Analysis of interrupted time series mortality trends: an example to evaluate regionalized perinatal care.

Abstract

Interrupted time series designs are frequently employed to evaluate program impact. Analysis strategies to determine if shifts have occurred are not well known. The case where statistical fluctuations (errors) may be assumed independent is considered, and a segmented regression methodology presented. The method discussed ia applied to the assessment of changes in local and state perinatal postneonatal mortality to identify historical trends and will be used to evaluate the impact of the North Carolina Regionalized Perinatal Care Program when seven years of post-program mortality data become available. The perinatal program region is contrasted with a control region to provide a basis for interpretation of differences noted. Relevant segmented regression models provided good fits to the data and highlighted mortality trends over the last 30 years. Considerable racial differences in these trends were identified, particularly for postneonatal mortality. Segmented regression is considered relevant for the analysis of interrupted time series designs in other applications when errors can be taken to be independent. Thus, the methodology may be regarded as a general statistical tool for evaluation purposes.