Self-reported awareness and use of the International Classification of Diseases coding of inflammatory bowel disease services by Ontario physicians

1Department of Medicine, Division of Gastroenterology, 2McQ Research Services and 3Intestinal Disease Research Programme, McMaster University, Hamilton, Ontario Correspondence: Dr E Jan Irvine, Room 4W8  Health Sciences Centre, McMaster University, 1200 Main Street West, Hamilton, Ontario L8N 3Z5. Telephone 905-521-2100 ext 76402, fax 905-521-4958, e-mail irvinej@mcmaster.ca Received for publication January 22, 2002. Accepted June 10, 2002 F Farrokhyar, K McHugh, EJ Irvine. Self-reported awareness and use of the International Classification of Diseases coding of inflammatory bowel disease services by Ontario physicians. Can J Gastroenterol 2002;16(8):519-526.

A s the global population ages and health expenditures rise, it is critical that estimates of the burden of certain diseases be optimized to improve the allocation of scarce resources.The cost of undertaking such studies can be substantially reduced by linking administrative databases.Health insurance databases, hospitalization files, physician billing files and prescription utilization databases are among those that can be analyzed.Countries that have been most successful in health services and epidemiological research have followed this path, with the diagnosis coded using the International Classification of Diseases (ICD) (1) as a critical field.This coding system is now in its 10th revision (ICD-10) (2), although most countries still use the ninth revision (ICD-9) system.Until now, very few countries have considered inflammatory bowel disease (IBD) to be an important condition.Nonetheless, the young age at onset and its considerable impact on health status, with the potential to affect the earning capacity and socioeconomic status of patients and their families, suggest that it imposes a significant health burden.Some countries such as Sweden (3) and Denmark (4) have developed record linkage systems that capture information on all diseases, including IBD.They already consider IBD to be a public health problem and have developed registries to identify all cases prospectively.This approach, however, is more difficult to assume in countries, such as Canada, with a large geographic expanse and heterogeneous health care systems.Bernstein et al (5) have used computerized linkage of existing data from administrative databases to identify patients and estimate the incidence and prevalence rates of IBD in Manitoba.Indeed, they suggest that these prevalence rates are among the highest reported in the world for IBD -at almost 400 per 100,000 person-years.
The validity of conclusions derived from linkage analysis research depends greatly on the accuracy of the data in these databases.Few studies have been undertaken to verify the accuracy of ICD-9 coding in administrative databases, but those that have been done suggest that there is 74% to 81% agreement between the database data and review of the patients' chart for the most responsible diagnosis (6)(7)(8)(9)(10).
No study has previously evaluated the awareness and accuracy of use by physicians of the ICD-9 coding of services provided to patients with IBD.In the present study, we surveyed Ontario physicians to assess both their familiarity with and current application of ICD-9 coding when billing for Crohn's disease (CD)-or ulcerative colitis (UC)-related services.

METHODS
All gastroenterologists and 10% of all internists, pediatricians, and pediatric or general surgeons were selected randomly from the College of Physicians and Surgeons of Ontario database.Ten per cent of family physicians were also selected from the College of Family Physicians of Ontario database.Each physician was mailed a survey questionnaire (Appendix 1) to evaluate whether they saw IBD patients, their knowledge of ICD-9 codes and self-reported accuracy of using ICD-9 codes 555 and 556 when billing the Ontario Health Insurance Plan for IBD or unrelated services.The ICD-9 codes for CD or UC are 555 and 556, respectively, and the frequency of use and 95% confidence intervals were calculated.χ 2 tests were performed for comparisons between physician groups.P<0.05 was considered to be statistically significant for all tests.Data were analyzed using SPSS.pcstatistical software (SPSS Inc, USA).

Survey sample
Table 1 shows the number of physicians sampled from each group, and the respective proportions who responded and were in practice at the time.Of the 614 physicians surveyed, 416 (67.7%) completed the questionnaire.However,
13 had moved out of the province and 12 were not in practice.The median year of graduation of respondents was 1981 (range 1951 to 1999), and 292 (72.5%) were male.A significantly greater proportion of female physicians graduated after 1990 (26.4%) than between 1951 and 1970 (9.4%) (P<0.001).

IBD patient exposure
Two hundred fifty-eight respondents (66%, 95% CI 61.3 to 71.7) saw IBD patients and 48% (95% CI 42.0 to 54.0) of those had a solo practice.Table 1 shows the proportion in each specialty who saw IBD patients.Family physicians and surgeons were the nongastroenterologist (non-GI) physicians who reported the highest exposure to IBD patients.Table 2 illustrates the practice patterns of the Ontario physicians who were surveyed, demonstrating that 100% of gastroenterologists, significantly more than 57% (51.8 to 62.8) of the non-GI physicians, saw IBD patients (P=0.001).Of the gastroenterologists, 94.9% (95% CI 90.1 to 99.7) claimed to have 10 or more patients with IBD in their practices.

