Table 2.

Evaluation of initial oxygen saturation measured by paramedics in COVID-19 patients in the community used as a binary classifier for predicting 30-day intensive care unit admission within 30 days of conveyance

Oxygen saturation (on air) threshold (%)	Sensitivity (95% CI)	Specificity (95% CI)	Number of observations	Cumulative sum of number of observations
85	0.294 (0.200 to 0.400)	0.947 (0.933 to 0.962)	8	76
86	0.316 (0.216 to 0.421)	0.941 (0.927 to 0.955)	8	84
87	0.320 (0.216 to 0.432)	0.935 (0.920 to 0.950)	6	90
88	0.370 (0.261 to 0.476)	0.916 (0.899 to 0.933)	23	113
89	0.413 (0.304 to 0.523)	0.894 (0.874 to 0.913)	25	138
90	0.512 (0.411 to 0.615)	0.870 (0.849 to 0.890)	32	170
91	0.590 (0.477 to 0.699)	0.841 (0.823 to 0.867)	31	201
92	0.655 (0.544 to 0.761)	0.817 (0.796 to 0.841)	33	234
93	0.706 (0.593 to 0.803)	0.776 (0.751 to 0.801)	45	279
94	0.742 (0.642 to 0.840)	0.706 (0.678 to 0.734)	74	353
95	0.808 (0.718 to 0.892)	0.634 (0.605 to 0.662)	76	429
96	0.848 (0.767 to 0.921)	0.508 (0.477 to 0.538)	129	558
97	0.898 (0.822 to 0.963)	0.357 (0.330 to 0.386)	156	714
98	0.911 (0.841 to 0.973)	0.226 (0.201 to 0.254)	132	846
99	0.961 (0.913 to 1)	0.091 (0.075 to 0.109)	139	985
100	1	0	95	1080

Each row denotes a different threshold for determining those at risk of an adverse outcome. We display the sensitivity and specificity for each threshold, equivalent to all possible intersections of the receiving operator curve, using thresholds between 85% and 100%. In total, 68 patients had an oxygen saturation of 84% or less (not shown). The column on the far right denotes the cumulative sum of the number of observations of the given oxygen saturation (row) or below. For example, 76 patients had an oxygen saturation of 85% or less recorded (top row) and 429 patients had an oxygen saturation of 95% or less recorded. CIs are estimated by bootstrapping.