Table 3 Prediction performance of the trained models on classifying 52 week outcome.

	N subjects	N sites	TP	TN	FP	FN	Sens (%)	Spec (%)	PPV	NPV	PSI	LR +	LR-	BAC	AUC	p-value
Pooled CV classifier EUFEST	338	46	47	187	73	31	60.3	71.9	39.2	85.8	24.9	2.1	0.6	66.1	0.75	<0.002
Validation EUFEST classifier in PSYSCAN sample	226	15	93	34	79	20	82.3	30.1	54.1	63.0	17.0	1.2	0.6	56.2	0.64	–
Pooled CV classifier PSYSCAN	226	15	79	67	46	34	69.9	59.3	63.2	66.3	29.5	1.7	0.5	64.6	0.70	<0.002
Validation PSYSCAN classifier in EUFEST sample	338	46	78	0	260	0	100.0	0.0	23.1	–	–	1.0	–	50.0	0.62	–
Leave-site-out / inner pooled CV classifier merged sample^†	433	25	101	216	73	43	70.1	74.7	58.0	83.4	41.4	2.8	0.4	72.4	0.79	<0.002
Leave-site-out / inner pooled CV classifier EUFEST sample^†	218	15	23	152	24	19	54.8	86.4	48.9	88.9	37.8	4.0	0.5	70.6	0.76	<0.002
Leave-site-out / inner pooled CV classifier PSYSCAN sample^†	210	12	51	73	38	48	51.5	65.8	57.3	60.3	17.6	1.5	0.7	58.6	0.65	0.086

Note. Classified poor outcomes (GAF < 65) are labeled as positive predictions and good outcomes (GAF ≥ 65) as negative predictions, i.e. sensitivity measures the classifier’s ability to correctly identify patients with poor outcomes as such. In all models, mean offset correction was applied as an extra preprocessing step. ^† Sites that included <10 participants were excluded from the analysis (see Suppl. Table 2). CV Cross-validated. TP True Positives. TN True Negatives. FP False Positives. FN False Negatives. Sens Sensitivity. Spec Specificity. PPV Positive Predicted Value. NPV Negative Predicted Value. PSI Prognostic Summary Index. LR+ Positive Likelihood Ratio. LR- Negative Likelihood Ratio. BAC Balanced Accuracy. AUC Area Under the Receiver Operating Characteristic Curve.

Quick links

Search