Aims/hypothesis: Deterioration and improvement in the electrocardiogram are important outcomes in cardiovascular disease progression assessment. We used a sample of serial records from the WHO Multinational Study of Vascular Disease in Diabetes (WHO MSVDD) to assess Minnesota coding variability.
Methods: A constructed subsample of 118 of the 352 paired (baseline and follow-up) and previously Minnesota-coded ECG records from the London cohort was randomised and re-read independently of the first code (respectively 11 and 0.5 years later) by the same two coders. Detailed Minnesota codes were summary coded into groups 1 (CHD unlikely), 2 and 3 (CHD possible and probable, respectively).
Results: Re-reading of the constructed sample for the baseline records (11 years later) generated 21 Summary code reassignments (2 unlikely to possible or probable; 19 possible or probable to unlikely); re-reading for the follow-up records (0.5 years later) generated only 8 summary code reassignments (21 vs 8p < 0.001) (3 unlikely to possible or probable; 4 possible or probable to unlikely; 1 probable to possible). Re-reading increased the estimated net ECG deterioration in the constructed sample from 11.8 % to 25.4%. Consistency analysis showed most variability in marginal baseline abnormalities.
Conclusion/interpretation: Coding variability is now small though re-reading suggests some time-dependent coding drift. Relative over-reading at baseline suggests that the change reported in the complete WHO MSVDD cohort at follow-up was underestimated and that almost all of the reported ECG deterioration and about half of the reported ECG 'improvement' was real.