Abstract
Diagnostic heterogeneity is a longstanding challenge for the classification of mental illness because people who receive the same diagnosis can differ markedly in their symptom presentations as well as in the mechanisms underlying those symptoms. High heterogeneity undermines both the clinical utility of diagnostic labels and what we can learn from research that treats disorders as coherent constructs. In this study, we examined heterogeneity across six internalizing disorders (specific phobia, social anxiety disorder, agoraphobia, generalized anxiety disorder [GAD], major depressive disorder [MDD], and posttraumatic stress disorder [PTSD]) using three complementary approaches to quantifying heterogeneity: 1) descriptive heterogeneity (how many symptom profiles can theoretically satisfy the diagnostic criteria, and how many are observed in practice), 2) person-focused heterogeneity (whether diagnostic labels capture people with similar symptom presentations), and 3) construct-focused heterogeneity (whether diagnostic labels group symptoms with similar patterns of association with external validators). Analyses drew on subsamples from a large convenience sample with elevated psychopathology, predominantly from Australia and the USA (n = 14,762, mean [SD] age = 38.0 [18.0]; 55% women). Results showed that specific phobia was consistently the least heterogeneous; social anxiety disorder, agoraphobia, and GAD had mixed results; and MDD and PTSD were highly heterogeneous across all metrics. Notably, empirically derived phenotypes nested within each disorder substantively reduced heterogeneity relative to the DSM diagnoses. The study provides both a systematic multi-method approach to characterising different types of diagnostic heterogeneity, and a practical path toward more precise and informative constructs that can be implemented by rescoring existing data.