OIG risk adjustment compliance audit showcases 3 clear takeaways
As Medicare Advantage plans push back against a proposed overhaul of the risk adjustment data validation (RADV) process that the Centers for Medicare & Medicaid Services (CMS) could finalize by November 1 of this year, the Office of the Inspector General (OIG) recently released its audit findings regarding one MA plan. OIG ultimately recommended that the plan refund $3.5 million in estimated net overpayments that occurred in 2015 and 2016.
In response to the agency’s draft report, this particular health plan objected to the agency’s conclusions, asserting that OIG’s audit methodology was flawed, that medical record documentation supported certain diagnoses, and that the agency improperly implied that MA organizations are expected to ensure 100% accuracy of provider-submitted codes. However, the health plan added that it is “in a continual process of evaluating and enhancing its compliance procedures and will consider these recommendations.”
In addition to the OIG’s audit findings, a new report published by the Urban Institute argues that risk adjustment coding contributes to overpayment to MA plans. The report further states that CMS’s approach to offsetting this coding specificity—by applying a uniform coding intensity adjustment across the MA program—is “disadvantaging plans engaging in more accurate coding while providing undeserved rewards to plans that aggressively game the system.”
Here are three takeaways that plans should keep in mind as scrutiny of the Medicare Advantage program continues to increase.
Pay special attention to high-risk diagnosis codes
OIG stated in its report that the medical records for more than 61% of the 250 sampled enrollee-years did not support the diagnosis codes, which fall under seven high-risk groups. Here is a summary of the issues OIG reported:
- Acute stroke: Diagnosis found on professional claim but not on inpatient claim. Proper documentation should have been pulled from past medical history, which does not map to an HCC.
- Acute heart attack: Diagnosis found on professional claim but not on inpatient claim; inpatient claim diagnosis missing 60 days before or after professional claim documentation. Diagnosis for a less severe disease manifestation should have been used.
- Acute stroke and acute heart attack combination: Medical records did not support submitted diagnosis codes. Although not part of this audit, a disease interaction with COPD would affect members’ disease burden and risk scoring or payment purposes.
- Embolism: Professional claim diagnosis without any anticoagulant medications indicates history of embolism versus active condition. A diagnosis should have been documented in past medical history, which does not map to an HCC.
- Major depressive disorder: Diagnosis found without antidepressant medications dispensed; documentation was generally not supportive of diagnosis.
- Vascular claudication: Diagnosis for vascular claudication with medication dispensed for neurogenic claudication. Documentation generally not supportive of diagnosis mapping to HCC for vascular disease.
- Potentially mis-keyed diagnosis codes: Numbers in diagnosis codes transposed or other data entry errors that were identified with automated OIG software that found unsupported HCC diagnoses.
Given OIG’s specific focus on red flags in these high-risk diagnosis codes, health plans should conduct additional coder training specific to these conditions and coding patterns to ensure that coding accuracy in the right setting exceeds industry standards of 95%.
Supplement processes with machine learning and natural language processing
If not already in place, health plans should add an extra layer of validation to their risk adjustment program by deploying machine learning and natural language processing (NLP) to ensure that submitted diagnoses are supported by an active diagnosis in both inpatient and outpatient claims. While human expertise is vital, given the volume of medical records that must be coded, expertise must be complemented by advanced analytics to improve coding completeness. Leveraging automation and machine learning may address concerns such as missing input claims and inconsistent and incomplete coding patterns.
Improve coding quality and electronic clinical data acquisition
MA plans have many tools at their disposal for monitoring coding quality and accuracy, but provider and vendor collaboration is critical. The three key pillars of coding quality are:
- Accuracy: Ensuring that identified conditions are properly supported in the medical record to reduce audit risks
- Specificity: Ensuring the proper combination of clinical conditions as it relates to the hierarchy of clinical coding
- Completeness: Ensuring the identification of the true quantity of supported conditions
Plans must collaborate with providers to ensure their documentation complies with HCC reporting requirements while implementing strict best practice guidelines for their own medical record coders to ensure accurate coding and reporting of HCCs annually. In choosing a risk adjustment vendor, focus not just on its automation and machine learning capabilities, but whether it can deliver robust reporting to identify which providers are submitting incorrectly coded claims, enabling the plan to proactively offer coding education support.
Finally, partner with a risk adjustment vendor that can establish and maintain API connections with the plan’s strategic provider locations, which reduces the need to chase charts to validate diagnoses, expedites clinical data acquisition, and improves operational efficiency.
In today’s ultra-competitive MA marketplace and amid increasing OIG scrutiny, plans must go beyond the fundamental activities of risk adjustment to achieve excellence. Read our eBook on Achieving risk adjustment excellence in the hyper-competitive Medicare Advantage landscape for key strategies to improve your results, including:
- Mitigating compliance audit exposure
- “Next best action” analytics to support comprehensive gap closure
- Increased digital capture, interoperability, and SDoH collection