استفاده از تکنیک های داده کاوی برای پیش بینی بستری شدن بیماران همودیالیزی
|کد مقاله||سال انتشار||مقاله انگلیسی||ترجمه فارسی|
|22192||2011||10 صفحه PDF||21 صفحه WORD|
Publisher : Elsevier - Science Direct (الزویر - ساینس دایرکت)
Journal : Decision Support Systems, Volume 50, Issue 2, January 2011, Pages 439–448
2. مواد و روش ها
2.2. مفهوم زماني
شکل 1. اپراتور زماني
شکل 2: طرح کلی سیستم تجزیه و تحلیل اطلاعات هوشمند
2.3 درخت تصمیم گیری
شکل 3. جریان تجزیه و تحلیل داده ها بیماران همودیالیزی.
2.4. قوانین استخراج شده وابسته به حداقل چندین معیار پشتیبانی
3. توسعه سیستم پشتیبان تصمیم گیری
شکل 4 نمودار پیش پردازش اطلاعات.
3.1 جمع آوری داده ها
شکل 5. جریان مفهوم زمانی
3.2 پیش پردازش اطلاعات
جدول 1. مقادیر آستانه آیتم های بیوشیمیایی مورد آزمایش برای تبدیل TA
3.3 تبدیل به مفهوم زمانی
3.3.1 تبدیل به TA پایه
3.3.2 تبدیل به TA پیچیده
شکل 6. TA پایه
شکل 7. نمودار شماتیک مفهوم زمانی.
3.4 استخراج قانون بستری شدن
3.4.1 مجموع قوانین استخراج با حداقل حمایت چندگانه
3.4.2 درخت تصمیم گیری
جدول 2. داده های جزئی پس از اجرای تحولات پیچیده TA برای مجموع قانون معادله استخراج.
جدول 3. داده های جزئی پس از اجرای تبدیل TA پیچیده برای درخت تصمیم گیری.
4. نتایج محاسباتی
4.2 نتایج داده کاوی
4.2.1 قوانین درخت تصمیم گیری
جدول 4. فرمول های ارزیابی عملکرد.
جدول 5. مقررات بالقوه معنی دار بستری بیماران HD از DT
4.2.2 قوانین الگوریتم MSapriori
جدول 6. از نظر بالینی معنی دار بودن قوانین بستری بیماران HD از الگوریتم MSApriori
5. نتیجه گیری
Hemodialysis patients might suffer from unhealthy care behaviors or long-term dialysis treatments and need to be hospitalized. If the hospitalization rate of a hemodialysis center is high, its service quality will be low. Therefore, decreasing hospitalization rate is a crucial problem for health care centers. This study combines temporal abstraction with data mining techniques for analyzing dialysis patients' biochemical data to develop a decision support system. The mined temporal patterns are helpful for clinicians to predict hospitalization of hemodialysis patients and to suggest immediate treatments to avoid hospitalization.
End stage renal disease (ESRD), commonly known as uremia, is a severe chronic state corresponding to the final stage of kidney failure. In ESRD, kidneys are not able to purify blood from metabolites or to exclude water from the body. Without medical intervention, ESRD patients may die or remain in intensive care unit (ICU) for a long time. These patients require either a kidney transplant or blood-filtering dialysis treatment. The former treatment is difficult to obtain because of a long waiting list and certain patients, such as the elderly, cannot undergo a transplant. The latter includes two main categories, hemodialysis (HD) and peritoneal dialysis (PD). In HD, the blood passes through an extra-corporal circuit where metabolites (e.g. urea) are eliminated. The acid-based equilibrium is re-established and excess water is removed . PD works on the same principles of solute diffusion and fluid ultra filtration as HD, but the blood is cleaned inside the body rather than through a machine . More than 80% of ESRD patients are currently treated with HD . HD patients typically undergo a dialysis session for 4 h, three times a week. During the long-term dialysis treatment, patients will likely receive hospitalization due to caregiver carelessness or other infections. This has been the main reason for HD patient hospitalization in previous years. High hospitalization rate for a hospital hemodialysis department (HHD) means low service quality in health care. Therefore, the HHD focuses on reducing hospitalization rate. Preventing hospitalization of HD patients from the perspective of preventive medicine is also very important. This paper develops a decision support system to predict hospitalization of HD patients based on a real dataset collected from a hemodialysis center in Taiwan. The HHD examines HD patients receiving long-term treatment to obtain biochemical data during hemodiaysis sessions, such as hematocrit (Hct), albumin, alkaline-p, cholesterol, triglyceride, blood urea nitrogen (BUN), creatinine, uric acid, Na, etc. . The accumulated data over time contains a set of patient variables that are monitored during each dialysis session. The collected data are sequences of multidimensional time series . For time series data, the temporal abstraction (TA) method proposed by Shahar  can be integrated with data mining techniques to support data analysis. For example, Bellazzi et al.  successfully applied temporal data mining techniques for assessing the clinical performance of HD services such as preprocessing, data reduction, multi-scale filtering, association rule discovery, etc. They found their approach to be suitable for knowledge discovery in clinical time series. Using an auditing system context for dialysis management helped clinicians improve their understanding of patients' behavior. Adlassnig et al.  proposed and discussed promising research directions in the field of TA and temporal reasoning in medicine. They identified