Importance of Coding in Hospital Benchmarking and Performance Evaluation

Clara Dismuke, PhD

Introduction

The Diagnosis Related Group (DRG) system was created in the US and has spread widely around the world. Currently, approximately 36 countries use some version of the system for either information collection, financing, or reimbursement for inpatient care.¹ This widespread adoption of DRG-type systems should facilitate international comparisons of inpatient medical practices and outcomes by either DRG, diagnosis, or procedure code in countries using the same coding sources.

DRG assignment is determined by the diagnosis and procedure codes reported in the inpatient claim. Coding of procedures is required when the presence of the procedure (usually surgical) affects DRG assignment. When a procedure is not required for DRG assignment, coding of the procedure may still be done for reasons unrelated to DRG assignment such as differences in healthcare system or hospital policies, or medical record coder training. Presently, at least seven countries use ICD-9-CM for coding procedures.² Researchers using inpatient data generated from DRG-based systems may not realize that variations in hospital coding practices between countries can cause bias in statistical results when they rely on one or more of the 1000 ICD-9-CM procedure codes that are not required for DRG assignment. We show the statistical implications of underreporting of ICD-9-CM procedure codes in one country (US) relative to another (Portugal) for analyzing differences in resource utilization (Computerized Tomography [CT]) for elderly patients hospitalized with cerebrovascular disorders.

Background

A limited number of studies have compared inpatient resource utilization between at least two countries. Klarenbach and Jacobs³ compared, among other things, hospitalization of US and Canadian residents with diabetes. Nashef et al.⁴ compared the national samples of patients who underwent isolated coronary artery bypass grafting (CABG) during the European System for Cardiac Operative Risk Evaluation (EuroSCORE) trial in order to evaluate national differences in surgical methods. Rapoport et al.⁵ compared the likelihood of intensive care unit (ICU) admissions between residents of Massachusetts, US and Alberta, Canada. Cherkin et al.⁶ compared rates of back surgery in 11 countries. Though there are a number of studies that have used ICD-9-CM procedure codes to compare inpatient utilization in the US,⁷ we could not find any published studies of international comparisons of procedures by ICD-9-CM codes.

The International Classification of Diseases, Ninth Revision, Clinical Modification (ICD- 9-CM), is a clinical system that categorizes related diseases and procedures for statistics reporting. The National Center for Health Statistics (NCHS) developed clinical modification of ICD-9 (the classification system used worldwide) for use in the US. ICD-9-CM serves many purposes including classification of morbidity and mortality information for statistics, diagnosis reporting by physicians, and diagnosis and procedure reporting in inpatient claims for the purposes of DRG assignment.⁸

The US government payer, Medicare, began reimbursing hospital inpatient stays for its beneficiaries on the basis of DRGs in 1983. Since then, many Medicaid programs and an increasing number of private third-party payers, have adopted DRGs as their method of inpatient reimbursement. Consequently, medical record coders are generally trained to code procedures in inpatient claims for reimbursement purposes and thus DRG assignment. DRG assignment begins with a classification of the patient's stay to one of 25 Major Diagnostic Categories (MDC). This is followed by a second classification, based either on medical diagnoses or procedures that are normally performed in the operating room (OR) in addition to patient characteristics such as age, gender, and disposition on discharge.

When assignment is based solely on principal diagnoses, coding of procedures performed during the hospital stay is not required. Hence, if medical record coders report strictly for reimbursement purposes, most procedures performed in settings other than the OR may not be coded in patient level claims and will not be found in national and state inpatient databases. For these reasons, non-DRG procedures are generally defined as non-OR procedures that are not required for assignment of DRGs. Since there currently are approximately 1599 non-OR procedures of which only about 116 determine DRG assignment, more than one-third of the approximately 4000 ICD-9-CM existing procedures are non-DRG procedures.⁹

To quantify underreporting of non-DRG procedures in ICD-9-CM in the US, an alternate measure is required for comparison. US hospitals have a parallel system of information known as Universal Billing Revenue Codes that co-exist with ICD-9-CM diagnosis and procedure codes on the Universal Billing (UB-92) Form (see Figure 1).

