Skip to main content
NCBI home page
Nature Portfolio logoLink to Nature Portfolio
. 2025 Oct 30;3(11):1417–1424. doi: 10.1038/s44220-025-00528-x

The effect of sertraline on networks of mood and anxiety symptoms: secondary analysis of the PANDA randomized controlled trial

Giulia G Piazza 1,✉, Andrea G Allegrini 1,2, Larisa Duffy 3, Gemma Lewis 3, Glyn Lewis 3, Jonathan P Roiser 4, Jean-Baptiste Pingault 1,2
PMCID: PMC12589115  PMID: 41211530

Abstract

Depression consists of heterogeneous symptoms that can occur in hundreds of possible combinations. However, intervention studies commonly operationalize depression as a homogeneous condition. Here we adopt a symptom-level approach to test the effects of the selective serotonin reuptake inhibitor sertraline on depression and anxiety symptoms and to test their associations. Using data from the PANDA randomized controlled trial, we use network models to estimate the effects of sertraline at different time points (contemporaneous networks at 2, 6 and 12 weeks) and across time (temporally lagged networks). Results show that sertraline has beneficial effects on core depression and anxiety symptoms as early as after 2 weeks of treatment, counteracted by detrimental effects on somatic symptoms of depression. This intricate pattern of treatment effects is typically masked when measuring depression on a single dimension. Focusing on individual symptoms of depression and anxiety may shed light on the nature, effectiveness and timing of antidepressant action.

Subject terms: Depression, Psychology

This research adopts a symptom-level approach to evaluate the effects of sertraline on depression and anxiety using network models from the PANDA trial, revealing early benefits for core symptoms while highlighting adverse impacts on somatic manifestations over time.

Main

Selective serotonin reuptake inhibitors (SSRIs) are a first-line treatment for depression and anxiety. Although meta-analytic evidence suggests that they have modest effect sizes compared with placebo1,2, SSRIs have been increasingly prescribed in recent years3. The response to antidepressants can take weeks to develop, and relatively little is known about the precise mechanism of action behind it4,5.

Multiple lines of evidence indicate considerable heterogeneity in symptoms of depression and anxiety. For example, some symptoms of depression, such as sad mood and concentration problems, show larger associations with functional impairment compared with other symptoms, such as weight and appetite problems6. Symptoms of depression are also differentially associated with environmental and genetic risk factors7,8; for instance, appetite changes and fatigue appear to have higher heritability estimates7. Isolation and grief have been associated with crying and sadness, while chronic stress is associated with fatigue and hypersomnia9.

Similarly, studies focusing on individual symptoms have reported differential treatment responses to SSRIs across symptom subgroups10–14. Commonly used SSRIs were found to be more effective at treating core emotional symptoms than somatic symptoms15, suggesting that they may simultaneously be effective in alleviating a subset of symptoms while failing to treat or even exacerbating others.

In addition, reciprocal causal associations between symptoms may lead to maladaptive cycles16. For example, insomnia might cause concentration problems, which could, in turn, reduce self-esteem. Separating the direct and indirect effects of SSRIs on individual symptoms has potentially important implications for understanding the mechanisms underlying interventions17.

Network analysis is a useful framework that allows for the statistical modeling and visualization of symptoms and their associations18. In networks, symptoms are represented by nodes, while their associations are represented as edges between nodes19. In this framework, SSRIs could exert direct effects on individual symptoms, for example, by directly improving mood. In addition, network analysis can examine network structures, that is, the presence or absence and magnitude of associations between symptoms. SSRIs could alter network structures20, for example, by reducing the strength of the association between feelings of sadness and feelings of guilt.

Network studies have suggested that antidepressant treatment is associated with improvements in individual symptoms of depression and anxiety, such as feelings of guilt21, anxiety and avoidance22,23, depressed mood24 and worry25. However, few such studies have included a placebo group22,24, which precludes drawing strong conclusions, and most have only compared pre- and posttreatment networks cross-sectionally17,23,25–28, neglecting potentially important temporal associations between symptoms. New insights into the effects of sertraline can emerge from modeling temporal associations between symptoms in both treatment and placebo groups.

Therefore, this study tests the direct effects of SSRI treatment on symptoms of depression and anxiety, relative to placebo, both at a single time point and across time and examines associations between these symptoms. Combining analytical approaches, we conduct a secondary analysis of a large placebo-controlled randomized trial on the effectiveness of sertraline for the treatment of depression (the PANDA trial29). First, using a standard regression approach, we investigate the effects of sertraline on individual depression and anxiety symptoms, compared with placebo. Second, we investigate these effects while accounting for associations between symptoms with network analyses, at each time point (contemporaneous networks) and across time (temporally lagged networks). Third, we compare the patterns of associations between symptoms (that is, network structures, both contemporaneously and across time) between sertraline and placebo groups. On the basis of the primary results of the PANDA trial (using sum-scores), we predicted a beneficial effect of sertraline on depression symptoms by 12 weeks of treatment, compared with placebo. At the symptom level, drawing on existing literature17,23,24,28,30–33, we anticipated direct beneficial effects of sertraline on depressed mood and worry, relative to placebo, when accounting for associations with all other symptoms. We expected these effects to be detectable both in contemporaneous and temporally lagged symptom networks, with changes emerging by 12 weeks of treatment.

