A multimodal predictive framework for early detection of anxiety instability and panic attacks

Although anxiety disorders and panic attacks belong to the types of the most widespread and, accordingly, the most debilitating mental disorders of the world, the current clinical practice is largely based on retrospective self-report and episodic assessment, which does not always include early warning signs of behaviour and physiological indicators. More recent evidence suggests that slight changes in autonomic regulation, namely decreases in heart-rate variability (HRV) is antecedent to the increment of anxiety and the beginning of the panic attacks. A combination of wearable sensing, digital phenotyping and machine-learning can be used to continuously track psychological vulnerability in the real world. The current research suggests using an algorithmic approach to digital predictive analysis of the changes in the intensity of anxiety and the onset of a panic attack based on continuous multimodal state measurements. This model combines physiological measures, such as heart rate, HRV, respiration, electrodermal activity and behavioural dynamics, linguistic cues and exposure to environmental stressors to produce a personalised Anxiety Stability Index (ASI). The hybrid Random Forest-Long Short -Memories model also reveals the nonlinear correlations and the time-specific trends, which in turn predicts the instability at hand and the pace of the anxiety development. The principles of multimodal fusion and interpretable modelling will be used to improve transparency and clinical usefulness. The given framework would aim at determining the typical patterns of instability prior to the onset of acute panic, transforming the unchanging questionnaire-based measurements into the actual predictions. It can promote digital psychiatry because it encourages proactive treatment, elimination of panic cases in case of an emergency, and enhancement of the provision of individualised mental-health services in ordinary environments.