Early Screening for Developmental Language Disorder: Predictive Validity of German Parent Reports

Abstract

ABSTRACT Background Developmental Language Disorder (DLD) is a prevalent neurodevelopmental condition associated with long‐term cognitive, academic, and psychosocial consequences. Early identification is critical for timely intervention, yet predictive frameworks within German‐speaking contexts remain limited. Parent‐report questionnaires such as the FRAKIS‐K and SBE‐2‐KT are widely used in the German paediatric U7 check‐up for two‐year‐olds, but their predictive validity for DLD has not been systematically evaluated. Aims This study assessed the predictive validity of the FRAKIS‐K and SBE‐2‐KT administered at age two for identifying risk for DLD at age four. A secondary objective was to test whether adjusting the screening cut‐off values could enhance predictive accuracy. Methods and Procedures A community sample of 500 families in Hanover, Germany, was contacted through the city register; 186 children (mean age = 2;0) participated at the first assessment. At age four (N = 145), expressive vocabulary and grammar were measured using the Potsdam‐Illinois Test of Psycholinguistic Abilities (P‐ITPA). Children scoring ≤ –1.5 SD on P‐ITPA subtests were classified as having language impairment consistent with risk for DLD. Predictive validity indices—sensitivity, specificity, positive and negative predictive values, relative improvement over chance (RIOC), and area under the curve (AUC)—were calculated for both parent‐report tools. ROC analysis was applied to explore optimal cut‐off adjustments. Outcomes and Results Prevalence of language impairment at age four was 5.5%. The SBE‐2‐KT showed moderate predictive power (AUC = 0.85; RIOC = 0.51), with sensitivity of 57% and specificity of 89%. The FRAKIS‐K vocabulary alone performed similarly (AUC = 0.83; RIOC = 0.58), achieving sensitivity of 63% and specificity of 91%, while including grammar criteria reduced predictive validity. ROC analysis indicated improved sensitivity (up to 71%) with minimal specificity loss when adjusting cut‐offs. Conclusion and Implications Both FRAKIS‐K vocabulary and SBE‐2‐KT demonstrate adequate predictive validity for identifying risk of DLD at age two within the German U7 screening context. They function effectively as first‐tier tools to rule out children not at risk, though positive screens should be followed by comprehensive assessments. The study's longitudinal design and evaluation of clinically relevant screening instruments using multiple predictive validity indices strengthen the findings. While refining vocabulary‐based cut‐offs may improve early detection, the small number of children with language impairment in the sample limits the generalizability of optimization suggestions, highlighting the need for further research in larger clinical populations. WHAT THIS PAPER ADDS What is already known on this subject Parent‐report questionnaires are commonly used for early language screening, but their predictive validity for DLD is often limited. In German‐speaking contexts, evidence on the predictive accuracy of brief tools such as FRAKIS‐K and SBE‐2‐KT is scarce. What this study adds to the existing knowledge This longitudinal study shows that both FRAKIS‐K (vocabulary component) and SBE‐2‐KT at age two moderately predict outcomes of language impairment at age four. Adjusting the screening cut‐offs increased sensitivity while maintaining acceptable specificity, indicating that predictive validity may be improved through threshold refinement. However, as the sample included few children with risk for DLD, recommendations for optimizing these tools should be interpreted with caution. What are the potential or actual clinical implications for this work? FRAKIS‐K vocabulary and SBE‐2‐KT can be used as first‐tier screening instruments during the U7 paediatric check‐up in Germany. Using adjusted cut‐offs may enhance early identification of children at risk for DLD, supporting earlier referral and intervention, though further validation in larger samples is required.