Audit of Automated Code Assignment in Israeli Hospital's Computerized Medical Information Systems

Rhona Gill and Rivka Hecht

Introduction

How accurate and complete is the coded diagnostic (and procedure) data collected on patients who are discharged from Israeli government-owned general hospitals that have implemented automated code assignment? Can the individual facilities and the Israel Ministry of Health rely on the quality of this coded data for public reporting and use? As a result of our initial audits of two hospitals’ coded data using the same automated coding program compared to that from trained professional hospital coders, we uncovered several areas of concern regarding the integrity of the data. This new coding environment challenged traditional roles of coders and physicians.

Background

Government-owned acute care hospitals in Israel have been gradually implementing computerized medical information systems (CMIS) during the last 10 years. The development of the CMIS began in the early 1990s with the computerization of the hospital admission, patient account, and financial department information, followed soon after by clinical departments (pathology, radiology, surgery, laboratory). The transition to the electronic health record (EHR) since 2000, integrates key data from these departments and results in the generation of a computerized patient discharge summary.   The physician is authorized to enter the final discharge diagnosis and procedure text in English on the basis of information contained in the body of the summary, which is documented in Hebrew, our native language. A software program, developed locally and used in two of these hospitals, assists the physicians when they enter the English text. As the physician begins to enter the first few letters of the desired diagnosis and/or procedure, this simple word-search engine program displays a built-in alphabetical list of all related diagnoses and procedures in ICD-9-CM text (modified for auto-assigned coding program), with corresponding ICD-9-CM codes. The physician then selects the most appropriate text, and the codes are automatically assigned and entered into the hospital’s database.

The rationale for developing this automated code assignment program was to perform two functions simultaneously: (1) physician documentation of the final list of diagnoses/procedures in the discharge summary, with (2) the assignment of ICD-9-CM codes without the need for a professional "human" coder. The corresponding ICD-9-CM code for the selected diagnosis/procedure is "automatically" entered into the hospital database from the program’s code table. Only the responsible physician is authorized to revise, as needed, the documented diagnosis and/or procedure and corresponding codes. At defined periods, the file with coded diagnoses and procedure information is transferred from the hospital to the Ministry of Health’s data warehouse and the national cancer registry.

Assuring the integrity and accuracy of this coded discharge data is one of the principal missions of the Ministry of Health (MOH). The MOH must be able to depend on reliable and timely information for assessing the quality and quantity of healthcare delivered to its citizens. This data is used for determining reimbursement for certain hospitalizations (our DRG groups are based on ICD-9-CM procedure codes), health planning and policy making, research, epidemiological studies, and for statistical analysis of the current health status of the Israeli population.   In addition, the discharge summary, including the list of final diagnoses and procedures, is routinely given to the patient upon discharge (as mandated by MOH regulations) and is used as an important source of information for continuing care.

Does the program allow for the most efficient method of documenting the final diagnoses and procedures into the electronic record and transference of this information into coded data? Is the computerized medical information system in these Israeli hospitals capturing accurate and complete data?  

In order for the coded data to be functional for national and international purposes, this automated coding program and its code tables must adhere to the same official coding guidelines and rules, definitions and standards. We performed two types of audits--on-site and trend analysis--and were surprised with our results. We also discovered that the roles of hospital coders and our physicians regarding documentation and coding, and the working relationship between these two groups have changed.

On-Site Audit: Hospital A

Selected teams of hospital coding experts visited "Hospital A"--one of the two government acute care hospitals using the automated code assignment program and performed an on-site audit. The objectives of this audit included:

  • Evaluate codes "automatically" assigned compared to those of the hospital coders and on-site coding auditors. The assignment of the principal diagnosis and procedure, and selection of relevant hospitalization diagnoses and procedures to be coded were also evaluated.
  • Evaluate the accuracy of the diagnosis/procedure code tables and extent of built-in edits contained in the automated coding program.
  • Assess the changing roles and relationships of hospital coders, physicians, and computer information staff.

Hospital A’s Current Coding Process  

The automated code assignment program was initiated in 2001. Since the implementation of the automated coding process at this hospital, the coding staff has continued to recode 100 percent of the discharge records. This 100 percent recoding effort began as a temporary measure during the transition period between the hospital coder’s manual coding process to the automated code assignment process to ensure the completeness and accuracy of coded data.   However, after only a short time, the hospital coding staff and the HIM department director discovered many inaccuracies and problems with the automated code assignment program. Consequently, the hospital has been maintaining two separate databases with discharge diagnostic and procedural data--one generated from the automated code assignment program and the other with the data from the professional hospital coders from the HIM department. An administrative decision was made to continue transferring only the hospital coder’s database to the Ministry of Health’s national repository of discharge information for the foreseeable future.

Site Visit Survey 

Our survey team reviewed discharged patient records and analyzed the discrepancies between the documented final diagnoses/procedures and corresponding coded data as generated from the automated assignment program compared to that from the auditors.   The survey team spent one day reviewing a random sample of 25 complete discharged patient records newly arrived in the HIM department for processing; 16 from General Medicine departments, 8 from the Orthopedic department, and one from General Surgery (Table 1).

