Overview: This paper is particularly relevant to the question of EF in ADHD. The idea that EF ability is varied across ADHD and not all impairments are equal is being teased apart in our data as well as others. The ability to identify a specific subtype related to a specific EF deficit could be very impactful for supportive programming. There is an established range of EF deficits in ADHD anywhere from 30-80% depending on the sample and study ie Nigg 2005. Recent work has suggested that the range is borne of varies and shallow measurement, one task or one report leading to under-identification since potentially only those with either a gross serious or a specific impairment that happens to match the measurement (Kofler, 2018). This calls for a. deeper measurement of EF and b. trying to unveil specific EF deficit subtypes that ultimately could be identified with targeted measurement and supported with targeted interventions. This paper takes an interesting approach to the subtype question by using data driven clustering/community detection approaches to sorting individuals based on EF reports and then validated them using SVM. Neural data for a subset was used to test the validity of the EF subtype vs. DSM diagnosis.
Methods:
Participants - participants were taken from two publicly available datasets the discover sample (N = 320) and a replication sample (N= ~600) individuals were 8-14 with and ADHD diagnosis ASD diagnosis or no diagnosis. A subsample of 84 had MRI data during a EF task that taps both ‘bottom up’ and ‘top down’ processing.
Measures - For the community detection the BRIEF (Inhibit, Shift, Emotional Control, Initiate, Working Memory, Plan/Organize, Organization of Materials, and Monitor), The CBCL (internalizing/externalizing), and the ADHD-RS (inattention/hyperactivity impulsive) sub-scales were used, 12 T-scores in all were used as the features for community detection.
Analyses - Louvain community detection was used to form the clusters. SVM was used to test the groups on held out cases.
Imaging Task - flickering shapes that participants had to respond to in the center of the screen with shapes in the periphery as well. Two runs were collected. The bottom up run had participants just responding to specific shapes. The top down run included monitoring peripheral shapes as well, so a dual task demand.
Imaging protocol- ~ 6 minutes for each task, 2 sec TR.
MRI analysis - bottom up significant regions were applied to the top down contrast and parameter estimates were extracted from them and then MANOVAS were done using both the EF model derived subtypes and DSM dx looking for differences in activity related to those groups
Results: Groups - the model found three groups with specific EF deficits, a meta cognitive group (WM and planning) an inhibition group and a flexibility/emotional group. SVM confirmed these groups with ~77-88% accuracy in predicting held out cases. These subtypes were found both across ADHD and ASD samples with high impairment of ASD in the emotional/flexibility group. The subtypes were found across non dx individuals as well as in individuals were lower in one than the other two however burden was higher across all subtypes in the DX group. Imaging analysis - MANOVA results revealed that the EF subtypes showed bigger differences in activity pulled from the EF demanding task than DSM dx groups.
Thoughts: The method and attempt at meaningfully subtyping individuals on EF ability is quite intriguing and promising. However when the results are dug into the big takeaways to me are 1. EF subtyping reveals a three group structure that is overwhelmingly reminiscent of a structure of EF that has been put forward time and time again. 2. EF ability relates to neural activity during an EF demanding task more than DX status which is cool but again not wildly unexpected right? A big open question I think is this is all based on parent reports, how would this look using behavioral task based measures of EF only and then comparing symptom burden. This is an open question I want to answer.