Skip to main content

Table 3 Sensitivity (95% CI) of prediction models from split-sample and entire sample estimation approaches in the development dataset and prospective validation dataset

From: Empirical evaluation of internal validation methods for prediction in large-scale clinical data with rare-event outcomes: a case study in suicide risk prediction

 

Split-sample prediction model

Entire-sample prediction model

Risk percentile cutpoint

Testing set, Development

Prospective validation

5-fold cross-validation, Development

Bootstrap optimism correction, Development

Prospective validation

 ≥ 99%

12.2% (7.5%, 17.8%)

19.9% (10.9%, 30.1%)

11.6% (8.4%, 15.2%)

17.5% (8.8%, 24.6%)

15.1% (7.5%, 24.2%)

 ≥ 95%

42.1% (32.6%, 51.0%)

40.1% (30.1%, 50.2%)

33.9% (27.1%, 40.5%)

47.2% (40.1%, 52.9%)

39.2% (28.7%, 50.5%)

 ≥ 90%

54.0% (45.4%, 62.1%)

52.0% (41.7%, 61.1%)

50.5% (43.3%, 57.2%)

63.2% (56.3%, 67.8%)

53.0% (43.0%, 62.3%)

 ≥ 75%

78.5% (72.9%, 83.4%)

75.1% (67.0%, 82.3%)

75.1% (70.4%, 79.8%)

84.4% (81.0%, 86.9%)

72.1% (63.8%, 79.7%)