機器學習理論預測乳癌病人術後十年生活品質:前瞻性世代追蹤研究

研究背景及目的乳癌婦女的情緒困擾及生活品質是醫護人員亟須關注的議題，藉由本研究瞭解身處台灣文化脈絡的乳癌婦女的生活品質及其影響因素，期望將研究結果提供相關研究人員或臨床醫護人員，作為乳癌照護介入措施或衛教指導的參考。因此本研究目的為探討機器學習理論於乳癌病人術後十年生活品質預測模式之準確度，並且利用全域敏感度分析評估乳癌病人術後十年生活品質影響因子之權重。研究方法本研究針對南台灣某三家醫學中心自2007年6月至2010年6月進行乳癌手術病患，為本研究之樣本。術後以電話進行追蹤訪問，測量工具以病歷審查(Chart review)與健康相關生活品質評估量表，本研究健康相關生活品質量表(SF-36)共有2個生理層面(PCS) 及心理層面(MCS)。本研究利用常見機器學習理論，包括:類神經網路(Artificial Neural Networks, ANN)、支援向量機(Support Vector Machines, SVM)、最近鄰居(K-neatest Neighbors, KNN)、線性複迴歸(Multiple Linear Regression, MLR)評估乳癌病人術後十年生活品質預測模式之準確性，進而利用全域敏感度分析探討乳癌病人術後十年生活品質影響因子之權重。研究結果本研究結果顯示類神經網路(ANN)預測模式優於最近鄰居演算分析模式(KNN)、支持向量機分析(SVM)及線性複回歸(MLR)預測模式。由研究結果發現在術後十年生活品質之生理層面(PCS)中，最重要預測因子為合併症指數(CCI)，其次為術後30天內再住院與術前健康相關生活品質(SF-36 PCS)；於心理層面(MCS)中，研究結果發現，最重要之預測因子為合併症指數(CCI)，其次為腫瘤期別(Stage)與術前健康相關生活品質(SF-36 MCS)。結論與建議整體而言，本研究運用資料處理藉由機器學習理論預測技術研究，針對乳癌進行長期的生活品質分析研究，透過生活品質預測模式了解乳癌術後的生活品質及其影響因素，以提供醫療照護者針對不同年齡族群擬定照護指引參考及未來臨床照護上的運用。也讓醫療照護體系能妥善及有效的應用，提升其醫療品質。幫助患者減緩治療疾病過程中的不適，有助於生活品質的提升。

關鍵字

乳癌；生活品質；機器學習理論；類神經網路；全域敏感度分析

並列摘要

Purpose Emotional distress and quality of life are the critical issues in women with breast cancer. The research results of the study can provide precious information to relevant researchers or clinical medical staff. Therefore, this study purposed to explore the accuracy of the machine learning theory in prediction of postoperative 10-year quality of life of breast cancer patients, and to evaluate the weight of the impact factors of postoperative 10-year quality of life of breast cancer patients. Methods This prospective cohort study recruited breast cancer surgery patients from three medical centers in Southern Taiwan undergoing breast cancer surgery from June 2007 to June 2010. During the study period, patient characteristics and clinical characteristics were collected from the chart review and preoperative and postoperative 10-year health-related quality of life was interviewed by the reviewers by using the SF-36 (PCS and MCS). The patient characteristics, clinical characteristics, quality of medical care, and preoperative quality of life are the significant predictors associated with postoperative 10-year quality of life of breast cancer patients. This study uses common machine learning algorithms, including: artificial neural networks (ANN), support vector machines (SVM), K-nearest neighbors (KNN), and multiple linear rgression (MLR) to evaluate the accuracy of 10-year postoperative quality of life and uses global sensitivity analysis to explore the important predictors on postoperative 10-year quality of life. Results The results showed that the ANN mode is superior to the KNN, SVM, and MLR modes in predicting postoperative 10-year quality of life. Moreover, according to the ANN model, the most important predictor of postoperative 10-year SF-36 PCS score is the Charlson comorbidity index (CCI) followed by re-hospitalization within 30 days after surgery and preoperative SF-36 PCS score. Additionally, the study also found that the most important predictor of postoperative 10-year SF-36 MCS score is the Charlson comorbidity index (CCI) followed by the tumor stage (Stage) and preoperative SF36 MCS score. Conclusions and Suggestions Overall, this long-term prospective cohort study found that the ANN model is superior to the other prediction medicals in predicting postoperative 10-year quality of life. It also found that Charlson comorbidity index (CCI), re-hospitalization within 30 days after surgery or cancer stage and preoperative quality of life were the most important predictors on postoperative 10-year quality of life after breast cancer surgery. It also allows the medical care system to be properly and effectively applied to enhance its medical quality. Helping patients ease the discomfort in the treatment of diseases, and help improve the quality of life.