Algorithms to predict cerebral malaria in murine models using the SHIRPA protocol

Malaria Journal

Table 2 Best predictive performance of different logistic regression models.

Score	Day	_auROC	S	E	PV+	PV-	Acur.	n	Cut off	Score Value
Item+P₄	4	79	68	89	95	47	73	38	80	-57
Item₅	5	92	64	100	100	64	78	59	95	-5
Item+P₅	5	96	81	100	100	77	88	59	75	-46
Item₆	6	94	76	100	100	73	86	57	90	-37
Item+P₆	6	95	77	100	100	73	86	57	90	-1244
Item²⁴	5	77	25	100	100	95	95	61	40	-11
Item+P²⁴	5	85	25	100	100	95	95	61	55	-37
Item⁴⁸	6	91	14	100	100	89	89	56	85	250
Item+P⁴⁸	6	93	14	100	100	89	89	56	80	123
Item⁷²	4	82	46	100	100	77	80	37	65	-16
Item+P⁷²	4	82	61	100	100	82	86	37	65	-83

Best predictive performance of different logistic regression models for correctly select mice that will develop CM in C57Bl/6 animals infected with P. berghei ANKA. Values are expressed in percents (%). auROC = area under the receiver operator characteristic; Se = sensitivity; Sp = specificity; PV+ = positive predictive value; PV- = negative predictive value; n = total of mice analysed; Cut-off = estimated probability of developing CM above which the score is considered positive; Score Value = score value corresponding to the respective cut off and above which a mouse can be considered positive in the score. Item_x = Item scores designed for prediction of CM using the most important prognostic factors of SHIRPA individual scores on days 5 - 6 of infection; Item+P_x = Item scores associated with parasitaemia on days 4 - 6; Item^y = Item scores for 24, 48, 72 or 96 hours prediction designed using the most important prognostic factors of SHIRPA individual scores; Item+P^y = Item scores associated with the levels of parasitaemia in the corresponding time.

ISSN: 1475-2875