Tool for Converting ADHD Rating Scales Scores Based on Individual Participant Data from 53 Randomized Controlled Trials of ADHD Medications.

作者：Christos Christogiannis, Miguel Garcia-Argibay, Anneka Tomlinson, Sulagna Roy, Luís C. Farhat, Guilherme Fusetto Veronesi, Valeria Parlatini, Alessio Bellato, Corentin J. Gosling, Dimitris Mavridis, Orestis Efthimiou, Edoardo G. Ostinelli, Andrea Cipriani, Samuele Cortese · 发表于：Open Access CRIS of the University of Bern · 年份：2026 · DOI：10.48620/98758 · 研究领域：Attention Deficit Hyperactivity Disorder、Autism Spectrum Disorder Research、Behavioral and Psychological Studies

Introduction A variety of rating scales are currently being used to assess symptom severity and quantify symptoms change in attention-deficit/hyperactivity disorder (ADHD) research and clinical practice. This poses difficulties in interpreting scores from different scales in clinical practice and synthesizing data from studies using different scales. We aimed to develop algorithms for converting scores across the ADHD scales most often used in randomized controlled trials (RCTs) of ADHD medications in children/adolescents and adults, and to develop an online tool for implementing the algorithms.Methods We analyzed individual participant data from RCTs of ADHD medications (32 RCTs in children/adolescents, 21 in adults), with data on at least two scales per participant at the same timepoint. We applied a series of competing models, that is, univariable and multivariable regression, random forests, and an equipercentile linkage approach, to link pairs of scales. To assess the error of the linking procedure and identify the optimal model, we calculated the median absolute error and R2 of all approaches by comparing the values predicted from the models to the observed ones. We subsequently developed a tool to implement the best algorithms.Results We linked six commonly used ADHD scales, such as the ADHD Rating Scale (ADHD-RS-IV; investigator-rated) and the Conners' Parent Rating Scale (CPRS-R:S). Spline models most frequently yielded the lowest prediction error, outperforming alte...