Table 1. Survey Summary

Hospital
Depart-ment

Total # Records Reviewed

Total # Dx./Proc. Selected
by MD

Avg. # Dx/Proc. per Discharge

Total of Accurate and Complete Dx/Proc. Code Assignment

Ratio of Accurate Codes / Total Codes Selected by MD

General Medicine

16

131

8

43

4/10

Orthopedic

8

48

6

11

3/10

General Surgery

1

13

13

5

4/10

The survey team found two basic categories of errors with the automated code assignment program:

  1. Problems with the code tables (see examples in Table 2).
  2. Problems with diagnosis/procedure text selected by doctor (see examples in Table 3).

Table 2. Examples of Problems with Code Tables

Description of Code Table Problem

Example

Incorrect Code Assigned for Designated Dx/Procedure Text

Code assigned for Other Heart Block Complete AV is 426.6; should be 426.0

Missing Code Edits and/or Use of Unfamiliar Abbreviation in Text Leads to Incorrect Selection of Dx/Proc. and Wrong Code Assignment

89-year-old patient was assigned Dx. NB Drug Reaction/Intox , coded 779.4
Diagnosis of Essential Hypertension Preg-Unsp given to an 82 year old patient

Additional Codes are Missing (code for organism with infection diagnosis)

Only code assigned for listed diagnosis Bacteremia NOS due to Staph. Aureus is 790.7. The code for staphylococcal aureus, code 041.11 is missing

Table 3. Examples of Problems with Physician Selection of Computerized Final Diagnoses and Procedures

Description of Physician Selection Problem

Example

Physician Selected Text for Dx./Proc. Inconsistent with Documentation in Record

Final Dx. of Breast Neoplasm NOS Rt ., code 239.3 given to patient who was hospitalized for secondary malignancy of liver with metastasis to brain and bones - primary site breast). Correct Dx. and corresponding code should be 174.9, Malignant Neoplasm of Breast

Missing Code - Procedure Not Listed Separately on Discharge Summary’s Final List of Procedures

Physician documented insertion of pacemaker as part of diagnosis text - Other Heart Block Complete AV Block, Insertion of Temporary Pacemaker.
Consequently, the pacemaker insertion was NOT assigned a separate procedure code. The discharge summary text documentation stated that a Single-chamber, Rate-responsive device was inserted, requiring code 37.82

Incorrect Diagnosis Assigned as Principal Diagnosis

Patient admitted with fever and diagnosed with upper respiratory tract infection during hospitalization is assigned final diagnosis/code for Fever.   (The URI was listed as subtext, and subsequently not coded).

Physician Incorrectly Selected General Description of Diagnosis Leading to Incorrect Code

Physician selected Ascites with assigned code 789.5, instead of Malignant Ascites, coded 197.6. The physician entered "malignant" as subtext under diagnosis of "ascites."

Trend Analysis of Reported Data

Hospital B’s Current Coding Process

This hospital initiated the automated code assignment process, the same simple word-search software program in 2000. As opposed to the HIM department in Hospital A, where they continue to recode 100 percent of the discharges and create their separate ICD-9-CM code database, this hospital decided to maintain only one database containing only the codes provided by the automated code assignment program. Consequently, this is the database from which the diagnostic and procedure codes are abstracted and transferred to the Ministry of Health. The HIM department’s coding staff’s responsibilities changed dramatically. They are now expected to audit (recode) only a random sample of 20-30 percent of discharges from each hospital department for auditing and educational purposes only. Any discrepancies found as a result of their recoding efforts were entered into a specially designed code audit software program for tracking purposes. Individual physicians are queried when coders recommend any revision to the diagnosis/procedure text and corresponding code(s) on the discharge summary.   There is no follow-up to verify that a corrected version of the discharge summary was created and whether the codes in the database were updated. The focus of the professional coder’s new role moved from assigning codes and managing the coding process to assisting physicians improve their general documentation practices and to select the most complete list of the patient’s final diagnoses and procedures.  

Trend Analysis Review Methodology

Tabulated reports of all hospital diagnosis ICD-9-CM codes submitted to the Ministry of Health for the same month-period during the years 1998-2001 from the Orthopedics and OB-GYN departments were generated and reviewed. These reports included the diagnoses as reported as "principal/main only" and also as "either-principal/main or other." The HIM department’s professional coding staff produced the codes from years 1998 and 1999. During years 2000 and 2001 codes were generated from the automated code assignment program according to the diagnosis and procedure text chosen by the physician upon completion of the discharge summary record.

In analyzing the code reports, we looked for significant changes in the trend (numbers) of similar diagnoses reported between the years, especially between years 1998-1999 (professional coder generated) and 2000-2001 (automated code assignment generated). The following are some of the highlights from some our review (see Figures 1 and 2).

