Introduction
Late-life depression (LLD) refers to a major depressive disorder (MDD) in individuals aged 60 years or older [1]. It includes a first episode of depression in older adults or recurrent depressive episodes that began earlier in life but re-emerged at an advanced age. LLD is one of the most prevalent mental disorders among older adults, with a prevalence of approximately 14% [Reference Andreas, Schulz, Volkert, Dehoust, Sehner and Suling2]. With the rapid growth of the aging population and the high prevalence of LLD, the demand for geriatric mental health care is expected to rise substantially [Reference Dafsari, Bewernick, Böhringer, Domschke, Elsaesser and Löbner3]. Beyond this burden, LLD is linked to higher relapse rates and severe health consequences, including accelerated somatic multimorbidity, cognitive decline, and increased mortality [Reference Choi, Kim, Marti and Chen4–Reference Van den Berg, Wiersema, Hegeman, van den Brink, Rhebergen and Merijnissen8]. Additionally, depressive symptoms vary across the lifespan, posing challenges for interventions. Older adults are more likely to experience somatic and vegetative symptoms, such as insomnia symptoms, including difficulties in sleep initiation, early morning awakening, difficulty staying asleep, and daytime dysfunctions, compared to younger depressed patients [Reference Schaakxs, Comijs, Lamers, Beekman and Penninx9, Reference Schaakxs, Comijs, Lamers, Kok, Beekman and Penninx10].
Existing evidence suggests that insomnia and depression exhibit a bidirectional relationship with shared underlying mechanisms, including shared genetic factors, hyperactivity of the arousal system, disruptions in synaptic plasticity, alterations in the biological timekeeping system, and shared macroscale neuroanatomical abnormalities [Reference Bagherzadeh-Azbari, Khazaie, Zarei, Spiegelhalder, Walter and Leerssen11–Reference Wang, Tian, Long, Dong, He and Qiu15]. Several meta-analyses and longitudinal studies demonstrated that insomnia symptoms and poor sleep are key risk factors for developing depression [Reference Hertenstein, Benz, Schneider and Baglioni16–Reference Yokoyama, Kaneita, Saito, Uchiyama, Matsuzaki and Tamaki18]. Conversely, the presence of depression may adversely affect sleep patterns and induce insomnia symptoms [Reference Pandi-Perumal, Monti, Burman, Karthikeyan, BaHammam and Spence19], which are one of the most prevalent residual symptoms following LLD treatment, with a prevalence of approximately 50% [Reference Patel, Steinberg and Patel20, Reference Irwin, Carrillo, Sadeghi, Bjurstrom, Breen and Olmstead21]. These symptoms have been demonstrated to be associated with the onset, worsening, and recurrence of depressive symptoms, as well as suicidal ideations, particularly in older adults [Reference Bao, Han, Ma, Wang, Shi and Wang22–Reference Wang, Cheng and Xu25]. More severe insomnia symptoms are associated with higher rates of non-response to antidepressant treatment, failed remission, and relapse in LLD [Reference Gebara, Kasckow, Smagula, DiNapoli, Karp and Lenze26, Reference Pigeon, Hegel, Unützer, Fan, Sateia and Lyness27]. To address the association between insomnia and depressive symptoms, multivariate approaches should be employed, as relying solely on the total scores of questionnaires may obscure changes at the individual symptom level of a very heterogeneous clinical condition such as LLD.
Current interventions for LLD include pharmacotherapy and psychotherapy (e.g., cognitive behavioral therapy (CBT)), although suboptimal response to antidepressants and an increased risk of adverse effects, such as hyponatremia, have been reported in older adults [Reference Ji, Sun, Zhou, Li, Wei and Wang28–Reference Nelson, Macaluso and Preskorn30]. Thus, the present study aimed to explore the multivariate interrelationship between insomnia and depressive symptoms before and after psychotherapy using canonical correlation analysis (CCA) as conducted previously [Reference Küppers, Bi, Nicolaisen-Sobesky, Hoffstaedter, Yeo and Drzezga31, Reference Nicolaisen-Sobesky, Mihalik, Kharabian-Masouleh, Ferreira, Hoffstaedter and Schwender32]. This method allows us to assess bidirectional associations between two sets of variables to identify 1) how much insomnia and depressive symptoms are related to each other in patients with LLD at baseline, and 2) to determine whether and how psychotherapy treatment may modify their overall multivariate association, as well as each individual variables of both insomnia and depressive scales.
