Improving prediction accuracy of hospital arrival vital signs using a multi-output machine learning model: a retrospective study of JSAS-registry data

Table 2 Comparison of predicted vital sign values at hospital arrival with actual measurements across different models

	Variable	HR	SBP	DBP	RR	GCS
HEMS contact values	MAE	7.3	18.8	13.5	2.9	0.38
	R²	0.63	0.1	-0.18	0.1	0.84
	SD	11	25.9	18.4	4.3	1.3
	Spearman correlation	0.83	0.65	0.52	0.54	0.93
Change-based predicted values	MAE	13.9	29.8	22.8	5.9	0.99
	R²	-0.68	-1.5	-2.7	-2.7	0.42
	SD	23.5	43	32.5	8.7	2.4
	Spearman correlation	0.68	0.51	0.37	0.42	0.77
NN predicted values	MAE	7.1	15.7	10.8	2.9	0.62
	R²	0.69	0.42	0.29	0.32	0.84
	SD	10	20.8	14	3.9	1.4
	Spearman correlation	0.83	0.68	0.55	0.54	0.86

Table 2 compares three methods for predicting hospital arrival vital signs: (1) HEMS contact values (initial HEMS measurements), (2) change-based predictions (extrapolated from GEMS and HEMS data), and (3) NN predictions from our multi-output regression model. Performance metrics (MAE, R², SD of residuals, and Spearman's correlation) were calculated by comparing each method’s predictions with the actual hospital arrival values. HR, heart rate; SBP, systolic blood pressure; DBP, diastolic blood pressure; RR, respiratory rate; GCS, Glasgow coma scale; MAE, mean absolute error; R², R-squared; SD, standard deviation; Spearman Correlation, Spearman's rank correlation coefficient

ISSN: 1471-227X