The Ability of a Computer Program Based on the Marquette Matrix-12 Short Measurement Matrix to Replicate Coding by the Minnesota ECG Coding Laboratory Vuyisile T. Nkomo, MD,* Thomas E. Kottke, MD,* Mark J. Brekke, MA, † Lee N. Brekke, PhD, † and Stephen C. Hammill, MD* Abstract: The study was undertaken to determine whether a computer program that uses “short measurement matrix” data from the Marquette Matrix-12 system can replicate Minnesota electrocardiogram (ECG) coding laboratory interpreta- tions. An agreement was found between coding of median complex ECGs at the Minnesota ECG coding laboratory and coding based on Marquette Matrix-12 short measurement matrix. The comparison was based on 763 ECGs plus chest pain history and serum enzyme values for a stratified random sample of 141 patients hospitalized in 1990 or 1991 for an event coded as HICDA 410.x (acute myocardial infarction), 411 (other acute and subacute forms of ischemic heart disease), 413 (angina pectoris), or 796.9 (other ill defined and unknown causes of morbidity and mortality). The population was reconstructed from the stratified random sample to enable population-based inferences. Exact agreement between Matrix-12 and Minnesota coding laboratory interpretation on 4 ECG patterns (evolving diagnostic, diagnostic, equivocal, or other ECG pattern) was 74.5% (Kappa = 0.63  0.05) for the stratified random sample and 78.8% (Kappa = 0.66  0.05) for the reconstructed population. For coding myocardial infarction based on the ECG, serum enzyme levels, and ischemic chest pain, agreement was 91.5% (Kappa = 0.85  0.04) for the stratified random sample and 90% (Kappa = 0.83  0.04) for the reconstructed population. Although ECG interpre- tation by a computer program based on the short measurement matrix of the Matrix 12 system results in better agreement than prior attempts to replicate the Minnesota coding laboratory, interpretation remains unacceptably discordant. Key words: Electrocardiograms, epidemiology, myocardial infarction, agreement. The Minnesota system of electrocardiographic coding (1), introduced 40 years ago to standardize interpretation for epidemiologic studies, has also become the standard electrocardiogram (ECG) in- terpretation system for clinical trials. At the time that it was developed, this system was a radical departure from the custom of using clinical judge- ment and unexplicated decision rules to interpret ECGs. Alternatively, it was based on the precise measurement and coding of Q waves, ST segments, and T waves with the application of explicit decision rules to these codes (see Appendix). Modifications From the * Departments of Medicine and Health Sciences Research, Mayo Clinic, Rochester, MN; and † Brekke Associates, Inc., Minneapolis, MN. Supported in part by NIH Grants HL 24326 and AR 30582. Reprint requests: Thomas E. Kottke, MD, Mayo Clinic and Foundation, 200 First St SW, Rochester, MN 55905. Copyright © 2000 by Churchill Livingstone ® 0022-0736/00/3304-0005$10.00/0 doi: 10.1054/jelc.2000.18359 Journal of Electrocardiology Vol. 33 No. 4 2000 341