and focused on fuzzy logic, temporal reasoning and data mining, health information systems, and temporal clinical databases and recommended developing decision support systems to properly manage the multifaceted temporal aspects of information and knowledge encountered by physicians in their clinical work. Stacey and McGregor  surveyed previous research in developing intelligent clinical data analysis systems that incorporate TA mechanisms and present research trends. They suggested the necessity of fusing data mining and TA processes to fully exploit new knowledge from stored clinical data through data mining and apply it to data abstraction. For TA rule mining, Sacchi et al.  proposed a new kind of TA rule and related algorithms for the extraction of temporal relationships between complex patterns defined over time series. Their approaches could be used in a variety of application domains, and they were already tested on two different biomedical problems. Concaro et al.  developed a general methodology for the mining of TA rules on sequences of hybrid events for Diabetes Mellitus. The method was capable to characterize subgroups of subjects, highlighting interesting frequent temporal associations between diagnostic or therapeutic patterns and patterns related to the patients' clinical condition. They concluded that the approach could find a practice for the evaluation of the pertinence of the care delivery flow for specific pathologies. Based on the literature review, this study integrates TA with data mining techniques for analyzing biochemical data of HD patients to discover temporal patterns resulting in hospitalization. This work develops a decision support system to provide clinicians with association rules and the probability of HD patients' hospitalization for implementing preventive medicine to decrease hospitalization incidence. This system will hopefully help to understand patients' changing biomedical data that leads to hospitalization and to improve service quality of the hemodialysis center. The remainder of this paper is arranged as follows: The Materials and methods Section describes hemodialysis and temporal abstraction, the Development of decision support system Section demonstrates the development of the decision support system used in this paper, the Computational results Section illustrates the experimental results using the combined approach for hemodialysis patients' data analysis, and the last section give the conclusions.
نتیجه گیری انگلیسی
Medical data analysis includes data gathering, preprocessing, result evaluation, favorable interaction, and discussions with clinicians for correct analytic results , ,  and . This study used data mining techniques for extracting professional knowledge. This method is an improvement over traditional face-to-face discussions with professional persons, and enables us to obtain important knowledge effectively and quickly. The experimental results show that different data mining methods can be combined effectively, and more abundant patterns can be found for practical applications. Furthermore, we can add domain knowledge prior to data analysis by combining the TA method, to make mining results more likely comprehended by the clinician. Therefore, TA is an indispensable method for future medical time series data analyses. Among the hospitalization patterns found in this study, albumin is the most important index for predicting patients' hospitalization. This index is also currently used clinically for predicting patients' death rate. The results of this study therefore have clinical significance. Predicting patients' hospitalization by biochemical value evolution of blood examination has been an undefined biochemical item in previous clinical applications. After validating by medical care personnel, the time evolution of this index value proves to have definite relevance for hospitalization. This study combined TA with data mining techniques to analyze dialysis patients' biochemical data. The mined temporal hospitalization patterns are helpful for doctors to diagnose patient hospitalization probability and to suggest some immediate treatments to avoid hospitalization. Finally, we hope this research will help hemodialysis centers to improve health care quality. Many relevant methods and concepts could be added for analysis results. Directions for future research include the following three points. First, Adlassnig et al.  indicated that combining TA with Fuzzy Logic is more coincident for describing actual temporal data situations and is also a direction worthy of study. Second, HD patients' hospitalization rules can be obtained either by one-class classification method  or by subgroup discovery algorithm . Third, for mined temporal patterns, implementing a system to assist medical care personnel to carry out medical intervention or treatment, namely to automate the overall flow from temporal mining preprocessing to rule generating, would assist medical personnel with daily business.