The use of hybrid statistical models to chart the heterogeneity of internalizing problem behavior in child- and adolescent psychiatric patients

Abstract

s ingediend voor het Amsterdam Kindersymposium 2013 97 The use of hybrid statistical models to chart the heterogeneity of internalizing problem behavior in childand adolescent psychiatric patients Mark Patrick Roeling (1), Gwen C. Dieleman (1), Jan van der Ende (1), Frank C. Verhulst (1), Danielle Posthuma (1), Danielle Posthuma (2), Tinca J.C. Polderman (2), Gitta H. Lubke (3) (1) Erasmus MC departement kinder en jeugdpsychiatrie (2) VU Amsterdam dep Functionele Genoomanalyse (3) University of Notre Dame en VU Amsterdam dep Biologische Psychologie INTRODUCTION Good phenotyping becomes increasingly important for genetic studies, as diff erences on a phenotypic level might be caused by diff erences in genetic architecture. Unobserved heterogeneity has to be inferred from the data with latent variable models such as latent class analysis (LCA) or factor analysis (FA). The drawback of using these models separately is that it is unclear whether the symptomatology defi nes a continuous or categorical structure. METHODS Here, we apply an innovative hybrid statistical method (Factor Mixture Modeling; FMM) to investigate heterogeneity in Child Behavior Checklist syndrome scales anxious depressed (AD), withdrawn depressed (WD) and somatic complaints (SC). FMM combines LCA and FA into one statistical model, testing whether the patient’s symptoms refl ect diff erences on a quantitative or a qualitative level. Data from a child-psychiatric outpatient sample of the Sophia Child Hospital. Data were available for 3271 (2600 diagnosed) children. RESULTS Per syndrome scale, a two factor model was created based on Eigenvalues and factorloadings. Within the specifi ed model, the FMM with continuous indicators and two latent classes fi tted best for AD and SC, whereas a FMM with three classes fi tted best in WD. All FMMs had better fi t indices with less parameters and revealed quantitatively diff erent classes. Class membership is a good predictor of internalizing disorders (mood, anxiety, somatoform, eating) and autism. High scores lower the chance of an externalizing disorder. CONCLUSION The outcomes indicate internalizing problem behavior refl ects a continuum, underlining the importance of dimensional measures in diagnostic instruments as diff erences between patients result from fl uctuations in symptom severity.