Patterns of activity correlate with symptom severity in major depressive disorder patients

Objective measures, such as activity monitoring via actigraphy, offer a potential complement to clinical assessments in psychiatry. Alterations in rest-activity patterns are frequently observed in MDD patients, characterized by lower overall activity, shorter diurnal activity periods, shorter activity bouts, and flattened circadian activity fluctuations. Prior research has shown correlations between symptom severity and physical activity levels, with structured exercise demonstrating antidepressant effects. Circadian activity patterns are regulated by the interplay of environmental cues (light, social interactions, meals, activity) and the internal biological clock. Studies have linked genetic alterations within the molecular clock mechanism and weaker coupling between the central and peripheral clocks to depression. However, direct correlations between activity patterns and symptom severity have been sparsely investigated. This study aimed to determine if actigraphy-derived features correlate with MDD symptom severity (assessed by MADRS) in untreated adult patients experiencing a major depressive episode. The researchers employed a range of non-parametric and non-linear approaches for feature extraction and trained/validated linear models to predict symptom severity.

Existing literature indicates a strong association between disrupted circadian rhythmicity and mood disorders, including depression. Studies using actigraphy have revealed characteristic alterations in activity patterns in MDD patients compared to healthy controls and other psychiatric disorders. These alterations include globally lower activity levels, shorter diurnal activity periods and bouts, and flattened circadian fluctuations. Furthermore, studies have linked symptom severity to the amount of moderate-intensity physical activity and sedentary bouts. Conversely, increasing activity through structured exercise has proven to be an effective antidepressant intervention. At the molecular level, alterations in circadian clock gene expression and weakened coupling between the central and peripheral clocks have been associated with depression. However, the correlation between specific activity patterns and symptom severity has received limited attention.

Actigraphy data from two independent clinical trials were used: one involving patients receiving cognitive behavioral therapy (CBT) and another examining the effects of ketamine on treatment-resistant depression. The first study used GENEActiv Original actigraphs, while the second employed Actiwatch 2 devices. Data preprocessing involved smoothing, high-pass filtering, and normalization across devices. A total of 12 recordings from the CBT study and 23 recordings from the ketamine study were included after quality control, which eliminated recordings with significant missing data, artifacts, or shift work. The recordings were cropped to include only the period before any intervention to isolate the effects of depression. Several features were extracted from the data, including: circadian period (Lomb-Scargle algorithm), scaling exponent (detrended fluctuation analysis), intradaily variability, interdaily stability, and relative amplitude of circadian rhythms. Multiple regression (MR) models were trained to predict MADRS scores from these features. A brute-force approach generated all possible models, while a stepwise machine learning (ML) algorithm trained models of increasing complexity. Model selection involved filtering based on R-squared, RMSE, and VIF. Internal validation was performed on the training dataset, followed by external validation on the independent test dataset. The performance of validated models was compared against a dummy model and a random prediction model to evaluate significance.

Three features showed significant correlations with MADRS scores: scaling exponent (alpha full, negative correlation), and intradaily variability at 5 and 30-minute intervals (IV5, IV30, positive correlations). The brute-force approach yielded 3837 models surviving internal validation (average RMSE = 1.84, R-squared = 0.67). After external validation, 192 models remained (average RMSE = 2.70, R-squared = 0.59). Stepwise ML generated 14 models that passed internal validation, with 5 surviving external validation. The models showed good consistency across predictors. Across both approaches, features such as scaling exponent (alpha full), interdaily stability (IS5, IS30), and relative amplitude (RA) were frequently included in the best-performing models. These results indicate that depression severity correlates with activity pattern features independently of activity levels. The best models accurately predicted MADRS scores within 2 units in approximately 54% of cases and within 3 units in about 75% of cases.

The findings demonstrate that depression symptom severity correlates with features of activity patterns, independent of overall activity levels. The study provides proof-of-concept evidence that actigraphy can predict symptom severity. The identified features—scaling exponent, interdaily stability, and relative amplitude—reflect the complexity of activity patterns, the strength of coupling between activity and circadian entrainers, and the robustness of circadian rhythms. The results suggest that more severe depression is associated with less complex activity patterns, stronger coupling to circadian entrainers, and less robust circadian rhythms. This aligns with previous findings of less complex activity patterns in depressed individuals. The methodology considered the 24-hour cycle as a continuum, avoiding discrete classifications of activity levels, offering a novel perspective compared to previous studies focusing on overall activity levels. This approach provides insights into the underlying mechanisms of depression and could facilitate connections with molecular mechanisms. The findings support the use of actigraphy as a minimally invasive objective measure for evaluating depression.

This study demonstrates a significant correlation between specific features of activity patterns and depression symptom severity. The findings support the potential of actigraphy as a valuable tool for objective assessment of MDD, providing a novel perspective beyond simple activity level measurements. Future research should focus on larger datasets, exploring diverse populations and considering the influence of treatments. Furthermore, investigating the dynamic changes in activity patterns in response to treatment could enhance the clinical utility of this approach.

The study population consisted solely of patients with unipolar MDD, limiting the generalizability of the findings to other mood disorders. The model was trained on data collected before interventions, hindering inferences about dynamic changes during treatment. The relatively small sample sizes and differences in inclusion criteria between datasets pose limitations. The use of different actigraphy devices in the two datasets could also influence the results, although the researchers addressed this issue by focusing on device-independent features.