Results

Effect of sertraline on individual symptoms

A maximum sample of 571 individuals with complete cases for each symptom was included in this analysis (Supplementary Table 3). Mixed models indicated significant main effects of sertraline on all symptoms (accounting for baseline score), with small effect sizes (η2 = 0.007–0.019) (Figs. 1 and 2), except for problems with appetite, crying, feelings of guilt, physical health, feelings of self-punishment, sleep and tiredness. The largest beneficial effects of sertraline were on feelings of self-loathing (false discovery rate (FDR) P value, PFDR < 0.001, η2 = 0.019, 95% confidence interval (CI) 0.006–0.038), feeling bad about oneself (PFDR < 0.001, η2 = 0.018, 95% CI 0.006–0.037) and anhedonia (PFDR < 0.001, η2 = 0.017, 95% CI 0.005–0.035). There were significant main effects of time on all symptoms except problems with libido, physical health, and suicidal thoughts (Supplementary Table 4). Following corrections for multiple comparisons, no treatment-by-time interactions achieved significance (Supplementary Table 4).

Fig. 1. Effects of treatment and time on mean symptoms.

Fig. 1

Open in a new tab

Means ( ± standard errors) of symptoms of depression and anxiety (derived in the node selection step) at baseline, 2 weeks, 6 weeks and 12 weeks (nmax = 571). AFR, feeling afraid; ANH, loss of interest and pleasure in everyday life; ANX, feeling nervous or anxious; APP, lack of appetite or eating too much; BAD, feeling bad about oneself; CON, concentration problems; CRY, crying; DIS, disliking oneself; FAI, past failure; GUI, guilt feelings; IMP, self-reported improvement; IND, indecisiveness; LIB, loss of interest in sex; PHY, general physical health; PUN, punishment feelings; RES, being restless or slow; SAD, feeling sad or depressed; SLE, sleep problems; SUI, suicidal thoughts; TIR, feeling tired; WOR, feeling worried.

Fig. 2. Effect sizes of the effect of sertraline on individual symptoms.

Fig. 2

Open in a new tab

a,b, Data are presented as effect sizes (η2) of time, treatment group and group by time interactions and associated confidence intervals (95% CI) in linear mixed models for each symptom. In these models, baseline scores for each symptom were included as fixed effects. Asterisks indicate a significant effect (based on FDR-corrected P values) (a). Data are presented as effect sizes (Cohen’s d) of sertraline on each symptom at 2, 6 and 12 weeks and associated confidence intervals (95% CI). Red lines indicate effect sizes derived from the main (sum-score) results of the PANDA trial (b).

Standardized differences in means between sertraline and placebo groups (Cohen’s d) indicated comparable effect sizes to the main results of the PANDA trial, with larger effect sizes on somatic symptoms (for example, libido and sleep) at week 6 (Fig. 2b).

Contemporaneous networks

We found beneficial effects of sertraline on symptoms across all assessments (n2weeks = 550, n6weeks = 523, n12weeks = 512) in contemporaneous networks (Fig. 3 and Supplementary Tables 5–7). Sertraline treatment caused lower feelings of sadness (r2weeks = −0.092), restlessness (r2weeks = −0.053), self-loathing (r2weeks = −0.044), suicidal thoughts (r2weeks = −0.039) and physical health problems (r2weeks = −0.028) at the 2-week time point; lower levels of feeling bad about oneself (r6weeks = −0.087), sadness (r2weeks = −0.027), feeling afraid (r6weeks = −0.041), restlessness (r6weeks = −0.098) and concentration problems (r6weeks = −0.0046) at the 6-week time point; and lower levels of anxiety (r12weeks = −0.057), physical health problems (r12weeks = −0.055), anhedonia (r12weeks = −0.103) and self-loathing (r12weeks = −0.061) at the 12-week time point. In addition, sertraline treatment caused higher self-reported improvement at 6 weeks (r6weeks = −0.036). However, sertraline also had detrimental effects at all time points, such as on problems with sleep (r6weeks = 0.219, r12weeks = 0.065), appetite (r2weeks = 0.089, r12weeks = 0.099), libido (r2weeks = 0.082, r6weeks = 0.235, r12weeks = 0.132), tiredness (r2weeks = 0.077), fatigue (r2weeks = 0.039) and indecisiveness (r2weeks = 0.065).

Fig. 3. Contemporaneous networks of symptoms of depression and anxiety.

Fig. 3

Open in a new tab

In all networks, thicker edges indicate stronger associations. To highlight connections to the treatment node, positive associations (detrimental effects) with the treatment node (center) are in dark blue and negative associations (beneficial effects) in red. Positive associations between symptoms are in light blue and negative associations in yellow. Networks were plotted with an identical layout to better compare results. TREAT, treatment allocation.

