Research Paper on the Algorithm for Monitoring Diabetes

Introduction

For this assignment, I chose diabetes as my chronic disease of interest. One of the best algorithms that can be used in monitoring chronic diseases is sparse Poisson convolution conditional autoregression (Baker, 2017; Duncan, White, & Mengersen, 2016; Lawson & Rotejanaprasert, 2014; Wang, Sontag, & Wang, 2014). According to Baker, White, and Mengersen (2015), this type of algorithm can be used to separately assess the geographical variation of diabetes mellitus I and diabetes mellitus II incidence as well as in the estimation of joint spatial correlation between diabetes mellitus I and diabetes mellitus II. Bayesian conditional autoregression refers to a spatial structure whereby the correlated random error of every geographical point on a map of interest is designated by a lattice of adjoining regions, with i-j showing that regions i and j are bordering each other.

Is your time best spent reading someone else’s essay? Get a 100% original essay FROM A CERTIFIED WRITER!

Save 25%

One of the variations of sparse Poisson convolution conditional autoregression, spare Poisson multivariate conditional autoregressive (MCAR) is useful in evaluating the joint spatial correlation between diabetes mellitus I and diabetes mellitus II. MCAR has been reported to be useful in simultaneous modelling of joint correlation between many outcomes while taking into consideration the correlated error existing between spatial neighbors. In sparse Poisson convolution conditional autoregression, the joint spatial correlation between diabetes mellitus I and diabetes mellitus II is assessed through empirical calculation of the correlation between the RR estimates derived from sparse Poisson convolution models using the Pearson correlation coefficient.

A spatial approach to disease modelling is appropriate to all the types of chronic diseases which can be attributed to environmental factors and particularly suitable for diabetes mellitus because recent research studies have established associations between diabetes mellitus prevalence and geographical factors. Examples of these factors include decreased opportunities for exercise, increased use of vehicle transport, increased availability of fast-food, walkability, and green space (Astell-Burt, Feng, & Kolt, 2014). Most of these factors are useful in health promotion. Consequently, spatial analysis generates vital data that can be used to guide public policy decisions and resource allocation.

The spatial approach to disease surveillance has also been reported to be the most effective because of four key features (Kulldorff, 2001). First, it is capable of adjusting both for the presence of confounding variables as well as for the inhomogeneous population density. Secondly, this method eliminates preselection bias because of its ability to search for clusters without specifying their location or sizes. Thirdly, it yields a single p-value and takes multiple testing into account. Lastly, if there is a rejection of the null hypothesis, there is a possibility of approximating the location of the cluster responsible for the rejection (Kulldorff, 2001).

Covariates Included in The Algorithm

Some of the covariates that have been included in previous studies that employed sparse Poisson convolution conditional autoregression include age, gender, and ethnicity. Covariates are included with the aim of capturing random effects and confounding effects that can be used to explain other variations that are not capturable by covariates (Vannieuwenhuyze, Loosveldt, & Molenberghs, 2014).

The Limitations of the Algorithm and the Implications for Public Health

One of the limitations of using Sparse Poisson convolution conditional autoregression in monitoring diabetes is that if the collected data contains sparse count data, then there will be an excessive number of zeros leading to the violation of assumptions of traditional Poisson models. Consequently, the Poisson model is not appropriate to apply the data when there are an excessive amount of zeros because the fundamental model assumption of the equality of the variance and the mean of the Poisson model cannot be met. However, in such cases, a sparse Poisson convolution model can be chosen in consideration of the sparseness of data. The sparse Poisson convolution model is a variation of the Bayesian conditional autoregression with an extra indicator variable which denotes zero or non-zero count in any area (Baker et al., 2015).

There are two other limitations associated with Poisson models, especially with frequent repetition of purely spatial analysis as a constituent of the time-periodic surveillance system. First, such a model has a low power of rapidly detecting merging clusters. Consequently, if a truly excessive risk is only found in the last few years of the time-periodic surveillance, then by doing a purely spatial analysis for the whole duration, the strength of the cluster is diluted. This is because such analysis includes random fluctuations of the rates for earlier years when the risk was low. Secondly, this method does not adjust for the multiple testing associated with repeated analysis every year despite its strength of being capable of adjusting for the multiple testing coming from various cluster sizes and cluster locations.

There are some public health implications associated with the use of Poisson models. The most important is that associated with dilution of the strength of the cluster. Because such analysis includes random fluctuations of the rates for earlier years when the risk was low, the findings of such surveillance can underestimate the prevalence or incidence of the disease. Consequently, findings can mis-inform healthcare policies.

References

Astell-Burt, T., Feng, X., & Kolt, G. S. (2014). Is neighborhood green space associated with a lower risk of type 2 diabetes? Evidence from 267,072 australians. Diabetes Care, 37(1), 197-201. https://doi.org/10.2337/dc13-1325

Baker, J. F. (2017). Bayesian spatiotemporal modelling of chronic disease outcomes (Doctoral dissertation, Queensland University of Technology). 61-72. https://doi.org/10.1111/1467-985X.00186

Baker, J., White, N., & Mengersen, K. (2015). Spatial modelling of type II diabetes outcomes: a systematic review of approaches used. Royal Society Open Science, 2(6). https://doi.org/10.1098/rsos.140460

Duncan, E. W., White, N. M., & Mengersen, K. (2016). Bayesian spatiotemporal modelling for identifying unusual and unstable trends in mammography utilisation. BMJ Open, 6(5), e010253. https://doi.org/10.1136/bmjopen-2015-010253

Kulldorff, M. (2001). Prospective time periodic geographical disease surveillance using a scan statistic. Journal of the Royal Statistical Society: Series A (Statistics in Society), 164(1),

Lawson, A. B., & Rotejanaprasert, C. (2014). Childhood brain cancer in florida: a bayesian clustering approach. Statistics and Public Policy, 1(1), 99-107. https://doi.org/10.1080/2330443X.2014.970247

Vannieuwenhuyze, J. T. A., Loosveldt, G., & Molenberghs, G. (2014). Evaluating mode effects in mixed-mode survey data using covariate adjustment models. Journal of Official Statistics, 30(1), 1-21. https://doi.org/10.2478/jos-2014-0001

Wang, X., Sontag, D., & Wang, F. (2014). Unsupervised learning of disease progression models. In Proceedings of the 20th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (pp. 85-94). New York, NY, USA: ACM. https://doi.org/10.1145/2623330.2623754