• HOME
  • NEWS
  • EXPLORE
    • CAREER
      • Companies
      • Jobs
    • EVENTS
    • iGEM
      • News
      • Team
    • PHOTOS
    • VIDEO
    • WIKI
  • BLOG
  • COMMUNITY
    • FACEBOOK
    • INSTAGRAM
    • TWITTER
Sunday, November 22, 2026
BIOENGINEER.ORG
No Result
View All Result
  • Login
  • HOME
  • NEWS
  • EXPLORE
    • CAREER
      • Companies
      • Jobs
        • Lecturer
        • PhD Studentship
        • Postdoc
        • Research Assistant
    • EVENTS
    • iGEM
      • News
      • Team
    • PHOTOS
    • VIDEO
    • WIKI
  • BLOG
  • COMMUNITY
    • FACEBOOK
    • INSTAGRAM
    • TWITTER
  • HOME
  • NEWS
  • EXPLORE
    • CAREER
      • Companies
      • Jobs
        • Lecturer
        • PhD Studentship
        • Postdoc
        • Research Assistant
    • EVENTS
    • iGEM
      • News
      • Team
    • PHOTOS
    • VIDEO
    • WIKI
  • BLOG
  • COMMUNITY
    • FACEBOOK
    • INSTAGRAM
    • TWITTER
No Result
View All Result
Bioengineer.org
No Result
View All Result
Home NEWS Science News Biology

Machine Learning Tracks Preterm Infant Development to Gauge Autism Risk

by
November 21, 2026
in Biology
Reading Time: 6 mins read
0
Machine Learning Tracks Preterm Infant Development to Gauge Autism Risk

Machine Learning Tracks Preterm Infant Development to Gauge Autism Risk

Share on FacebookShare on TwitterShare on LinkedinShare on RedditShare on Telegram

Researchers tracking infants born before 32 weeks of gestation have found that early developmental trajectories across cognitive, communication, and motor domains can help identify children at a low likelihood of developing autism spectrum disorder by age five. In a longitudinal, population-based cohort study conducted in southern Taiwan, a multidisciplinary research team evaluated domain-specific developmental paths during the first two years of life using standardized assessments. By integrating these developmental patterns with neonatal medical risk factors into machine learning classifiers, the investigators demonstrated that tracking milestones through 24 months of corrected age improves risk stratification compared to relying on neonatal risks alone. The resulting models exhibited high negative predictive value, pointing to an objective surveillance approach that could help clinical follow-up programs prioritize specialized evaluation resources while reassuring families of infants demonstrating low statistical risk.

Very preterm infants face an elevated likelihood of neurodevelopmental impairments, making structured long-term follow-up essential for evaluating clinical care quality and supporting early childhood development. Children born very preterm show an autism spectrum disorder prevalence between seven and eight percent, compared to two to three percent in the general pediatric population, with some recent cohorts reporting rates ranging from 12.7 percent to 20.8 percent. Despite this heightened susceptibility, early identification of autism spectrum disorder in very preterm toddlers remains challenging. Standard screening tools designed for general populations frequently lack sufficient sensitivity and specificity when applied to preterm children, whose developmental presentations are often complicated by co-occurring delays and vulnerabilities across multiple neurodevelopmental domains. While earlier research connected low or declining cognitive trajectories between six and 24 months to increased autism risk at age five, the precise multi-domain trajectories across communication and motor abilities had remained insufficiently characterized in very preterm populations.

To address this diagnostic challenge, the investigators conducted a population-based multicenter study enrolling infants born at less than 32 weeks of gestation between January 2011 and December 2018 across a defined geographic area in southern Taiwan. Among 914 very preterm infants admitted to participating neonatal intensive care units, 785 survived to discharge. Long-term developmental follow-up was completed through age five by 625 children, representing an 80 percent follow-up rate. Following the protocol’s exclusion criteria, children diagnosed with congenital syndromes, brain malformations, or severe sensorimotor impairments—specifically including blindness, hearing impairment, or severe cerebral palsy classified as Gross Motor Function Classification System level four or higher—were excluded to prevent primary sensory and physical disabilities from confounding the developmental and diagnostic assessments. The final analytical cohort comprised 583 children whose early development and preschool outcomes were systematically evaluated.

Neurodevelopmental progress was prospectively evaluated at six, 12, and 24 months of corrected age using the Bayley Scales of Infant and Toddler Development, Third Edition, administered by child psychologists and examined by pediatric neurologists. The assessment tool measured performance across five distinct domains: cognitive function, receptive communication, expressive communication, fine motor function, and gross motor function. Raw scores from each domain were transformed into age-appropriate standardized scaled scores with a normative mean of 10 and a standard deviation of three. At age five years, comprehensive diagnostic evaluations for autism spectrum disorder were conducted by a cross-disciplinary clinical team comprising pediatric neurologists, child psychologists, and child psychiatrists. Diagnoses were determined based on the Diagnostic and Statistical Manual of Mental Disorders, Fifth Edition, integrating standardized evaluations from the Autism Diagnostic Interview–Revised and the Autism Diagnostic Observation Schedule alongside clinical judgment. General intelligence was assessed at the five-year evaluation using the Wechsler Preschool and Primary Scale of Intelligence.

