Refined Capture-Recapture Methods for Surveilling Cancer Recurrence

点击翻译按钮获取中文摘要

Project Summary/Abstract The monitoring of disease prevalence and estimation of the number of affected individuals in a defined population are among the crucial goals of epidemiologic surveillance for chronic and infectious diseases. This proposal aims to provide novel and reliable statistical tools to improve best practices for design and analysis of such surveillance studies. We take specific motivation from timely challenges associated with the registry- based monitoring of cancer recurrences in the state of Georgia Cancer Registry (GCR). We focus on customizing capture-recapture (C-R) methods, which are ever increasingly used tools for estimating total numbers of cases or deaths based on multiple epidemiologic surveillance streams. We clarify underappreciated pitfalls associated with widely popular log-linear model-based C-R techniques, and propose an accessible approach to sensitivity analysis with data visualization that promotes a general strategy for more appropriate propagation of uncertainty into ultimate estimates of case totals. This in turn provides a gateway to a broad class of useful models, whereby practitioners can transparently encode assumptions about how surveillance streams operate relative to one another at the population level. As a next step, we consider the case in which one surveillance stream is implemented by means of a well-controlled sampling design. Under appropriate conditions, this provides what we refer to as an “anchor stream”, whereby otherwise ever-present inherent uncertainties in specifying a defensible C-R model are overcome. In this setting, we will promote best statistical practices for estimating case totals by means of a novel C-R estimator that harnesses the power of the principled sampling behind the anchor stream while offering markedly enhanced precision. We propose to extend this approach to account for misclassification, which is inevitable in the case of our motivating study of cancer recurrence and in any setting in which surveillance streams identify cases in an error-prone manner. We will tailor proposed methodology toward breast and colorectal cancer recurrence monitoring via the ongoing Cancer Recurrence Information and Surveillance Program (CRISP), based on the GCR. CRISP is actively compiling informative but potentially false-positive recurrence signals from up to 6 data streams, and conducts validation sampling through protocol-based medical record review to confirm true cases among signaled recurrences. We will use such validation data to adjust for misclassification in estimating C-R-based recurrence counts. In particular, the current project will implement a principled “anchor stream” random sample of 200 GCR patients for validation through medical record review, leading to valid and demonstrably precise estimates of true recurrence counts over the study period that are free of misclassification bias.