On Scene Injury Severity Prediction (OSISP) model for trauma developed using the Swedish Trauma Registry

Table 4 Model performance for different points on the ROC curve when predicting the risk of severely injured (NISS > 15)

Model			LR				RF				XGBoost				SVM				ANN
Dataset	A	B	C	D	A	B	C	D	A	B	C	D	A	B	C	D	A	B	C	D
Undertriage^a	1	1	1	1	1	1	1	1	1	1	1	1	1	1	1	1	1	1	1	1
Overtriage	67	67	67	66–66	1	1	1	1–1	64	60	64	59–60	90	86	91	85–86	66	65	68	63–64
Precision	20	22	22	23–23	15	17	16	17–17	21	25	23	25–25	16	19	17	19–19	21	24	22	24–24
Undertriage^a	5	5	5	5	5	5	5	5	5	5	5	5	5	5	5	5	5	5	5	5
Overtriage	49	47	49	47–48	1	55	1	54–55	46	43	45	43–44	70	61	66	60–62	49	47	50	45–47
Precision	25	29	27	28–29	15	26	16	26–26	26	31	28	31–31	19	24	21	24–24	25	29	26	29–30
Undertriage	11	10	11	10–10	17	13	16	14–14	10	8	10	8–9	17	14	17	15–15	13	11	12	10–11
Overtriage^a	35	35	35	35	35	35	35	35	35	35	35	35	35	35	35	35	35	35	35	35
Precision	31	35	33	34–34	29	33	31	33–34	30	34	33	34–35	29	33	31	33–33	30	34	32	34–35
Undertriage	19	17	19	17–17	25	22	24	22–22	18	16	18	16–16	24	21	24	22–22	21	19	21	18–18
Overtriage^a	25	25	25	25	25	25	25	25	25	25	25	25	25	25	25	25	25	25	25	25
Precision	36	40	38	40–40	34	39	36	39–39	37	41	38	41–41	34	39	36	39–39	36	40	37	39–40

Values are presented as percentage and ^adenotes fixed metric values. Dataset D presented with an interval across the five imputed datasets (D1–D5)

ISSN: 1472-6947