運用GPS移動性和即時情緒數據的多元分析揭示情緒模式

Abstract

背景： 重度憂鬱症與雙極性情感疾患患者的病程往往呈現反覆發作的特性。傳統的情緒監測方法多仰賴主觀評估，不僅頻率與便利性有限，亦難以進行長期、連續性的偵測。為因應此限制，近年來逐漸發展出以數位表型進行被動資料蒐集的技術，特別是透過全球定位系統所獲得的移動模式，結合生態瞬時評估所提供的即時性主動自我回報機制，以更即時且高頻率的方式監測情緒波動。儘管已有相當多針對情緒障礙相關數位表型的研究，但情緒狀態與移動性之間的雙向關聯性與時間延遲效應，仍屬相對較少探討的領域。此外，雖然傅立葉轉換等頻域分析方法已廣泛應用於身體活動的研究並展現諸多優勢，其在反映參與者移動行為的週期性與穩健性方面，仍存在一定的應用限制。研究目標為:（1）比較工作日與週末之移動特徵，評估其是否具有區分情緒狀態與診斷憂鬱症與雙極性疾患的潛力 （2）檢驗移動特徵與情緒之間的雙向時間關係，評估其時間延遲效應的影響 （3）探索由傅立葉轉換萃取之移動頻域特徵，是否可有效區分情緒狀態與診斷類別 （4）分析每日四個時段所計算的位置變異，並評估其與生態瞬時評估情緒狀態及臨床診斷之間的關聯性。 方法： 本研究為前瞻式世代追蹤研究，招募了15位健康受試者、20位雙極性情感障礙患者及27位重度憂鬱症患者，進行為期六個月的追蹤調查。我們透過楊氏躁症量表（YMRS）、漢密爾頓憂鬱量表（HAMD）、病人健康問卷-4（PHQ-4）、貝克焦慮量表（BAI）與貝克憂鬱量表（BDI）評估受試者的症狀嚴重程度。此外，本研究亦透過智慧型手機應用程式的生態瞬時評估問卷，評估受試者的疲憊、憂鬱、躁動及易怒情緒的主觀感受。另一方面，使用手機全球定位系統功能持續記錄受試者於追蹤期間每日的移動模式。 結果： 移動行為特徵可有效反映情緒狀態：平日的位置變異與移動時間與憂鬱顯著相關，週末則以移動熵為主；疲勞與移動時間、平均速度、在家時間呈正相關；躁狂與位置變異/在家時間有正相關；易怒則與移動時間/在家時間有正相關；此外，就診斷而言，重度憂鬱患者在週末呈現較低的平均速度、群集數目、在家時間、移動熵。 在患有情緒障礙的參與者中，我們觀察到了雙向時間滯後關聯。第一天到第二天位置變異和移動熵的變化與第二天的憂鬱情緒呈現正相關，而第一天到第二天在家時間的變化與第二天的憂鬱情緒呈現正相關，並觀察到雙向影響的關聯性。 傅立葉轉換分析顯示，在一個月的觀察期間內，移動熵的頻譜強度在區分雙極性情感障礙患者與重度憂鬱症患者的診斷上尤為有效；雙極性情感障礙患者在移動行為的頻譜強度與穩健度上均高於重度憂鬱症患者。此外，位置變異的頻譜強度也提供了情緒狀態的輔助資訊。相較於週期性波動，位置變異與移動熵的頻率振幅更能反映移動強度，進一步揭示了更細緻的行為模式。 即使憂鬱症狀程度相近，雙極性情感疾患患者在早晨仍展現出較高的位置變異。晚間位置變異與瞬時評估之憂鬱情緒呈現最顯著的相關性。雙極性情感疾患患者在跨時段的移動變化從睡眠時間到早晨、下午到晚上比重鬱症患者更為劇烈，顯示其日內行為模式的變化性高於重鬱症患者。 結論： 整體而言，將以全球定位系統移動模式為基礎的數位表型技術與生態瞬時評估方法整合，對於即時情緒波動的監測以及提升雙極性情感疾患與重度憂鬱症之診斷流程，展現出顯著的應用潛力。日常使用智慧型手機作為監測工具，提供一種便捷且低干擾的方式，有助於持續追蹤病患的病程變化；未來介入措施亦應優先考量促進身體移動的可能性。 本研究的重要貢獻在於應用了創新的方法──傅立葉轉換的頻域分析技術，深入解析情緒障礙患者在移動行為上的週期性與強度變化。此方法不僅補足傳統時間序列分析的不足，也為未來在數位表型與生態瞬時評估領域的研究提供了可參考的分析架構與實證基礎。

