Detection and Estimation of Change Points in the Incidence Rate of a DISEASE

Predicting changes in individual and population health represents a significant challenge for public health professionals. Understanding and identifying primary risk factors related to abrupt changes in one's health are critical components in prevention and intervention. Thus, the purpose of this presentation is to demonstrate the use of detection algorithms to predict whether, and at what point, abrupt changes in individual and population health can be anticipated. In this presentation a number of detection algorithms will be discussed. Nonparametric techniques for the detection and location of change points as they occur will be presented. Several existing data sets were used to examine the problem of retrospective estimation retrospectively to identify critical change points and to indicate whether the change was gradual or abrupt, permanent or transient, and how long the change could be expected to last. The CUSUM bootstrap iterative algorithm search we used begins from both ends of the data to identify the boundaries of the change. The results obtained from simulation and practical datasets show nearly 99% success rate in the exact determination and extraction of regions where the change occurred.

Thus, it appears that this technique holds promise for identifying and detecting future changes in health. When changes were detected in real-time processing, the region, boundaries, and nature of the change was identified in < 1/3 of a second. Based on simulation and practical data sets, it appears this algorithm has unique applicability in the development of public health prevention and intervention efforts.