In this report, we assess the predictive capacity of an environmental risk score (E-score) used to predict colorectal cancer (CRC) in caucasian men and women between 48 and 78 years old. We use the PLCO screening trial to assess the calibration and validation of this previously developed E-score.

The report is organized as follows. The PLCO tab above reports the distribution of key variables in the PLCO data set used in the formulation of the E-score (demographics), and information about the outcome of interest (CRC outcome).

Under the E-score tab drop-down menu, we show:

• About: briefly describes how the score was developed, and how we can use the E-score to obtain estimates of absolute n-year risk of CRC.

• Calibration: Model calibration assesses the reliability of risks estimated using the E-score. For example, for individuals with a predicted absolute risk of approximately 10%, do we observe that 10% of these individuals go on to develop CRC? We assess model calibration by showing visual displays of observed vs. expected risk across a range of risk values in the PLCO data set.

• Validation: We assess model discrimination–how well the model assigns higher risks to those that will go on to develop CRC, and lower risks to those who will not. To assess model validation, we look at measures of sensitivity, specificity, and area under the ROC curve (AUC) for several landmark prediction times.

In the future, we will assess the calibration/validation of a risk score that incorporates genetic information as well (G-score).