The multiverse of developmental dyslexia subtyping: The "Mapping Dyslexia Subtypes" (MAP-Dys) app for analytical decision-making

Biomedical researchers frequently use subtyping methods to classify individuals into subgroups based on shared attributes, aiming to design more precise and targeted interventions and improve treatment effectiveness. However, variations in subtyping methodologies, arising from differences in theoretical frameworks, data preprocessing, and model specifications, can lead to inconsistent subgroup identification. To date, we know little about such variability which may affect the comparability of findings across studies. Our study aims to help researchers systematically review and reflect on the variability in subtyping methods, in order to address this issue. Using developmental dyslexia subtyping as a case study, we systematically reviewed the analytical decisions used in the literature listed in four databases and created a multiverse of dyslexia subtyping methods. We found considerable variability in key decision points, including theoretical models, data preprocessing (e.g., score standardisation), performance indices (e.g., accuracy vs. reaction time), statistical methods (e.g., cluster analysis, quantile classification), and subgroup prevalence. Next, we developed a Shiny app, the "Mapping Dyslexia Subtypes" (MAP-DyS) app, which can be used to visualise and interactively explore the multiverse of subtyping methods, enabling users to understand the variability in methodological choices. By making the analytical variability transparent and navigable, our app supports researchers in critically assessing existing practices and making more informed decisions when designing or interpreting subtyping studies. While focused on developmental dyslexia, the framework has broader applicability to subtyping research in biomedicine.