Accuracy of ICD‐9‐CM Codes by Hospital Characteristics and Stroke Severity: Paul Coverdell National Acute Stroke Program

Abstract

Background

Epidemiological and health services research often use International Classification of Diseases, Ninth Revision, Clinical Modification ( ICD‐9‐ CM) codes to identify patients with clinical conditions in administrative databases. We determined whether there are systematic variations between stroke patient clinical diagnoses and ICD‐9‐ CM codes, stratified by hospital characteristics and stroke severity.

Methods and Results

We used the records of patients discharged from hospitals participating in the Paul Coverdell National Acute Stroke Program in 2013. Within this stroke‐enriched cohort, we compared agreement between the attending physician's clinical diagnosis and principal ICD‐9‐ CM code and determined whether disagreements varied by hospital characteristics (presence of a stroke unit, stroke team, number of hospital beds, and hospital location). For patients with a documented National Institutes of Health Stroke Scale score at admission, we assessed whether diagnostic agreement varied by stroke severity. Agreement was generally high (>89%); differences between the physician diagnosis and ICD‐9‐ CM codes were primarily attributed to discordance between ischemic stroke and transient ischemic attack ( TIA), and subarachnoid and intracerebral hemorrhage. Agreement was higher for patients in metropolitan hospitals with stroke units, stroke teams, and >200 beds (all P<0.001). Agreement was lowest (60.3%) for rural hospitals with ≤200 beds and without stroke units or teams. Agreement was also lower for milder (94.9%) versus more‐severe (96.4%) ischemic strokes ( P<0.001).

Conclusions

We identified disagreements in stroke/ TIA coding by hospital characteristics and stroke severity, particularly for milder ischemic strokes. Such systematic variations in ICD‐9‐ CM coding practices can affect stroke case identification in epidemiological studies and may have implications for hospital‐level quality metrics.