Presently, the UB-92 is used to submit more than 98 percent of hospital claims to Medicare and 80 percent of hospital claims to all payers.¹⁰ In addition to revenue codes, corresponding units of service for groups of procedures and a small number of individual procedures are present on the form. This facet is especially important for attaining a more accurate estimate of resource use because ICD-9-CM codes are normally coded only once in the claim and thus do not indicate the quantity used. Moreover, revenue codes are automatically generated from hospital accounting chargemasters and are thus not subject to the discretion of medical record coders. For this reason, revenue codes are more likely to reflect actual resource utilization of non-DRG procedures. Nevertheless, revenue codes do have a major disadvantage since there are very few codes identifying specific procedures. Consequently, each individual claim potentially includes very few procedures that can be measured by both ICD-9-CM and revenue codes.

Portugal was the first country with a national health system (NHS) to incorporate DRG case-mix adjustment in formulating hospital budgets on a nationwide basis.¹¹ Portugal's public hospitals have been collecting what is known as the European minimum basic data set (MBDS) since 1989 for each inpatient discharge. The MBDS includes the following: hospital identification, patient's number, gender, age, marital status, place of residence, month and year of admission, duration of stay, discharge status, main diagnosis, other diagnoses, surgical and obstetric procedures, and other significant procedures.¹²

Coder education and background is very different between the US and Portugal. The majority of US coders do not possess a clinical background, and many do not have a four-year university degree. Portuguese medical record coders are licensed physicians who often code in the specialty for which they were trained. The effect of both coding purpose and coder training probably results in a more clinically oriented coding in Portugal. Also, since hospitals in Portugal have no alternative information system to account for resource use, codes are more likely to record procedures that are not required for DRG assignment.

Data and Methods

We began by comparing the frequency of ICD-9-CM reporting of CT in all discharges classified to DRG 14-Cerebrovascular Disorders Except TIA for patients aged 65 and older in Portugal and Washington state, US, during 1997. The Washington State Inpatient Database (SID) is one of 29 state databases composing the Health Care Cost and Utilization Project (HCUP). HCUP is sponsored and disseminated by AHRQ and contains the largest all-payer database in the US, the National Inpatient Sample (NIS). The NIS contains inpatient claims for hospitals in approximately 29 states, representing about 20 percent of all inpatient stays in the US, and it consists of more than 100 clinical and nonclinical variables for each claim (for example, primary and secondary ICD-9-CM diagnosis and procedure codes, admission and discharge status, patient demographics, expected payer, total charges, and length of stay). It also contains a hospital identification code that permits linkage to the American Hospital Association (AHA) Annual Survey database, which includes a wealth of information on hospital characteristics. However, the NIS does not contain revenue codes, possibly due to the heterogeneity of individual states' reporting of these codes. Washington is one of only four states reporting revenue codes in a format that allows comparison of ICD-9-CM and revenue codes for the CT. We limit our data to one DRG, Cerebrovascular Disorders Except TIA, in order to limit potential differences in the likelihood that a patient receives a CT due to illness heterogeneity. We also limit the age group to those 65 and older to limit potential differences due to payer. Since US residents age 65 and older are Medicare eligible, we are comparing two groups covered by a government payer.

To evaluate the hypothesis that ICD-9-CM codes are underreported for non-DRG procedures, all claims with an ICD-9-CM or revenue code in either the primary procedure or one of the six secondary procedure fields for each of these technologies were extracted. CT use was identified by ICD-9-CM codes 87.03, 87.41, 87.71, 88.01, and 88.38 and by revenue codes 350, 351, 352, and 359 in the Washington data. CT as identified in the Portuguese data uses the same ICD-9-CM codes. Frequencies of CT coding in ICD-9-CM and revenue codes were calculated as were means of patient and hospital characteristics by whether or not a code for CT was found in the claim.