Network structure comparison in contemporaneous networks

The network comparison test revealed no significant differences in network structure between placebo and sertraline networks (all P > 0.05).

Temporally lagged networks

The estimated cross-lagged panel model had adequate fit according to standard fit indices (comparative fit index (CFI) 0.965, root mean square error of approximation (RMSEA) 0.043). Sertraline caused lower symptoms of depression compared with placebo at all time points (n = 550) when controlling for temporal associations at previous time points (Fig. 4 and Supplementary Tables 8 and 9). For example, when accounting for symptoms at 2 weeks, sertraline caused, at 6 weeks, a reduction in feeling sad (β6weeks = −0.096), bad about oneself (β6weeks = −0.090), afraid (β6weeks = −0.114), restlessness (β6weeks = −0.091), anxiety (β6weeks = −0.110), worry (β6weeks = −0.083) and indecisiveness (β6weeks = −0.086). Moreover, even when accounting for symptoms at 6 weeks, sertraline still caused, at 12 weeks, a reduction in feeling sad (β12weeks = −0.106), anxiety (β12weeks = −0.092), anhedonia (β12weeks = −0.105), self-loathing (β12weeks = −0.084) and indecisiveness (β12weeks = −0.081). Notably, sertraline treatment consistently caused self-reported improvement over time (β6weeks = −0.121, β12weeks = −0.130) but also caused problems with libido (β6weeks = 0.116) and sleep (β6weeks = 0.113) during the middle of treatment.

Fig. 4. Temporally lagged networks of symptoms of depression and anxiety.

Fig. 4

Open in a new tab

Thicker edges indicate stronger associations. Directed arrows represent cross-lagged associations and looped arrows represent autoregressive associations. To highlight connections to the treatment node, positive associations (detrimental effects) with the treatment node (center) are in dark blue and negative associations (beneficial effects) in red. Positive associations between symptoms are in light blue and negative associations in yellow. Networks were plotted with an identical layout to better compare results.

Network structure comparison in temporally lagged networks

We found no significant structural network differences between sertraline and placebo groups. The cross-lagged model where edges were set to be equal across groups (model 1) had better support than the model where edges were free to vary across groups (model 2) (Bayesian information criterion (BIC)Model1 57,444, BICModel2 61,960, Akaike information criterion (AIC)Model1 49,298, AICModel2 50,013; Supplementary Table 10).

Discussion

We examined the effects of sertraline on symptoms of anxiety and depression and their associations in a sample drawn from a large placebo-controlled randomized trial. First, we found beneficial effects of sertraline on most symptoms of depression and anxiety when using typical analytical approaches (linear mixed models). Second, by accounting for associations between symptoms in network analyses, we found early effects on core emotional and volitional symptoms of depression and all symptoms of anxiety at ~2 weeks of treatment. These early beneficial effects may be masked when outcomes are measured using a single summary score by concurrent detrimental effects on somatic symptoms, which were also clear at 2 weeks. Finally, we found no evidence of differences in patterns of associations between symptoms (network structures), either at each time point or across time, between treatment groups.

Adjusting for associations between symptoms showed that antidepressants may act more rapidly on some symptoms of depression than has previously been suggested using a single summary score of symptoms5. When accounting for associations at each time point (contemporaneous networks), we found a rapid, albeit small, effect of sertraline on sad mood compared with placebo, appearing at ~2 weeks. In addition, analyses that account for temporal associations (temporally lagged networks) suggested that sertraline caused a reduction in all included anxiety symptoms, which is consistent with the (sum-score) results of the PANDA trial. However, importantly, using temporally lagged networks, we found an additional clear reduction in core symptoms of depression, such as feeling sad and bad about oneself, at as early as 6 weeks. While these network findings are partially mirrored by typical analyses that do not account for associations between symptoms, the network results suggest sertraline may have an early effect on core symptoms of depression (such as sadness).

Importantly, our findings point to a pattern of contrasting effects of sertraline, with both beneficial and detrimental effects compared with placebo. Although the main results of the PANDA trial indicated no differences in adverse effects between groups, somatic symptoms of depression included in our analyses are also frequently reported side effects of SSRIs5. While we did not observe detrimental effects on sleep, tiredness and appetite when only examining symptoms individually (in linear mixed models), taking into consideration associations between symptoms provided additional sensitivity, revealing some detrimental effects of sertraline on libido, tiredness and appetite at as early at 2 weeks, and on sleep and libido at 6 weeks. However, we did not detect additional effects at 12 weeks of treatment beyond those at 6 weeks. By contrast, we found a continued beneficial effect of sertraline on depression and anxiety symptoms beyond 6 weeks of treatment, independent of its prior effects, and consistent with our predictions, on the basis of sum-score results of the PANDA trial. Therefore, the adverse impact on somatic symptoms may peak and stabilize within 6 weeks of continued sertraline administration, and it may be counteracted by improvements in other symptoms.

