Summary: Researchers have developed a predictive model that combines genetic information with early developmental milestones to estimate which autistic children are at risk of later developing an intellectual disability (ID). In an analysis of 5,633 children from three large North American cohorts, the model correctly identified about 10% of those who later received an ID diagnosis and improved risk stratification compared with current approaches.
This work represents a shift from passive “wait and see” monitoring toward earlier, more personalized support for families. Although the method is still evolving, it demonstrates how genetic data plus developmental history can inform earlier decision-making as genetic testing becomes more affordable and computational tools grow more powerful.
Key Facts:
- Predictive step forward: The combined model correctly predicted 10% of intellectual disability cases among autistic children in the study cohort.
- Better risk stratification: It doubled the ability to distinguish between lower- and higher-risk children compared with typical clinical approaches, particularly among children with delayed milestones.
- Supports early intervention: Results suggest combining genetics with milestone data may enable more proactive, individualized support instead of waiting for later, clearer signs of ID.
Source: University of Montreal
Can clinicians predict whether a child assessed for autism will later be diagnosed with an intellectual disability? New research from Quebec points to a promising approach.
Researchers affiliated with Université de Montréal analyzed 5,633 autistic children drawn from three North American cohorts to create a prognostic model that integrates a wide range of genetic variants with detailed records of developmental milestones. Their aim was to provide families and clinicians with earlier, evidence-based estimates of developmental trajectories so that targeted support can begin sooner.
Rather than using genetics only to explain an already-observed condition, the team used genetic profiles together with early developmental data to forecast future cognitive outcomes. The goal is to give clinicians and families more reliable, individualized guidance well before school age, when intellectual disability is typically diagnosed.
“Difficult to foresee the future”
“Signs of autism often appear between 18 and 36 months,” explained Dr. Vincent-Raphaël Bourque, the study’s first author and a PhD candidate at the Centre de recherche Azrieli du CHU Sainte-Justine. Yet predicting medium- and long-term development remains challenging.
Many families wonder whether their young child will later meet criteria for intellectual disability, which affects an estimated 10 to 40 percent of autistic children and usually becomes apparent around ages six to eight. Current clinical practice relies mainly on tracking milestones such as language acquisition and walking, but these alone do not provide consistently accurate long-term predictions—especially for very young children. That often leads to a passive “wait and see” approach while developmental needs and support requirements diverge.
The authors argue that earlier, individualized intervention would be preferable. To do that responsibly, clinicians need tools that can estimate a child’s likely challenges—ideally with clear measures of certainty—so families can make informed decisions sooner.
The new model attempts to fill that gap by combining genetic signals with developmental timing and regression events to produce clinically relevant risk estimates.
10 percent correctly predicted
By integrating many genetic variations—including some that single-gene diagnostic labs often consider to have limited clinical value—with detailed milestone histories, the model achieved an overall correct identification of approximately 10 percent of children who later received an ID diagnosis. In addition, the model improved the ability to separate lower- and higher-probability cases, particularly among children who showed significant delays in early milestones.
The full model produced an area under the receiver operating characteristic curve (AUROC) of 0.653, indicating modest but meaningful predictive performance that generalized across the three cohorts studied. The modest overall performance reflects that only a subset of children carried strong-effect genetic variants, very high polygenic scores, or marked developmental delays—factors that most clearly drive prediction.
Notably, combinations of genetic variants that are often treated as clinically marginal produced positive predictive values (PPVs) around 55% for identified high-risk cases and correctly flagged about 10% of those who developed ID. Adding polygenic scores improved negative predictive values (NPVs), helping to rule out higher risk in many children.
The researchers also emphasize the importance of communicating uncertainty: the model quantifies how often predictions are expected to be accurate or inaccurate, which is essential for shared decision-making with families. “It is crucial to explain both the degree of confidence and the limits of our predictions so parents can make informed choices,” Bourque said.
Senior author Dr. Sébastien Jacquemont noted that ongoing discovery of neurodevelopmental variants and advances in computational modeling mean predictive power should improve over time, and that broader access to affordable genetic testing will make these tools more practical in clinical care.
About this genetics and autism research news
Author: Jeff Heinrich
Source: University of Montreal
Contact: Jeff Heinrich – University of Montreal
Image: The image is credited to Neuroscience News
Original Research: “Genomic and Developmental Models to Predict Cognitive and Adaptive Outcomes in Autistic Children” by Vincent-Raphaël Bourque et al., published in JAMA Pediatrics (open access).
Abstract
Genomic and Developmental Models to Predict Cognitive and Adaptive Outcomes in Autistic Children
Importance
Early signs of autism are often visible between 18 and 36 months, but clinicians lack reliable tools to predict which children will later be diagnosed with co-occurring intellectual disability (ID).
Objective
To develop and validate models that predict ID among children diagnosed with autism by integrating genetic variants and developmental milestones.
Design, Setting, and Participants
This prognostic study built and tested models across three autism cohorts—SPARK, the Simons Simplex Collection, and MSSNG—using data from participants assessed for ID after age 6. Models were trained, cross-validated, and tested for generalizability. Analysis covered data from January 2023 to July 2024.
Exposures
Predictors included ages at early milestones, language regression, polygenic scores for cognitive ability and autism, rare copy number variants, and de novo loss-of-function and missense variants in constrained genes.
Main Outcomes and Measures
Model performance was evaluated using AUROC, positive predictive values (PPVs), and negative predictive values (NPVs) on out-of-sample data.
Results
The analysis included 5,633 autistic participants (4,574 male [81.2%]). Median age at autism diagnosis was 4 years (IQR, 3–7), and participants were assessed for ID at a median age of 11 years (IQR, 8–14). A total of 1,159 participants (20.6%) were diagnosed with ID. The integrated model produced an AUROC of 0.653 (95% CI, 0.625–0.681) and generalized across cohorts. While only a minority carried large-effect variants or marked delays, combinations of genetic variants not typically prioritized in clinical diagnostics achieved PPVs around 55% and identified roughly 10% of individuals who developed ID. Including polygenic scores mainly improved NPVs. Genetic-based stratification of ID risk was up to twofold better among children with delayed milestones versus those with typical development.
Conclusions and Relevance
Although many neurodevelopmental variants alone are insufficient for accurate prediction, combining multiple classes of genetic variation with developmental milestone data yields clinically useful, individualized predictions. These models could support earlier, targeted interventions for children at higher risk of intellectual disability.