Methods
Participants
The current study analyzed a subset of data from a larger, multicenter, randomized controlled trial conducted at seven trial sites in Germany (Cognitive Behavioral Therapy for Late-Life Depression – CBTlate trial) [Reference Dafsari, Bewernick, Böhringer, Domschke, Elsaesser and Löbner3]. From the original cohort of 251 participants with moderate to severe LLD, we included 185 participants with available scores for assessing insomnia and depressive symptoms at three time points (baseline, end-of-treatment, and follow-up sessions) (Figure 1a,b). Inclusion criteria were the following: 1) outpatients with age ≥ 60 years; 2) the Geriatric Depression Scale (GDS-30) score > 10; 3) the Quick Inventory of Depressive Symptomatology, Clinician Rating (QIDS-C) score > 10; 4) the Mini-Mental Status Test (MMST) score > 25, to ensure not including patients with severe cognitive impairment; 5) not received or been stable on pharmacological treatment for at least six weeks before and during the eight weeks of treatment (Table 1). In view of conditions that could additionally alter sleep patterns, exclusion criteria were: 1) existing comorbidities, including bipolar depression or psychotic disorders; 2) regular use of hypnotics or benzodiazepine medications; 3) other chronic medical/somatic conditions. For ethical reasons, 38% of patients continued to take their current antidepressants without restrictions on medication type. Any change in dosage or medication led to exclusion from the study. More details on the in−/exclusion criteria and the interventions in the original trial are available elsewhere [Reference Dafsari, Bewernick, Böhringer, Domschke, Elsaesser and Löbner3]. This clinical trial is registered at ClinicalTrials.gov (NCT03735576) and DRKS (DRKS0013769). The study was approved by the Institutional Review Board/Institutional Ethical Committee (IRB/IEC) of University Hospital Cologne (Nr. 18–129) and the participating sites prior to trial initiation. All participants provided written informed consent before all study procedures.
Figure 1. Overview of the study procedure: (a) Distribution of participants across seven trial sites and available data; (b) Assessment and intervention time points; (c) rCCA model representation of insomnia and depressive symptoms domains; (d) Machine learning framework (outer and inner split of data); (e) rCCA output: canonical correlation coefficients, canonical loadings and cross-loadings. LLD-CBT: Late-life depression cognitive behavioral therapy; SUI: Supportive unspecific intervention.
Table 1. Patient characteristics
Note: Participants had an average of 69.7 years old (SD = 7.1). Abbreviations: LLD-CBT: Late-life-depression specific cognitive behavioral therapy, SUI: supportive unspecific intervention, GDS: Geriatric Depression Scale (possible score range 0–30), QIDS-C: Quick Inventory of Depressive Symptomatology-Clinician rating (possible score range 0–27), ISI: Insomnia Severity Index (possible score range 0–28), NA: not available.
a Reflects the number and percentage of participants answering “yes” to this question.
* These patients were among the 38% who received stable pharmacological treatment before and during the study, with some patients partially represented in multiple medication categories.
Assessments and interventions
Three different assessment time points were considered: baseline scores, at the end of the treatment period (week 10), and the six-month follow-up (Figure 1b).
Depressive and insomnia symptoms: The German versions of the Geriatric Depression Scale (GDS) questionnaire [Reference Yesavage, Brink, Rose, Lum, Huang and Adey33, Reference Gauggel and Birkner34] and the Quick Inventory of Depressive Symptomatology (QIDS) [Reference Rush, Trivedi, Ibrahim, Carmody, Arnow and Klein35, Reference Roniger, Spaeth, Schweiger and Klein36] were used to measure depressive symptoms. Insomnia symptoms were assessed using the German version of the Insomnia Severity Index (ISI) [Reference Bastien, Vallieres and Morin37, Reference Gerber, Lang, Lemola, Colledge, Kalak and Holsboer-Trachsler38]. For more details, see Supplementary Material.
