Thompson Wesley K, Gershon Anda, O'Hara Ruth, Bernert Rebecca A, Depp Colin A
Stein Center for Research on Aging, University of California, San Diego School of Medicine, La Jolla, CA, USA; Department of Psychiatry, University of California, San Diego, La Jolla, CA, USA.
Bipolar Disord. 2014 Nov;16(7):669-77. doi: 10.1111/bdi.12218. Epub 2014 Jun 5.
Bipolar disorder is associated with idiosyncratic precursors of clinically important states such as suicidal ideation. Ecological momentary assessment (EMA) - high frequency data collection in a subject's usual environment - provides the potential for development of temporal, individualized prediction of risk states. The present study tested the ability of EMA data to predict individual symptom change in clinician-rated suicidal ideation.
Thirty-five adults diagnosed with inter-episode bipolar disorder completed daily measures of affect in their home environments using diaries administered over an eight-week assessment timeline. Suicidal ideation was assessed monthly at in-person visits using the Inventory of Depressive Symptomatology-Clinician Rated. We used a novel application of functional linear models (FLMs) to generate prospective predictions of suicidal ideation at in-person clinician assessments based on intensively sampled trajectories of daily affect.
Eight instances of suicidal ideation scores > 0 were recorded during the study period on six participants. Utilizing trajectories of negative and positive affect, cross-validated predictions attained 88% sensitivity with 95% specificity for elevated suicidal ideation one week prior to in-person clinician assessment. This model strongly outperformed prediction models using cross-sectional data obtained at study visits alone.
Utilizing EMA data with FLM prediction models substantially increases the accuracy of prediction of study-emergent suicidal ideation. Prediction algorithms employing intensively sampled longitudinal EMA data could sensitively detect the warning signs of suicidal ideation to facilitate improved suicide risk assessment and the timely delivery of preventative interventions.
双相情感障碍与自杀观念等具有临床重要意义状态的特异前驱因素相关。生态瞬时评估（EMA）——在个体日常环境中进行高频数据收集——为发展对风险状态的个体化、基于时间的预测提供了可能。本研究测试了EMA数据预测临床医生评定的自杀观念中个体症状变化的能力。
35名被诊断为发作间期双相情感障碍的成年人，在八周的评估时间内，通过日记在家中完成每日情感测量。使用临床医生评定的抑郁症状量表每月进行一次面对面访视，评估自杀观念。我们使用功能线性模型（FLM）的一种新应用，基于密集采样的每日情感轨迹，对临床医生面对面评估时的自杀观念进行前瞻性预测。
在研究期间，6名参与者记录到8次自杀观念评分>0的情况。利用消极和积极情感轨迹，交叉验证预测在临床医生面对面评估前一周对自杀观念升高的预测敏感性达到88%，特异性为95%。该模型明显优于仅使用研究访视时获得的横断面数据的预测模型。
将EMA数据与FLM预测模型相结合可大幅提高对研究中出现的自杀观念的预测准确性。采用密集采样的纵向EMA数据的预测算法能够灵敏地检测出自杀观念的预警信号，以促进改善自杀风险评估并及时提供预防性干预措施。