This pattern of opposing effects on symptoms would be overlooked in analyses using summary scores on depression scales (for example, the Beck Depression Inventory (BDI-II) and Patient Health Questionnaire (PHQ-9)) as primary outcomes. On the contrary, summary scores on anxiety scales (for example, the Generalized Anxiety Disorder Assessment (GAD-7)) may be more sensitive to certain improvements, as they may not include somatic symptoms associated with medication. Therefore, it is possible that the effects on depression reported in the primary analyses of the PANDA trial were partially attenuated by the inclusion of physiological indicators in main outcome measures (for example, sleep and appetite items in the PHQ-9). Finally, although we found direct effects of active treatment, we do not find evidence of different patterns of associations between symptoms across treatment groups (that is, differences in network structures). This suggests that antidepressant treatment may not alter the associations between symptoms. In other words, although sertraline may cause an improvement in core symptoms, it may not change their reciprocal associations. For example, sertraline may, on average, improve sad mood and worry, but it may not alter the extent to which these two symptoms are associated with each other. Therefore, sertraline does not seem to operate via interrupting maladaptive reinforcement cycles between symptoms.

Our findings align with the results of previous studies examining the effects of SSRIs on individual symptoms, which suggest that the likelihood of detecting an effect of an SSRI is greater when using depressed mood as the sole outcome (as opposed to sum-scores)30. Our findings also add to evidence that beneficial effects can be detected early in treatment31, along with detrimental effects on somatic symptoms32. In addition, our results are consistent with previous cross-sectional network studies indicating that SSRIs have effects on both affective and somatic symptoms17,22,33. We present a longitudinal, placebo-controlled analysis that captures associations between symptoms in a heterogeneous sample not typically included in randomized controlled trials, which provides an important demonstration of the above findings, in a population typical of that presenting to primary care for depression treatment. Importantly, our findings provide evidence of symptom-specific effects that generalize across trials and patient characteristics, suggesting that these effects reflect core features of the antidepressant response.

The interpretation of these findings has some limitations. First, psychological networks are dependent on the choice of network nodes18,34,35. Therefore, our findings are conditional on the selection of symptoms from commonly used depression and anxiety scales. However, the PHQ-9, BDI-II and GAD-7 include all the common symptoms of both depression and anxiety. Second, our findings should be further confirmed and replicated in independent samples. Third, some symptoms of depression and anxiety may be measured more reliably than others and are therefore more likely to be detected in network edges.

In conclusion, we show that sertraline has direct effects on individual anxiety and depression symptoms, as early as ~2 weeks into treatment, although it does not change associations between symptoms. Although the PANDA study found no evidence for an effect on depression at 6 weeks after starting sertraline, we observed effects of sertraline on depression symptoms at as early as 2 weeks. These beneficial effects may have been masked by detrimental effects on somatic symptoms such as libido and sleep. Using a network approach can reveal insights into the effectiveness, timing, and direct pathways of antidepressant action by taking into consideration individual symptoms and their associations.

Methods

Sample and measures

The sample included patients from the PANDA trial29 (Supplementary Table 1). In this trial, 653 adult patients (384 female, mean age 39.7 ± 14.96 years) with depressive symptoms were recruited in a primary care setting (ISRCTN ref. no. ISRCTN84544741). Participants received either sertraline—50 mg, once daily for 1 week, then 100 mg daily for up to 11 weeks—(n = 324, 203 female, mean age 39.7 ± 14.6 years) or placebo (n = 329, 181 female, mean age 39.7 ± 15.4 years), in a double-blind, randomized design. Details on recruitment, treatment allocation and randomization are described in detail by Lewis et al.29 and Salaminios et al.36. Ethics approval was obtained from the National Research Ethics Service Committee, East of England—Cambridge South (ref. no. 13/EE/0418). All participants provided written informed consent.

In the current analysis, we used the PHQ-9 (ref. 37), BDI-II (ref. 38) and GAD-7 (ref. 39) as measures of anxiety and depression symptoms; the physical health component of the Short Form Health Survey40; and a single item reflecting subjective improvement (‘Compared to 2 weeks ago, how have your moods and feelings changed?’, rated 1 for ‘I feel a lot better’ to 5 for ‘I feel a lot worse’). Depression severity was assessed with total scores on the Clinical Interview Schedule—Revised41, divided into three categories (0–11, 12–19 and ≥20). Patients were assessed at baseline and followed up at 2 weeks, 6 weeks, and 12 weeks after baseline.

Statistical analysis

Analyses were carried out in R, version 4.2.0 (ref. 42) and are shown in Extended Data Fig. 1. Complete cases were used in each analysis step.

Extended Data Fig. 1. (a-g): Symptom-level analyses included in the study.

Extended Data Fig. 1

Open in a new tab

Further discussed in Methods.

