A novel mixed model for the analysis of adherence to screening invitations
The mixed nature of the models will allow the prediction of the individual risk of non-adherence above and beyond measured risk factors, so that the more non-adherence-prone women may be identified as targets for compliance-inducing interventions, as well as for the further study of the determinants of their behavior.
The new models will represent an alternative to traditional binary GLMM models, and will allow for closed form analytical expressions and computationally efficient and accurate estimation algorithms. As part of this project we will compare the two modeling approaches.
The questionnaires' data will be used as risk modifiers for non-adherence, and any information on diagnoses and deaths will be used to allow the identification of structural reasons for non-adherence.
We will ensure that all necessary privacy safeguards be in place, and that only authorized researchers have access to the data.
1. Implementation of the authorization process for the HIP study data transfer, both with NCI (through the Cancer Data Access System) and with the Ethics board of our local institution. Transfer of the data files and first descriptive analyses, also with reference to existing publications based on the same data.
2. Fine-tuning of the R software being developed for fitting the new class of models, and fitting of the models until convergence (through model selection) on a parsimonious yet informative model for the individual adherence process.
3. Prediction of the subject specific risk of non-adherence.
4. Comparison of the new models to binary GLLMs with respect to computational effort, stability, interpretation, and prediction of risk of non-compliance.
5. Preparation of manuscript for publication, and dissemination at scientific conferences.
Edoardo Ratti, University of Milan-Bicocca, Milan, Italy