Table 3.

Mean absolute error of Quick Inventory of Depressive Symptomatology score estimation.

Model	HCa, mean (SD)	BDb patients, mean (SD)	BPDc patients, mean (SD)	Overall, mean (SD)	Significance of reduction in overall mean absolute error compared to reference models
					Population-level model (P value)	Fully personalized model trained on calibration data (P value)
Population-level model	4.86 (2.54)	4.74 (2.07)	6.43 (2.58)	5.27 (2.48)	—d	—
Fully personalized model using cross-validation validation over all data points	0.80 (0.76)	2.27 (1.44)	3.05 (1.67)	1.94 (1.60)	<.001	<.001
Fully personalized model trained on calibration data	1.06 (0.75)	3.67 (2.60)	4.38 (2.43)	2.90 (2.49)	<.001	—
Clustering based on Lane et al [15] and Abdullah et al [16]	5.33 (3.11)	4.50 (2.30)	6.15 (2.88)	5.29 (2.82)	—	—
Group-personalized model with optimized clusters	0.83 (0.52)	2.30 (1.96)	2.82 (1.28)	1.90 (1.60)	<.001	<.001
Group-personalized model with clusters allocated using calibration data	0.86 (0.46)	3.30 (3.09)	3.75 (2.07)	2.52 (2.47)	<.001	.02

^aHC: healthy control.

^bBD: bipolar disorder.

^cBPD: borderline personality disorder.

^dNot applicable.