The late stages of diabetic retinopathy are characterized by a proliferation of abnormal new retinal blood vessels which may bleed into the vitreous gel of the eye leading to a decrease in vision. Panretinal photocoagulation has been shown to reduce the risk of severe visual loss that is associated wit the hemorrhaging. However, without proper detection, the condition may progress to a point where this treatment is less efficacious. Because of this and the risks associated wit the treatment, there is great interest in modeling the natural progression of diabetic retinopathy to investigate the cost- benefit of treatment under different screening and treatment programs. In today's medical world, however, it is impossible to observe this natural progression directly and so it must be inferred from observational studies. One such data set is the Wisconsin in Epidemiologic Study of Diabetic Retinopathy. It is complicated by multi-year observations at unequal intervals and the fact that some subjects received treatment intervention during the period of the study. We propose to fit non-homogeneous discrete-time Markov models which describe the natural history of this condition under different types of diabetes and account for these """"""""non-natural"""""""" occurrences. Bayesian inference, enabled by Markov chain Monte Carlo, combines these models with the cohort data set. Having obtained estimates of the model parameters, posterior predictive distributions will be used as a prognostic tool to assist researchers in evaluating costs and benefits of treatment protocols. Unlike other estimation procedures, these cost-benefit estimates will be conditional on particular parameter values and an assessment of the uncertainty of these estimates is very straightforward. We will investigate several difference screening and treatment programs for each type of diabetes.
| Craig, B A; Fryback, D G; Klein, R et al. (1999) A Bayesian approach to modelling the natural history of a chronic condition from observations with intervention. Stat Med 18:1355-71 |
