.png){kind=icon}
.png){kind=icon}
Return
Identifying Transdiagnostic Predictors of Depression across Psychoses: Informing Stratified Antidepressant Treatments
.png){kind=icon}
OA
AI
S
F
J
G
E
N
L
L
C
A
R
W
J
A
A
P
N
DOI：10.1159/000551070.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Introduction: Depression frequently co-occurs with psychosis and is associated with poor outcomes. Early identification of patients at risk of persistent depression remains challenging, limiting opportunities for stratified treatment planning. This study aimed to evaluate the transdiagnostic generalizability of machine learning (ML) models predicting depressive episodes across the affective-psychotic spectrum and whether model-informed predictions could identify patients who may benefit from antidepressant treatment. Methods: Support vector machine models were trained to predict depressive episodes within 6 months using clinical and physiological data from two large, multisite first-episode psychosis (FEP) trials: EUFEST (n = 447) and RAISE-ETP (n = 288), totalling 735 participants. A nested cross-validation framework was used to evaluate model performance. Generalizability was tested in psychotic depression (PD) patients from the STOP-PD trial (n = 142), which compared olanzapine plus sertraline versus olanzapine plus placebo. Results: Models predicted depressive episodes in the FEP sample with a balanced accuracy (BAC) of 69% (sensitivity: 65.7%, specificity: 72.4%). When applied to STOP-PD patients treated with olanzapine plus placebo, FEP-trained models achieved a BAC of 65.2% (sensitivity: 58.3%, specificity: 72.0%) in predicting 3-month non-remission. In the olanzapine plus sertraline group, predictions were at chance levels (BAC: 47.2%, sensitivity: 48.4%, specificity: 46.0%), reflecting sertraline's therapeutic effects. Conclusion: ML models can identify shared risk signatures for depression across the psychosis-affective spectrum. Patterns of depressive episodes in FEP patients share predictive features with PD patients not receiving antidepressants, while adjunctive antidepressant treatment improves remission outcomes beyond model expectations. These findings support ML-informed treatment stratification to identify patients unlikely to benefit from antipsychotic monotherapy. Depression is common in people experiencing psychosis, and when both conditions occur together, recovery is often more difficult. Doctors would benefit from tools that can identify early on which patients are likely to continue experiencing depression, so that treatment can be adjusted sooner. In this study, we tested whether computer-based models could help predict which patients will still have depression months after their first episode of psychosis. We used information from two large studies of people experiencing their first episode of psychosis. These studies collected clinical information as well as simple physiological measures, such as blood lipids. Using these data, we trained ML models; a type of computer models that looks for patterns in large datasets and uses those patterns to make predictions about new cases. After the models were trained, we tested whether they would work in a different group of patients - individuals with psychotic depression. Psychotic depression refers to major depression occurring together with symptoms such as delusions or hallucinations. In this group, the models were able to correctly identify many of the patients who did not improve when they were receiving antipsychotic medication alone. However, when patients were treated with a combination of an antipsychotic and an antidepressant, the models no longer predicted outcomes well because the antidepressant treatment helped more patients recover. These findings suggest that computer models may help identify patients who are less likely to recover with standard treatment, supporting earlier decisions to add or adjust treatments to improve overall outcomes.
Keywords:
First-episode psychosis
Depressive symptoms
Psychotic depression
Machine learning
Treatment stratification
Antidepressants
Clinical prediction
Transdiagnostic
Precision psychiatry
Journal
IF:
17.4
Papers:
2.6K
Citations:
6.5K
Organization
Cited Papers