Psychotherapeutic treatment: Scores on the GDS and ISI were considered at the baseline assessment before treatment and after receiving either the LLD-specific cognitive behavioral therapy (LLD-CBT) or the supportive unspecific intervention (SUI). Both groups underwent a structured manual-based treatment consisting of 15 sessions, conducted twice weekly over an eight-week period. Further details on the interventions have been published elsewhere [Reference Dafsari, Bewernick, Böhringer, Domschke, Elsaesser and Löbner3, Reference Dafsari, Bewernick, Biewer, Christ, Domschke and Froelich39].
Multivariate and statistical analyses
Regularized canonical correlation analyses (rCCA): We applied CCA and conducted three separate analyses for the baseline, end-of-treatment, and follow-up individual scores of insomnia and depressive symptoms. CCA is a multivariate, data-driven statistical method integrating various symptoms into a single model to identify latent dimensions where specific items from both sets are maximally correlated [Reference Hotelling40]. A “canonical variate” represents the specific linear combination of variables within each set, and a pair of canonical variates constitutes a mode (Figure 1c). To interpret each mode, we computed the “canonical correlation,” which quantifies the overall strength of association between the paired canonical variates. In addition, we calculated “canonical loadings,” defined as the Pearson correlations between each original variable and its corresponding canonical variate. These loadings indicate how much each symptom contributes to shaping its domain in a way that maximizes its connection to the other domain. Symptoms with high canonical loadings play a central role in the shared relationship between the two domains, highlighting core features within each domain that are clinically relevant for understanding cross-domain interactions and informing targeted treatment strategies. We also computed “cross-loadings,” which are the correlations between each original variable and the canonical variate from the opposite set. Cross-loadings reflect how strongly a symptom from one domain (e.g., insomnia symptoms) is associated with the broader pattern of variance in another domain (e.g., depressive symptoms) (Figure 1c). Such cross-domain associations are particularly relevant in the context of comorbid conditions such as insomnia and depression, as they can help identify bridging symptoms (i.e., symptoms that link two domains), may reflect shared underlying mechanisms, and could serve as important targets for transdiagnostic interventions.
One set of variables (domain X) represented the insomnia dimension, which consisted of scores from seven items of ISI. The other set of variables (domain Y) represented depressive symptoms, comprising scores from 26 GDS items and eight QIDS-C items (Figure 1c). Four items of the GDS (sadness, feeling downhearted, feeling without energy, not feeling clear-minded) were excluded, as they showed an extremely skewed distribution in our sample. Notably, the sleep disturbance criterion domain (which includes initial, middle, and late insomnia or hypersomnia) from the QIDS-C was excluded, as performed previously [Reference Olfati, Samea, Faghihroohi, Balajoo, Küppers and Genon41]. This decision was made to avoid including any sleep-related items within the depressive domain, as doing so could result in an inaccurately higher association between the two sets. Since both LLD-CBT and SUI significantly reduced depressive symptoms without evidence for the superiority of LLD-CBT in the initial trial [Reference Dafsari, Bewernick, Böhringer, Domschke, Elsaesser and Löbner3, Reference Dafsari, Bewernick, Biewer, Christ, Domschke and Froelich39], all participants were pooled for the current analyses, regardless of the allocation to the treatment groups. The analyses were implemented by using the CCA/PLS Toolkit (v1.0.0, https://github.com/anaston/cca_pls_toolkit/) in MATLAB Runtime (v9.14 (R2023b)) [42]. Age, gender, and trial site were included as confounding variables in the analyses.
