以生活型態資料與醫療資訊輔助之自殺防治決策系統概念架構

Abstract

自殺問題是社會、國家及全球的重要挑戰。治療有自殺意圖或行為的個案是臨床上的挑戰。由於其高度複雜性，預測自殺的因子在短期風險上預測能力並不佳。本計畫將以改善自殺病患照護為研究目標，透過行動科技，機器學習與臨床結合，以降低自殺發生。 生理活動與情緒之間的關聯，在神經科學中，近年來由人工智慧演算法得到了不同的啟發，人類大腦不停產生預測模型，並且將此預測與實際收到的感官經驗進行抽樣觀測及比較。大腦會擷取環境中的統計規律性，形成貝氏機率模型，此種預測編碼亦會影響到代謝需求的預測，Karl Friston之自由能計算理論對神經科學帶來影響，本研究亦嘗試將理論帶到臨床，解釋心理治療之神經機制。 本研究與醫院自殺防治中心團隊協同合作，運用穿戴式裝置收集個案的生理數據與生活型態相關資料，進行數據分析及AI技術研發，並進行個案追蹤。建置資料庫之軟體及硬體資源，發展自殺防治輔助決策系統，使照護團隊得以有效照顧病人並與個案維持良好治療關係。 初步結果探討重複自殺住院病人的自主神經系統，靜止時，迷走神經活性與貝克自殺意念指數 (Beck suicide scale index) 呈正相關 (r = 0.33 P = .019)。控制其他變數後，靜止迷走神經活性與住院天數相關 (beta coefficient = 3.00; P = .030).在長期住院病人 (大於30天)，有更高的迷走神經活性 (odds ratio 5.08, P = .017)。另外，亦發現治療者迷走神經活性在治療自殺個案時有更高的活性。 提出之假設用長新冠症狀 (long COVID) 造成之憂鬱自殺個案為例，活動的減少與日夜節律的變化可能與腦中預測代謝能量的失調有關。我們同時用大腦的預測編碼模型說明辯證行為治療的機轉。大腦的網路，包含預設模式網路(Default Mode Network, DMN)、杏仁核、恐懼迴路及酬賞迴路，都與大腦預測的生成相關，治療的過程以預測錯誤 (prediction errors) 信號回饋，逐步修正腦中酬賞反應及預測模型。 使用人工智慧以辨識台灣自殺個案之特徵。將建立模型以預防重複自殺，幫助臨床人員長期追蹤並掌握個案。研究者預計設計「緊急救生包」手機應用程式，在個案面臨緊急狀況時，減輕壓力與負面情緒。應用程式可進行心情及自殺意念回報，亦可提供資源，使用情形也將進行資料收集。 我們希望藉由分析遠距醫療平台的資料，可以對自殺背後複雜的神經生物學機制與因子間的互動進行闡釋，提出理論。期望此臨床研究能帶動人工智慧與自殺防治研究的創新。

The issue of suicide is a major challenge faced by the society, the country and the whole world. Treating patients with ideations or attempts of suicide is a clinical demanding work. Due to the high complexity of suicide, factors predicting long-term suicide risk had poor performance in predicting short-term risk. The purpose of this project is to improve the quality of medical care with smart technology. It is expected that the combination of information communication technology, machine learning and clinical work will save lives. The neuroscientific association between physiological activity and emotions are inspired by recent progress in Artificial Intelligence and algorithms. The human brain keeps generating prediction models and simultaneously sampling, comparing with the incoming sensory information and experiences. The brain will take the statistical predictability of the environments and build a Bayesian model. The predictive coding process also participates in expecting the metabolic need. The free energy principle proposed by Karl Friston has brought great influence on neuroscience. This study also attempts to apply it to clinical practice in elucidating the neurobiological mechanism of psychotherapy. This project integrates lifestyle data and physiological information of real-world patients of suicide, using wearable devices to assist the hospital suicide prevention center to manage and track suicidal cases. We also establish a database with software and hardware resources, build a suicide decision-making system, assist the mental healthcare team effectively in taking care the patient and maintaining a good therapeutic relationship between the patients and the clinicians. The preliminary results suggest the resting vagal activity was positively associated with a higher Beck Scale for Suicidal Ideation score (r = 0.33, P < 0.019). Stepwise multiple regression analysis showed a significant correlation between resting vagal activity and hospitalization length after adjusting other variables (beta coefficient = 3.00; P < 0.030). Patients with a higher resting vagal activity had more prolonged hospitalizations (hospitalization beyond 30 days) after controlling for other variables (odds ratio = 5.08, P < 0.017). In addition, we also observed the higher vagal tone in therapist when interacting with suicidal patients. We propose our hypothesis using a long COVID patient with depression and suicide as an example. The reduction of activities and circadian rhythm change might be related with allostatic dysregulation in predicting energy expenditure. We also explain the mechanism of Dialectical Behavior Therapy (DBT) by the predictive coding process of the brain. The brain networks, including Default Mode Network (DMN), amygdala, fear circuitry and reward circuitry, are all related to the generation of prediction. The treatment process may involve the feedback of prediction error signals, gradually update the brain reward pathway and prediction models. The core technology is to utilize Artificial Intelligence to identify the characteristics and patterns of suicide cases in Taiwan. We would establish prediction models to prevent repetitive suicide attempts and help clinicians to track and comprehend the situation of the clients in the long term. We will develop a crisis survival mobile application to assist patients to alleviate pressure and negative emotions in case of emergency. The monitoring of mood and suicidal ideations, as well as provision of resources that can be used will be implemented via the mobile application and the record of user condition will also be collected. We hope by analyzing the data of the telemedicine platform, we may propose a better explanation of the mechanism behind the transactional characteristics of the neurobiology of suicidality and real-world clinical variables and bring innovations and advances in the fields of psychiatry and Artificial Intelligence in Taiwan.