Identification of subtypes of depression using remote measurement technologies

尽管它的异质性，它已被证明是难以确定的亚型重性抑郁症（MDD）。这限制了新疗法的开发，新疗法需要更精确地将治疗与特定患者表现相匹配，同时也阻碍了潜在生物标志物的鉴定（Baumeister &帕克，2012）。在文献中，区分非典型抑郁症和典型抑郁症是常见的，以将患者特征分层，然而，探索两组疾病轨迹的研究发现了相互矛盾的证据（Lamers等人，2016）。研究人员认为，大规模前瞻性研究的数据挖掘方法可能有助于更好地了解这些亚型（Chekroud et al. 2016和货车Loo et al. 2014）。然而，这需要具有高时间分辨率的长期症状跟踪方法，以便在各种变量中获得更明确的抑郁症表型。远程测量技术（RMT）从智能手机和可穿戴设备收集数据，可以提供个人日常生活的360度全景，包括睡眠、活动、心率、位置、认知、言语和情绪/压力源。疾病和复发的远程评估-重度抑郁症（RADAR-MDD）研究收集了三个临床研究中心（英国，西班牙，荷兰）623名患者的此类数据。该研究利用智能手机应用程序和可穿戴健身设备在2年内纵向跟踪MDD症状（更多细节参见Matcham等人，2019），为进一步研究抑郁症结果及其轨迹的表型聚类提供了肥沃的土壤。拟议的博士项目旨在填补文献中的这一空白，并通过利用整个RADAR-MDD研究中收集的数据设计一项独特的研究，以了解症状聚类。该项目的主管提供临床和流行病学专业知识（Hotopf）和AI/计算机科学（康明斯）的合作伙伴关系。Hotopf是RADAR-MDD项目的PI，因此对数据质量有全面的了解，可以确保数据访问。主要的研究问题是确定与抑郁症不同轨迹相关的行为/生理亚型。该研究将结合联合收割机的“自上而下”的数据驱动的方法，利用机器学习工具来聚类症状概况，以及假设驱动的方法来确定预定组之间的表型差异：确定相同参与者中的抑郁发作是否在表型上相似。确定是否可以使用RMT数据区分预定义组（轻度、中度和重度发作;典型模式与非典型模式）。

Despite its heterogenous nature, it has proved difficult to identify sub-types of major depressive disorder (MDD). This has limited the development of novel therapeutics, that require more precise matching of treatments to specific patient presentations, while also hindering the identification of potential biomarkers (Baumeister & Parker, 2012). Distinguishing between atypical and typical depression is frequent in the literature to stratify patient profiles, however, research exploring disease trajectories across the two groups have found contradicting evidence (Lamers et al., 2016). Researchers have suggested that data mining methods on large scale prospective studies could be useful to better understand these subtypes (Chekroud et al. 2016 and van Loo et al. 2014). Yet, this requires long-term symptom tracking methodologies with high temporal resolution, to get a more defined picture of depression phenotypes across a wide variety of variables. Remote measurement technologies (RMT) harvest data from smartphones and wearable devices and can provide a 360-degree picture of an individual's day-to-day life, including sleep, activity, heart rate, location, cognition, speech and mood/stressors. The Remote Assessment of Disease and Relapse- Major Depressive Disorder (RADAR-MDD) study collected such data on 623 patients across three clinical sites (UK, Spain, Netherlands). The study utilized smartphone applications and wearable fitness devices to track MDD symptoms longitudinally over 2 years (for more detail see Matcham et al., 2019), providing fertile ground for further investigation into phenotypic clustering of depression outcomes and their trajectories. The proposed PhD project aims to fill this gap in the literature and design a unique study by utilizing the data collected throughout the RADAR-MDD study, to understand symptom clustering. The supervisors for this project provide a partnership of clinical and epidemiological expertise (Hotopf) and AI/computer science (Cummins). Hotopf is the PI for the RADAR-MDD project, and therefore has a comprehensive understanding of data quality and can ensure data access. The primary research question is to identify behavioural/physiological subtypes which are associated with different trajectories of depression. The research will combine a "top-down" data driven approach, utilizing machine learning tools to cluster symptom profiles, as well as hypothesis driven approaches to identify differences in phenotype across predetermined groups: Identify whether depressive episodes within the same participants are phenotypically similar. Determine whether pre-defined groups (mild, moderate and severe episodes; typical versus atypical patterns) can be distinguished using the RMT data.