Accuracy of coding for possible warfarin complications in hospital discharge abstracts

Abstract

Background

Hospital discharge abstracts could be used to identify complications of warfarin if coding for bleeding and thromboembolic events are accurate.

Objectives

To measure the accuracy of International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9CM) codes for bleeding and thromboembolic diagnoses.

Setting

University affiliated, tertiary care hospital in Ottawa, Canada.

Patients

A random sample of patients discharged between September 1999 and September 2000 with an ICD-9-CM code indicating a bleeding or thromboembolic diagnosis.

Methods

Gold-standard coding was determined by a trained chart abstractor using explicit standard diagnostic criteria for bleeding, major bleeding, and acute thromboembolism. The abstractor was blinded to the original coding. We calculated the sensitivity, specificity, positive, and negative predictive values of the original ICD-9CM codes for bleeding or thromboembolism diagnoses.

Results

We reviewed 616 medical records. 361 patients (59%) had a code indicating a bleeding diagnosis, 291 patients (47%) had a code indicating a thromboembolic diagnosis and 36 patients (6%) had a code indicating both. According to the gold standard criteria, 352 patients experienced bleeding, 333 experienced major bleeding, and 188 experienced an acute thromboembolism. For bleeding, the ICD-9CM codes had the following sensitivity, specificity, positive and negative predictive values [95% CI]: 93% [90–96], 88% [83–91], 91% [88–94], and 91% [87–94], respectively. For major bleeding, the ICD-9CM codes had the following sensitivity, specificity, positive and negative predictive values: 94% [91–96], 83% [78–87], 87% [83–90], and 92% [88–95], respectively. For thromboembolism, the ICD-9CM codes had the following sensitivity, specificity, positive and negative predictive values: 97% [94–99], 74% [70–79], 62% [57–68], and 98% [96–99], respectively. By selecting a sub-group of ICD-9CM codes for thromboembolism, the positive predictive value increased to 87%.

Conclusion

In our centre, the discharge abstract could be used to identify and exclude patients hospitalized with a major bleed or thromboembolism. If coding quality for bleeding is similar in other hospitals, these ICD-9-CM diagnostic codes could be used to study population-based warfarin-associated hemorrhagic complications using administrative databases.

Introduction

Warfarin is a commonly used anticoagulant that has many clinical indications. Although warfarin is effective, there is also considerable risk associated with its use including bleeding and treatment failures. The annual risks are estimated to be between 2–8% for bleeding and 1–3% for treatment failures [1]. Estimating risk of bleeding and treatment failure for a particular patient is an important consideration when deciding to initiate or continue anticoagulant therapy.

Most data regarding bleeding and treatment risks with warfarin come from clinical trials or specialized warfarin clinics. Such patients are often at low risk for complications because they tend to be carefully selected and are assessed at a greater frequency. The paucity of studies measuring community-based risks could be due to prohibitive costs of following large cohorts of patients. Administrative databases could address this issue if they are truly population-based. With administrative databases, warfarin-treated patients could be identified with pharmacy claims and bleeding or thromobembolism outcomes data could be identified from diagnoses within hospital discharge abstracts.

Databases of pharmacy claims are reliable since their creation requires relatively minimal human input. This is because many pharmacies can automatically create a claim for a medication when it is dispensed to a patient. However, diagnostic coding in discharge abstracts requires extensive human input. Getting a diagnostic code into the discharge abstract typically requires the physician to identify and document the diagnosis, the health records analyst to recognize and correctly interpret the physician's documentation, determine the appropriate code for that diagnosis, and then correctly input the code into the discharge abstract. Given the multiple steps involved in this process, one should expect that the accuracy of diagnostic codes in the discharge abstract is suboptimal.

The accuracy of diagnostic coding in the discharge abstract coding varies considerably between diagnoses. For example, the positive predictive value for acute myocardial infarction is 96.9% [2] while that for pelvic inflammatory disease it is only 18.1% [3]. Some data are available for potential warfarin complications such as upper gastrointestinal tract bleeding [4], [5], [6]. There are also some data for potential warfarin failures such as venous thromboembolism [7] and stroke [8], [9], [10], [11] However, we are unaware of any published study describing coding accuracy for a wide range of bleeding or clotting outcomes.

We performed this study to determine the accuracy of bleeding and thromboembolic diagnostic codes in the hospital discharge abstract. We reasoned that if such codes were sufficiently accurate, we could use administrative data to study patient outcomes related to warfarin treatment.