Development and validation of a prognostic model and risk calculator for the estimation of bipolar-spectrum disorder risk in hospitalised adolescents with non-psychotic/non-bipolar mental disorders

We aimed to develop and validate a risk estimation model for developing bipolar-spectrum disorders (BSD) in psychiatrically hospitalized adolescents based on clinical characteristics, including putatively prodromal symptoms. Adolescent inpatients (ages = 12–18 years) with non-psychotic/non-BSD diagnoses were recruited for the Adolescent Mood Disorder and Psychosis Study (AMDPS), a longitudinal, prospective 5-year follow-up cohort. We assessed prevalence and severity of syndromal/subsyndromal psychopathology at baseline using the validated Bipolar Prodrome Symptom Interview and Scale–Prospective. We carried out machine learning analyses (Lasso-Cox regression analyses, LCR) to create a calculator to estimate the risk of developing BSD based on baseline demographic/comorbidity/illness/treatment characteristics. Of 105 adolescents (age = 15.6 ± 1.3 years, females = 72.4%), we observed that 18 developed BSD. The cumulated estimated risk of BSD was 5/22/29/36% at 1/2/3/4 years. BSD development was associated with presence of persistent depressive disorder (HR = 4.0, p < 0.018) at baseline, treatment with mood stabilizers (hazard ratio (HR) = 3.9, p = 0.006), and ADHD medications (HR = 3.3, p = 0.023). BSD development risk estimation calculator included the prevalence of inflated self-esteem/grandiosity (β = 0.83) and racing thoughts (β = 0.08) and the severity of overtalkativeness (β = 0.03) and increased energy (β = 0.04). For predicting BSD onset within the first 20 months, the area under the receiver operating characteristic curve (AUC) indicated acceptable to strong discrimination (cross-validation AUC = 0.72; bootstrap out-of-bag validation AUC = 0.86). Codes used in this study are provided in the R package “easy.glmnet”. In conclusion, in this prognostic model/calculator, presence and severity of subthreshold (hypo)mania-like symptoms conferred increased risk of BSD development in youth, informing preventive efforts to identify individuals at risk for BSD and improve their outcomes.