Learning-to-learn as a metacognitive correlate of functional outcomes after stroke: a cohort study

Abstract

Background: Meta-learning is a metacognitive function for successful, efficient learning in various tasks. While it is possible that meta-learning is linked to functional recovery in stroke, it has not been investigated in previous clinical research on metacognition.

Aim: Examine if individual meta-learning ability is associated with functional outcomes.

Design: Cohort study.

Settings: Rehabilitation ward in Fujita Health University Hospital.

Population: Twenty-nine hemiparetic people after stroke.

Methods: The study measured individual sensorimotor adaptation rate, meta-learning (acceleration of adaptation through training), and Functional Independence Measure (FIM) motor effectiveness, an index of functional outcome measuring improvement in proficiency of activity of daily living (ADL). Participants performed visuomotor adaptation training sessions with their less-affected arm. They made arm-reaching movements to hit a target with cursor feedback, which was occasionally rotated with regard to their hand positions, requiring them to change the movement direction accordingly. Initial adaptation rate and meta-learning were quantified from pre- and post-training tests. The relationship between these indices of adaptation ability and FIM motor effectiveness was examined by multiple linear regression analyses.

Results: One participant was excluded before data collection in the motor task. In the remaining 28 individuals, the regression analyses revealed that FIM motor effectiveness positively correlated with meta-learning (µ=0.90, P=0.008), which was attenuated by age (µ=-0.015, P=0.005), but not with initial adaptation rate (P=0.08). Control analyses suggested that this observed association between FIM motor effectiveness and meta-learning was not mediated by patients' demographics or stroke characteristics.

Conclusions: This study demonstrates that those who can accelerate adaptation through training are likely to improve ADL, suggesting that meta-learning may be linked with functional outcomes in some stroke individuals. Meta-learning may enable the brain to keep (re-)learning motor skills when motor functions change abruptly due to stroke and neural recovery, thereby associated with improvement in ADL.

Clinical rehabilitation impact: Meta-learning is part of metacognitive functions that is positively associated with functional outcomes.

Figure 1

—Task design. A) Time flow. Stroke onset was labelled as T0. FIM motor was measured at admission to (T1) and discharge from (T3) the rehabilitation ward. The motor task was performed once during hospital stay at T2. The rate of motor learning (β) was measured before and after a training task. B) Motor task and environment. Participants moved the arm to cross a virtual target in front of them. C) Test task to measure β. Participants crossed the target with the cursor, which was rotated from the actual hand position. D) Training task. Exposure to rotation and target reach were separated into two trials. A score was given for successful target reach as encouragement. Gray lines (faded) and navy lines indicate individual data and the group means, respectively.

Figure 2

—Results of the multiple linear regression. A) Individual FIM scores in the motor domain and motor learning speed (β). FIM effectiveness and Δβ were calculated for each individual. B) Individual values of the independent and the dependent variables. C) AIC values and estimated coefficients of predictive factors in multiple regression models with top AIC values. Note that age (as a main factor) is omitted in the right panel because it is not included in any of these models. D) Plots of residuals to examine the normality (by quantile-quantile [Q-Q] plot and histogram), the center, and the homoscedasticity. E) Estimated marginal effects of Δβ for younger and older groups (N.=14 per group). Marginal effects were calculated at their mean ages (51 and 72 years). Dots represent individual data. Lines represent individual data in A, group means in C and E, and a reference line of normal distribution in the Q-Q plot and regression slope in the scatter plot in D. Bars represent single estimate values in C and number of data in the histogram in D. Error bars and shades represent 95% confidence intervals, except the error bars in D that represent a standard error of mean.