Ikäheimonen Arsi, Luong Nguyen, Baryshnikov Ilya, Darst Richard, Heikkilä Roope, Holmen Joel, Martikkala Annasofia, Riihimäki Kirsi, Saleva Outi, Isometsä Erkki, Aledavood Talayeh
Department of Computer Science, Aalto University, Espoo, Finland.
Department of Psychiatry, University of Helsinki, Helsinki, Finland.
J Med Internet Res. 2024 Dec 3;26:e56874. doi: 10.2196/56874.
Clinical diagnostic assessments and the outcome monitoring of patients with depression rely predominantly on interviews by professionals and the use of self-report questionnaires. The ubiquity of smartphones and other personal consumer devices has prompted research into the potential of data collected via these devices to serve as digital behavioral markers for indicating the presence and monitoring of the outcome of depression.
This paper explores the potential of using behavioral data collected with smartphones to detect and monitor depression symptoms in patients diagnosed with depression. Specifically, it investigates whether this data can accurately classify the presence of depression, as well as monitor the changes in depressive states over time.
In a prospective cohort study, we collected smartphone behavioral data for up to 1 year. The study consists of observations from 164 participants, including healthy controls (n=31) and patients diagnosed with various depressive disorders: major depressive disorder (MDD; n=85), MDD with comorbid borderline personality disorder (n=27), and major depressive episodes with bipolar disorder (n=21). Data were labeled based on depression severity using 9-item Patient Health Questionnaire (PHQ-9) scores. We performed statistical analysis and used supervised machine learning on the data to classify the severity of depression and observe changes in the depression state over time.
Our correlation analysis revealed 32 behavioral markers associated with the changes in depressive state. Our analysis classified patients who are depressed with an accuracy of 82% (95% CI 80%-84%) and change in the presence of depression with an accuracy of 75% (95% CI 72%-76%). Notably, the most important smartphone features for classifying depression states were screen-off events, battery charge levels, communication patterns, app usage, and location data. Similarly, for predicting changes in depression state, the most important features were related to location, battery level, screen, and accelerometer data patterns.
The use of smartphone digital behavioral markers to supplement clinical evaluations may aid in detecting the presence and changes in severity of symptoms of depression, particularly if combined with intermittent use of self-report of symptoms.
抑郁症患者的临床诊断评估和结果监测主要依赖专业人员的访谈以及自我报告问卷的使用。智能手机和其他个人消费设备的普及促使人们研究通过这些设备收集的数据作为数字行为标志物来指示抑郁症的存在和监测其结果的潜力。
本文探讨使用智能手机收集的行为数据来检测和监测抑郁症患者抑郁症状的潜力。具体而言，研究这些数据能否准确分类抑郁症的存在情况，以及监测抑郁状态随时间的变化。
在一项前瞻性队列研究中，我们收集了长达1年的智能手机行为数据。该研究包括164名参与者的观察数据，其中有健康对照者（n = 31）以及被诊断患有各种抑郁症的患者：重度抑郁症（MDD；n = 85）、合并边缘性人格障碍的MDD（n = 27）和双相情感障碍的重度抑郁发作（n = 21）。数据根据使用9项患者健康问卷（PHQ - 9）评分得出的抑郁严重程度进行标记。我们对数据进行了统计分析，并使用监督式机器学习来分类抑郁严重程度并观察抑郁状态随时间的变化。
我们的相关性分析揭示了32个与抑郁状态变化相关的行为标志物。我们的分析对抑郁症患者的分类准确率为82%（95%置信区间80% - 84%），对抑郁状态变化的分类准确率为75%（95%置信区间72% - 76%）。值得注意的是，用于分类抑郁状态的最重要的智能手机功能是屏幕关闭事件、电池电量水平、通信模式、应用使用情况和位置数据。同样，对于预测抑郁状态的变化，最重要的特征与位置、电池电量、屏幕和加速度计数据模式有关。
使用智能手机数字行为标志物来补充临床评估可能有助于检测抑郁症症状的存在及其严重程度变化，特别是如果与间歇性使用症状自我报告相结合。
