The limits of agreement and the intraclass correlation coefficient may be inconsistent in the interpretation of agreement

Abstract 

Objective

To compare the interpretation of agreement in the prediction of neonatal outcome variables, using the limits of agreement (LA) and the intraclass correlation coefficient (ICC).

Study Design and Setting

Three obstetricians were asked to predict neonatal outcomes independently based on the evaluation of intrapartum cardiotocographic tracings. Interobserver agreement was assessed with the LA and the ICC, and the results obtained were interpreted by six clinicians and six statisticians on a scale that established agreement as very poor, poor, fair, good, or very good.

Results

Interpretation of the LA results was less consensual than the ICC results, with proportions of agreement of 0.36 (95% confidence interval [CI]: 0.28–0.44) vs. 0.63 (95% CI: 0.54–0.73), respectively. LA results suggested a fair to good agreement among obstetricians, whereas interpretation of ICC results suggested a poor to fair agreement. LA results were more plausible with reality, suggesting that obstetricians predicted neonatal outcomes better than randomly generated values, whereas it was not always the case with the ICC.

Conclusions

LA and ICC can provide inconsistent results in agreement studies. Accordingly, in the absence of better strategies to assess agreement, both should be used for this purpose, but their results need to be interpreted with caution keeping their respective limitations in mind.

Keywords: Agreement, Reproducibility of results, Observer variation, Apgar score, Umbilical artery blood pH, Statistical data interpretation