Table 2 Summary of the different statistical approaches
From: Using multiple agreement methods for continuous repeated measures data: a tutorial for practitioners
Statistical Approach | Advantages/Strengths | Disadvantages | Key summary results (COPD study example) |
|---|---|---|---|
Concordance correlation coefficient | - A widespread and frequently used method. - Can still be used in cases where defining an appropriate CAD is either very difficult or impossible. | - Heavily influenced by the degree of between-subject and between-activity variability and the range of the data. - Can be very difficult to determine if the CCC is large enough to constitute acceptable agreement. - Can be very difficult to interpret clinically: interpretation not in terms of original measurement unit. | CCC 0.68 (95% CI 0.60 to 0.72) |
Limits of agreement | - Simplicity of application: relatively straightforward to compute limits. - Clinical interpretation is based on the original measurement scale. - Estimate of mean bias. - Easy to understand and interpret. | - Standard approach is highly dependent on the normality assumption for validity. - High variability in residual errors may mask the fact that a device could measure the true value more precisely than the gold-standard. - Easy to apply method incorrectly without explicitly specifying a clinically acceptable difference. | Mean bias −1.60 95% LoA − 11.57 to 8.38 |
TDI | - Easy to compute. - Easy to interpret. - Clinical interpretation is based on the original measurement scale. | - Can be difficult to determine if the TDI is large enough to constitute acceptable agreement. - Does not explicitly calculate the mean bias. | TDI 10.9 (95% CI 9.4 to 12.7) |
CP | - Easy to interpret. - Easy to compute. - Method cannot be used without explicitly specifying a clinically acceptable difference, which is based on the original measurement scale. | - Does not explicitly calculate the mean bias. | CP of 0.63 (95% CI 0.56 to 0.70) for boundary of ± 5 |
CIA | - Directly compares the disagreement between devices against the disagreement within devices and within subjects. - Much less dependent on the between-subject and between-activity variability compared to the CCC. - Can still be used in cases where defining an appropriate CAD is either very difficult or impossible. | - Depends heavily on the within-subject within-device variance. - Relies on data which has acceptable replication error. | CIA 0.68 (95% CI 0.57 to 0.75) |