Predicting end-stage renal disease: Bayesian perspective of information transfer in the clinical decision-making...

Predicting outcomes such as end-stage renal disease (ESRD) by integration and better utilization at individual level of epidemiologic data may facilitate clinical decision-making processes.

To predict individual ESRD risk in an average patient in the United States, ESRD prevalence and levels of uncertainty and conditional risk factors independence were considered by population data (1998) and pooled analysis of 11 randomized trials. Data integration and input were by decision-tree simulation approach (simple, parallel, and sequential scenarios) and Bayes' theorem. Sensitivity analysis and risk profiles were employed to address uncertainty and assess different risk factor combinations. A health state values, associated with ESRD outcome levels, were taken from the literature.

In this theoretical study, we provided a scholarly example about the use of two known risk factors (urinary protein >/=3 g/day and systolic blood pressure >/=140 mm Hg) to predict individual ESRD risk in an average patient in the United States. The highest posterior (decisional) probability of ESRD occurrence (risk of 3.61% to 5.07%) in the individual patient was associated with the worst health state, as assessed by multidimensional scenarios when both risk factors were present.

Decision tree models through an empirical Bayesian approach may serve to predict the individual ESRD risk on the basis of simple epidemiologic, demographic, and clinical information that is easily available already at the first patient evaluation.