Node selection

To reduce the number of network nodes, both for interpretability and to avoid collinearity issues, we examined items of the selected scales for content overlap, using a combination of data-driven analysis and conceptual inspection of item similarity. First, using the ‘goldbricker’ function in the R package networktools43 (version 1.5.0), we identified correlated pairs of items that also showed a low proportion of statistically different correlations with other nodes (that is, variable pairs with correlations r ≥ 0.5 and <40% of significantly different correlations at α = 5% were flagged, using the ‘threshold’ argument in the goldbricker function). The identified pairs were then inspected for content overlap and, when appropriate, combined by taking mean values (rounded to the next integer; Supplementary Table 2). The selection procedure resulted in 21 symptoms.

Change in symptoms over time

We used standard linear mixed regression models to analyze the effects of time, treatment and their interaction on the 21 symptoms derived by node selection, using the R package lmerTest (version 3.1.3), restricted maximum likelihood estimation and Satterthwaite’s method for approximating degrees of freedom44. These models included time (2, 6 and 12 weeks) and individuals as random effects, allowing for random slopes. Site, the corresponding baseline symptom score, depression duration and treatment allocation were included as fixed effects, with an interaction between treatment and time. Effect sizes (η2, that is, the amount of variation in each item explained by predictors) and associated 95% CIs were obtained using the R package effectsize (version 0.7.0)45. P values were adjusted for multiple comparisons (21 tests) with FDR using the Benjamini–Hochberg method (α = 5%) and the R package stats (version 4.2.0)42. In addition, we calculated Cohen’s d for all symptoms at each time point and compared our estimates to the main PANDA trial results.

Network analyses

To compare our analyses with prior studies, we separately modeled each time point at which symptoms were measured (‘Contemporaneous networks’) (Extended Data Fig. 1). We then included associations between symptoms across time (‘Temporally lagged networks’). Within both network types, we modeled treatment allocation as a network node to estimate the direct effect of sertraline on individual symptoms while accounting for all other associations in a network. For example, we estimated the association between the treatment node and feelings of sadness, while accounting for all associations between symptoms. We then focused on a comparison of network structures between sertraline and placebo groups (‘Network structure comparisons’) in both contemporaneous and temporally lagged networks. This allowed us to establish whether individuals in either group had a greater number of nonzero associations between symptoms or showed stronger associations between symptoms. For example, we estimated whether there was a weaker association between feelings of sadness and low self-esteem in the sertraline group, relative to the placebo group, at the 2-week time point.

All item-level data used in networks were adjusted for covariates and baseline variables associated with missingness (identified in the main PANDA trial results) using linear regression models. In these models, each item was predicted by sex, age, surgery site, baseline item values, depression severity (Clinical Interview Schedule—Revised) and duration, ethnicity (‘White’ or ‘Ethnic minority’), financial difficulty (‘Comfortably/Alright’, ‘Just about coping’ or ‘Finding it difficult’), previous antidepressant use (‘Yes’ or ‘No’), marital status (‘Married/Living as married’, ‘Single’ or ‘Separated, divorced or widowed’) and notable life events (number of life events in the past 6 months). Standardized residuals obtained from linear regressions were then used in network analyses.

Contemporaneous networks

We estimated one network per time point using the mgm R package (version 1.2.13)46, modeling the selected symptoms and a node indicating treatment allocation (0 = placebo, 1 = sertraline)47. The least absolute shrinkage and selection operator was used to minimize the number of spurious edges, and cross-validation was used to select the least absolute shrinkage and selection operator tuning parameter (Supplementary Methods). In the resulting networks, edges represent partial correlations (r), and nodes represent symptoms at each time point. For network structure comparison in contemporaneous networks, we tested the null hypothesis that network edges were equal across sertraline and placebo groups for each contemporaneous network with a resampling-based permutation test (network comparison test with 1,000 iterations48).

Temporally lagged networks

We estimated a cross-lagged panel model including all symptoms (as observed variables) with the R package lavaan (version 0.6.12)49 using full information maximum likelihood estimation, including treatment allocation as a predictor50 (Supplementary Fig. 1). In this model, each symptom at one time point was regressed on all symptoms at the previous time point, allowing us to model the association of one symptom with another later symptom (cross-lagged paths) and with itself over time (autoregressive paths), while controlling for the associations with all other symptoms at the previous time point51. For example, we modeled the effect of concentration problems at the 2-week time point on sleep problems at the 6-week time point, while controlling for associations with all other symptoms at the 2-week time point.

The resulting standardized regression coefficients (β) were visualized as a network of directed edges. We report the model fit of the cross-lagged panel model according to standard fit indices (CFI and RMSEA, with CFI ≥0.95 and RMSEA ≤0.05 considered adequate model fit52).

For network structure comparison in temporally lagged networks, we compared groups by testing whether all edges between network nodes had comparable weights in the sertraline and placebo groups. We constructed a cross-lagged panel model without including treatment allocation as a variable (Supplementary Fig. 2). We then compared a model where all regression coefficients were set to be equal between groups (model 1) to a model where all coefficients were allowed to freely vary between groups (model 2) using common fit indices (CFI, RMSEA, AIC and BIC).

