Statistical methods for surveillance of spatial health data are of critical importance to public health practitioners. Yet, prospective surveillance for changes in disease risk over in space and time is a relatively undeveloped arena of statistical methodology. Most methods for space-time surveillance have been developed for retrospective analyses of complete data sets. However, data in public health registries accumulate over time and sequential analyses of all the data collected so far is a key concept to early detection of emerging trends or differences in disease risk. The impact derived from timely treatment and control measures can be dramatic, especially when monitoring maps of disease incidence of chronic diseases such as cancer, one of the leading causes of death worldwide. The goal of this proposal is to develop statistical methodology for prospective spatio-temporal disease surveillance, with cancer surveillance being our primary focus. The conditional predictive ordinate is a Bayesian diagnostic tool that detects unusual observations. Although it has never been applied in a surveillance context, we hypothesize it is a powerful technique, in a modified form, for detection of unusual aggregations of disease in space and time. We will also extend our approach to the analysis of multiple diseases, as surveillance systems are often focused on more than one disease. This extension, incorporating correlation between diseases, is likely to improve cluster detection capability. We propose three specific aims.
In Specific Aim 1 we will adapt the conditional predictive ordinate for a surveillance setting. Publicly available small area cancer count data and simulated data mimicking possible true disease relative risk changing patterns will be used to test the performance of the proposed methodology in different scenarios.
In Specific Aim 2 we will generalize this approach to a multivariate setting which allows for inclusion of correlation between diseases. Different types of cancer will be monitored simultaneously to assess the performance of the multivariate extension in comparison to the individual analyses.
In Specific Aim 3, the implementation of the surveillance conditional predictive ordinate in an R package, a free statistical programming language available in many public health departments, will enable use by public health practitioners. Upon the completion of this project, we will have a Bayesian surveillance technique that will be used to detect areas of increased disease incidence as quickly as possible in an effort to reduce morbidity and mortality. The multivariate extension of the proposed surveillance technique will fill in a major gap on the current literature. This extension, allowing for inclusion of correlation between diseases, may contain important clues for the early detection of changes. Finally, the implementation of the surveillance methodology in a user-friendly package within the R software environment will facilitate dissemination.
In this project we will develop a novel model-based surveillance technique to monitor a map of disease over time. This technique will enable early detection of changes in disease risk helping to reduce undue morbidity and mortality. The implementation of the proposed technique in a user-friendly package within the R software environment will facilitate dissemination and use by public health practitioners.
|Corberan-Vallet, A (2012) Prospective surveillance of multivariate spatial disease data. Stat Methods Med Res 21:457-77|
|Corberan-Vallet, Ana; Lawson, Andrew B (2011) Conditional predictive inference for online surveillance of spatial disease incidence. Stat Med 30:3095-116|