentional and can be addressed on the plan side with minimal provider abrasion.
THE KEYS TO DATA MINING ROI
When the following principles are integrated into program development, health plans reap efficiencies such as fewer demands on internal staff to resolve issues, a better recovery percentage and reduced provider abrasion.
Any data mining partner should begin with a learning phase that enables the data team to understand the health plan’s internal processes and systems. In some cases, inefficiencies can be identified right away, while in others internal processes may overwrite apparent patterns in the data. This step is also key to program customization, which is critical to achieving high accuracy and a favorable ROI.
CUSTOM CONCEPT DEVELOPMENT
Overpayment identification programs should always be customized to a given health plan. So although program development begins with industry- standard concepts, it should also integrate contract- and policy-specific concepts as well as standards based on which type of overpayment a health plan is and is not willing to collect on.
Program development should also include aggressive data testing on the part of the payment integrity partner before scaling up. This process is key to ensuring any patterns identified in the data are valid and actionable. Any data-mining partner that takes a health plan data set at face value is not doing its due diligence and will likely return a high rate of false positives. Health plans should be looking for an accuracy rate greater than 95%.
HYBRID ANALYTICS ADDS VALUE
Advanced analytics is revolutionizing other payer programs, and plans that bring this thinking to their overpayment identification programs will uncover new sources of claims errors and additional opportunities to recoup inappropriate payments. A hybrid analytics approach builds on data mining to maximize the crossover between human intellect and machine learning.
Hybrid analytics is a data-driven process that begins with a clear understanding of the business problem it is meant to solve. Patterns in the data – which become evident through a unified data strategy and longitudinal data mining – inform hypothesis generation and concept development. Testing validates concepts, and then they are applied at scale.
The next generation of hybrid analytics involves machine learning, in which models are trained to understand the data. As data models are optimized, additional patterns and insight emerge, and then those insights feed back into further model optimization. As models are refined and scaled up, overpayment recovery and root cause analysis both gain efficiency.
THE BUILDING BLOCKS OF HYBRID ANALYTICS
Hybrid analytics integrates the capability to detect:
- Rules-based patterns: Identify expected abnormalities, such as overlapping services.
- Supervised patterns: Anomalies such as unusual allowed amounts for the same service.
- Complex patterns: Identify complex patterns in data, such as member risk for long-term disability.
- Linkage patterns: Link analysis identifies associative patterns such as benefits administered differently for the same plan.
A CUSTOMIZED OVERPAYMENTS SOLUTION
Health plans that adopt a customized data mining program will benefit from fewer false positives, reduced provider abrasion and ultimately a seamless process that requires far fewer internal resources. Meanwhile, longitudinal data mining uncovers patterns that traditional programs might miss, including at key and costly points on the care continuum through care transitions analysis. Hybrid analytics takes data mining to the next level by relying on the crossover between human intelligence and machine learning. With hybrid analytics, root-cause analysis is further enabled, and overpayment identification becomes a selfoptimizing process. Additional value is unlocked.
Overpayments are a source of frustration across the industry, but the right approach will reduce the pain for payers and providers. In a challenging market where excess costs translate to higher premiums that could put care out of reach, it is a problem worth the investment in a smarter solution for recovery.
DATA MINING ONBOARDING
When health plans and data-mining partners invest time in preparation and testing before scaling up programs, they eliminate many of the problems of overpayment identification.
Initial steps include:
- System access: Claims processing systems, policies, procedures, contracts, editing systems and member benefits.
- Data management: Review, feedback and correction of sample files.
- Setup and learning phase: Establish overpayment threshold, notification timeline, exclusions, format and frequency for results. Prepare staff to manage the process.
- Query development: Identify, customize and validate queries based on health plan requirements.
- Review and approve overpayments: Turn on queries incrementally, continually verifying.
- Optimize: Review denials and adjust queries as needed, assess top-paid providers, develop and test new queries based on existing data and results.
Written by EXL Health Team