Development and evaluation of uncertainty quantifying machine learning models to predict piperacillin plasma concentrations in critically ill patients

Table 5 GUH a priori classification performance of the ML and PopPK models

Model	Range	Precision	Specificity	Sensitivity	F1-score	Support
A priori
GBT new	Sub.	0.88	0.88	0.89	0.88	99
	Ther.	0.62	0.58	0.69	0.65	35
	Sup.	0.67	0.77	0.47	0.55	17
GP new	Sub.	0.88	0.88	0.85	0.87	99
	Ther.	0.53	0.47	0.60	0.56	35
	Sup.	0.56	0.58	0.53	0.55	17
PopPK	Sub.	0.71	0.60	0.98	0.82	99
	Ther.	0.50	0.89	0.11	0.19	35
	Sup.	0.83	0.94	0.29	0.43	17
A posteriori
GBT prev	Sub.	0.93	0.93	0.92	0.93	76
	Ther.	0.63	0.54	0.79	0.70	24
	Sup.	0.75	0.89	0.33	0.46	9
GP prev	Sub.	0.92	0.92	0.92	0.92	76
	Ther.	0.59	0.53	0.67	0.63	24
	Sup.	0.50	0.67	0.33	0.40	9
PopPK	Sub.	0.84	0.86	0.75	0.79	76
	Ther.	0.35	0.15	0.46	0.40	24
	Sup.	0.50	0.44	0.56	0.53	9

Subtherapeutic (Sub.): < 91.43 mg/L, Therapeutic (Ther.): \(\ge\)91.43 mg/L and < 160 mg/L, Supratherapeutic (Sup.): \(\ge\)160 mg/L. Support indicates the number of samples in that range. Bold indicates the best model for that metric and case

ISSN: 1472-6947