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Supplementary information

Supplementary Information (1MB, pdf)

Supplementary Tables 1–10 and Figs. 1 and 2.

Reporting Summary (2.1MB, pdf)

Acknowledgements

We thank the co-applicants, participants, general practice surgeries, researchers, research nurses and all staff involved in this study and in the PANDA study. The PANDA trial was supported by the National Institute for Health Research and University College London Hospitals Biomedical Research Centre (NIHR Program Grants for Applied Research; grant no. RP-PG-0610-10048 awarded to Glyn Lewis). G.G.P. is the recipient of a Wellcome Trust PhD studentship (grant no. 224920/Z/22/Z). The funders of the study had no role in the study design, data collection, data analysis, interpretation of the data, writing of the report or decision to submit the paper for publication.

Extended data

Author contributions

G.G.P., A.G.A., J.P.R. and J.-B.P. contributed to the conceptualization and design of the study. G.G.P., A.G.A., J.P.R., Glyn Lewis, Gemma Lewis and J.-B.P. contributed to the statistical analyses. G.G.P. and L.D. provided technical, administrative or material support. All authors contributed to the acquisition, analysis or interpretation of the data. All authors contributed to the drafting of the manuscript or to critical evaluation of the manuscript for intellectual content. J.P.R. and J.-B.P. jointly supervised this work.

Peer review

Peer review information

Nature Mental Health thanks Taylor Braund and the other, anonymous, reviewer(s) for their contribution to the peer review of this work.

Data availability

All de-identified individual participant data collected in the PANDA trial and related documents (study protocol, analysis plan and code) are available, with no end, from the publications of the original trial paper. To gain access, researchers will need to enter a data access agreement with University College London (London, UK), providing a proposal for the use of data and a request for access (glyn.lewis@ucl.ac.uk).

Code availability

All code used for the analyses is available via GitHub at https://github.com/giuliapiazza18/PANDAnet-2/.

Competing interests

Glyn Lewis reports funding from the National Institute of Health Research (NIHR), the Wellcome Trust, Mental Health Research UK (MHRUK) and University College London Hospitals Biomedical Research Centre (UCLH BRC), as well as travel and subsistence to the European College of Neuropsychopharmacology (ECNP) 2023. The other authors report no competing interests.

Footnotes

Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

These authors jointly supervised this work: Jonathan P. Roiser, Jean-Baptiste Pingault.

Extended data

is available for this paper at 10.1038/s44220-025-00528-x.

Supplementary information

The online version contains supplementary material available at 10.1038/s44220-025-00528-x.

