Medical Coding Data Mining: A Must-Have Skill for Coding Professionals

By
Original story posted on: November 11, 2019

Data mining skills will be a prerequisite in ICD-11.  

Medical coding has long been identified as the “signature” skill set or area of expertise of the health information management (HIM) profession, but the new signature skill set is quickly becoming data mining – due in large part to changes in the health information technology (HIT) associate degree programs, and in how ICD-11 diagnosis codes will be assigned and managed.

HIT Associate Degree Programs
No later than Sept. 1, 2021, all two-year HIT associate degree-level programs that are accredited by the Commission on Accreditation for Health Informatics and Information Management Education (CAHIM) must elect to teach a:

  • Revenue management (RM) track and/or
  • Data management (DM) track.

In the past, HIT associate degree-level program curricula primarily focused on RM content (i.e., medical coding and reimbursement); however, while conducting surveys and research for their HIM Reimagined initiative, the American Health Information Management Association (AHIMA) found that data management is a much-needed competency for both the future and sustainability of HIM professionals, and for the HIM profession overall.   

So, AHIMA used the results of their HIM Reimagined research to work with CAHIM on revising the curricula competencies for HIT associate degree-level programs – and now these programs must elect to teach an RM and/or DM track. Both tracks are, however, required to include a math statistics course, because math statistics is now considered to be a core competency for all HIT students. Those HIT students who choose the DM track will be required to learn how to manage data within a database system, and as a result, they will learn about data mining, an invaluable tool for data management.

ICD-11 Diagnosis Codes
The inclusion of a data management track in HIT associate degree-level programs is appreciated even more when we consider how ICD-11 will forever change the roles and responsibilities of coding professionals.

When ICD-11 is implemented in the United States, “coders will become auditors of computed codes, and guardians of the code data quality,” as was stated during a recent ICD-11 expert roundtable in Washington, D.C.

To be effective and impactful auditors of ICD-11 coding data, coding professionals (e.g., senior coders, coding quality coordinators, coding supervisors, coding managers, coding directors, etc.) will need to utilize data management, specifically, data mining, to determine, for example:

  • If some types of cases should be audited prior to billing (pre-billing);
  • The specific types of cases that must be audited (e.g., complex conditions and procedures, new technology payment groups, etc.); and
  • If the audit results identify opportunities for physician clinical documentation improvement (CDI) and/or coder continuing education/re-training.

Now that you understand why it’s imperative that coding professionals add the data mining tool to their toolkit, let’s define data mining, in the context of medical coding.

Data Mining
Data mining is defined as sorting through data to identify patterns and establish relationships, such as association (when one event is connected to another event), sequence (when one event leads to another later event), classification (discovering new patterns by assigning items in a collection to target categories or classes), clustering (identifying and grouping related facts that were not previously known), forecasting (identifying patterns in data that lead to reasonable predictions).

Let’s review examples of real-life medical coding-related patterns and relationships, to get a better understanding of data mining (note that the details of these real-life examples have been simplified, to stay within the scope of this article).

Association
A coding quality coordinator generated a report listing all of the cases that were coded by a new emergency department (ED) coder on her first day of work (because all new coders are subject to 100 percent coding quality reviews). While reviewing the report, the coding quality coordinator noticed that most of the cases included an adverse effect diagnosis code listed as a secondary code (see Figure 1 below for an excerpt from the report). 

While it’s not uncommon for ED patients to be seen for adverse drug effects, the coding quality coordinator reviewed the electronic health records for those cases, and she found that the new ED Coder inappropriately reported drug allergy status as an adverse drug effect (e.g., “patient is allergic to penicillin” was incorrectly coded as a current episode of an allergic reaction to penicillin). This is an example of association data mining – the unusually high volume of adverse drug effect diagnosis codes was connected to the new coder’s incorrect understanding and coding of drug allergy status.

Figure 1.

CODER INITIALS

ACCOUNT

NUMBER

DISCHARGE

DATE

ICD-10-CM

DIAGNOSIS

CODE 

DIAGNOSIS CODE DESCRIPTION

 

CPT

CODE 

CPT CODE DESCRIPTION

MODIFIER(S)

NC

Xxx

Xxx

Z04.1

ENCOUNTER FOR EXAM AND OBS FOLLOWING TRANSPORT ACCIDENT

99283

EMERGENCY DEPT VISIT

 

 

 

 

M25.552

PAIN IN LEFT HIP

 

 

 

 

 

 

T36.0X5A

ADVERSE EFFECT OF PENICILLINS, INITIAL ENCOUNTER

 

 

 

 

Sequence
A coding manager and a senior outpatient coder worked with the denials management team to investigate the reason for the increased number of Medicare denials for outpatient colonoscopies. They generated a report listing all of the Medicare outpatient colonoscopy denials for the last three months (see Figure 2 below for an excerpt from the report). The senior outpatient coder reviewed the electronic health records for those cases, and she found that most of the cases were return visits, i.e., the patients’ prior encounter for a colonoscopy was unsuccessful, and had to be terminated/aborted (e.g., poor bowel preparation). 

