Colorectal cancer is one of the most common malignancies in the United States. There are an increasing number of studies using recurrent colorectal adenomas to evaluate the prevention effect for some promising agents. The number of recurrent colorectal adenomas is often measured by performing colonoscopy, which is known to miss a small percentage of existing adenomas and result in misclassification on recurrence status. In addition, some participants might not comply with the schedule of follow-up colonoscopy, which is scheduled to be performed once at the end of the study and, therefore, have variable followup lengths compared to the compliant participants. The reasons that a participant cannot comply with the schedule of follow-up colonoscopy could be informative of risk of recurrence and then bias the results derived from statistical methods that do not adjust for noncompliance. Conventional statistical methods for colorectal adenoma prevention trials cannot simultaneously incorporate misclassification and variable follow-up into analysis and cannot adjust for informative non-compliance without strong assumptions and, furthermore, may incorrectly produce equivocal results for some promising nutritional or chemopreventive agents. The purpose of this application is to develop sophisticated and appropriate statistical models to describe the relationship between the preventive agents and recurrence of colorectal adenomas. We will use a latent variable recurrence model, which assumes a portion of non-recurrent participants were misclassified due missing existing adenomas at follow-up colonoscopy, to handle misclassification (Aim 1) and a weight function to incorporate the length of follow-up into analysis (Aim 2). The prognostic factors for risk of recurrence can be incorporated into the weight function to adjust for potential informative non-compliance. If the research in this application is successful, a better understanding of the relationship between preventive agents and recurrence of colorectal adenomas can be obtained and will then allow clinical investigators to identify an agent that truly reduces recurrence of colorectal adenomas.
Colorectal cancer is one of the primary causes of cancer mortality in the United States and the second most expensive cancer to Medicare;therefore, the identification of modifiable risk factors for this disease is imperative. There has been a great deal of research regarding testing the preventive effect of some promising agents, e.g. fiber, in colorectal neoplastic. Approximately 25 million research dollars are directed to prevention of colorectal cancer annually. However, the results remain equivocal for some preventive agents. Inappropriate statistical methods could be one of possible explanations for the equivocal results. As of today, there are only few statistical methods designated to analyze the data from colorectal polyp prevention trials. Therefore, it is imperative and important to now delve deeper into developing statistical methods that might account for the apparently conflicting results reported from a number of large and well-designed prevention studies of colorectal adenomas, which are subject to misclassification and variable follow-up. The underlying rationale for the research in this proposal is that: (i) a better understanding of the relationship between preventive agents and recurrence of colorectal adenomas will allow clinical investigators to identify an agent that truly reduces recurrence of colorectal adenomas, and (ii) proper and efficient statistical analyses are needed to evaluate prevention effects using recurrent colorectal adenomas. The research is, therefore, at the interface of colorectal cancer prevention and statistics. The aim is to develop sophisticated and appropriate statistical models to describe the relationship between the preventive agents and recurrence of colorectal adenomas.
| Long, Qi; Hsu, Chiu-Hsieh; Li, Yisheng (2012) Doubly Robust Nonparametric Multiple Imputation for Ignorable Missing Data. Stat Sin 22:149-172 |
| Hsu, Chiu-Hsieh; Li, Yisheng; Long, Qi et al. (2011) Estimation of recurrence of colorectal adenomas with dependent censoring using weighted logistic regression. PLoS One 6:e25141 |
| Long, Qi; Zhang, Xiaoxi; Hsu, Chiu-Hsieh (2011) Nonparametric multiple imputation for receiver operating characteristics analysis when some biomarker values are missing at random. Stat Med 30:3149-61 |
| Hsu, Chiu-Hsieh; Long, Qi; Alberts, David S (2009) Estimation of colorectal adenoma recurrence with dependent censoring. BMC Med Res Methodol 9:66 |
