| Summary: | 碩士 === 慈濟大學 === 醫學資訊研究所 === 95 === According to the data from Taiwan Society of Nephrology, there are more than 800000 chronic renal failure patients in Taiwan and the number increase day after day. Besides the data from United States Renal Data System (USRDS) also revealed Taiwan had the highest incidence of end stage renal disease in the worldwide on 2006. The health insurance budget for ESRD was about 24.5 billion NT dollars on year 2002, and the amount is higher than the budget for Cancer 23.6 billion. However, as we know, cancer is the leading cause of death in Taiwan for many years, but the renal disease only on the eighth. Hence, renal disease really becomes a heavy load to the government. [1]
In year 2006, the cerebralvascular diseases and heart diseases are the second and third causes of death in Taiwan; moreover, CVD is the main reasons for uremia patients to induce mortality.
According to the clinical data, uremia patients commonly have two main categories of risk factors for cardiovascular disease (CVD) : (1) Traditional risk factors include hypertension, dyslipidemia, age, hyperglycemia, smoking and physical inactivity; The risk factors that related to renal failure include sodium water overloading, anemia, hyperhomocystinemia, hypoalbuminemia, inflammation（high sensitive C-reactive protein）, vascular calcification, elevated CaXP and increase pulse pressure. All of them are regularly followed up in uremia patients. [2]
Artificial neural networks (ANNs) are mathematical models of neurons which try to simulate human neurons to handle highly complex, nonlinear problems. The method is successfully applied in a wide range of medical areas to assist diagnosis. We investigated several ANN structures to predict CVD by using routine clinical data.
According to the DBNN method, the accuracy, sensitivity and specificity of predicting CVD are 29.4%、52.63% and 24.10% respectively in the training group, and 30.0%、57.14% and 25.58% respectively in the testing group. Under the BP method, the accuracy,sensitivity, and specificity of predicting CVD are 73.53%、74.19% and 66.67% in the training group, and 66.00%、72.50% and 40.00% in the testing group.
Conclusions: Our method may help health providers to screen the potential CVD patients by using regular, low-cost clinical and biochemical data.
|
|---|
