腦中風為國內常見疾病,依據行政院衛生福利部資料統計2012年腦血管疾病居國內十大死因第三位,根據國內衛生福利部推廣急性後期照護模式(Post-Acute Care)計畫中指出腦中風病人有超長住院的情況,約10.4%有超長住院情形,佔所有腦中風住院個案病患天數總和之38.9%、住院費用總和之47.8%,計畫中指出在社區醫院提供專業團隊的急性後期照護,可以顯著促進病人功能的恢復和減少住院天數,急性後期照護確實有其顯著的成效,因此,如何有效評估腦中風病患實施PAC的照護,對醫院、病患及家屬及健保資源之運用與規劃有一定重要性。 本研究以腦中風疾病之病患資料庫為研究對象,彙整出影響腦中風病患超長住院及再住院之影響因子,採用粒子群最佳化演算法、交叉熵演算法、基因邏輯斯迴歸演算法、倒傳遞類神經網路、支援向量機、案例式推理系統交互建構十一種評估模型用於預測腦中風病患的住院情況,並以案例式推理技術建構腦中風病患超長住院評估系統。研究結果顯示在再住院預測模型方面,大部分模型皆有88%以上平均測試準確率,其中以交叉熵演算法結合支援向量機模型為最佳,平均準確率達88.873%;平均ROC曲線下面積達0.8349。而在超長住院評估系統方面,以交叉熵演算法之權重結合案例式推理系統有最佳表現,其平均準確率達88.536%;平均ROC曲線下面積達0.7988。本研究結果可做為醫師與相關醫療人員對於腦中風病患再住院與超長住院相關議題之參考依據,且對於醫療機構之醫療資源規劃與住院品質有實質的幫助。
Stroke is a common domestic disease. According to the statistics of Ministry of Health and Welfare, Executive Yuan (MHW), cerebrovascular disease ranks third among the ten leading causes of death for 2012. According to the Post-acute Care(PAC) Model Promotion Plan of MHW, extended hospital stay has been seen in stroke patients, about 10.4% had extended hospitalization, accounting for 38.9% of the total number of hospitalization days for stroke patient cases and 47.8% of the total hospitalization costs. It was pointed out in the plan that community hospitals with a professional team to provide acute post-care significantly contributed to patients’ functional recovery and reduced number of hospitalization days. Hence, acute post-care is said to be significantly effective. How to effectively assess stroke patients’ PAC is of considerable importance for hospitals, patients and their families, as well as National Health Insurance (NHI) resource use and planning. The database of stroke patients was adopted as the research participants in this study. The influential factors of the stroke patients’ extended hospitalization and re-admission were compiled. The particle swarm optimization algorithm, cross entropy algorithm, genetic logistic regression algorithm, back propagation neural network, support vector machine, and case-based reasoning system combined were used to construct 11 assessment models for predicting stroke patients’ hospitalization situations. Additionally, the case-based reasoning technique was applied to construct the stoke patients’ extended hospital stay assessment system. Findings show that in terms of the re-admission prediction model, most models had mean test accuracy exceeding 88%. In particular, the cross entropy algorithm, combined with the support vector machine model, produced the best results, with the mean accuracy reaching 88.873% and the mean area under curve of ROC (AUC) reaching 0.8349. As for the extended hospitalization assessment system, the cross entropy algorithm weights, combined with the case-based reasoning system, had the best performance, with the mean accuracy of 88.536% and the mean AUC of 0.7988. The research results shall serve as a reference for physicians and relevant medical personnel when exploring issues related to brain stroke patients’ re-admission and extended hospitalization. The results are also expected to be substantially helpful for medical institutions in medical resource planning and hospitalization quality improvement.