Estimating the prognostic value of cross-sectional network connectivity for treatment response in depression.

BACKGROUND: Tightly connected symptom networks have previously been linked to treatment resistance, but most findings come from small-sample studies comparing single responder v. non-responder networks. We aimed to estimate the association between baseline network connectivity and treatment response in a large sample and benchmark its prognostic value against baseline symptom severity and variance. METHODS: N = 40 518 patients receiving treatment for depression in routine care in England from 2015-2020 were analysed. Cross-sectional networks were constructed using the Patient Health Questionnaire-9 (PHQ-9) for responders and non-responders (N = 20 259 each). To conduct parametric tests investigating the contribution of PHQ-9 sum score mean and variance to connectivity differences, networks were constructed for 160 independent subsamples of responders and non-responders (80 each, n = 250 per sample). RESULTS: The baseline non-responder network was more connected than responders (3.15 v. 2.70, S = 0.44, p < 0.001), but effects were small, requiring n = 750 per group to have 85% power. Parametric analyses revealed baseline network connectivity, PHQ-9 sum score mean, and PHQ-9 sum score variance were correlated (r = 0.20-0.58, all p < 0.001). Both PHQ-9 sum score mean (ß = -1.79, s.e. = 0.07, p < 0.001), and PHQ-9 sum score variance (ß = -1.67, s.e. = 0.09, p < 0.001) had larger effect sizes for predicting response than connectivity (ß = -1.35, s.e. = 0.12, p < 0.001). The association between connectivity and response disappeared when PHQ-9 sum score variance was accounted for (ß = -0.28, s.e. = 0.19, p = 0.14). We replicated these results in patients completing longer treatment (8-12 weeks, N = 22 952) and using anxiety symptom networks (N = 70 620). CONCLUSIONS: The association between baseline network connectivity and treatment response may be largely due to differences in baseline score variance.

Are central eating disorder network symptoms sensitive to item selection and sample? Implications for conceptualization of eating disorder psychopathology from a network perspective.

Item selection is a critical decision in modeling psychological networks. The current preregistered two-study research used random selections of 1,000 symptom networks to examine which eating disorder (ED) and co-occurring symptoms are most central in longitudinal networks among individuals with EDs (N = 71, total observations = 6,060) and tested whether centrality changed based on which items were included in the network. Participants completed 2 weeks of ecological momentary assessment (five surveys/day). In Study 1, we obtained initial strength centrality values by estimating an a priori network using eight items with the highest means. We then estimated 1,000 networks and their centrality from a random selection of unique eight-item symptom combinations. We compared the strength centrality from the a priori network to the distribution of strength centrality estimates from the random-item networks. In Study 2, we repeated this procedure in an independent longitudinal dataset (N = 41, total observations = 4,575) to determine if our results generalized across samples. Shame, guilt, worry, and fear of losing control were consistently central across networks, regardless of items included in the network or sample. Results suggest that these symptoms may be important to the structure of ED psychopathology and have implications for how we understand the structure of ED psychopathology. Existing methods for item inclusion in psychological networks may distort the structure of ED symptom networks by either under- or overestimating strength centrality, or by omitting consistently central symptoms that are nontraditional ED symptoms. Future research should consider including these symptoms in models of ED psychopathology. (PsycInfo Database Record (c) 2023 APA, all rights reserved).

Elevated emotion network connectivity is associated with fluctuations in depression.

Elevated emotion network connectivity is thought to leave people vulnerable to become and stay depressed. The mechanism through which this arises is however unclear. Here, we test the idea that the connectivity of emotion networks is associated with more extreme fluctuations in depression over time, rather than necessarily more severe depression. We gathered data from two independent samples of N = 155 paid students and N = 194 citizen scientists who rated their positive and negative emotions on a smartphone app twice a day and completed a weekly depression questionnaire for 8 wk. We constructed thousands of personalized emotion networks for each participant and tested whether connectivity was associated with severity of depression or its variance over 8 wk. Network connectivity was positively associated with baseline depression severity in citizen scientists, but not paid students. In contrast, 8-wk variance of depression was correlated with network connectivity in both samples. When controlling for depression variance, the association between connectivity and baseline depression severity in citizen scientists was no longer significant. We replicated these findings in an independent community sample (N = 519). We conclude that elevated network connectivity is associated with greater variability in depression symptoms. This variability only translates into increased severity in samples where depression is on average low and positively skewed, causing mean and variance to be more strongly correlated. These findings, although correlational, suggest that while emotional network connectivity could predispose individuals to severe depression, it could also be leveraged to bring about therapeutic improvements.

Machine learning of language use on Twitter reveals weak and non-specific predictions.

Depressed individuals use language differently than healthy controls and it has been proposed that social media posts can be used to identify depression. Much of the evidence behind this claim relies on indirect measures of mental health and few studies have tested if these language features are specific to depression versus other aspects of mental health. We analysed the Tweets of 1006 participants who completed questionnaires assessing symptoms of depression and 8 other mental health conditions. Daily Tweets were subjected to textual analysis and the resulting linguistic features were used to train an Elastic Net model on depression severity, using nested cross-validation. We then tested performance in a held-out test set (30%), comparing predictions of depression versus 8 other aspects of mental health. The depression trained model had modest out-of-sample predictive performance, explaining 2.5% of variance in depression symptoms (R2 = 0.025, r = 0.16). The performance of this model was as-good or superior when used to identify other aspects of mental health: schizotypy, social anxiety, eating disorders, generalised anxiety, above chance for obsessive-compulsive disorder, apathy, but not significant for alcohol abuse or impulsivity. Machine learning analysis of social media data, when trained on well-validated clinical instruments, could not make meaningful individualised predictions regarding users' mental health. Furthermore, language use associated with depression was non-specific, having similar performance in predicting other mental health problems.

Using language in social media posts to study the network dynamics of depression longitudinally.

Network theory of mental illness posits that causal interactions between symptoms give rise to mental health disorders. Increasing evidence suggests that depression network connectivity may be a risk factor for transitioning and sustaining a depressive state. Here we analysed social media (Twitter) data from 946 participants who retrospectively self-reported the dates of any depressive episodes in the past 12 months and current depressive symptom severity. We construct personalised, within-subject, networks based on depression-related linguistic features. We show an association existed between current depression severity and 8 out of 9 text features examined. Individuals with greater depression severity had higher overall network connectivity between depression-relevant linguistic features than those with lesser severity. We observed within-subject changes in overall network connectivity associated with the dates of a self-reported depressive episode. The connectivity within personalized networks of depression-associated linguistic features may change dynamically with changes in current depression symptoms.