Table 6.
Prediction model updates
| Key publications | Sample (training set) | Key findings |
|---|---|---|
| Brodey et al. (2019) [118] | CHR (N = 182), CHR-C/FEP (N = 76), CHR-NC (N = 106) | The EPSI-SR tool achieved a PPV of 86.6% when combined with clinician-administered SIPS in differentiating psychosis |
| Ciarleglio et al. (2019) [14] | CHR (N = 199), CHR-C (N = 64), CHR-NC (N = 135) | Visual perceptual abnormalities, dysphoric mood, unusual thought content, disorganized communication, and violent ideation predicted conversion in the model, C-Index = 0.73 |
| Kegeles et al. (2020) [120] | CHR (N = 19), CHR-C (N = 7), CHR-NC = 12 | Striatal glutamate 1H MRS and visual perceptual abnormalities performed with an AUC of 0.87 in a multivariate regression model |
| Zhang et al. (2019) [113] | SHARP: CHR (N = 196), CHR-C (N = 51) at 24 months | The smartphone-based SHARP-RC achieved high discriminatory accuracy of predicting conversion to psychosis using four clinical predictors, AUC of 0.78 |
| Kristensen et al. (2021) [29] | Denmark: CHR (N = 110), CHR-C (N = 10) | Global FA in a multivariate prediction model was predictive of conversion after 12 months (sensitivity 0.70, specificity of 0.88, AUC of 0.87) |
| Worthington et al. (2020) [131] | NAPLS2: CHR (N = 417), CHR-C (N = 54) at 24 months | Inclusion of salivary cortisol into the original eight-predictor NAPLS Psychosis Risk Calculator improved its predictive accuracy by 7%, C-index 0.78 |
| Mongan et al. (2021) [109] | EU-GEI: CHR (N = 133), CHR-C (N = 49), CHR-NC (N = 84) | Model included proteomic and clinical predictors AUC 0.95 |
| Dickens et al. (2021) [107] | CHR (N = 263), CHR-C (N = 50), CHR-NC (N = 213) | CHR-C vs CHR-NC distinguished based on lipid profile in model with AUC 0.81 (95% confidence interval = 0.69–0.93) |
| Koutsouleris et al. (2021) [116] | PRONIA: CHR (N = 167), ROD (N = 167), CHR-C (N = 23), CHR-NC (N = 144), ROD-C (N = 3), ROD-NC (N = 164) | Alongside clinician input, model consisting of structural MRI, schizophrenia PRS, clinical and neurocognitive predictors achieved a balanced accuracy of 85.5% in predicting conversion among CHR and ROD |
| Cadenhead et al. (2020) [64] | NAPLS2: CHR (N = 543), CHR-C (N = 58), CHR-NC (N = 255) | CHR-C vs CHR-NC had slower startle response latency that was more predictive of conversion than clinical symptoms (AUC 0.65 vs 0.55) in female CHR participants |
| Perkins et al. (2020) [111••] | NAPLS2: CHR (N = 764), CHR-C (N = 80), CHR-NC (N = 248) | Incorporating PRS into NAPLS psychosis risk calculator contributed 15% risk prediction in Europeans and 7% in non-Europeans |
SHARP Shanghai at Risk for Psychosis, NAPLS North American Prodrome Longitudinal Studies, EU-GEI European Network of National Schizophrenia Networks Studying Gene-Environment Interactions, PRONIA Personalised Prognostic Tools for Early Psychosis Management, CHR clinical high risk, CHR-C clinical high risk converted, CHR-NC CHR non-converted, ROD recent onset depression, ROD-C ROD converted, ROD-NC ROD non-converted, PPV positive predictive power, PRS polygenic risk score