Agreement and coding reliability of the Minnesota and Mayo electrocardiographic coding systems

To determine whether diagnoses of myocardial infarction assigned by a system that uses Marquette 12SL electrocardiographic (ECG) codes with manual over-reading agree with diagnoses assigned by Minnesota ECG codes.

Agreement and recode reliability of Minnesota and Mayo coding systems based on 768 ECGs plus chest pain history and serum enzyme values were analyzed for a stratified random sample of 141 patients with an event in 1990 or 1991 coded as HICDA 410.x, 411, 413 or 796.9. The population was reconstructed from the stratified random sample so that population-based inferences could be made from the analysis.

For the stratified random sample, exact agreement on 4 categories (evolving diagnostic, diagnostic, equivocal, or other ECG) between Mayo and Minnesota ECG coding was 53.9% (kappa = 0.37 +/- 0.05). Code-recode agreement was higher for Minnesota coding (83.0%; kappa = 0.74 +/- 0.05) compared with Mayo coding (73.8%; kappa = 0.64 +/- 0.05). The same pattern was present for the reconstructed population. For coding myocardial infarction based on the ECG, serum enzyme levels, and the presence or absence of ischemic chest pain, agreement between Mayo and Minnesota coding was 84.4% (kappa = 0.72 +/- 0.05) based on the stratified random sample and 81.7% (kappa = 0.67 +/- 0.06) based on the reconstructed population. For the stratified random sample, reliability of diagnosis of myocardial infarction was 93.6% (kappa = 0.88 +/- 0.04) for the Minnesota system and 94.3% (kappa = 0.90 +/- 0.03) for the Mayo system.

ECG interpretation by the Mayo and Minnesota coding systems differs significantly, and Mayo ECG coding is less reliable than Minnesota ECG coding. Coding of myocardial infarction on the basis of ECGs, serum enzymes, and ischemic chest pain, however, is equally reliable for both systems.