February 17, 2014

Mining is More than Just Digging a Hole

By

“Big data” – we have all heard the term, as it points to the degree of ICD-10 specificity that will provide more comprehensive and complete data for outcome analysis and decision-making. As the industry is preparing for the clinical documentation and coding changes inherent in ICD-10, how are we preparing to understand the outcomes of this enhanced data specificity?

Mining means more than just digging a hole and hoping you find gold. The bigger the data, the easier it is to produce meaningless trends and data outcomes.How prepared will you be to identify meaningful trends, understand the impact of ICD-10, and analyze the Medicare case mix index?

Data is different from information

Analyzing big data and converting it into actionable items is more challenging than people think. More information is not always the answer and can easily lead to inaccurate conclusions. Some outcome analyses produce large volumes of data without representing concrete findings reflecting our logical conclusions. Have you ever seen a Medicare case mix index analysis that has pages of information, highlighting myriad deviations from benchmarks, only to demonstrate variances without merit? Does a 2 percent deviation from MedPAR data represent a truly aberrant finding?

I find that with many current ICD-10 reimbursement impact analyses, the predictability model used is logically flawed. A prime example of a flawed predictability model is an ICD-10 reimbursement impact analysis that predicts that a hospital will receive increased reimbursement for DRGs that are not dependent on clinical documentation, coding, or DRG grouping hierarchy changes. Further investigation of the constructs of these predictability models usually uncovers that the logic used as it pertains to reimbursement impacts were derived from general equivalency mappings (GEMs) enhanced with additional statistical calculations applied across the entire DRG population. This yields predictions of increased ICD-10 reimbursement for DRGs that are not dependent on ICD-10 coding specifics, but rather on the patient population. Understanding the framework of the predictability model, but more importantly understanding the framework of what you are analyzing, is critical for deriving meaningful outcomes.

Know What Influences the Outcomes


Coded data outcomes are influenced by several contributing factors, including clinical documentation, coding quality, patient populations, and regulatory changes. Don’t take the data at face value. Understanding what impacts the outcomes will provide you with a starting point to begin your analysis. Data analytics, without the understanding of the impact these components can have on the data, can easily result in misguided conclusions.

Another key is to gain trust in the data outcomes. How do you do that?

Incorporating the right personnel in data interpretation is critical. Analyzing coded data without having the subject matter knowledge to interpret the outcomes (and what contributes to these outcomes) will only produce information and not actionable results.

  • Start with the end in mind. What are the outcomes you want to prove or disprove with the data analysis? Start with a conceptual framework and begin analyzing the influencing factors to identify trends impacting outcomes.
  • Apply reasonability to the outcomes. If you are a 250-bed hospital, benchmark your outcomes to other similar hospitals in your peer group.
  • Question the outcomes with skepticism; don’t have blind faith that they are correct. Prove out the results to gain confidence in them. Remember that flawed data will provide flawed conclusions.
  • Focus on the process and regulatory changes to understand the impact these will have on outcomes.

A prime example of this is the new “two-midnight rule.” The change was intended to address widespread complaints that Medicare's rules were too vague regarding when a moderately sick patient should be admitted for expensive inpatient care instead of outpatient observation. Medicare's recovery auditors, which will employ sophisticated data mining to locate questionable claims, will begin to review short stays on a limited basis starting Sept. 30and will deny payment if the patient record doesn't support medical necessity for the setting. Hospitals’ reporting of the wrong setting continues to be the most common reason for RAC denials following complex reviews.

The two-midnight rule will require hospitals and physicians to work much more closely to ensure that the admitting physicians’ documentation, diagnoses, and codes support the appropriate setting for the stay. If the RAC denial trend continues and the physicians’ documentation doesn’t support the inpatient short stays, according to the two-midnight rule hospitals will see fewer short-stay inpatient admissions. This means that the less sick patient population, which Medicare agrees should be treated in an observation setting, will be removed from hospitals’ case mix index, resulting in a recalculation of the CMI and a LOS shift. There are also several DRG grouping logic changes that will also take effect in October 2014. Without understanding how these changes will impact outcomes, one can see how misinterpretation of them can lead to incorrect conclusions.

You will find that coded data outcomes have multiple factorial influences that will impact results. Being aware of these complexities and enhancing your analytic skills will be invaluable in identifying opportunities for improvement with big data.

Digging a bigger hole doesn’t necessarily mean you will find more gold, but looking in the right location will yield better outcomes.

About the Author

John Pitsikoulis, RHIA, is the ICD-10 practice leader and an AHIMA ICD-10-Approved Trainer for Nuance Communications. John has more than 28 years of revenue cycle, health information management, coding, and compliance consulting experience. John has developed and led several corporate and client strategic engagements for managing the conversion to ICD-10, including ICD-10 assessments, implementation planning, integrated testing, education plan management and revenue preservation strategies. 

Contact the Author

To comment on this article please go to

John Pitsikoulis, RHIA, is the ICD-10 practice leader and an AHIMA ICD-10-Approved Trainer for Nuance Communications. John has more than 28 years of revenue cycle, health information management, coding, and compliance consulting experience. John has developed and led several corporate and client strategic engagements for managing the conversion to ICD-10, including ICD-10 assessments, implementation planning, integrated testing, education plan management and revenue preservation strategies.