Estimation of Change Points from Regression models of Count Data

Abstract

Change point problem is an important issue to detect the change in the trend of count data over time. Polynomial and spline regression models are used to estimate change points and give the indication of the most important interventions which may impact the rates of healthcare associated infections (HAIs). The aim of this research is to estimate change points from regression models and construct their confidence intervals to give the recommendation on healthcare interventions. The data of HAIs was collected by Health Protection Scotland (HPS) and includes methicillin resistant staphylococcus aureus (MRSA). Some interventions took place in National Health Service (NHS) in Scotland were reported by HPS from 2003. Polynomial and spline models are fitted with consideration of seasonal effect and used to detect the change points where the trend of the rates changed significantly. Bootstrap method is used to construct confidence intervals for estimated change points. The research showed that polynomial and spline regression methods estimated approximately similar one change point for MRSA bacteraemia when the rate starts to decrease during 2005. To conclude, polynomial and spline models have similar results of estimated change points however, spline model presents few numbers of change points than polynomial model. Screening MRSA in patients prior to hospital admission and applying antibiotic policy such as hand hygiene help to reduce and prevent the infection in hospitals and healthcare systems.