Familiarity with the ICD-9 code
The majority of participants (81.0%, 95% CI 76.3 to 85.7) who saw IBD patients had a secretary or external billing service submit their claims.There were no significant dif-ferences in the rates of awareness (93.9% versus 84.7%, P=0.09) or use (81.6% versus 89.1%, P=0.16) of ICD-9 codes 555 or 556 between physicians who submitted their bills themselves and those who did not.Of the 258 physicians who saw IBD patients, 86.5% (95% CI 82.4% to 90.6%) were familiar with the ICD-9 coding system and 91.4% (95% CI 88.1% to 95.6%) of those reported using ICD-9 IBD codes 555 and 556.There was no significant difference in familiarity by sex (87.4% men versus 83.3% women, P=0.41).Gastroenterologists reported the highest rate of familiarity with the ICD-9 codes 555 and 556 at 91.1% (Table 2).

ICD-9 IBD code use and self-reported accuracy
Gastroenterologists also reported usage of 555 and 556 (Table 2) significantly more than did non-GI physicians (95.8% versus 83.9%, P=0.01). Figure 1 shows the frequency of ICD-9 555 and/or 556 use for IBD services by medical specialty.Physicians who had graduated after 1980 (93%) also used the IBD ICD-9 codes significantly more than those who had graduated before 1980 (P=0.02) (Figure 2).Table 3 illustrates the frequency with which physicians reported using ICD-9 codes 555 and 556 for billing IBD-related services.The majority of physicians stated that they had used ICD-9 codes always or most of the time when billing for IBD-related services.Few physicians  (10%) reported using these codes to bill for non-IBDrelated services for their IBD patients.Gastroenterologists were significantly more likely to use the IBD codes for IBD or non-IBD services than non-GI physicians.There were no significant differences in the reported use of codes between men (92.6%) and women (96.4%) (P=0.32).

DISCUSSION
This survey is the first to assess physician self-reported ICD-9 coding for IBD services.Over 85% of Ontario physicians surveyed were familiar with ICD-9 codes, and 91.4% of those reported using ICD-9 codes 555 or 556 for billing when they saw IBD patients.Recent graduates were more likely to use these codes.Not surprisingly, gastroenterologists (91%) were most familiar with the use of ICD-9 555 and 556 codes than other specialists (84%).The selfreported accuracy with which all of the physicians surveyed used 555 or 556 always or most of the time when billing for IBD-related services, irrespective of specialty, was over 75%, while only a few (10%) used these codes to bill for non-IBD-related problems in patients with IBD.The majority of the physicians (81%) who saw IBD patients did not bill for the services themselves, but rather secretaries (64.3%) or an agency (16.7%) performed these tasks.However, there was no significant difference in the awareness of or use of ICD-9 codes between these two groups.Such accuracy would be reasonable for undertaking some epidemiological studies, such as estimating the burden of disease, but might be less acceptable for more specific research questions, such as estimating the resources needed for IBD cancer surveillance.Irrespective of the research question, a key data field for linkage analysis is the disease diagnosis, which in Canada is generally coded using the ICD-9 classification (6).
Bernstein et al (5), in an epidemiological study, examined the Manitoba health care databases to develop valid case definitions for IBD.The accuracy of diagnosis from health care data was compared with two reference measures -a self-report postal questionnaire sent to the IBD patients identified in the databases and a standardized chart review.The database accuracy increased with the consistency of coding, and the number of health contacts, being greatest (approximately 90%) if subjects had been in the system for two or more years and had five or more of the last nine con-   In Canadian studies (7)(8)(9)(10)(11) that have used provincial databases for assessing the accuracy of diagnosis for a number of medical conditions, the accuracy for the most responsible diagnosis has ranged from 74% to 98%.Discrepancies were attributed to the vagueness of the diagnosis, hospital policy and incomplete diagnostic documentation at the time of coding.Hospital service abstracts were more likely to be accurate than physician claims data because of the detailed documentation required in a hospital compared with an office setting, and because the recording hospital physicians were usually specialists (11).Studies from the United States (12,13) have reported similar diagnostic accuracy from hospital abstracts (76% to 90%) and suggested that further improvements could be achieved by increasing the number of diagnoses abstracted and carefully selecting diagnostic codes of interest.

