Adult Lifespan Cognitive Variability in the Cross-Sectional Cam-CAN Cohort

Abstract

This study examines variability across the age span in cognitive performance in a cross-sectional, population-based, adult lifespan cohort from the Cambridge Centre for Ageing and Neuroscience (Cam-CAN) study (n = 2680). A key question we highlight is whether using measures that are designed to detect age-related cognitive pathology may not be sensitive to, or reflective of, individual variability among younger adults. We present three issues that contribute to the debate for and against age-related increases in variability. Firstly, the need to formally define measures of central tendency and measures of variability. Secondly, in addition to the commonly addressed location-confounding (adjusting for covariates) there may exist changes in measures of variability due to confounder sub-groups. Finally, that increases in spread may be a result of floor or ceiling effects; where the measure is not sensitive enough at all ages. From the Cam-CAN study, a large population-based dataset, we demonstrate the existence of variability-confounding for the immediate episodic memory task; and show that increasing variance with age in our general cognitive measures is driven by a ceiling effect in younger age groups.

Keywords: MMSE; adult lifespan; ceiling effects; cognitive variability; episodic memory; heterogeneity; variance confounders; verbal fluency.

Figure 1

Box plots of Cam-CAN data by age-groups: (a) Wechsler immediate story recall; (b) Wechsler delayed story recall; (c) Mini-Mental State Examination; (d) Addenbrooke’s Cognitive Examination Revised; (e) category fluency; (f) letter fluency. Note that the verbal fluency measures are not within a bounded range, unlike the MMSE (0–30) or story recall (0–24); the plot ranges from 0–60 to reflect a limit of one response per second. Definition of box plot: box indicates the inter-quartile range, mid-line indicates the median, box-width proportional to square-root of sample size (i.e., $\sqrt{n_i}$), whiskers extend to largest data value within 1.5 × IQR of the respective quartile, circles correspond to outliers (i.e., data values beyond the whiskers).

Figure 2

Box plots of simulated data by age-groups: (a) declining mean and fixed variance; (b) fixed mean and increasing variance; (c) declining mean and increasing variance; (d) Simulated Scenario D is a mixture of two sub-groups (≤ GCSE, ≥A-level) with declining mean and fixed mean respectively, and both with fixed variance; as illustrated in (e). As a result of this mixture of distributions, the combined Scenario D appears similar to Scenario C; within (d) there is an unmeasured confounder.

Figure 3

Box plots of delayed story recall, subdividing age-groups by: (a) sex; (b) education attainment.

Figure 4

Box plots of immediate story recall, subdividing age-groups by: (a) sex; (b) education attainment.

Figure 5

Illustration of the effect of truncating the range of simulated Scenario A to [0,33] (original scale was unbounded, i.e., [0,∞]). (a) Box plot of truncated score, contrast with original in Figure 2a; (b–g) Comparisons of the histograms for the truncated and original scores over the first six age-groups. All original scores above 33 are truncated, i.e., recoded, as 33. Hence the large spikes in the truncated score histograms.

Figure 6

(a) Repetition of Figure 1c, note the extended y-axis to indicate the likely ceiling effect on the MMSE; (b–g) Histograms of the MMSE scores over the first six age-groups. The MMSE range is [0,30]. The 28–37 age-group (c) have a particularly large “spike” at the maximum score, indicative of a ceiling effect.