Manual vs. automatic assessment of the QT-interval and corrected QT

Sudden cardiac death (SCD) is challenging to predict. Electrocardiogram (ECG)-derived heart rate-corrected QT-interval (QTc) is used for SCD-risk assessment. QTc is preferably determined manually, but vendor-provided automatic results from ECG recorders are convenient. Agreement between manual and automatic assessments is unclear for populations with aberrant QTc. We aimed to systematically assess pairwise agreement of automatic and manual QT-intervals and QTc.

A multi-centre cohort enriching aberrant QTc comprised ECGs of healthy controls and long-QT syndrome (LQTS) patients. Manual QT-intervals and QTc were determined by the tangent and threshold methods and compared to automatically generated, vendor-provided values. We assessed agreement globally by intra-class correlation coefficients and pairwise by Bland-Altman analyses and 95% limits of agreement (LoA). Further, manual results were compared to a novel automatic QT-interval algorithm. ECGs of 1263 participants (720 LQTS patients; 543 controls) were available [median age 34 (inter-quartile range 35) years, 55% women]. Comparing cohort means, automatic and manual QT-intervals and QTc were similar. However, pairwise Bland-Altman-based agreement was highly discrepant. For QT-interval, LoAs spanned 95 (tangent) and 92 ms (threshold), respectively. For QTc, the spread was 108 and 105 ms, respectively. LQTS patients exhibited more pronounced differences. For automatic QTc results from 440-540 ms (tangent) and 430-530 ms (threshold), misassessment risk was highest. Novel automatic QT-interval algorithms may narrow this range.

Pairwise vendor-provided automatic and manual QT-interval and QTc results can be highly discrepant. Novel automatic algorithms may improve agreement. Within the above ranges, automatic QT-interval and QTc results require manual confirmation, particularly if T-wave morphology is challenging.