Predictive Modeling Shortfalls for Assessing ICD-10 Financial Risk
By Kerry Martin
Predictive modeling is commonly used to determine the financial risks associated with a provider transitioning from the ICD-9 code set to ICD-10. While predictive modeling lays the foundation for assessing financial risk, it is critical that organizations realize that it should be utilized only as the starting point for determining true ICD-10 financial risk.
The process of predictive modeling generally utilizes a crosswalk, such as general equivalence mappings (GEMs), for mapping one year’s worth of ICD-9 claims to ICD-10 pseudoclaims. In some cases, Medicare Provider Analysis and Review (MEDPAR) data are used and, in other cases, a facility will provide its previous year’s claims data for mapping and analysis. Through this mapping process, claims are created, grouped, and analyzed against the original ICD-9 claims.
Using predictive modeling alone, organizations can assess potential financial risk but with a limited view. Certainly, predictive modeling of either MEDPAR or actual claims data will assist in identifying diagnosis-related groups (DRGs) and the types of cases that are at risk for shift between ICD-9 and ICD-10. However, because of the limitations of MEDPAR data, organizations will not be able to determine the granularity and detail of that risk. By modeling actual claims data, organizations can further use predictive modeling analytics to drill down and assess potential risk by physician provider as well as diagnosis and procedure code use.
For many, predictive modeling is a quick way to get a general idea of financial risk. However, there are risks in relying solely on predictive modeling. In this case, what you don’t know can hurt you.
Predictive Modeling Limits
Predictive modeling techniques give hospitals a false sense that all of their submitted ICD-9 claims will be mapped and analyzed through the process.
In reality, because of flaws in GEMs and DRG groupers, the existence of combination codes, and the enormity of potential ICD-10 pseudoclaim permutations in translating ICD-9 claims (which, in some cases, can top out at more than 100,000 ICD-10 claims from one ICD-9 claim), not all ICD-9 claims can be translated to an ICD-10 claim without human intervention. In fact, it’s likely that only 40% to 50% of an organization’s claims will be able to be translated to ICD-10 using a crosswalk.
To get to the real financial risk, ICD-10 financial risk categories should be identified at the onset of any project. Risks will vary from facility to facility but should include claims with potential DRG shift, claims with high risk of documentation deficiency based on diagnosis or procedure codes, claims that could not be mapped to ICD-10 because of permutation limits, and ungroupable ICD-10 pseudoclaims.
Although all DRG shift can’t be avoided, documentation audits and ten-coding a subset of claims contained in the above-mentioned risk categories will provide the data and analytics to assess true financial risk.
The number of claims an organization will need to review and ten-code also will vary depending on its size and its case mix. However, experience shows that up to 30% of claims will need to be reviewed. The ten-coding exercise will provide additional analytics and actionable data to assist in a smooth transition to ICD-10.
Benefits of Ten-Coding
In most cases, organizations will find that their highest risk DRGs may not be their top 10 or even top 20 grouped DRGs, and they should consider reviewing a subset of claims in all risk categories. The updated Medicare severity-DRG (MS-DRG) groupers were developed by the Centers for Medicare & Medicaid Services with payment neutrality as a core goal. Generally, the top 20 DRGs across the country have been thoroughly analyzed and modified through the ICD-10 code set to be payment neutral. In effect, analyzing these common top DRGs is not getting at the real risk.
Through my company’s ten-coding projects, we’ve found that it is the remaining DRGs—and in many cases, those that contain complex comorbidities—that have a higher risk of a significant DRG shift if not completely and accurately documented.
Only with claim reviews and ten-coding efforts can hospitals discover documentation deficiency opportunities because of the specificity and granularity required in ICD-10. So although there is risk, we also have found that opportunity exists in current documentation and coding practices.
Case Study: Going Beyond Predictive Modeling
A large teaching hospital in the Midwest sought an effective means to identify potential ICD-10 risk areas because of current practices, including an analysis of reimbursements and case mix. To help achieve its objectives, the hospital requested a more intensive ICD-10 clinical documentation gap and analytics review.
Over the course of 11 weeks, the project team analyzed more than 2,000 of the hospital’s clinical documentation records to determine inefficiencies under ICD-10 and identify MS-DRG shifts to determine potential impacts on the organization’s financial health.
The goal was to uncover the hospital’s MS-DRG shifts that would create the strongest impact. With this information in hand, the executive team would have the tools necessary to customize coder education and perform informed clinical documentation improvement.
The organization also wanted to quantify unavoidable shifts and better equip its chief financial officer and chief information officer to manage ICD-10’s anticipated financial impacts.
The project included a review of 10 to 20 cases for the top 100 surgical DRGs with approximately 12 months of claims data. As part of the methodology, the hospital’s ICD-9 Medicare inpatient claims were mapped to ICD-10. ICD-10 pseudoclaims were created based on GEMs and grouped through the MS-DRG grouper version 28.
First, coding claims in the following key areas were analyzed:
• cases with potential DRG shift (In this case, reports showed that 33% of the hospital’s MS-DRGs were going to shift under ICD-10.);
• cases producing the highest volume and highest revenue; and
• cases from high-production physicians.
Then the team conducted a clinical documentation review to accomplish the following:
• identify at-risk claims and prepare them to be recoded in ICD-10;
• evaluate documentation deficiencies to determine missed opportunities for more accurate DRGs, code assignment, and clinically appropriate reimbursement; and
• create a baseline of documentation deficiencies by physician and diagnosis/procedure code or DRG.
Specifically, my company not only ten-coded a subset of claims, but the team also divided claims into the following five categories:
• claims with insufficient documentation to ten-code (One or more grouper-significant diagnosis or procedure codes cannot be coded without additional documentation.);
• claims that experienced DRG shift because of grouper functionality (All grouper-significant diagnoses and procedures can be coded in ICD-10, but DRG shift occurs because of grouper translation in ICD-10.);
• claims that shifted, positively or negatively, because of documentation deficiencies (All grouper-significant diagnoses and procedures can be coded in ICD-10, but DRG shift occurred because of documentation deficiency.);
• claims that did not shift but contained opportunities for improved clinical documentation and coding education; and
• claims with no DRG shift with no documentation deficiency.
When completed, outcome reports showed the hospital’s top 20 DRG shifts, the number of claims shifts per DRG, and a detailed financial analysis showing the financial impact of positive and negative DRG shifts.
By ten-coding a subset of claims, the hospital now can adjust coder and physician training as necessary; hone in on DRGs that may experience an increase or decrease in payment, which is helpful for budgeting and payer negotiations; and discover and resolve specific documentation deficiencies.
In addition to being able to determine true ICD-10 financial risk and create laser-focused training road maps, the data and analytics available from actual ten-coding efforts also can assist with payer negotiations by managed care professionals in a health care organization prior to ICD-10 implementation. By knowing what DRGs will be most relevant to your organization after the ICD-10 transition, your team will be able to effectively negotiate reimbursement terms.
Predictive modeling may seem efficient and fast, but it gives hospitals a fuzzy picture, at best, of true financial risk. For reliable information on actual DRG shift and the top DRGs that have either a positive or negative financial impact in a specific health care organization, it’s necessary to go beyond predictive modeling.
— Kerry Martin is CEO and founder of VitalWare.