E model Calibration:

The E-score risk model calculates the cumulative risk of experiencing a CRC dx by a fixed landmark prediction time \(t\):

\[ r_t(X) = P(T < t | X) \] Where \(T\) is survival time and covariate information \(X\) includes the E-score, family history, and endoscopy history. See here for more information about the construction of the E-score risk model.

We check the calibration of a risk model in order to assess if the risk estimates output from \(r_t(X)\) are meaningful. Specifically, we assess ‘weak’ calibration that asks the question:

Among people with \(r_t(X) = r\), is the cumulative \(t-year\) event rate observed to be approximately r?

Description of results

We create calibration plots for the E-score absolute-risk estimates at years t = 2, 5, 10 as follows:

  • Sort individuals into 10 groups (by percentiles) based on the absolute risk score.
  • Calculate cumulative incidence rate at t year for each group with 95% CI.
  • Plot empirical rates (observed) and model based rates (predicted).

Observed cumulative incidence rates are shown by points. The solid black curve shows the predicted t-year absolute risk.

Age stratified calibration

Below, we show calibration plots stratified on categorized age and gender.

Sample size by age/gender is shown in the table below:

  Female Male
<= 60 31517 28541
60-65 21185 21905
65-70 15745 15899
70+ 7597 7342

Males

Females