Skip to main content

Table 4 AUC and C-statistics for DRSF models with different number of ensemble windows and benchmark models on the testing window covering days from March 1st, 2015 to August 31st, 2015

From: Dynamic prediction of hospital admission with medical claim data

Models

AUC at the 60th day

AUC at the 180th day

Harrell’s C statistics

Batch-mode RSF

0.67a (0.61,0.72)b

0.66 (0.63,0.70)

0.67 (0.63,0.71)

Batch-mode Cox1

0.72 (0.65,0.76)

0.72 (0.69,0.76)

0.72 (0.69,0.76)

Cox1 with 1 window

0.63 (0.63,0.74)

0.70 (0.66,0.74)

0.71 (0.67,0.74)

PLS with 1 window2

NA

NA

0.71 (0.67,0.74)

DRSF with 1 window

0.61 (0.55,0.67)

0.65 (0.61,0.69)

0.68 (0.64,0.71)

DRSF with 2 windows

0.67 (0.62,0.71)

0.68 (0.65,0.71)

0.70 (0.67,0.72)

DRSF with 3 windows

0.69 (0.66,0.72)

0.68 (0.63,0.72)

0.70 (0.67,0.73)

DRSF with 4 windows

0.67 (0.63,0.72)

0.7 (0.67,0.73)

0.70 (0.67,0.73)

DRSF with 5 windows

0.71 (0.66,0.75)

0.71 (0.68,0.74)

0.71 (0.68,0.74)

  1. Note: A larger AUC or C-statistics represents a better model prediction performance
  2. 1Cox: Here we used penalized Cox proportional hazard model
  3. 2PLS with 1 window: Cumulative time-dependent AUC at a specific time point is not available in penalized logistic regression model. Thus, we put NA as Not Available here
  4. aScore obtained with the original dataset
  5. b95% confidence interval of the score obtained with 500 bootstrapped datasets