The research team, led by Yuheng He, Ping Long, and Rui Chen of the State Key Laboratory of Cognitive Neuroscience and Learning at Beijing Normal University, together with colleagues including Vince Calhoun of the Tri-institutional Center for Translational Research in Neuroimaging and Data Science in Atlanta, and corresponding authors Sha Tao and Jing Sui, set out to address a long-standing fragmentation in the literature. Evidence on the predictors of children’s academic achievement has accumulated in parallel silos: perinatal epidemiology, cognitive psychology, educational research on family and school contexts, and more recently educational neuroscience. Few syntheses have brought these traditions together within a common developmental framework, and fewer still have asked which predictors are shared across reading and mathematics and which are domain-specific.

To build the synthesis, the team organized evidence across several nested levels. At the broadest level sit perinatal and early-life conditions — birth weight, prenatal exposures, nutrition, and health status — which the review characterizes as distal starting conditions. These factors matter, the authors conclude, but their influence is largely indirect, setting the stage on which later capacities develop rather than determining academic outcomes outright. At the most proximal level sit cognitive skills: phonological awareness, vocabulary, morphological knowledge, number sense, symbolic numerical processing, and spatial abilities. These, the review finds, are the most consistent and powerful near-term predictors of achievement, because they are the very tools children deploy when learning to decode text or manipulate quantities.

Between these poles, the review places behavioral, emotional, and motivational processes — attention, self-regulation, anxiety, motivation, and self-beliefs — which it interprets as regulatory pathways through which cognitive capacities are expressed. A child with strong phonological skills who cannot sustain attention in the classroom, or a child with solid numerical understanding who is paralyzed by math anxiety, may fail to translate underlying competence into measured achievement. This framing carries practical weight: it suggests that intervening on regulation and affect may unlock cognitive potential that would otherwise remain untapped.

One of the review’s central contributions is its careful separation of shared from domain-specific prediction. Executive function and general cognitive ability emerge as a common scaffold supporting both reading and mathematics, helping to explain the well-documented correlation between achievement in the two domains. Beyond that shared core, however, the domains diverge sharply. Reading achievement is more consistently predicted by language-related and sound-symbol skills — phonological awareness, vocabulary breadth, morphological and semantic knowledge, and the ability to map sounds onto written symbols. Mathematics achievement, by contrast, is more consistently tied to numerical concepts, symbolic processing, spatial resources, and, notably, math anxiety, which appears as a distinctive emotional predictor in the mathematical domain.

The technical backbone of the review’s neuroimaging illustration is a multimodal data fusion approach grounded in independent component analysis. The supervised fusion algorithm, known as multimodal fusion with reference, integrates structural and functional brain data — such as gray matter morphology from structural MRI and intrinsic network connectivity from resting-state functional MRI — by decomposing each modality into spatially independent components and then identifying joint patterns that co-vary with a reference variable of interest, in this case academic performance. The method, originally developed to identify joint neuromarkers of cognitive deficits in clinical populations, allows researchers to detect convergent evidence across imaging modalities rather than relying on any single measure. The authors have released the algorithm through the freely available Fusion ICA Toolbox, making the approach accessible to labs worldwide.

Crucially, the authors frame the neuroimaging section as an illustrative application rather than an evidential extension of the systematic review. Brain-based measures, they argue, may eventually serve as child-proximal markers — indicators measured in the child that sit closer to the learning process than questionnaire-based family or school variables — but the current evidence base does not yet license strong causal or interventional claims. Neural predictors of reading ability, including findings linking frontoparietal connectivity to early math skill and striatal-prefrontal resting-state connectivity to reading performance, remain preliminary and often study-specific. The review is explicit about the interpretive limits of translating predictive evidence into intervention: knowing that a brain feature correlates with achievement does not mean that manipulating that feature, or remediating it directly, will improve learning.

This caution distinguishes the review from a growing body of “brain-based learning” claims in popular media and commercial education products. By calibrating the strength of evidence at each level of its framework, the authors offer researchers and practitioners a kind of confidence map: cognitive predictors rest on extensive, replicated longitudinal evidence; contextual predictors such as family socioeconomic status and home literacy environment show robust but more heterogeneous associations; and neuroimaging predictors remain exploratory. The framework thus serves a second function as a methodological guide, indicating where the field’s claims are well-supported and where they outrun the data.

The developmental architecture the review proposes is also notable for what it implies about timing. If early biological and contextual conditions are distal starting points, and cognitive skills are the proximal engines of achievement, then the highest-leverage interventions may differ depending on when they occur. In infancy and early childhood, addressing perinatal health, nutrition, and the home literacy and numeracy environment may shape the trajectory; once children are in school, strengthening domain-specific cognitive skills — phonological training for reading, number-magnitude mapping for mathematics — and supporting attention and emotional regulation may yield more direct returns. The review’s synthesis of longitudinal studies, some following children from preschool through adolescence, supports this staging logic.

The work also carries implications for identifying children at risk. Dyslexia and developmental dyscalculia research has long pointed to deficits in sound-symbol learning and visuo-spatial working memory respectively, and the review’s framework consolidates these findings within a broader predictive structure. Multilevel screening that combines cognitive measures with attentional and emotional indicators may identify struggling learners earlier and more reliably than any single measure, while neuroimaging markers — once validated — could add a biologically grounded layer to risk models without replacing behavioral assessment.

The study was conducted under the approval of the Beijing Normal University Ethics Committee and supported by China’s Brain Science and Brain-Inspired Intelligence Technology major project, the National Natural Science Foundation of China, and the Beijing Municipal Science and Technology Commission. The multimodal neuroimaging data can be accessed upon request from the corresponding authors, and the fusion code is freely downloadable, lowering the barrier for other teams to test whether the framework’s neural layer holds up across cohorts and cultures.

For a field often divided between those who see achievement as destiny written in biology and those who see it as the product of environment and instruction, the review offers a more disciplined synthesis: every level matters, but not equally at every moment, and not for every outcome. Reading and mathematics draw on a shared cognitive scaffold, yet each has its own most reliable predictors — language for one, numbers and space for the other — and the brain, for now, remains a promising but unproven source of markers. The framework gives educators, clinicians, and scientists a common map for the terrain ahead, and a reminder that prediction is not prescription.