Figure 1. Malunion and Nonunion of Fractures (ICD-9-CM codes 733.8, 733.81, 733.82)

Figure 2. Menorrhagia and Postmenopausal Bleeding (ICD-9-CM codes 627.1 and 626.2)

^Professional code
# Automated code

Note: Further investigation is needed to determine why there was a decrease of at least half the reported number these above diagnoses by physicians during the years 2000 and 2001.

 

Impact of Audit Findings

Loss of Data that Can Never Be Recovered

Diagnoses and procedures that were not selected/reported and corresponding codes for 2000-2001 will never be recovered by Hospital B. This impacts both the local hospital database as well as the MOH’s Discharge Data Warehouse, including the national registries (cancer registry). For example:

  • Cases of patients with malignant neoplasm may not have been referred to the cancer registry.
  • Statistical analysis of cases by diagnoses and/or procedures for national and international reports may have been biased.
  • Researchers (students, physicians, epidemiologists) who depend upon the diagnosis/procedure databank of ICD-9-CM codes loose the ability to access cases.

Accuracy/Integrity of Reported Data

Without certainty that the code tables are accurate and follow national guidelines, the ability to perform reliable and consistent data comparisons nationally and internationally is problematic.

Discharge Summary Released to Patient May Contain List of Final Diagnoses /Procedures That Do Not Correlate to Documentation

Discharge summaries with inaccurate, incomplete, and missing final diagnoses or procedures selected by physicians were released to patients upon discharge without review. This may lead to problems with follow-up care and reimbursement. For example:

  • Physician selected an obstetrical diagnosis for an 80-year-old patient due to the lack of understanding of the abbreviation "OB" in the text of the chosen diagnosis. A built-in code edit for age-by-diagnosis is missing.
  • Omission of chronic conditions on final list of diagnoses, even though treatment for these conditions were documented within text of summary.

Ability of System to Be Independent of "Human" Coders Not Yet Realized

To assure accurate documentation and reliable, qualified databases with the current automated code assignment program, 100 percent of patient discharge summaries need to be checked and re-coded by human coders.   Concurrent, on-site cooperation between the physicians and trained coders prior to patient discharge can significantly improve the quality of the automated coding data.

Changing Roles of the "Human" Coder   

As the patient record moves from a paper to paperless version, electronic documentation and the use of automated code assignment programs pose new challenges to the traditional roles of the human coder. We see coders’ roles in their computerized environment changing in the ways outlined below.

Re-engineering the Coder’s Working Relationship

The professional coder must work with Computer Information department personnel to develop and maintain a mutually supportive relationship in the design and establishment of all diagnostic and procedure code files, including:

  • Text descriptions
  • Integration of rules, guidelines, code table edits
  • Official updates
  • Data mappings between different code sets
  • Investigation of future natural language processing (NLP) technology possibilities

Coders need to assist physicians with the selection of complete final diagnoses/procedure statements and automated code assignment on-site prior to patient discharge. Health information management professionals and the coding staff need to be involved in the orientation and continuous training of staff physicians in documentation and coding requirements.

Empowerment of Auditing

As we move from assigning codes to the automated code assignment programs, the human coder performs advanced and expertise audits. This extended auditing function is dynamically correlated to previous audit results, orientation of staff, turnover of physicians, implementation of new automated coding programs, and updates/changes to code sets.

Conclusion

The potential of automated code assignment programs will be realized as an integral part of the electronic patient record. However, as we saw with the one simple word-search program that was integrated into two of our hospital’s electronic health record systems:

  • Working with the implementation team to design the databases, formats, definitions, and procedures is essential to ensure accurate capture of data. Cooperation must continue on an on-going basis to make necessary changes and updates to the code files and programs.
  • All automated code programs, related code files, and mappings that are to be integrated into electronic medical record systems will need to be reviewed and approved by the Ministry of Health by a formal standardized process.
  • "Human" coders will need to acquire and continuously enrich skills in clinical system analysis, database administration, database reporting analyst, data quality analysis/management, team building, and training/teaching.
  • Coding experts will need to develop the characteristics of perseverance, adaptability, and patience.

References

(1) Kloss, Linda. "Data Quality: Part of the e-HIM Agenda." Journal of AHIMA, no. 10 (November 2003): 19.   

(2) Olsen, Brenda and Patricia Thierry. "HIM Professionals in the Information Systems Trenches." 2000 AHIMA National Convention Proceedings, September 23-28, 2000, pages 265-274.

(3) Johns, Merida. "A Crystal Ball for Coding." Journal of AHIMA 71, no.1 (2000): 26-33.

(4) Schnitzer, Gregory L. "Natural Language Processing: A Coding Professional’s Perspective." Journal of AHIMA 71, no.9 (2000): 95-98.

(5) Schnitzer, Gregory and Mary Stanfill. "Coding Notes: Outwit, Outlast, Outcode: Surviving in the Autocoding Era." Journal of AHIMA 72, no.9 (2001): 102-104.

(6) Warner Jr., Homer. "Can Natural Language Processing Aid Outpatient Coders?"   Journal of AHIMA 71, no.8 (2000): 78-81.


Source: 2004 IFHRO Congress & AHIMA Convention Proceedings, October 2004