使用可穿戴設備資料和臨床評估資料的恐慌症機器學習預測模型

Machine Learning Prediction Models for Panic Disorder Using Wearable Device Data and Clinical Evaluation Data

陳珮甄(Pei-Chen Chen)

指導教授 ： 賴飛羆

國立臺灣大學/電機資訊學院/生醫電子與資訊學研究所/碩士(2021年)

https://doi.org/10.6342/NTU202101687

摘要

恐慌症是一種焦慮症，在全球的患病率約為2-6%。典型的恐慌發作是突然和反覆的強烈恐懼發作，並在恐懼感幾分鐘內達到高峰。恐慌症患者通常會擔心下一次發作的時間，並試圖避免與驚恐發作相關的地點、情況或行為來預防發作。病患為了避免恐慌症的發作經常會導致他們生活各方面出現嚴重問題，甚至可能導致恐懼症。
本研究招募了初步診斷為恐慌症且年齡在 20-75 歲之間的參與者。本研究使用的數據包括生理數據、環境數據和問卷數據。收集這些數據的工具包括穿戴式裝置（Garmin Vivo smart4）、手機應用程式和政府的環境開放數據平台。我們嘗試了六個機器學習的模型來達到預測恐慌發作的目的。這些模型包括隨機森林、決策樹、線性判別分析、自適應提升、極端梯度提升和正則化貪婪森林。此外，本研究也使用了基於深度學習的模型，具有四個全連接層。
我們使用生理數據（生活型態數據）、環境數據和問卷數據作為預測模型的輸入。在這項研究中，總共招募了 59 名參與者。對7天內恐慌發作預測，結果顯示最佳的模型是具有 94.6% 敏感性和 96.8% F1-score 的隨機森林模型。接收者操作特徵曲線分析表明，預測恐慌發作的模型曲線下面積大於0.9。最重要的特徵是貝克焦慮量表值、貝克抑鬱量表值、平均心率和休息心率。
我們通過使用可穿戴設備、環境開放數據、臨床問卷和有監督的預測算法，在預測未來7天內發生恐慌症取得不錯的效果。通過本研究開發的預測模型，可以產生恐慌症的早期預警，提醒患者和個案管理師提前預防。

關鍵字

恐慌症 ； 恐慌發作預測 ； 穿戴式裝置 ； 智慧型手機應用程式 ； 開放環境數據 ； 機器學習 ； 深度學習

並列摘要

Panic disorder (PD) is a kind of anxiety disorder, with a life prevalence of around 2-6% worldwide. The typical panic attack (PA) is unexpected and repeated intense fear attacks, appearing suddenly and reaching a peak within a few minutes. Patients suffered from PD usually worry about the time of the next attack and actively try to prevent future attacks by avoiding locations, situations, or behaviors related to the panic attack. Worrying about PA and avoiding them can cause severe problems in all aspects of the patient's life and may even cause phobias.
Participants with the primary diagnosis of panic disorder and between 20-75 years old were recruited in the study. The data used in this study contained physiological data, environment data, and questionnaire data. The tools to collect these data were wearable devices (Garmin Vivo smart4), smartphone application, and the government's environmental open data platform. Six machine-learning-based models were implemented to predict PA. The models included random forest, decision tree, linear discriminant analysis (LDA), adaptive boosting (AdaBoost), extreme gradient boosting (XgBoost), and regularized greedy forest. Furthermore, the deep-learning-based model was also used in this study with four fully connected layers.
The combinations of physiological data (lifestyle data), environment data, and questionnaire data were used as the input feature. In the study, a total of 59 participants were recruited. For 7-day PA prediction, the result shows the best performance model is the random forest with 94.6% sensitivity and 96.8% F1-score. Receiver operating characteristic curve (ROC) analysis showed that the area under the model's curve predicting PA was greater than 0.9. The most important features are BAI value, BDI value, average heart rate, and resting heart rate.
We achieved excellent performance in predicting PA in the next 7 days using wearable devices, environment open data, clinical questionnaires, and supervised prediction algorithms. Through the prediction model developed in this study, an early warning of PA could be generated to remind patients and case managers to prevent them in advance.

並列關鍵字

panic disorder ； panic attack prediction ； wearable device ； smartphone application ； environment open data ； machine learning ； deep learning

參考文獻

[1] D. J. Robinaugh et al., "Assessing vulnerability to panic: a systematic review of psychological and physiological responses to biological challenges as prospective predictors of panic attacks and panic disorder," General psychiatry, vol. 32, no. 6, 2019.

[2] D. Caldirola and G. Perna, "Toward a personalized therapy for panic disorder: preliminary considerations from a work in progress," Neuropsychiatric disease and treatment, vol. 15, p. 1957, 2019.

[3] M. Elgendi and C. Menon, "Assessing anxiety disorders using wearable devices: Challenges and future directions," Brain sciences, vol. 9, no. 3, p. 50, 2019.

[4] H.-A. Chang, C.-C. Chang, N.-S. Tzeng, T. B. Kuo, R.-B. Lu, and S.-Y. Huang, "Decreased cardiac vagal control in drug-naïve patients with posttraumatic stress disorder," Psychiatry investigation, vol. 10, no. 2, p. 121, 2013.

[5] L. Cruz, J. Rubin, R. Abreu, S. Ahern, H. Eldardiry, and D. G. Bobrow, "A wearable and mobile intervention delivery system for individuals with panic disorder," in Proceedings of the 14th International Conference on Mobile and Ubiquitous Multimedia, 2015, pp. 175-182.