COMPSTAT 2022: Start Registration
View Submission - COMPSTAT2022
Title: Risk predictions using panel count data with informative observation times Authors:  Qing Pan - George Washington University (United States) [presenting]
Abstract: In epidemiology studies of screening-detected disease, researchers often face screening data in the form of interval censored panel count data. Furthermore, observation times are usually informative about the disease risks. We analyze the Study of Colonoscopy Utilization within the PLCO Cancer Screening Trial, which followed patients for up to 15 years on colorectal cancer screening results including both cancer and its non-advanced/advanced adenoma precursors. Screening times strongly depend on past screening results. Recurrent adenoma processes are defined by the numbers and sizes of adenoma. Furthermore, the recurrent adenoma processes are reset to zero at each screening because colonoscopy removes all detected adenomas. We model the recurrent times to screening and recurrent adenoma at each screening jointly. Correlations between the screening and adenoma processes are modeled by subject-specific frailty terms. The baseline intensity function and regression coefficients for the recurrent adenoma processes are estimated using estimating equations for interval censored panel count data under the piecewise baseline intensity assumption. Probabilities of advanced adenoma at the next fixed or expected screening time are predicted. Performance of the risk prediction is examined through extensive simulation studies and illustrated on the PLCO clinical trial data.