References

  • 1.Slee, A. et al. Pharmacological treatments for generalised anxiety disorder: a systematic review and network meta-analysis. Lancet393, 768–777 (2019). [DOI] [PubMed] [Google Scholar]
  • 2.Cipriani, A. et al. Comparative efficacy and acceptability of 21 antidepressant drugs for the acute treatment of adults with major depressive disorder: a systematic review and network meta-analysis. Lancet391, 1357–1366 (2018). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 3.Prescription cost analysis - England, 2018 [PAS]. NHS Digitalhttps://digital.nhs.uk/data-and-information/publications/statistical/prescription-cost-analysis/2018 (2019).
  • 4.Otte, C. et al. Major depressive disorder. Nat. Rev. Dis. Primers2, 1–20 (2016). [DOI] [PubMed] [Google Scholar]
  • 5.Rang, H. P., Dale, M. M., Ritter, J. M., Flower, R. J. & Henderson, G. Rang and Dale’s Pharmacology (Elsevier, 2012).
  • 6.Fried, E. I. & Nesse, R. M. The impact of individual depressive symptoms on impairment of psychosocial functioning. PLoS ONE9, e90311 (2014). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 7.Thorp, J. G. et al. Genetic heterogeneity in self-reported depressive symptoms identified through genetic analyses of the PHQ-9. Psychol. Med.50, 2385–2396 (2020). [DOI] [PubMed] [Google Scholar]
  • 8.Piazza, G. G. et al. Polygenic scores and networks of psychopathology symptoms. JAMA Psychiatry10.1001/jamapsychiatry.2024.1403 (2024). [DOI] [PMC free article] [PubMed]
  • 9.Keller, M. C., Neale, M. C. & Kendler, K. S. Association of different adverse life events with distinct patterns of depressive symptoms. Am. J. Psychiatry164, 1521–1529 (2007). [DOI] [PubMed] [Google Scholar]
  • 10.Bondar, J., Caye, A., Chekroud, A. M. & Kieling, C. Symptom clusters in adolescent depression and differential response to treatment: a secondary analysis of the Treatment for Adolescents with Depression Study randomised trial. Lancet Psychiatry7, 337–343 (2020). [DOI] [PubMed] [Google Scholar]
  • 11.Hieronymus, F., Emilsson, J. F., Nilsson, S. & Eriksson, E. Consistent superiority of selective serotonin reuptake inhibitors over placebo in reducing depressed mood in patients with major depression. Mol. Psychiatry21, 523–530 (2016). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 12.Hieronymus, F., Lisinski, A., Nilsson, S. & Eriksson, E. Influence of baseline severity on the effects of SSRIs in depression: an item-based, patient-level post-hoc analysis. Lancet Psychiatry6, 745–752 (2019). [DOI] [PubMed] [Google Scholar]
  • 13.Fournier, J. C. et al. Differential change in specific depressive symptoms during antidepressant medication or cognitive therapy. Behav. Res. Ther.51, 392–398 (2013). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 14.Uher, R. et al. Differential efficacy of escitalopram and nortriptyline on dimensional measures of depression. Br. J. Psychiatry194, 252–259 (2009). [DOI] [PubMed] [Google Scholar]
  • 15.Chekroud, A. M. et al. Reevaluating the efficacy and predictability of antidepressant treatments: a symptom clustering approach. JAMA Psychiatry74, 370–378 (2017). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 16.Ebrahimi, O. V. et al. Towards precision in the diagnostic profiling of patients: leveraging symptom dynamics as a clinical characterisation dimension in the assessment of major depressive disorder. Br. J. Psychiatry224, 157–163 (2024). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 17.Boschloo, L., Hieronymus, F., Lisinski, A., Cuijpers, P. & Eriksson, E. The complex clinical response to selective serotonin reuptake inhibitors in depression: a network perspective. Transl. Psychiatry13, 1–7 (2023). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 18.Borsboom, D. et al. Network analysis of multivariate data in psychological science. Nat. Rev. Methods Primers1, 58 (2021). [Google Scholar]
  • 19.Epskamp, S., Borsboom, D. & Fried, E. I. Estimating psychological networks and their accuracy: a tutorial paper. Behav. Res.50, 195–212 (2018). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 20.Borsboom, D. A network theory of mental disorders. World Psychiatry16, 5–13 (2017). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 21.Boschloo, L. et al. Symptom-specific effectiveness of an internet-based intervention in the treatment of mild to moderate depressive symptomatology: the potential of network estimation techniques. Behav. Res. Ther.122, 103440 (2019). [DOI] [PubMed] [Google Scholar]
  • 22.Cervin, M. et al. Symptom-specific effects of cognitive-behavioral therapy, sertraline, and their combination in a large randomized controlled trial of pediatric anxiety disorders. J. Child Psychol. Psychiatry61, 492–502 (2020). [DOI] [PubMed] [Google Scholar]
  • 23.Zhou, J. et al. The network analysis of depressive symptoms before and after two weeks of antidepressant treatment. J. Affect. Disord.299, 126–134 (2022). [DOI] [PubMed] [Google Scholar]
  • 24.Komulainen, K. et al. Network dynamics of depressive symptoms in antidepressant medication treatment: secondary analysis of eight clinical trials. Mol. Psychiatry26, 3328–3335 (2021). [DOI] [PubMed] [Google Scholar]
  • 25.Bekhuis, E. et al. Symptom-specific effects of psychotherapy versus combined therapy in the treatment of mild to moderate depression: a network approach. Psychother. Psychosom.87, 121–123 (2018). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 26.Madhoo, M. & Levine, S. Z. Network analysis of the Quick Inventory of Depressive Symptomatology: reanalysis of the STAR*D clinical trial. Eur. Neuropsychopharmacol.26, 1768–1774 (2016). [DOI] [PubMed] [Google Scholar]
  • 27.Bos, F. M. et al. Cross-sectional networks of depressive symptoms before and after antidepressant medication treatment. Soc. Psychiatry Psychiatr. Epidemiol.53, 617–627 (2018). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 28.Berlim, M. T., Richard-Devantoy, S., Santos, N. R. dos& Turecki, G. The network structure of core depressive symptom-domains in major depressive disorder following antidepressant treatment: a randomized clinical trial. Psychol. Med.51, 2399–2413 (2021). [DOI] [PubMed] [Google Scholar]
