Skip to main content

Table 4 The importance of the accuracy of the weighting of the risk prediction tool

From: Multivariable risk prediction can greatly enhance the statistical power of clinical trial subgroup analysis

 

RF's #1 & #2*

RF's #3 & #4*

RF's #5 & #6*

Risk Index's Predictiveness (AUROC) †

Statistical Power (p < 0.05)‡

True Predictiveness

1.5

2.0

2.5

-

-

Perfect Weighting

1.5

2.0

2.5

.69

.82

Uniform Weighting

1

1

1

.66

.75

Reverse Weighting

2.5

2.0

1.5

.63

.69

Incomplete Reverse Weighting

2.5

0.0

1.5

.60

.51

  1. * Each of the 6 risk factors (RF's) has a prevalence of 25% and the 5-year CER for those without any risk factors is .75% (see Table 1 and Figure). The true predictiveness (relative risk) of the risk factor is shown as well as the relative weight used in the risk index.
  2. † The Area Under the Receiver Operator Characteristic (AUROC) curve is a measure of the overall predictiveness of a model for predicting a dichotomous outcome (i.e, event occurred vs. event did not occur).
  3. ‡ Statistical subgroup comparisons tests the power to detect whether the treatment's relative benefit varies as a function of the risk score.
Back to article page

BMC Medical Research Methodology

ISSN: 1471-2288

Contact us