Background: Patients with major depressive disorder (MDD) and bipolar disorder (BP) often experience recurrent episodes throughout the course of their illness. Traditional methods for monitoring emotional states primarily rely on subjective self-report measures, which are limited in both frequency and convenience and are unsuitable for long-term, continuous tracking. To address these limitations, digital phenotyping—particularly passive data collection through Global Positioning System (GPS)–derived mobility patterns—has emerged as a promising approach. When combined with ecological momentary assessment (EMA), which captures real-time active self-reports, this integration allows for more timely and high-frequency monitoring of emotional fluctuations. Although a growing body of research has explored digital phenotyping in mood disorders, the bidirectional temporal relationship and time-lagged effects between EMA-reported emotional states and GPS-based mobility patterns remain underexplored. Furthermore, while frequency-domain approaches such as Fourier transform have shown utility in physical activity research, their application to understanding the periodicity and stability of mobility behaviors in mood disorders remains limited. The study aims included: (1) Comparing GPS-derived mobility features between weekdays and weekends to differentiate emotional states and diagnosis. (2) Examining the bidirectional temporal associations between GPS mobility features and EMA-reported mood, focusing on time-lagged effects. (3) Exploring the utility of frequency-domain features extracted via the Fourier transform in distinguishing EMA mood states and diagnosis. (4) Analyzing location variance across four daily time intervals to investigate its associations with EMA mood and clinical diagnoses. Methods: This prospective cohort study recruited 15 healthy controls, 20 individuals with BP, and 27 individuals with MDD for a six-month longitudinal follow-up. Symptom severity was assessed using the Young Mania Rating Scale (YMRS), Hamilton Depression Rating Scale (HAMD), Patient Health Questionnaire-4 (PHQ-4), Beck Anxiety Inventory (BAI), and Beck Depression Inventory (BDI). In addition, EMA surveys via a smartphone application are used to evaluate subjective mood of fatigue, depression, mania, and irritability. On the other hand, the smartphone's GPS function was used to continuously record participants’ daily mobility patterns throughout the follow-up period. Results: GPS-derived mobility features effectively reflect mood states: location variance and transition time were linked to depressive symptoms on weekdays, while entropy was more relevant on weekends. Fatigue correlated positively with transition time, speed mean, and home stay. Manic and irritable moods were associated with higher location variance/home stay and transition time/home stay, respectively. MDD patients exhibited lower speed mean, number of clusters, home stay, and entropy on weekends. Among participants with mood disorders, we observed significant bidirectional time-lagged associations: changes in location variance and entropy from day 1 to day 2 were positively associated with depressive mood on day 2. Similarly, changes in homestay duration between day 1 and day 2 also positively correlated with next-day depressive symptoms, with reciprocal relationships observed as well. Fourier transform analysis revealed that entropy frequency features over a one-month period were particularly effective in distinguishing BP from MDD. BP patients exhibited higher power spectrum intensity and robustness in their mobility patterns than MDD patients. Additionally, the spectral characteristics of location variance offered supplementary insights into affective states. Compared to cyclical waveforms, the amplitude power spectrum of location variance and entropy frequencies more accurately captured the intensity of movement, allowing for a more nuanced understanding of behavioral patterns. BP patients exhibited greater morning location variance, even when depressive symptoms were comparable to MDD. Evening location variance showed the strongest association with EMA-reported depressed mood. Cross-period mobility shifts were more pronounced in BP than in MDD, especially from sleep to morning and afternoon to evening. Conclusion: Integrating GPS-based digital phenotyping with EMA presents considerable potential for real-time monitoring of emotional fluctuations and for improving diagnostic processes for BP and MDD. The use of smartphones as passive monitoring tools offers a convenient and non-invasive method to continuously track symptom trajectories. Future interventions should prioritize strategies that promote physical mobility. A key contribution of this study lies in its innovative use of frequency-domain analysis via Fourier transform to investigate the periodicity and intensity of mobility behaviors in mood disorders. This approach not only complements traditional time-series methods but also provides a foundational framework for future research in digital phenotyping and EMA.