Among the 583 children evaluated at age five, 75 children, or 12.9 percent of the cohort, received a confirmed diagnosis of autism spectrum disorder. The group diagnosed with autism was significantly more likely to be male, accounting for 69 percent of cases compared to 49 percent in the group without autism. Aside from sex distribution, the two groups did not differ significantly in gestational age, birth weight, family socioeconomic status, maternal educational attainment, or proportions of severe neonatal morbidities at hospital discharge. At age five, children in the autism spectrum disorder group exhibited significantly higher scores in reciprocal social interaction, communication difficulties, and restricted, repetitive, and stereotyped behaviors on standardized diagnostic instruments. Furthermore, children diagnosed with autism spectrum disorder demonstrated a slightly lower median full-scale intelligence quotient of 91 compared to a median of 94 among children without an autism diagnosis.

Longitudinal analysis using locally estimated scatterplot smoothing revealed distinct domain-specific developmental trajectories that diverged well before the preschool diagnostic evaluations. Children later diagnosed with autism spectrum disorder demonstrated persistently lower cognitive performance throughout the first two years of life, with statistically significant differences evident at six months, 12 months, and 24 months of corrected age. In contrast, trajectories for gross motor function did not differ significantly between the two groups across the entire 24-month observation period. In receptive communication, expressive communication, and fine motor skills, performance was comparable between groups during the first year of life from six to 12 months; however, trajectories diverged sharply between 12 and 24 months. During this second year, infants later diagnosed with autism exhibited a significantly slower rate of skill acquisition, showing marked developmental divergence by 24 months of age across communication and fine motor domains.

The researchers observed that fine motor evaluation on the Bayley Scales of Infant and Toddler Development often involves interactive imitation tasks between the child and the examiner. Consequently, the reduced fine motor scores recorded during the second year of life in infants later diagnosed with autism spectrum disorder may partly reflect emerging social communication and interactive difficulties alongside manual dexterity challenges. Identifying this pattern highlights that general developmental surveillance tools routinely administered in neonatal follow-up clinics can capture meaningful behavioral phenotypes. Rather than relying solely on dedicated autism screening questionnaires that can struggle in preterm populations, clinicians can track longitudinal deviations across standard developmental scaled scores to gather early objective evidence of divergent developmental pathways.

To evaluate whether early developmental phenotypes could forecast five-year diagnostic outcomes, the researchers applied machine learning classifiers trained on neonatal risks and longitudinal assessment data. Feature selection was conducted using univariate logistic regression adjusted for false discovery rate, followed by the least absolute shrinkage and selection operator method to minimize multicollinearity. Six machine learning algorithms were optimized and tested using leave-one-out cross-validation: Decision Trees, Naïve Bayes, Support Vector Machines, K-Nearest Neighbors, Discriminant Analysis, and Ensemble Learning. Class weights and misclassification costs reflecting the 1:6.8 ratio of autism cases to non-autism cases were incorporated to account for dataset class imbalance and prevent models from prioritizing majority-class accuracy at the expense of sensitivity.

Model performance was compared across four feature sets: neonatal risks alone; neonatal risks combined with six-month developmental scores; neonatal risks combined with six- and 12-month scores; and neonatal risks combined with six-, 12-, and 24-month scores. When relying solely on neonatal variables—such as gestational age, sex, small-for-gestational-age status, and maternal education—predictive performance was low, with the top-performing K-Nearest Neighbors model achieving an accuracy of 60.2 percent and an area under the receiver operating characteristic curve of 0.57. Adding developmental scores at six and 12 months yielded only marginal enhancements. However, incorporating developmental scores through 24 months substantially improved predictive accuracy, resulting in a finalized 10-feature model comprising three neonatal factors and seven developmental scores across cognition, receptive and expressive communication, and motor domains.

Among the algorithms evaluated on the 10-feature set, the Support Vector Machine classifier achieved the strongest overall performance. It reached an overall accuracy of 71.8 percent and an area under the receiver operating characteristic curve of 0.69. Using an optimal classification threshold determined by maximizing Youden’s Index, the model achieved a sensitivity of 64.2 percent and a specificity of 72.9 percent. Notably, the model demonstrated a negative predictive value of 93.6 percent, alongside a positive predictive value of 24.7 percent. This statistical balance indicates that while the model has modest precision for confirming positive cases, it performs exceptionally well at identifying children who are at a low likelihood of receiving an autism diagnosis at preschool age.

The study authors highlighted several methodological limitations that provide context for interpreting these machine learning findings. Although the Support Vector Machine model demonstrated high negative predictive utility, its overall predictive performance remains modest, characterized by moderate sensitivity and a low positive predictive value that limits its reliability as an independent diagnostic tool. Furthermore, the model was trained and evaluated within a single regional cohort in southern Taiwan, which may restrict generalizability to healthcare systems with differing neonatal care practices or follow-up frameworks. The researchers also noted that autism diagnostic evaluations were conducted universally at age five, a timing chosen because earlier diagnostic evaluations in preterm toddlers aged two to four years frequently exhibit lower diagnostic stability, though clinical interventions were initiated whenever developmental concerns emerged during infancy.

