Change point analysis detection of trends in Malaria cases
Abstract
In this paper, we present CUSUM plus bootstrapping change point detection algorithm for offline inferences for time series data using malaria cases. Change point detection is the identification of abrupt changes in time series (sequential) data. As an online and offline signal processing tool, it has proven to be useful in applications such as process control. Change point detection can be done using SPC or statistical change point detection methods. Statistical Process Control involves ongoing checks to ensure that neither mean nor variability changes.
The dataset used in this study consist of weekly confirmed malaria cases and historical routinely collected data from five health facilities in epidemic prone area (Eldoret East) where interventions (IRS and LLITNs) have been put to reduce malaria incidences. The objective of the study was to detect points of change within the trends of cases. A change in the trends of weekly malaria cases indicated an epidemic while a change the trends in the historical malaria cases indicated points where an impact of the intervention measure may have been felt. The initial analyses using control charts gave the visual impression of where the change was likely to occur and detected change points.
Using the offline change point analysis we were able to locate the significant point of change, the magnitude of change and the significance levels of the change points in the trends of malaria cases. The significant change points detected for Moiben Health Center were at week 30 and week 45 with significance level of 99.8% and 98.6% respectively. The magnitude of those change points were 561.575 and 159.818 respectively. The CPA also detected some changes which were not significant too. The other change points detected for Eldoret East were at week 28, week 48, week 38, week 44 and week 45 for the year 2011 with significant levels of 99.6%,99.6%,99%, 96.3% and 94.5% respectively.
The magnitude of changes were 2494.058,805.25,639.65,219.4 and 46.9% respectively. The significant change point at week 28, week 38, week 44, week 45, and week 51 may have been attributed to the interventions. It is suggested that change point analysis procedure is an effective tool for detecting significant change point can therefore be applied in epidemic surveillance alongside other existing procedures to enhance detection of significant rising cases of diseases, for process monitoring process and for analysis of disease trends.
Keywords: Change-point analysis, Control charts, Statistical Process Control, Time-series.
Publisher
University of Nairobi, Kenya