Cancer is the second-leading cause of death among Americans. One of every four deaths in the United States (US) is due to cancer. The American Cancer Society (ACS) estimates that in 2010, about 1,479,350 Americans will receive a new diagnosis of invasive cancer, and 562,340 Americans will die of this disease. These estimates do not include in situ cancers or the more than 1 million cases of basal and squamous cell skin cancers expected to be diagnosed this year. According to a recent report by the Centers for Diseases Control and Prevention (CDC), the cost of cancer care in the United States has nearly doubled in the past 20 years and the rising costs are mainly driven by the increase in cancer prevalence, not cost per patient. The National Cancer Institutes (NCI) estimated that for 2009 the overall annual cost of cancer would be about $243.4 billion, which includes direct medical costs of $99.0 billion for health and care expenditures;indirect costs of $19.6 billion associated with lost productivity due to illness and indirect costs of $124.8 billion associated with lost productivity due to premature death. These costs are likely to increase because of the anticipated growth and aging of the US population. The population-based cancer statistics are crucial for health planners, policy makers, and cancer information providers to prioritize investments in cancer control and prevention. Researchers may use the cancer statistics to investigate the effect of cancer control planning, to characterize the heterogeneity of geographical areas and demographic groups and to examine the health disparities among different groups. The ACS publishes predicted numbers of cancer incidences and deaths in the current year for the whole US and individual states in the annual publication Cancer Facts and Figures since 1960 and in Cancer Statistics since the early 1970s. As a joint effort by NCI, ACS, CDC and NACCR, Annual Report to the Nation on the Status of Cancer, 1975-2006 provides an update on the trends in cancer incidence (new cases reported) and death rates in the US. This report also includes trends in colorectal cancer incidence and death rates and highlights the use of micro-simulation modeling as a tool for interpreting past trends and projecting future trends to assist in cancer control planning and policy decisions. Among the commonly used cancer statistics, cancer incidence is the direct and one of the most important measures of cancer burden. Cancer incidence count is the number of newly diagnosed cases of a specific site occurred in the population. The cancer incidence rate is defined as the rate of new cancer cases in a specified population during a year, usually expressed as the number of cancers per 100,000 population at risk. Both cancer incidence rate and count are important in the perspective of public health. The cancer incidence counts are helpful in determining the cancer burden and specific needs for services for a given population and cancer incidence rates, on the other hand, can be used to evaluate the trends and effects of cancer control and prevention. The statistical methods of predicting cancer incidence rates and counts have involved over the years. Various methods, e.g., statistical projection method, state-space model, age-period-cohort model , generalized additive model, have been used to predict the number of cancer incidences and deaths in the US and other countries. Kim et al. applied joinpoint regression models (JPM) to fit the trends of cancer incidence rates and proposed a permutation test to select the number of joinpoints of the optimal model. Since then, the joinpoint model has been widely used to predict cancer incidence and mortality rates and counts and to describe the trends of cancer survival. By far, the use of the joinpoint model is restricted to cancer incidence and mortality data from the whole country and individual states. For local policy-makers, health researchers, it may also be of interest to seek detailed county-level cancer incidence data to assess the burden of cancer in local regions. Now several state or local governments start to estimate the current and future county-level cancer incidences. However, most estimations and predictions for county-level data are limited to descriptive analysis and simple linear regression. In this project, we focus on the prediction of county-level cancer incidence rates and counts. We review the current standard method, i.e., the joinpoint model, for modeling cancer incidences and extend it to county-level data. For prediction of county-level cancer incidences, a direct extension is to apply the joinpoint model, which has been validated for national and state level data, to the county level data. However, there are a few challenges of using county-level data. First, the numbers of county-level cancer incidences are usually small, especially for rare cancer. It is possible that the cancer incidence counts for a county are zero for certain years. Second, county-level cancer incidence counts tend to have large variation and depends heavily on the population of that county. Third, it is important to consider the temporal trends with respect to calendar year and the spatial correlations of multiple counties in the same geographical area. Instead of estimating cancer incidence individually for each county, we develop a spatial random-effect joinpoint model (SRJM), in which a joinpoint model is used for the cancer incidence trends and the random effects incorporate correlations of spatially adjacent counties. The performances of the proposed model and direct application of the current JPM are evaluated by a validation study using data from the population-based cancer registry data from NCI.
Showing the most recent 10 out of 17 publications
