Expanding the predictive horizon in developmental coordination disorder screening: Promise and gaps

Developmental coordination disorder (DCD) remains a significantly underdiagnosed neurodevelopmental condition, which has long-term impact on children's functional, academic, and psychosocial outcomes. Dai et al.'s research contributes to this field by introducing a data-driven approach for the early screening of DCD utilizing a large sample of preschool-aged children.1 This aligns with global efforts to integrate predictive analytics in paediatric care, offering both clinical relevance and visions for future research. The authors adopt the bio-ecological model to guide the integration of multifaceted factors, recognizing that child development is shaped by the interaction between individual characteristics and environmental systems over time. Under this theoretical foundation, the construction of the screening tools in this study has incorporated in itself a wide range of bio-psycho-social factors that shape children's growth and development trajectories. Systematically combining multilevel predictors not only improves the validity of the screening tools for identifying children at risk of DCD but also enhances the applicability of the screening tools to the real-world complexity beyond exclusively clinical markers. This study adopted an analytical technique, which is frequently used for identifying important predictors in clinical and epidemiological research, and supporting easy use in practice. By using the large data and the advanced analytical techniques, the model developed by Dai et al. achieved a satisfactory prediction capacity for DCD, with an area under the curve (AUC) of 0.70 and sensitivity and specificity exceeding 70% for children aged 3 to 5 years, potentially enabling early identification of at-risk children. There was a decrease in the model performance for those aged 6 to 10 years (AUC = 0.58), indicating that different predictive features may exist by children's age. Thus, future research is needed to enhance the accuracy and clinical utility by tailoring predictive features for different age groups. The model was developed using data from one country, China. Cross-cultural validation may help increase its utility for populations of different cultural backgrounds and healthcare systems. Additionally, most predictive features used in this study may have already been routinely collected in the healthcare systems, especially in some high-income countries. Therefore, the predictive tools, if validated in these countries, may have provided a useful venue for screening of children at risk of DCD based on these routine data collections. Future studies should focus on additional key risk factors, the lack of which may limit the model's predictive performance in Dai et al.'s research. These potentially important factors include genetic, behavioural, social, and built environment. These again need to be studied using prospectively collected data. In summary, this model can be thought of as a screening rather than a diagnostic tool, identifying children at risk of DCD that will require further detailed assessment.