TABLE 3 Frequency (always or most of the time) of use of the International Classification of Diseases, Ninth Revision (ICD-9) codes for billing inflammatory bowel diseaserelated services
These studies have also illustrated some potential sources of error when using administrative databases.A major error is the omission of subjects who have not entered the health care system or who have been well and had no visits during the period of interest.A second source of errors is physician or diagnostic test error.In Canadian databases, the accuracy of diagnosis on the medical record appeared to vary with the number of criteria for a specific disease (6).For conditions such as stroke or rheumatoid arthritis, when the diagnosis can be somewhat uncertain, the agreement was 35% to 80% and 45%, respectively.For other conditions, for which diagnostic criteria are more explicit or based on objective evidence, the accuracy rates increased accordingly, to 78% to 94% for myocardial infarction and 93% for diabetes.A third source of error in the diagnosis of hospitalized patients occurs when medical records abstracters identify a secondary diagnosis as the primary diagnosis in patients with multiple disorders or omit a diagnosis if there is a limited number of diagnoses that can be recorded.This could lead to an underestimation of less frequent disorders, particularly in complex cases.Fourth, transcription errors may occur while entering diagnostic codes into the database.In the present survey, secretaries or agencies performed 81% of the billing entries and submissions to the Ontario Ministry of Health.Limited training or supervision of billing staff, substitution of missing codes or inappropriate use of nonspecific codes by the physicians could also lead to undercoding of some diagnoses.Faciszewski et al (14) suggested that other nonrandom sources of error, such as a reimbursement incentives, might cause some billers to favour particular codes or grade certain disorders as more severe.
Our study has several limitations.First, only physicians were surveyed, and results were not compared with those of billing staff, patient or chart data.Second, a restricted group of physicians was assessed, based on pilot data (unpublished), identifying groups that most frequently billed for IBD-related services (internists 39%, family physicians 28.5%, gastroenterologists 15%, general surgeons 13%, others 2.5% and pediatricians 2%).Indeed, physicians who see very few IBD patients are more likely to bill inconsistently than those who see many IBD patients.Third, this was a self-reported survey and, therefore, subject to possible recall or tendency of agreement biases, which could result in an overestimation of the accuracy of use.Nevertheless, our rates of 77% accuracy lie within the range of those of most studies that have reported 50% to 95% accuracy.The aim of our survey was to determine the feasibility of undertaking linkage studies in IBD by assessing sources of error in Ontario administrative databases, which involve a large population (over 12 million subjects) and a large geographic area.We recommend that investigators undertaking linkage research validate the accuracy of their case ascertainment methods during the study.This would require the examination of a random sample of the population of interest (15) and of a sample of patients who might be easily confused with the study population.In the context of IBD, patients with 'infectious colitis' or irritable bowel syndrome could represent 'negative controls', permitting the evaluation of the sensitivity and specificity of case ascertainment.There are also problems of disease reclassification, which occur when a patient's diagnosis is changed from CD to UC or vice versa.This should also be taken into consideration in the case validation process (16).In the present study, physicians were not asked whether they consistently used the same ICD-9 codes over time.This would require a longitudinal assessment of administrative or prospectively collected data.

CONCLUSIONS
The use of data linkage research in IBD in Ontario is feasible.The accuracy of IBD diagnostic coding seems to be comparable with that observed for other important diagnoses.However, it is crucial that the case ascertainment methods be validated in the study population and that accuracy estimates be carefully examined in the context of the research question.

Figure 1 )Figure 2 )
Figure 1) Frequency of use of International Classification of Diseases, Ninth Revision 555 and/or 556 codes for inflammatory bowel disease services, by medical specialty evaluate physicians' knowledge of International Classification of Diseases, Ninth Revision (ICD-9) codes, and self-reported accuracy of using ICD-9 codes 555 and 556 when billing the Ontario Health Insurance Plan (OHIP) for Inflammatory Bowel Disease (IBD) or unrelated services continued on next page APPENDIX 1 (continued) Questionnaire used to evaluate physicians' knowledge of International Classification of Diseases, Ninth Revision (ICD-9) codes, and self-reported accuracy of using ICD-9 codes 555 and 556 when billing the Ontario Health Insurance Plan (OHIP) for Inflammatory Bowel Disease (IBD) or unrelated services CCFC Crohn's and Colitis Foundation of Canada

TABLE 1 Population of Ontario physicians who were surveyed, by specialty
Percentage within each subspecialty who had seen patients with inflammatory bowel disease (IBD) in their practice; † Thirteen moved, 12 were not in practice *

TABLE 2 Practice patterns of surveyed Ontario physicians
*One pediatrician and seven surgeons saw more than 50 patients with inflammatory bowel disease (IBD) each year.ICD-9 International Classification of Diseases, Ninth Revision tacts for the same IBD diagnosis.Sensitivity estimates were similar in CD using either self-report (88.9%; 95% CI 87.0% to 90.6%) or chart review (89.2% 95% CI 84.2 to 92.8%) and for UC, were slightly lower, at 87.7% (95% CI 85.8% to 89.5%) for self-report and 74.4% (95% CI 67.3% to 80.5%) for chart review.Our results are predictably less accurate than those derived from the Manitoba database, which is made up of multiple files that are carefully crosschecked.Our survey data represent only physicians' estimates of their billing practices for IBD visits, and suggest an approximate error rate in a single database file.It is, therefore, likely that applying validated criteria and linking multiple files could improve the accuracy of the data obtained from linkage research.Other studies, cited below, have not specifically assessed IBD coding but have evaluated other medical conditions.