Upon review of the code assignments for the terminated/aborted colonoscopies, the senior outpatient coder found that the modifier -74 (Discontinued Procedure Prior to Anesthesia Administration) was not appended next to the colonoscopy CPT/HCPCS code that was assigned and billed (i.e., the claims didn’t reflect terminated colonoscopies). This is an example of sequence data mining – the incorrect omission of the discontinued procedure modifier -74 during the unsuccessful colonoscopy encounter led to a denial of the return visit/successful colonoscopy encounter, because the Medicare Administrative Contractor (MAC) didn’t know that the patient’s prior/recent colonoscopy was aborted/discontinued.  It appeared that the patients were undergoing unnecessary repeat colonoscopies.

Figure 2.

CODER INITIALS

ACCOUNT

NUMBER

DISCHARGE

DATE

ICD-10-CM

DIAGNOSIS

CODE

DIAGNOSIS CODE DESCRIPTION

CPT

CODE

CPT CODE DESCRIPTION

MODIFIER(S)

KY

xxx

xxx

Z12.11

ENCOUNTER FOR SCREENING FOR MALIGNANT NEOPLASM OF COLON

45380

COLONOSCOPY AND BIOPSY

-PT

 

 

 

K51.00

ULCERATIVE (CHRONIC) PANCOLITIS WITHOUT COMPLICATIONS

 

 

 

EU

xxx

xxx

K64.8

OTHER HEMORRHOIDS

45398

COLONOSCOPY W/BAND LIGATION

 

 

Classification
The innovative medical and surgical services being provided at a multi-hospital health system has prompted the coding director to create an Inpatient Coder IV position (currently, there are Coder I, II, and III positions). Before she can submit a new job requisition to the chief financial officer (CFO), she must first identify/predict the most complicated inpatient medical and surgical cases that the Coder IV position would be responsible for coding. 

The coding director generated two reports, using the facility’s 2018 inpatient data: “Top 50 Most Common ICD-10-CM Codes for 2018 Inpatient Discharges” and “Top 50 Most Common ICD-10-PCS Codes for 2018 Inpatient Discharges.” She then met with her inpatient coding manager, coding supervisor, coding quality coordinators, and senior coders, and together, they used the two reports to establish the following inpatient coding complexity criteria:

  • Clinicians’ knowledge about the condition or the technology for performing the procedure is constantly evolving;
  • Coding the condition/procedure will increase the overall chart coding time by at least 10 minutes;
  • Coding the condition/procedure will require the coder to re-review more than two official coding guidelines, or there are no official coding guidelines for the condition/procedure;
  • There are several code options with similar descriptions to consider when coding the condition/procedure.

Using the inpatient coding complexity criteria, the coding director and her team were able to identify the most complicated inpatient medical and surgical cases that the Coder IV position would be responsible for coding, which included, for example: craniomaxillofacial surgery, fetal surgery, multi-organ transplant, necrotizing fasciitis (flesh-eating bacteria), open heart surgery, pediatric-congenital heart surgery, severe sepsis with septic shock, and third-degree burns. This entire process is an example of classification data mining, or discovering new patterns – i.e., the most complicated inpatient medical and surgical cases – after developing and applying the inpatient coding complexity criteria to the facility’s top 50 diagnosis and procedure codes for inpatient discharges.

Clustering
A coding supervisor received complaints that some inpatient coding specialists were selecting the less complex cases in the coding work queues (because the less complex cases are generally easier/faster to code, thus allowing a coding specialist to consistently meet daily coding productivity standards).

The coding supervisor investigated the complaints by first generating a report for all inpatient discharges coded in the last quarter, and the report contained the following data fields:

  • Coder Initials
  • Account Number
  • Admission Date
  • Discharge Date
  • Length of Stay
  • Hospital Service
  • On-Hold/Pending History*
  • All ICD-10-CM Diagnosis Codes and Descriptions
  • All ICD-10-PCS Procedure Codes and Descriptions
  • Payment Group Number Description
  • Payment Group Dollar Amount

*The accounts that had on-hold/pending history were briefly reviewed and then excluded from the analyses, because the day on which the case is put on hold (e.g., missing path report, physician query) and the day on which the case is completed are often two different calendar days, so it’s extremely difficult to include these cases when comparing daily coding productivity data.  This entire process is an example of clustering – identifying and grouping related facts, such as coding productivity data – that were not previously known.        

