Contact Person

Dr. Peter Henning

Managing Editor

Phone: +49 (0)711 - 2 29 87 37
Fax: +49 (0)711 - 2 29 87 65
send an Email

Archive (2016–2006)

Predictive Value of ICD-9-CM Codes Used in Vaccine Safety Research

Journal: Methods of Information in Medicine
Subtitle: A journal stressing, for more than 50 years, the methodology and scientific fundamentals of organizing, representing and analyzing data, information and knowledge in biomedicine and health care
ISSN: 0026-1270
Issue: 2008 (Vol. 47): Issue 4 2008
Pages: 328-335

Predictive Value of ICD-9-CM Codes Used in Vaccine Safety Research

J. P. Mullooly 1, J. G. Donahue 2, F. DeStefano 3, J. Baggs 3, E. Eriksen4 for the VSD Data Quality Working Group

1Center for Health Research, Kaiser Permanente Northwest, Portland, OR, USA 2Marshfield Clinic, Marshfield, WI, USA 3Centers for Disease Control and Prevention, Atlanta, GA, USA 4UCLA Center for Vaccine Research, Los Angeles Biomedical Research Institute, Torrance, CA, USA


bias, ICD-9-CM codes, positive predictive values, vaccine safety, power reductions


Objectives: To assess how well selected ICD-9-CM diagnosis codes predict adverse events; to model bias and power loss when vaccine safety analyses rely on unverified codes. Methods: We extracted chart verification data for ICD-9-CM diagnosis codes from six Vaccine Safety Datalink (VSD) publications and modeled biases and power losses using positive predictive value (PPV) estimates and ranges of code sensitivity. Results: Positive predictive values were high for type 1 diabetes (80%) in children, relative to WHO criteria, and intussusception (81%) in young children, relative to a standard published case definition. PPVs were moderate (65%) for inpatient and emergency department childhood seizures and low (21%) for outpatient childhood seizures, both relative to physician investigator judgment. Codes for incident central nervous system demyelinating disease in adults had high PPV for inpatient codes (80%) and low PPV for outpatient codes (42%) relative to physicians’ diagnoses. Modeled biases were modest, but large increases in frequencies of adverse events are required to achieve adequate power if unverified ICD-9-CM codes are used, especially when vaccine associations are weak. Conclusions: ICD-9-CM codes for type 1 diabetes in children, intussusception in young children, childhood seizures in inpatient and emergency care settings, and inpatient demyelinating disease in adults were sufficiently predictive for vaccine safety analyses to rely on unverified diagnosis codes. Adverse event misclassification should be accounted for in statistical power calculations.

You may also be interested in...

J. Marienhagen, Ch. Eilles

Nuklearmedizin 2003 42 4: 129-134


C. Vogel, O. Gefeller

Methods Inf Med 2002 41 4: 342-348

Call for Papers “New Methodologies for Patient Rehabilitation”

Over the past few years there has been increasing interest in the application of Virtual Reality...

Methods issue 6/2015

John D. Ainsworth and Iain E. Buchan submitted the paper "Combining Health Data Uses to Ignite...

Methods issue 5/2015

At last year's annual conference of the German Association for Medical Informatics, Biometry and...

Methods issue 4/2015

Elske Ammenwerth submitted the paper "Evidence-based Health Informatics: How Do We Know What...

Methods issue 3/2015

The focus theme in this issue on "Biosignal Interpretation II – Advanced Methods for Studying...