The reliability of electrocardiogram interpretation in critically ill patients

Background

Critically ill patients are generally unable to report ischemic chest pain due to decreased consciousness, endotracheal intubation, and use of sedatives and narcotics. The diagnosis of cardiac ischemia is therefore usually based on elevated cardiac enzymes and typical ischemic changes on an electrocardiogram (ECG). Although key management decisions depend on ECG interpretation, the reproducibility of ECG interpretation is unclear in the ICU.

Objective

To estimate the inter-rater and intra-rater reliability of ECG interpretation for the presence of myocardial ischemia in critically ill patients, with and without knowledge of troponin T values.

Design

A prospective cohort study.

Setting

A 15-bed medical–surgical university-affiliated ICU.

Patients

Consecutive adults admitted to the ICU over a 2-month period.

Methods

We collated all consecutive 12-lead ECGs performed on all patients during the study period. Using a structured, pre-tested form, and blinded to patient data and each other's ratings, two internists interpreted each ECG. Two weeks later, both raters reinterpreted the same ECGs, blinded to their previous interpretation and unblinded to the troponin T values. We report results using chance-independent agreement (phi), which generates less biased agreement measures than chance-corrected agreement (kappa). Discrepancies for the two primary outcomes (presence of pathologic Q waves and the overall assessment of the presence of myocardial ischemia) and 13 secondary outcomes (other ECG abnormalities, such as ST segment changes) were resolved by a third internist blinded to the two previous ratings but unblinded to troponin T values.

Results

We found variable inter-rater reliability of ECG interpretation when raters were blinded to troponin T levels. We report all values with the associated 95% confidence interval (CI). For the two primary outcomes, agreement was substantial (phi 0.61 [95% CI 0.17, 0.85]) for the detection of pathologic Q waves and slight (phi 0.10 [-0.06, 0.25]) for the overall assessment of the presence of myocardial ischemia. For secondary specific ECG changes, agreement was highest (phi 0.79 [0.64, 0.89]) for detection of a new left bundle branch block (LBBB) and lowest (phi 0.30 [0.003, 0.54]) for ST segment depression. When unblinded to troponin T values, inter-rater reliability increased slightly for each primary and secondary outcome. Intra-rater reliability, assessed prior to and following knowledge of troponin T levels, was high for detection of pathologic Q waves (phi 0.44 [0.08, 0.69] and 0.78 [0.68, 0.85], respectively), and new LBBB (phi 1.0 and 0.95 [0.83, 0.98], respectively), but was lower for the overall assessment of the presence of ischemia (phi 0.21 [0.06, 0.36] and 0.47 [0.34, 0.58], respectively).

Conclusions

We found that the inter-rater and intra-rater reliability of ECG interpretation for the overall detection of specific ECG changes in critically ill patients was good but variable. However, for the overall assessment of the presence of myocardial ischemia, reliability was poor. Knowledge of troponin T levels appears to improve the reliability of ECG interpretation. Further studies evaluating the difficult diagnosis of myocardial ischemia in the critically ill are required.