Should child pain anxiety be modeled as a multidimensional latent construct or as a network of interrelated symptoms? Apsychometric comparison in spanish clinical and community populations

Pain anxiety is a clinically significant psychological process in children and adolescents, associated with greater functional disability and poorer adjustment beyond pain intensity. How this construct is conceptualised and assessed has direct implications for clinical evaluation and intervention. A long-standing question is whether the symptoms captured by pain anxiety mesures reflect a common latent construct or a system of directly interacting, conditionally associated symptoms. This distinction has both theoretical and clinical implications but has not been examined in a pediatric pain instrument. Using the Spanish version of the Child Pain Anxiety Symptoms Scale (CPASS), we directly compared latent variable and psychometric network models to examine how pain anxiety in children is best understood and measured. In 415 children and adolescents aged 8–18 years, were recruited from clinical (n = 195) and community (n = 220) settings. We applied a fully cross-validated analytic pipeline conducting exploratory factor and exploratory network analysis in an initial random split, followed by confirmatory factor and confirmatory network analysis in an independent held-out split. Four competing latent models (one-factor, two-factor, four-factor and second-order) and a confirmatory network model were compared using identical data and standard fit indices. Measurement invariance and network-structure invariance were examined across sex, age and clinical status. Two estimation packages (lavaan and psychonetrics) were also contrasted for ordinal data. The choice of psychometric framework proved consequential. The confirmatory network model provided the best relative fit to the observed covariance structure among all tested specifications with higher CFI and TLI, lower RMSEA with non-overlapping confidence intervals, and lower AIC. Several items that would have been excluded under a four-factor solution emerged as clinically meaningful bridge symptoms within the network, and both measurement and network invariance broke down between clinical and community samples—conclusions a conventional CFA-only evaluation would have missed. Within the multidimensional latent framework, a parsimonious two-factor structure (ω = 0.90) showed the strongest cross-sample congruence. Finding suggest that, pediatric pain anxiety symptoms were better represented statiscally by a network of conditionally associated indicators than by latent variable models, although this finding pertains to the data representation and does not in itself establish causal relationships among symptoms. Highly central and bridge symptoms may represent promising targets for clinical assessment and intervention. The Spanish CPASS demonstrated good psychometric properties and can be parsimoniously represented by two-factor structure Fear and Catastrophizing about Pain; Anxiety and Avoidance Response to Pain. Nevertheless the breakdown of measurement and network invariance between clinical and community samples means that scores require contextualised interpretation across these populations. More broadly, the findings demonstrated the value of comparing latent and network modelsather than defaulting to confirmatory factor analysis.