臺灣博碩士論文加值系統

居家血壓監測為近幾年台灣高血壓治療指南中最推薦的血壓追蹤之監測方式，但此方法有無法追蹤夜間睡眠期間血壓之缺點。然而，夜間睡眠期間相較於白天活動期間之血壓的下降幅度若低於正常值，會使心血管事件和死亡發生風險提高。因此本研究使用台灣高血壓相關心臟疾病臨床試驗聯盟資料庫，透過機器學習方法利用居家血壓及其他變數來預測病人的夜間血壓下降情形是否正常，並提供模型建議和分析策略協助臨床上的醫師提早判斷病人的狀況。

本研究分別對於多波重複測量、單波重複測量、橫截面資料三種不同角度進行模型建立與預測，總共使用二元混合模型森林(Binary Mixed Model Forest, BiMM Forest)系列模型、隨機森林、Logistic Regression以及XGBoost 做比較，在不同資料形態下挑出最合適的模型，在資料上，也分為無納入共變數(Covariates)以及有納入共變數兩種資料情況做討論。

最終發現針對台灣高血壓相關心臟疾病臨床試驗聯盟資料庫而言，在所有情況下有納入共變數之預測能力皆會優於沒有納入共變數之預測能力，而重複測量資料表現也優於橫截面資料，預測能力最好之情形為多波重複測量且有納入共變數之Visit1預測Visit2，使用 BiMM Forest 系列模型之 1 次迭代模型，準確度為 81.7%，表示居家血壓及其他變數有部分能力可以預測夜間血壓下降情形。

In recent years, home blood pressure monitoring (HBPM) is the most recommended monitoring method for blood pressure tracking in Taiwan's hypertension treatment guidelines. Still, this method has the disadvantage of monitoring blood pressure during nighttime sleep. However, an irregular nocturnal dipping rate would increase the risk of cardiovascular events and death. Therefore, this study uses the Taiwan Consortium of Hypertension associated Cardiac Disease (TCHCD) database and uses machine learning methods to predict nocturnal dipping status with home blood pressure and other variables and provide model recommendations analysis strategies to assist clinical physicians in early assess the patient's condition.

This study predicted the model from three different perspectives: multi-wave repeated measures, single-wave repeated measures, and cross-sectional data. Uses Binary mixed model forest (BiMM Forest) series models, random forest, Logistic Regression, and XGBoost model to compare and select the most suitable model under different data forms. In terms of the confounding effects, we implemented the analysis with and without covariates.

At the end of the study, we found out that for the TCHCD database, the performance of the included covariates was better than that of the non-included covariates in all cases, and the repeated measures data also performed better than the cross-section data. According to the data, the best prediction ability is the case of multi-wave repeated measures, and the inclusion of covariates in Visit1 predicts Visit2. We were using one iteration of the BiMM Forest series model, with 81.7% of accuracy, indicates that home blood pressure and other variables can predict. Predict nocturnal blood pressure drop.

誌謝 i
中文摘要 ii
Abstract iii
圖目錄 v
表目錄 vi
第一章 緒論 1
1.1 研究背景 1
1.2 研究目的 2
第二章 文獻探討 3
2.1 機器學習 3
2.2 隨機森林(Random Forest, RF) 4
2.3 重複測量資料機器學習預測模型 8
2.4 夜間血壓下降情形預測 9
2.5 程式語言 10
第三章 研究資料與方法 11
3.1 研究資料 11
3.1.1 應變數 13
3.2 研究方法 14
3.2.1 資料預處理 17
3.2.2 隨機森林插補法 22
3.2.3 隨機森林 25
3.2.4 二元混合模型森林 25
3.2.5 羅吉斯迴歸 28
3.2.6 XGBoost 29
第四章 研究結果 33
第五章 討論 36
參考文獻 37
附錄 39

Breiman, L., Friedman, J., Olshen, R., & Stone, C. J. C. S. P. S. (1984). Classification and
regression trees. Wadsworth & Brooks.
Breiman, L. J. M. l. (1996). Bagging predictors. 24(2), 123-140.
Breiman, L. J. M. l. (2001). Random forests. 45(1), 5-32.
Chen, T., & Guestrin, C. (2016). Xgboost: A scalable tree boosting system. Proceedings of
the 22nd acm sigkdd international conference on knowledge discovery and data
mining,
Cho, M.-C. J. K. C. J. (2019). Clinical significance and therapeutic implication of nocturnal
hypertension: relationship between nighttime blood pressure and quality of sleep.
49(9), 818-828.
Lin, H.-J., Wang, T.-D., Chen, M. Y.-C., Hsu, C.-Y., Wang, K.-L., Huang, C.-C., . . . Chuang,
W.-P. J. A. C. S. (2020). 2020 consensus statement of the taiwan hypertension society
and the taiwan society of cardiology on home blood pressure monitoring for the
management of arterial hypertension. 36(6), 537.
Mitchell, T. M. J. B. R., IL: McGraw Hill. (1997). Machine learning. 1997. 45(37), 870-877.
Ngufor, C., Van Houten, H., Caffo, B. S., Shah, N. D., & McCoy, R. G. J. J. o. b. i. (2019).
Mixed Effect Machine Learning: a framework for predicting longitudinal change in
hemoglobin A1c. 89, 56-67.
Polikar, R. J. I. C., & magazine, s. (2006). Ensemble based systems in decision making. 6(3),
21-45.
38
Radha, M., De Groot, K., Rajani, N., Wong, C. C., Kobold, N., Vos, V., . . . Velthoven, N. J.
P. m. (2019). Estimating blood pressure trends and the nocturnal dip from
photoplethysmography. 40(2), 025006.
Speiser, J. L., Wolf, B. J., Chung, D., Karvellas, C. J., Koch, D. G., Durkalski, V. L. J. C., &
Systems, I. L. (2019). BiMM forest: A random forest method for modeling clustered
and longitudinal binary outcomes. 185, 122-134.
Speiser, J. L., Wolf, B. J., Chung, D., Karvellas, C. J., Koch, D. G., Durkalski, V. L. J. C. i.
S.-S., & Computation. (2020). BiMM tree: A decision tree method for modeling
clustered and longitudinal binary outcomes. 49(4), 1004-1023.
Stekhoven, D. J., & Bühlmann, P. J. B. (2012). MissForest—non-parametric missing value
imputation for mixed-type data. 28(1), 112-118.
Speiser, Jaime Lynn, "Decision Tree and Random Forest Methodology for Clustered and
Longitudinal Binary Outcomes" (2017). MUSC Theses and Dissertations. 381.