We then combined the US and Portuguese data ensuring that information was available on all similar variables and estimated the following utilization equation for CT using first ICD-9-CM and then, revenue codes for the US:

P(CT=1)=f (age, number of diagnoses, male, bedsize, teaching hospital, CT on site, located in Portugal).

Results

The frequency of CT use in ICD-9-CM and revenue codes for Washington and ICD-9-CM for Portugal as well as means by patient and hospital characteristics are shown in Table 1. The magnitude of underreporting in the US data is evident with only 33 percent of elderly patients with cerebrovascular disorder receiving a CT according to ICD-9-CM, in contrast with 84 percent in revenue codes. While there are not large differences in patient characteristics by coding source, there is an important difference by bed size and teaching status with average bed size declining from 296 to 226 in those receiving a CT as we move from ICD-9-CM to the revenue code measure. Also, while it appears that 39 percent of individuals receiving CT were in teaching hospitals in ICD-9-CM, only 27 percent were in teaching hospitals according to the revenue code. This is likely to reflect greater coding of CT in ICD-9-CM in larger teaching hospitals in the US for reasons not related to reimbursement.

Table 1. Frequency of CT Use

Table 2. Results from CT Equation

The results of the utilization equation for CT are shown in Table 2. We find that the coefficients on the patient characteristics are relatively similar in both the ICD-9-CM and revenue code equations. Though there are some differences in the magnitudes of coefficients, we also find the coefficients have the same sign for hospital characteristics. However, the coefficient on the variable reflecting location in Portugal, not only changes magnitude but also reverses sign from positive in ICD-9-CM to negative in revenue codes; thus, based on ICD-9-CM codes, we would conclude that an individual has a higher likelihood of having a CT in Portugal than in Washington. However, if we use US revenue codes, we find the opposite. We have thus shown the importance of understanding the reasons for coding differences between countries for use of inpatient data.

Endnotes

1. 2003 Country Update. Patient Classification Systems Europe: Capitalizing on Case Mix: Gaining a Global Perspective. October 2003.

2. Kulinshaya E. “International Case-mix Research: Why and How.” Patient Classification Systems Europe: Capitalizing on Case Mix: Gaining a Global Perspective. October 2003.

3. Klarenbach SW and Jacobs P. International comparison of health resource utilization in subjects with diabetes: an analysis of Canadian and American national health survey. Diabetes Care. April 2003.

4. Nashef SA, Roques F, Michel P, Cortina J, Faichney A, Gams E, Harjula A, Jones MT. “Coronary surgery in Europe: comparison of the national subsets of the European system for cardiac operative risk evaluation database.” European Journal of Cardiothoracic Surgery 2000: 17(4).

5. Rapoport J, Teres D, Barnett R, et al. “A comparison of intensive care unit utilization in Alberta and western Massachusetts.” Critical Care Medicine 1995; 23:1336-1346.

6. Cherkin DC, Deyo RA, Loeser JD, Bush T, Waddell G. “An international comparison of back surgery rates.” Spine 1994; 19:1201-1206.

7. A review of these studies is contained in Dismuke CE. The Implications of Procedure Code Underreporting in Inpatient Claims Data for Health Care Utilization Studies. Center for Health Economic and Policy Studies, Medical University of South Carolina Working paper series. http://www.musc.edu/cheps/.

8. More information on the ICD-9-CM system can be found on the Web site for the National Center for Health Statistics: http://www.cdc.gov/nchs.

9. US Department of Health and Human Services. Generic ICD-9-CM 2002, Volume 1,2,3. Reno, Nevada: Channel Publishing, Ltd. (2002).

10. Information on the NUBC can be obtained on the site’s Web page: http://www.nubc.org.

11. Bentes M, Mateus MC, Goncalves ML. “DRGs in Portugal: a decade of experience.” The Eighth Case Mix Conference, Case Mix and Change: International Perspectives. September 1999.

12. Fetter RB, Brand D, Gamache D. eds. DRGs: Their Design and Development. 1991. Ann Arbor, Michigan: Health Administration Press, p. 254.