Table 4.

Calibration of the prediction probability of the deep learning–based model using internal and external test data sets. P values were calculated with the Spiegelhalter z test; significant values indicate inappropriate calibration. Probability 1 and probability 2 were the prediction values of the deep learning–based model without and with symptom probability, respectively.

Data set and outcome				P value (Spiegelhalter z test)
Internal test set (n=75)
	Hospital length of stay ≤2 weeks
		Probability 1	.72
		Probability 2	.57
	Oxygen supplementation
		Probability 1	.91
		Probability 2	.80
	Development of acute respiratory distress syndrome
		Probability 1	.002
		Probability 2	.81
External test set (n=467)
	Hospital length of stay ≤2 weeks
		Probability 1	<.001
		Probability 2	<.001
	Oxygen supplementation
		Probability 1	<.001
		Probability 2	<.001
	Development of acute respiratory distress syndrome
		Probability 1	.08
		Probability 2	.86