TL;DR
This study used unsupervised machine learning to identify distinct executive function (EF) profiles in neurotypical children based on their performance on a design fluency task. The findings reveal two profiles with different strategies for balancing speed and accuracy over time, suggesting EF is a dynamic process rather than a static trait.
What Problem Does This Paper Address?
Executive functions (EFs) are crucial for self-regulation and successful adaptation, but there's significant variability in how children develop and utilize them. Traditional methods often assess EFs in isolation or rely on broad composite scores, potentially missing nuanced individual differences. This study aims to address this by using a data-driven approach to uncover naturally occurring EF profiles in neurotypical children, moving beyond predefined categories and static measures to understand the dynamic interplay of EFs during a task.
Methods
The researchers employed unsupervised machine learning, specifically K-means clustering, to analyze performance data from a tablet-based design fluency task. This task involved children connecting dots to create novel designs over five minutes, with their performance (designs, repetitions, strategies) tracked minute-by-minute. The K-means algorithm was used to identify distinct clusters (profiles) within the data, and a silhouette coefficient was used to determine the optimal number of clusters (two were identified). Permutation tests were used to compare these clusters across various EF metrics, and Spearman correlations were used to examine age-related trends within each profile.
Datasets
The study utilized data from 113 neurotypical Canadian children (aged 7-13 years). The data was collected as part of a previous study at the Laboratoire d'études en neuropsychologie de l'enfance et de l'adolescence (LÉNEA). No new benchmarks were introduced; rather, existing cognitive assessment tools were used.
Key Insights and Ideas
- EF is Dynamic, Not Static: The study strongly suggests that executive functions are best understood as a time-sensitive, adaptive process that evolves during task performance, rather than a fixed ability measured at a single point.
- Two Distinct Profiles Emerge: Instead of a simple high-vs-low performer dichotomy, two distinct profiles with different strategies for managing speed and accuracy were identified.
- Process Measures Matter: Tracking performance over time (process measures) was crucial for differentiating these profiles, highlighting that how children perform is as important as how much they produce.
- Individual Variability is Key: The identified differences in EF profiles were more strongly linked to individual cognitive control strategies than to factors like age or sex, underscoring the importance of interindividual variability.
Key Results
The K-means clustering identified two distinct EF profiles:
- Profile A (Cluster 1): Characterized by a more controlled, accuracy-focused approach. These children produced fewer designs overall but showed a more stable production pattern with a gradual increase in repetition errors over time. Their cognitive flexibility (percentage of strategies) was lower initially but increased progressively.
- Profile B (Cluster 2): Exhibited a high-output, speed-prioritizing approach. They generated more designs early on but showed a steeper increase in repetition errors and a progressive decline in cognitive flexibility and planning. Their percentage of repetitions increased significantly over time, especially from the second minute onwards.
While age and sex did not significantly differ between the clusters, Profile B showed a slightly higher verbal IQ. Age-related correlations were found in Profile A, suggesting a more typical developmental trajectory for strategy use, which was absent in Profile B.
Limitations and Future Work
The authors acknowledge several limitations:
- Beta Version of App: The AppFLO task used was a beta version, which might have slightly influenced metric counts. Future research should use the prevalidated version.
- Sample Demographics: The sample was relatively modest (n=113) and predominantly French-speaking White children from Quebec. Replication with larger and more diverse samples is needed.
- Unmeasured Factors: Cognitive fatigue and self-regulation were not directly measured, making it difficult to distinguish preference-based from deficit-based profiles. Future studies should investigate fatigue susceptibility, especially in clinical populations.
Future work should explore:
- Transdiagnostic Components: Whether these EF profiles have implications across different clinical conditions.
- Impact on Academic/Problem-Solving: How these profiles relate to academic performance, problem-solving, and daily executive function challenges.
- Alternative ML Techniques: Exploring other machine learning techniques (e.g., neural networks) for a deeper understanding of gradual changes between profiles.
- Longitudinal Stability: Whether these profiles remain stable over time or shift with experience or training.
Summary updated