Dynamic updating of clinical survival prediction models in a changing environment

Table 2 Performance of intercept recalibration (Recal), refitting (Refit), and Bayesian dynamic updating (Bayes) methods to update the prediction model for 70-day COVID-19-related death. The original model was fit using data from period 1 and evaluated using data from period 2. The original model was then updated each period with new data and evaluated using the following period’s data

Model fit w/	Evaluated w/	Recal	Refit	Bayes	No	Recal	Refit	Bayes	No
data from:	data from:				update\(^\dagger\)				update\(^\dagger\)
		C-index				Brier Score
Period 1	Period 2				0.95				3E−04
Period 2	Period 3	0.92	0.93	0.76	0.92	3E−05	3E−05	3E−05	3E−05
Period 3	Period 4	0.94	0.91	0.92	0.94	8E−04	7E−04	8E−04	7E−04
Period 4	Period 5	0.91	0.91	0.90	0.91	4E−04	4E−04	4E−04	4E−04
		Calibration intercept				Calibration slope
Period 1	Period 2				− 0.82				1.10
Period 2	Period 3	-0.79	0.30	0.15	− 2.12	0.92	0.82	0.86	0.92
Period 3	Period 4	3.16	− 0.04	− 0.01	0.56	0.93	0.90	0.86	0.93
Period 4	Period 5	− 0.66	− 1.10	− 1.12	− 0.49	0.85	0.89	0.88	0.85

\(^\dagger\)No update refers to the original model fit in period 1 and evaluated in each subsequent period without any updating

ISSN: 2397-7523