Predicting the risk and timing of major mood disorder in offspring of bipolar parents: exploring the utility of a neural network approach

International Journal of Bipolar Disorders

Table 3 Mean accuracy, specificity, sensitivity, and positive predictive value (PPV) across time intervals and 10 CV-folds for 1-year, 3-year, and 5-year predictions for PLANN and the discrete survival model

Model	Threshold	Accuracy	Specificity	Sensitivity	PPV
One year
PLANN^a
	0.05	0.751	0.762	0.272	0.064
	0.10	0.948	0.986	0.019	0.051
	0.15	–	–	–	–
	0.20	–	–	–	–
	Optimal^b	0.681	0.683	0.488	0.081
Survival
	0.05	0.772	0.783	0.360	0.073
	0.10	0.895	0.924	0.186	0.100
	0.15	0.927	0.960	0.128	0.133
	0.20	0.940	0.976	0.083	0.160
	Optimal^b	0.556	0.560	0.383	0.043
Three year
PLANN
	0.05	0.269	0.186	0.822	0.107
	0.10	0.588	0.583	0.443	0.111
	0.15	0.762	0.808	0.257	0.167
	0.20	0.818	0.899	0.096	0.143
	Optimal^b	0.591	0.591	0.506	0.134
Survival
	0.05	0.270	0.181	0.791	0.101
	0.10	0.533	0.513	0.483	0.111
	0.15	0.760	0.813	0.218	0.125
	0.20	0.841	0.919	0.128	0.134
	Optimal^b	0.572	0.582	0.474	0.130
Five year
PLANN
	0.05	0.222	0.070	0.934	0.162
	0.10	0.425	0.335	0.709	0.169
	0.15	0.582	0.546	0.565	0.210
	0.20	0.687	0.722	0.360	0.213
	0.607	0.617	0.577	0.230	0.607
Survival
	0.05	0.264	0.116	0.893	0.161
	0.10	0.410	0.300	0.662	0.143
	0.15	0.548	0.511	0.508	0.160
	0.20	0.709	0.738	0.420	0.238
	Optimal^b	0.601	0.615	0.515	0.224

^aOne-year predictions made by PLANN were all below 0.15 and therefore, thresholds of 0.15 or greater could not be evaluated for accuracy metrics
^bThe Optimal rows present the average accuracy metrics across time intervals when the optimal threshold of each time interval is used (see Additional file 1)