Predicting psychosis based on clinical, neurocognitive, and linguistic factors

Abstract

Predicting the onset of psychosis is crucial for early intervention and improved outcomes. This review examines the current state of prediction models based on clinical, neurocognitive, and linguistic factors. Clinical predictors, including sociodemographic characteristics, family history, and subthreshold psychotic symptoms, have shown promise in identifying people at risk, and some models achieve concordance indices of 0.79-0.80 in external validation. Neurocognitive evaluation, particularly of verbal learning, processing speed, and attention/vigilance, has emerged as a cost-effective predictor, although the effect sizes remain modest. Recent advances in natural language processing have enabled automated analysis of speech patterns, with reduced semantic coherence and specific linguistic features predicting the transition to psychosis with precisions of up to 83%. Although these approaches show promise individually, the integration of multiple predictors may maximize predictive accuracy. Current limitations include small sample sizes in many studies, especially for linguistic analyses, and the need for broader population-level applicability beyond clinical high-risk groups. Dynamic prediction models that account for temporal changes in risk factors show improved performance over static approaches. More research is needed, particularly external validation studies in diverse populations, to develop comprehensive preventive strategies that can be implemented at the primary level. The field continues to evolve with emerging variables and advanced analytical methods, working toward an individualized application of prediction tools.