Development of a Practical Model for Targeting Patient Decision Support Interventions to Promote Colorectal Cancer Screening in Vulnerable Populations.

We sought to develop a practical model for predicting probability of colorectal cancer (CRC) screening completion in a diverse safety-net population and a subsequent framework for targeting screening promotion interventions. Data used to construct the models was prospectively collected from a CRC screening intervention. Models were cross-validated by randomly splitting the data into training and validation samples. The predicted probabilities of screening completion from the final model were trichotomized into framework groups and cross-validated by defining cut-points in the training sample, applying them to the validation sample, and comparing across samples for homogeneity. The final model included demographic and simple psychosocial measures and predicted screening behavior adequately (AUROC=0.63). The framework groups (cut-points 0.62 (low), 0.74 (medium), and 1.0 (high)) were homogeneous across the two samples. The model and framework may be useful for designing and delivering targeted interventions to promote CRC screening. Future research should validate the framework groups.