Use of Poisson regression and time series analysis for detecting changes over time in rates of child injury following a prevention program

Abstract

The use of two statistical methods to quantify time trends (Poisson regression and time series analysis) is illustrated in analyses of changes in child injury incidence after implementation of a community-based injury prevention program in Central Harlem, New York City. The two analytical methods are used to quantify changes in the rate of injury following the program, while taking into account the underlying annual and seasonal trends. Rates of severe injury during the period from 1983 to 1991 among children under the age of 17 years living in Central Harlem and in the neighboring community of Washington Heights are analyzed. The two methods provide similar point estimates of the effect of the intervention and have a good fit to the data. Although time series analysis has been promoted as the method of choice in analysis of sequential observations over long periods of time, this illustration suggests that Poisson regression is an attractive and viable alternative. Poisson regression provides a versatile analytical method for quantifying the time trends of relatively rare discrete outcomes, such as severe injuries, and provides a useful tool for epidemiologists involved with program evaluation.