Forecasting
A coding quality coordinator has found that several outpatient surgery coders have been violating the American Medical Association’s coding guideline that prohibits the reporting of an adjacent tissue transfer (CPT code 14000 or 14001) when adjacent tissue transfer (e.g., advancement flap) is used to close a partial mastectomy surgical defect site. The AMA has stated that the partial mastectomy CPT code 19301 includes any form of adjacent tissue transfer closure of the surgical site.

The coding quality coordinator generated a report of all outpatient surgery cases with a discharge date of Nov. 1, 2018 and later in which both CPT code 19301 and either code 14000 or code 14001 were reported (see Figure 3 for an excerpt from the report). She surmised that a coding error likely existed on accounts that contained the 19301/14000 or 19301/14001 code pair. This is an example of forecasting, or identifying patterns in data – i.e., CPT codes 19301/14000 or 19301/14001 – that lead to reasonable predictions (in this instance, a coding error).

Figure 3.

CODER INITIALS

ACCOUNT

NUMBER

DISCHARGE

DATE

ICD-10-CM

DIAGNOSIS

CODE

DIAGNOSIS CODE DESCRIPTION

 

CPT

CODE

 

 

CPT CODE DESCRIPTION

 

MODIFIER(S)

TE

Xxx

xxx

C50.912

MALIGNANT NEOPLASM OF UNSPECIFIED SITE OF LEFT FEMALE BREAST

19301

PARTIAL MASTECTOMY

-LT

 

 

 

F17.210

NICOTINE DEPENDENCE, CIGARETTES, UNCOMPLICATED

14001

TIS TRNFR TRUNK 10.1-30SQCM

 

 

Whether the data pattern and relationship is one of association, sequence, classification, clustering, or forecasting, the goal of data mining is to obtain constructive information to be applied in answering questions, formulating conclusions, predicting outcomes, and supporting decision-making. 

Medical Coding Data Mining Articles
As seen in the forecasting data mining example above, data patterns and relationships may reveal inaccurate medical coding, which can adversely impact, for example, continuity of patient care, clinical research data quality, regulatory compliance, and the optimal management of revenue. We are launching this series of articles to show you how to use data mining techniques to identify potentially high-risk patterns in medical coding data. Our overall goal is to help you improve your organization’s medical coding accuracy and to achieve compliance with federal rules and regulations (including official coding guidelines). 

Some of the topics that we’ll discuss in this series include lesion excision coding, breast reconstruction revision coding, and rotator cuff tears and repairs coding. Our first topic will be lesion excision coding data mining, so stay tuned.

Programming Note:
Listen to Lolita Jones report this story live today during Talk Ten Tuesday, 10-10:30 a.m. EST.

Lolita M. Jones, MSHS, RHIA, CCS

Lolita M. Jones, MSHS, RHIA, CCS has provided Product Consultant services to a warehousing and analytics start-up that developed and marketed decision support software, health outcomes services, and regulatory compliance toolsets. Her goal is to combine her medical coding expertise with data mining-pattern recognition, to help improve data accuracy and compliance in medical coding and reimbursement (i.e., ICD-10-CM, ICD-10-PCS, CPT, HCPCS Level II, modifiers, DRGs, APCs, and eAPGs). Ms. Jones also provides remote and on-site training/consulting in her newly developed Healthcare Data Mining Clinic educational series. She is currently pursuing a Graduate Certificate in Healthcare Data Analytics from a top university. Ms. Jones is based in New York and can be reached at .

Related Stories

  • The 10-Year Coder Challenge: Growing Your Skills beyond CCs and MCCs
    Today’s coder must embrace APR-DRGs and HCCs in order to stay relevant. At the beginning of the year, Facebook’s “10-year challenge” hit the web, with users posting comparison profile pictures from 2009 alongside their 2019 selfies. I’ve since seen a…
  • Data Carries Great Importance for the ICD-11 Transition
    It is important to understand the data needs for your organization from a content perspective as you prepare for ICD-11. As we talk about preparing for the transition from ICD-10 to ICD-11, it is important to realize the importance of…
  • Implementation: ICD-10 Lessons Learned
    EDITOR’S NOTE: George Vancore appeared during a recent edition of Talk Ten Tuesdays focused his report on lessons learned during the implementation of ICD-10 in preparation for ICD-11. This article is a summary of his remarks. When Chuck Buck recently…