Data Matching to Allocate Doctors to Patients in a Microsimulation Model of the Primary Care Process in New Zealand

Abstract

The authors aimed to use existing data to create a microsimulation model of the primary care process in New Zealand, including realistically simulating the allocation of general practitioners (GPs) to a population sample. This is important because GP behavior is likely to be a major determinant of future cost and service outcomes. Two nationally representative data sets were matched: a sample of GPs and their patients from the National Primary Medical Care Survey (NPMCS) and a population sample from the New Zealand Health Survey (NZHS). Matching involved first dividing the data sets into cells based on common variables. Further variables were then included in a distance function to guide matching within cells. A transportation optimization algorithm allocated GPs based on these—on similarities in patients’ attributes. Statistical matching performed well with high correlations for patient attributes and reduced average absolute rank differences on proportions of patients among GPs compared to random matching. Low Kullback–Leibler (K–L) divergences confirmed that our method of statistical matching had allocated GPs realistically. Models of primary care too frequently omit the role of the practitioner in driving health service outcomes. The authors developed a method to impute characteristics of GPs to a population-based microsimulation model of primary care.