Using Machine Learning to Identify Latent Profiles of Somatosensory Amplification and Catastrophic Misinterpretation of Bodily Cues in Psychosomatic Patients

Abstract

The study aimed to identify distinct latent profiles of psychosomatic patients based on their levels of somatosensory amplification and catastrophic misinterpretation of bodily cues using a person-centered machine learning approach. This cross-sectional study involved a sample of psychosomatic patients recruited from clinical settings in South Africa. Participants were assessed using the Somatosensory Amplification Scale (SSAS) and the Body Sensations Questionnaire (BSQ). Statistical analysis was performed using Gaussian Mixture Modeling (Latent Profile Analysis) to identify unobserved subpopulations. Model selection was determined by fit indices, including the Akaike Information Criterion (AIC), Bayesian Information Criterion (BIC), and the Lo-Mendell-Rubin Likelihood Ratio Test (LMR-LRT). A strong significant positive correlation was observed between somatosensory amplification and the catastrophic misinterpretation of bodily cues ( , ). Latent Profile Analysis identified an optimal three-profile solution: “Low Symptom Burden” ( ; ), “Moderate Symptom Burden” ( ; ), and “Severe Symptom Burden” ( ; ). Comparative analyses revealed that the “Severe Symptom Burden” profile was characterized by a significantly higher proportion of female participants, higher unemployment rates, and longer illness durations ( ), whereas age and educational background did not significantly differentiate the groups. The identification of three distinct latent profiles underscores the clinical heterogeneity of psychosomatic populations and highlights the necessity for tailored, precision-based psychological interventions that address specific levels of cognitive-perceptual distortion.