  • 29.Lewis, G. et al. The clinical effectiveness of sertraline in primary care and the role of depression severity and duration (PANDA): a pragmatic, double-blind, placebo-controlled randomised trial. Lancet Psychiatry6, 903–914 (2019). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 30.Hieronymus, F., Nilsson, S. & Eriksson, E. A mega-analysis of fixed-dose trials reveals dose-dependency and a rapid onset of action for the antidepressant effect of three selective serotonin reuptake inhibitors. Transl. Psychiatry6, e834 (2016). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 31.Hieronymus, F., Lisinski, A., Østergaard, S. D. & Eriksson, E. The response pattern to SSRIs as assessed by the Montgomery–Åsberg Depression Rating Scale: a patient-level meta-analysis. World Psychiatry21, 472–473 (2022). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 32.Lisinski, A., Hieronymus, F., Näslund, J., Nilsson, S. & Eriksson, E. Item-based analysis of the effects of duloxetine in depression: a patient-level post hoc study. Neuropsychopharmacoly45, 553–560 (2020). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 33.Boschloo, L. et al. The symptom-specific efficacy of antidepressant medication vs. cognitive behavioral therapy in the treatment of depression: results from an individual patient data meta-analysis. World Psychiatry18, 183–191 (2019). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 34.Borsboom, D. & Cramer, A. O. J. Network analysis: an integrative approach to the structure of psychopathology. Annu. Rev. Clin. Psychol.9, 91–121 (2013). [DOI] [PubMed] [Google Scholar]
  • 35.Epskamp, S. & Fried, E. I. A tutorial on regularized partial correlation networks. Psychol. Methods23, 617–634 (2018). [DOI] [PubMed] [Google Scholar]
  • 36.Salaminios, G. et al. A randomised controlled trial assessing the severity and duration of depressive symptoms associated with a clinically significant response to sertraline versus placebo, in people presenting to primary care with depression (PANDA trial): study protocol for a randomised controlled trial. Trials18, 496 (2017). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 37.Kroenke, K., Spitzer, R. L. & Williams, J. B. W. The PHQ-9. J. Gen. Intern. Med.16, 606–613 (2001). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 38.Beck, A., Steer, R. & Brown, G. Beck Depression Inventory–II (APA PsycTests, 1996).
  • 39.Spitzer, R. L., Kroenke, K., Williams, J. B. W. & Löwe, B. A brief measure for assessing generalized anxiety disorder: the GAD-7. Arch. Intern. Med.166, 1092–1097 (2006). [DOI] [PubMed] [Google Scholar]
  • 40.Jenkinson, C. et al. A shorter form health survey: can the SF-12 replicate results from the SF-36 in longitudinal studies?. J. Public Health19, 179–186 (1997). [DOI] [PubMed] [Google Scholar]
  • 41.Lewis, G., Pelosi, A. J., Araya, R. & Dunn, G. Measuring psychiatric disorder in the community: a standardized assessment for use by lay interviewers. Psychol. Med.22, 465–486 (1992). [DOI] [PubMed] [Google Scholar]
  • 42.R Core Team. R: A Language and Environment for Statistical Computing (R Foundation for Statistical Computing, 2022).
  • 43.Jones, P. networktools: Tools for Identifying Important Nodes in Networks (2018).
  • 44.Kuznetsova, A., Brockhoff, P. B. & Christensen, R. H. lmerTest package: tests in linear mixed effects models. J. Stat. Soft.82, 1–26 (2017). [Google Scholar]
  • 45.Ben-Shachar, M. S., Lüdecke, D. & Makowski, D. effectsize: estimation of effect size indices and standardized parameters. J Open Source Softw.5, 2815 (2020). [Google Scholar]
  • 46.Haslbeck, J. M. B. & Waldorp, L. J. mgm: estimating time-varying mixed graphical models in high-dimensional data. J. Stat. Softw.93, 1–46 (2020). [Google Scholar]
  • 47.Blanken, T. F. et al. Introducing network intervention analysis to investigate sequential, symptom-specific treatment effects: a demonstration in co-occurring insomnia and depression. Psychother. Psychosom.88, 52–54 (2019). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 48.van Borkulo, C. D. et al. Comparing network structures on three aspects: a permutation test. Psychol. Methods28, 1273–1285 (2023). [DOI] [PubMed] [Google Scholar]
  • 49.Rosseel, Y. lavaan: an R package for structural equation modeling. J. Stat. Softw.48, 1–36 (2012). [Google Scholar]
  • 50.Mulder, J. D. & Hamaker, E. L. Three extensions of the random intercept cross-lagged panel model. Struct. Equ. Modeling28, 638–648 (2021). [Google Scholar]
  • 51.Wysocki, A., McCarthy, I., Van Bork, R., Cramer, A. O. J. & Rhemtulla, M. Cross-lagged panel networks. adv.in/psych.2, e739621 (2025).
  • 52.Schermelleh-Engel, K., Moosbrugger, H. & Müller, H. Evaluating the fit of structural equation models: tests of significance and descriptive goodness-of-fit measures. Methods Psychol. Res. Online8, 23–74 (2003). [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

Supplementary Information (1MB, pdf)

Supplementary Tables 1–10 and Figs. 1 and 2.

Reporting Summary (2.1MB, pdf)

Data Availability Statement

All de-identified individual participant data collected in the PANDA trial and related documents (study protocol, analysis plan and code) are available, with no end, from the publications of the original trial paper. To gain access, researchers will need to enter a data access agreement with University College London (London, UK), providing a proposal for the use of data and a request for access (glyn.lewis@ucl.ac.uk).

All code used for the analyses is available via GitHub at https://github.com/giuliapiazza18/PANDAnet-2/.

Articles from Nature. Mental Health are provided here courtesy of Nature Publishing Group

RESOURCES