Measures of clinical significance are critical for meaningful interpretation of treatment outcome research on Alzheimer’s disease. A common method of quantifying clinical significance is to calculate a minimal clinically important difference (MCID), which represents the smallest numerical change on an outcome measure that corresponds to an added benefit in a patient’s life. Often the MCID is calculated based on an anchor response. Individuals who report a meaningful change serve as the “anchors”, and the mean level of change for this group serves as the MCID. In research on Alzheimer’s disease, there are several possible raters to provide anchors, including patients, family observers, and clinicians, who may or may not agree on whether there has been a meaningful change in outcome. The goal of this study was to examine the extent to which agreement among anchors impacts MCID estimation and whether this relationship is moderated by cognitive severity status.

Analyses were completed on a longitudinal sample of 2,247 adults, age 50-103, from the Uniform Data Set 3.0. Outcome measures included the Clinical Dementia Rating - Sum of Boxes (CDR-SB), Functional Activities Questionnaire, and Montreal Cognitive Assessment.

For all of the outcomes, the MCID estimate was significantly higher when meaningful decline was endorsed by all of the raters compared to situations in which there was disagreement among the raters. For example, on the CDR-SB, agreement significantly impacted MCID estimates (F(1, 2241)=168.80, p<0.001; partial h2 = 0.07), such that the agreement group had greater CDR-SB change score (mean=1.29, SD1.98) than the no agreement group (mean=0.37, SD=1.38; Tukey HSD: p<0.001). In addition, the MCID estimate increased with increasing levels of cognitive impairment. For instance, on the CDR-SB, MCID estimates were significantly different across the severity groups (F(2, 2241)=138.27, p<0.001; partial h2 = 0.11), such that increase in CDR-SB was highest for the mild dementia group (mean=1.84, SD=2.42), moderate in the MCI group (mean=0.71, SD=1.30), and lowest for the cognitively normal group (mean=0.07, SD=0.55; Tukey HSD; all p’s < 0.001). Finally, cognitive severity status moderated the influence of agreement among raters on MCID estimation for the CDR-SB and FAQ, such that rater agreement demonstrated less influence on the MCID as disease severity increased. For example, on the CDR-SB, post-hoc tests revealed that there was a significant difference across agreement groups in the cognitively normal (p<0.001; Cohen’s d = 0.96) and MCI groups (p<0.001; Cohen’s d = 0.49), but agreement did not impact MCID estimates for the mild dementia group (p=0.065).

MCID estimates based on one anchor may underestimate meaningful change, and researchers should consider the viewpoints of multiple raters in constructing MCIDs. Consideration of agreement appears most important in the early stages of cognitive decline, which are the focus of most modern clinical trials.