Registration gives the benefit of site update e-mails and additional information from Takeda on new education materials and events.
ADHD Institute Register

23 Jun 2021

Cervantes-Henriquez ML et al. J Atten Disord 2021; Epub ahead of print

Machine learning algorithms and multivariate analyses enables identification and classification of individuals with subtle differences in symptoms. Several genetic variants have been associated with ADHD and/or the association of ADHD, most of which have been examined in individuals of Asian, Caucasian and Latino descent but remains to be investigated in those of predominantly African ancestry. The authors of this study hypothesised that a new phenotypic construct of ADHD severity based on machine learning algorithms may enhance the performance of diagnostic tools for ADHD and reduce intrinsic biases associated with clinical heterogeneity. Specifically, the aim of this study was to investigate whether single nucleotide polymorphisms (SNPs) in the ADGRL3, DRD4 and SNAP25 genes are associated with, and predict, ADHD severity in families from a Caribbean community, which has one of the largest African genetic components in the Caribbean and in Central and South America.

Over 12 years, 408 individuals (57% male) aged 6‒60 years were prospectively recruited and clinically characterised from 120 nuclear families with ≥1 child with ADHD (proband) who were born in Barranquilla, Colombia and its metropolitan area. In total, 136 children and adolescents aged 6‒18 years (25% female) and 97 adults aged ≥18 years (14.5% female) included in the study had ADHD. Data from only 113 of the 120 nuclear families were included in this study as genotyping data were not available for seven families. The mean (standard deviation) family size was 3.4 (0.65). Unsupervised machine learning algorithms aimed to identify latent subgroups of individuals and distinct symptom profiles based on clinical data to define ADHD symptom severity. Based on clinical symptoms that are likely to occur within specific clinical profiles, individuals were classified as ‘not severe’ or ‘severe’ using latent cluster analysis of Diagnostic and Statistical Manual of Mental Disorders, 4th Edition (DSM-IV) symptomatology (i.e. global, inattention, hyperactivity and impulsivity). Using family-based association tests, associations between SNPs and ADHD severity latent phenotypes were detected. Machine learning algorithms were then used to create predictive models of ADHD severity according to demographic and genetic data.

Severity of symptoms

Individuals exhibiting a complex severity phenotype for ADHD had a higher number of DSM-IV symptoms; those with ADHD were 8.9 times more likely to display a severe phenotype based on global items (95% confidence interval [CI] 4.9‒16.6). Moreover, children (odds ratio [OR] 7.3 [95% CI 4.1‒13.4]) and adolescents (OR 4.9 [95% CI 2.1‒11.7]) had the highest risk of developing global-based severe symptoms. Within each domain, individuals with a positive ADHD diagnosis were more likely to exhibit severe inattention (OR 14.9 [95% CI 8.6‒26.5]), severe hyperactivity (OR 6.9 [95% CI 3.2‒17.7]) and severe impulsivity (OR 4.8 [95% CI 2.7‒8.7]).

Genetic markers conferring susceptibility to severe ADHD symptoms

Family-based association tests revealed significant linkage and association of either global or domain-specific severity with markers ADGRL3-rs2122642, ADGRL3-rs10001410, DRD4-rs916457 and SNAP25-rs362990. Under different models of genetic inheritance, DRD4-rs916457 and SNAP25-rs362990 were shown to be associated with both global and inattention symptom severity; ADGRL3-rs2122642, ADGRL3-rs10001410 and DRD4-rs916457 were associated with hyperactivity symptom severity; and ADGRL3-rs2122642, DRD4-rs916457 and SNAP25-rs362990 were associated with impulsivity symptom severity.

Predictive genomics framework for severity of symptoms

In this study, machine learning algorithms provided the highest accuracy for predicting global, inattention, hyperactivity and impulsivity severity. Receiver-operating characteristic (ROC) analysis showed that in all cases, the machine learning algorithms had a moderate ability in discriminating severe from non-severe individuals based on demographic and genetic data with accuracy values between ~70% and 82%. Analyses also revealed that age, sex and the severity-associated SNPs are important predictors of ADHD symptom severity.

Potential limitations of this study include that the individuals with ADHD were not medicated and the representativeness of the sample across different age groups due to family-based design may prevent extrapolating these results to the general population. Additionally, the same clinical assessment methods were used across all age groups and family members were used for the genetic analysis that the authors acknowledge may not have been ideal. Finally, these data were based on nuclear families characterised by predominantly African ancestry, which the authors consider may be both a limitation and strength of this study. Therefore, the authors suggest that the severity patterns observed in this study are compared with those reported in other populations around the world.

To conclude, in the authors’ opinion, this study supports the role of ADGRL3, DRD4 and SNAP25 genes in the aetiology of ADHD severity in a Caribbean community of predominantly African ancestry. Moreover, the authors provide a new prediction framework for determining ADHD symptom severity using machine learning algorithms which may have potential clinical use.

Read more about the use of machine learning algorithms for determining the severity of ADHD symptoms here

Disclaimer: The views expressed here are the views of the author(s) and not those of Takeda.

Cervantes-Henriquez ML, Acosta-López J, Martinez AF, et al. Machine learning prediction of ADHD severity: association and linkage to ADGRL3, DRD4, and SNAP25. J Atten Disord 2021; Epub ahead of print.

Filter content by:

ADHD Institue logo

You’re now being transferred to

and are leaving the ADHD Institute site

Takeda has no influence or control over the content of this third party website.

Continue Cancel