Despite these limitations, the authors emphasized that the structured utilization of standardized Bayley Scales of Infant and Toddler Development scores offers a practical, accessible framework for neonatal follow-up programs worldwide. Because these developmental assessments are already routinely administered to very preterm infants in clinical practice, machine learning models could be integrated into clinical workflows as triage and risk-stratification aids rather than standalone diagnostic tools. By objectively identifying very preterm children who exhibit a low likelihood of developing autism spectrum disorder, such models could help clinicians allocate intensive diagnostic resources, such as specialized observation schedules, toward infants showing persistent developmental divergence, while providing evidence-based reassurance to families during early developmental follow-up.

Subject of Research: Machine learning prediction of autism spectrum disorder risk in very preterm infants using longitudinal developmental trajectories

Article Title: Early developmental trajectory phenotypes for risk stratification of autism spectrum disorder in very preterm infants: a machine learning approach

Article References: Chen, L.-W., Li, Y.-T., Chu, C.-H., Wu, C.-C., Chu, C.-L., Wang, L.-W., Tsai, H.-Y., Chiang, C.-H., & Huang, C.-C. (2025). Early developmental trajectory phenotypes for risk stratification of autism spectrum disorder in very preterm infants: a machine learning approach. Molecular Autism, 17(1), Article 3. https://doi.org/10.1186/s13229-025-00692-y

Image Credits: AI Generated

DOI: 10.1186/s13229-025-00692-y

Keywords: very preterm infants, autism spectrum disorder, machine learning, developmental trajectories, Bayley Scales, BSID-III, risk stratification, Support Vector Machine, neonatal follow-up, cognitive development, receptive communication, expressive communication, fine motor skills, cohort study, pediatric neurology, neurodevelopmental assessment

News Source: Harold Sullivan. (November 21, 2026). Machine Learning Tracks Preterm Infant Development to Gauge Autism Risk. Scienmag.

Tags: autism spectrum disorderBayley ScalesBSID-IIICognitive developmentCohort Studydevelopmental trajectoriesexpressive communicationfine motor skillsMachine Learningneonatal follow-upneurodevelopmental assessmentPediatric Neurologyreceptive communicationrisk stratificationsupport vector machinevery preterm infants
Share12Tweet7Share2ShareShareShare1

Related Posts

GEN1 Resolvase Protects Genome Integrity and Self-Renewal in Mouse Embryonic Stem Cells

GEN1 Resolvase Protects Genome Integrity and Self-Renewal in Mouse Embryonic Stem Cells

November 22, 2026
Engineered Vaccine Platforms Transform Disease Prevention and Targeted Immunotherapy Across Clinical Applications

Engineered Vaccine Platforms Transform Disease Prevention and Targeted Immunotherapy Across Clinical Applications

November 22, 2026

Single-Cell Sequencing Uncovers RNA Methylation Pathway Driving Kidney Cancer Metastasis

November 22, 2026

Lung Growth Factors Fail to Protect Stem Cells from Hyperoxia

November 21, 2026

POPULAR NEWS

  • West African Study Finds No Detectable Link Between Carbon Intensity and Mortality

    West African Study Finds No Detectable Link Between Carbon Intensity and Mortality

    29 shares
    Share 12 Tweet 7
  • GEN1 Resolvase Protects Genome Integrity and Self-Renewal in Mouse Embryonic Stem Cells

    29 shares
    Share 12 Tweet 7
  • Malawi Simulation Shows Health System Strengthening Outperforms Disease-Specific Healthcare Funding

    29 shares
    Share 12 Tweet 7
  • Leadership Support and Adapted Workflows Shape Rural Wisconsin Police Deflection Programs

    29 shares
    Share 12 Tweet 7

About

We bring you the latest biotechnology news from best research centers and universities around the world. Check our website.

Follow us

Recent News

Congo Ebola Outbreak Linked to Spillover Clustered Near 1976 Lineage

West African Study Finds No Detectable Link Between Carbon Intensity and Mortality

GEN1 Resolvase Protects Genome Integrity and Self-Renewal in Mouse Embryonic Stem Cells

Subscribe to Blog via Email

Enter your email address to subscribe to this blog and receive notifications of new posts by email.

Join 89 other subscribers
  • Contact Us

Bioengineer.org © Copyright 2023 All Rights Reserved.

Welcome Back!

Login to your account below

Forgotten Password?

Retrieve your password

Please enter your username or email address to reset your password.

Log In
No Result
View All Result
  • Homepages
    • Home Page 1
    • Home Page 2
  • News
  • National
  • Business
  • Health
  • Lifestyle
  • Science

Bioengineer.org © Copyright 2023 All Rights Reserved.