Analysing behavioural risk factor surveillance data by using spatially and temporally varying coefficient models

Summary: The study of temporal and spatial trends in large databases, such as behavioural risk factor surveillance data, can be a great challenge, especially when the intent is to study the time-related effects of multiple independent variables; this is an issue which is not usually addressed in trend analysis in epidemiological studies. This study demonstrates the use of varying coefficient models using non-parametric techniques, which can show how coefficients vary in time or space; it is a useful statistical tool that is applied for the first time to health surveillance data. Using the US 'Behavioral risk factor surveillance system', a varying coefficient model is constructed using obesity as an outcome measure. Odds ratio plots and probability maps illustrate the temporal or spatial changes in coefficients of the independent variables; these results can be used to identify changes in at-risk subgroups of the population for the odds of obesity.

Background Behaviour risk factor surveillance (BRFS) data can be an important source of information for studying changes in various health outcomes and risk factors. Results obtained from surveillance data analysis are vital for informing health policy interventions, particularly with regards to evolutionary aspects. The objective of this analysis was to recommend a method that can be used for analysing trends in the association among variables from large public health data sets. This was demonstrated by examining the changing effects of various covariates, representing different sub-populations, on smoking status over time. Methods In our work, we propose the use of varying coefficient models (VCM) with non-parametric techniques to catch the dynamics of the evolutionary processes under study. This is a useful method, which allows coefficients to vary with time using smooth functions. Italian BRFS data from 2008-2012 was used with a sample size of 185,619 observations. In the application, a time VCM is fit for a smoking status binary outcome variable using the P-spline estimation method. The model includes ten independent variables comprising socio-demographic, health risk and behaviour variables. Results The VCM fit for the data indicates that the coefficients for some of the categories for the age and the alcohol consumption variables varied with time. The main results show that Italians aged 18-29 and 40-49 had higher odds of being smokers compared to those aged 60-69; however, these odds significantly decreased in the period 2008-2012. In addition, those who do not drink had lower odds for being a smoker compared to high risk drinkers and these odds decreased further during the observation period. Conclusion The application of the VCM to the BRFS data in Italy has shown that this method can be useful in detecting which sub-populations require interventions. Although the results have shown a decrease in the odds of being a smoker for certain age groups and non-drinkers, other sub-populations have not decreased their odds and health inequalities remain. This observation indicates that efforts and interventions are still required to target these non-changing sub-populations in order to modify their smoking behaviour.