استخراج دانش استفاده از دارو با استفاده از نقشه خودسازمانده و نظریه مجموعه راف
|کد مقاله||سال انتشار||مقاله انگلیسی||ترجمه فارسی||تعداد کلمات|
|29498||2007||10 صفحه PDF||سفارش دهید||محاسبه نشده|
Publisher : Elsevier - Science Direct (الزویر - ساینس دایرکت)
Journal : Expert Systems with Applications, Volume 33, Issue 2, August 2007, Pages 499–508
Cardiovascular disease is becoming the major cause of death in many industrialized countries. People who receive long-term treatments usually ignore the progress of the disease states. Therefore, it is critical and necessary to evaluate drug utilization and laboratory test in order to discover the knowledge that is beneath and can be extracted from those raw data. This paper utilizes techniques of self-organizing map (SOM) and rough set theory (RST) to discover the trend of individual patient’s condition. With 10-fold cross-verification, the proposed SOM–SOM–RST process successfully and effectively detects patients whose diagnosis codes have been changed during the period of investigation and attains an accuracy of approximate 98%. This method can remind physicians to reevaluate the disease conditions of their patients.
Pharmaceutical care is the most common treatment for ambulatory patients. When drugs are administered to a patient, the clinical goal is to achieve certain effectiveness. A modern clinician is expected to justify drug selection, hence is expected to know how potential beneficial effect are brought about (Hyfte, Maas, Tjandra-Maga, & Robbé, 2001). Although laboratory tests are tools helpful in evaluating the health status of individuals, many patients who receive long-term care do not reflect the disease state on laboratory tests significantly. Physicians, hence, might ignore progressive information revealed by these data. Upon this, it is crucial to find the trend, relationship, and knowledge among clinical data. Cardiac disease has become the major cause of death in industrialized countries. Clinical medicine researchers have indulged themselves in the field of etiology that is based on the evidence of medicine (Department of Health, 2005 and Kannel, 1997). Many researches have been done regarding the relationship between individual variables and disease states. Therefore, it is beneficial if the vascular disease is evaluated with the mapping of experimental data and treatment data (Department of Health, 2005). Drug utilization review concerns drugs interaction, drug-disease contraindication, therapeutic duplication, and early/late refill. It is rare to evaluate the laboratory values and treatment outcomes concurrently. A physician would find it more appropriate to describe his or her knowledge by means of rules like “if fever is high and cough is moderate then disease is X” rather an “if fever is 38 °C and cough is 5–10 times a day then disease is X”. Decision trees from symbolic rules are widely used to separate and classify elements in a field of domain knowledge. The rule-based method such as ID3 algorithm (Han & Kamber, 2001) aims to minimize the complexity of the tree and thereby maximizing the information contained in each decision node. If the case is large, the decision trees tend to get quite complicated and difficult for human interpretation. Furthermore, classical IF–THEN rules can only separate the domain parallel according to a single component at a time. And decision tree would be little use where the separation rules are non-linear. Thus, it could be beneficial to utilize neural network methods in order to explore the relationship and its trend. Techniques of utilizing soft computing in medical diagnostic support system (MDSS) are proven to be outperforming the physicians. Salchenberger, Venta, and Venta (1997) proved that the accuracy of neural network on detecting breast implant rupture than the radiologists. Baxt (1990) noted that the neural network correctly identifies 92% of patients with acute myocardial infarction and 96% of patients without infarction. Researches also stated that neural network can be more accurate than experienced cardiologist in diagnosing acute myocardial infarction (Josefson, 1997). West and West (2000) demonstrated the SOM can be an effective tool for MDSS model selection that identify optimal target values for a breast cancer MDSS for two conditions, the lowest overall error rate and minimize the false negative error. Recently, SOM network also has been applied on medical images (Coppini, Diciotti, & Valli, 2004) and automatic ultrasound contouring for breast tumors (Huang & Chen, 2004). Laboratory tests are the most common methods for physicians to investigate the disease state of patients. The result of laboratory test is also the evidence to claim health insurance reimbursement. For example, diabetes or cardiac disease patients, whose TG level over 200 mg/dl can receive treatment to achieve the goal of lowering than 160 mg/dl. In this paper, laboratory tests that relate to the disease under evaluation were extracted from clinical database. This paper utilizes techniques of two-level self-organizing map (SOM) (Kiviluoto & Bergius, 1998) and rough set theory (RST) (Pawlak, 1982 and Pawlak, 1991) to discover the trend of individual patient’s condition. The process of SOM–SOM–RST is to perform SOM for laboratory data and medicine treatments in Taiwan, and then IF–then rules are extracted by RST. Following Section 1, the related literatures are briefly reviewed in Section 2. Section 3 introduces proposed model in knowledge discovery. In Section 4, the experimental results are described, and conclusions are finally made in Section 5.
نتیجه گیری انگلیسی
The arena of medicine is full of opportunity, and challenge. With the increasing number of chronic disease in industrialized countries, medical budget is getting concerned on prevention and evaluating/auditing systems more and more. This paper has proposed a model, which is based on the combination of laboratory tests (by first SOM), medicine treatments (by second SOM), and extracting drug utilization knowledge (by RST). The results of SOM–SOM–RST show that it can successfully and effectively detect patients whose diagnosis codes have been changed during the period of investigation and attains an accuracy of approximate 98% with 10-fold cross-verification.