Intraclass correlation coefficient (ICC) is a measure of agreement between two continuous variables. It takes values between 0 (= no agreement at all) and 1 (= perfect agreement). Note that it differs from the correlation coefficient that measures linear association rather than agreement. Specifically, agreement between two variables refers to the extent the variables are numerically the same, whereas correlation refers to the extent a change of one variable implies a change of the other variable.

Example 1

The Morse Fall Scale is a questionnaire used to measure the likelihood of patient falls in hospitals. A Chinese version was developed and administered to 20 patients twice of 1-2 days apart to assess the stability over time (Chow et al., 2006). A quantitative score was computed for each assessment. To examine the agreement between the test and retest scores, the ICC was 0.98 (95% CI = 0.98 to 0.99). Since the ICC is fairly close to 1, the agreement is very high. In general, an ICC > 0.7 is often recommended.

Example 2

The Activity Measure for Postacute Care (AM-PAC) is a tool to assess changes in 1) movement and physical capability, 2) personal care, and 3) cognition in patients undergoing rehabilitation. Scores of these scales can be computed quantitatively. The questionnaire may be administered to a patient or a proxy (such as a care giver). Haley et al. (2006) interested in comparing the agreement between scores obtained from patients and their proxies. 60 subjects and their proxies were recruited.

To examine the patient-proxy agreement, the ICC was obtained for the three scales of the tool. They were 0.91 (95% CI = 0.85 to 0.95), 0.90 (95% CI = 0.88 to 0.96), and 0.77 (95% CI = 0.60 to 0.88), respectively. The agreement in general is quite acceptable.