Open population capture-recapture models and diabetes in Otago

Abstract

The aim of my research was to develop a capture-recapture model to estimate both prevalence and incidence and to monitor those estimates over time. The standard way of estimating prevalence of diabetes using lists is through closed population capture-recapture methods. These may use two lists or may use several. Prescribed capture-recapture models can be used for the multi-list analysis or, alternatively, loglinear models which allow for the modelling of specific dependencies between lists. They are, typically, analyses of lists from a particular point in time (to meet the assumption of a closed model - that is, no one joins or leaves the lists). Access was granted to lists of people with diagnosed diabetes resident in Otago, New Zealand. My new capture-recapture model allows information to be exploited that has previously not been included in standard models. Data covering several years were used rather than just looking at a single snapshot. This means that for each person there are several observations over a period of years allowing us to monitor estimated prevalence. Also, we can allow for people who join the lists subsequent to the start of the study period and those who may leave the lists before the study ends. That means that we can estimate the probability of remaining on the lists from one time period to the next. We also had diagnosis date which allowed us to estimate incidence and monitor that over time. The results from the analysis (both through a conventional dosed capture-recapture model and using the new model) were consistent with results reported nationally for diabetes. The new model was better suited to modelling five years of data than two. The excessive computing time required to fit the five year models impacted on their utility, but this new method could assist with monitoring chronic diseases in a cost effective way.