To address potential limitations of CCA (i.e., overfitting and multicollinearity among variables in each behavioral domain), a regularized version of CCA (rCCA) was implemented within a machine learning framework [Reference Küppers, Bi, Nicolaisen-Sobesky, Hoffstaedter, Yeo and Drzezga31, Reference Nicolaisen-Sobesky, Mihalik, Kharabian-Masouleh, Ferreira, Hoffstaedter and Schwender32]. A multiple hold-out framework was utilized by dividing the dataset into an 80% optimization set and a 20% hold-out set, using 100 random repetitions. This approach evaluates the model’s generalizability by examining how well the associations identified in the training set can apply to an independent test set. In this process, the hold-out set was projected onto the optimal rCCA model obtained from the optimization set, aiming to estimate the canonical correlation of the hold-out set. The optimization set was further split into an 80% training set and a 20% test set through an inner split, repeated 50 times. During this inner split, optimal hyperparameters were estimated using L2 norm regularization (Figure 1d). This additional constraint forces the weights to be small but not zero. By mitigating multicollinearity, these hyperparameters help enhance the stability of the rCCA weights [Reference Mihalik, Chapman, Adams, Winter, Ferreira and Shawe-Taylor43].
We further assessed the longitudinal stability and alteration of the associations across time points by testing whether pre-treatment associations persist post-treatment and whether new relationships emerge after treatment. Therefore, the end-of-treatment and follow-up data were projected onto the weights derived from the baseline rCCA model. Additionally, the baseline and follow-up data were projected onto the weights of the end-of-treatment model. The model was initially trained at the baseline, after which it was tested on the hold-out set from the baseline time point. Subsequently, the model was tested on the hold-out sets from the other two time points, and this process was repeated for the end-of-treatment model. This process was repeated for all 100 hold-out repetitions to ensure the robustness of rCCA correlation results.
Fisher’s Z transformation and statistical comparisons: We used Fisher’s Z transformation to compare the correlation coefficients from our rCCA analyses before and after treatment, assessing whether the treatment significantly altered the association between insomnia and depressive symptoms. First, we normalized canonical correlation coefficients, loadings, and cross-loadings by computing their Fisher’s Z scores. Using Fisher’s Z-test, we then compared the canonical correlation coefficients from baseline to the end of treatment and follow-up, applying the Holm–Bonferroni correction for multiple comparisons. By comparing these coefficients, we assessed whether treatment alters the strength of the relationship between domains.
To further identify symptom-specific changes and detect treatment-sensitive or persistent symptoms, we examined shifts in canonical loadings and cross-loadings. Delta scores (Δ; end of treatment – baseline Fisher’s Z) were tested using Fisher’s Z-test to detect significant changes, highlighting symptom-level interactions influenced by treatment. Changes in canonical loadings indicate which symptoms become more or less central to the part of a domain that is linked to the other. Changes in cross-loadings reveal how treatment affects the bridging symptoms. Together, these “symptom-level” changes provide clinical insight into how the structure and interaction of symptom networks evolve in response to treatment. This information could not be captured by “domain-level” canonical correlation coefficients alone.
Results
Descriptive statistics
The data set comprised 185 participants, with an average age of 69.7 years old (SD = 7.1) of whom 66% were female. Patients were recruited from seven university hospitals in Germany (Table 1).
Alteration of multivariate association between insomnia and depressive symptoms after psychotherapy
At baseline, a hold-out set correlation of r = 0.24 (SD = 0.13) was observed between insomnia symptoms and depressive symptoms. By the end of the treatment, this correlation slightly increased (r = 0.42, SD = 0.12) and remained consistently high at the six-month follow-up session (r = 0.48, SD = 0.12) (Figure 2a). Notably, the increase in the hold-out correlation from baseline to the end of treatment was not significant after applying the Holm-Bonferroni correction (p corrected = 0.064). However, the increase from baseline to follow-up was significant (p corrected = 0.018). Testing the baseline model on the post-treatment hold-out set shows a similar pattern, reflecting the stability of the identified associations (Figure 2b). Conversely, when testing the end of the treatment model on the baseline hold-out set, a decrease in the correlation can be observed, indicating new association patterns that were not present at baseline (Figure 2c).
Figure 2. Hold-out correlations of each regularized canonical correlation analysis (rCCA) model: (a) rCCA model between insomnia and depressive symptoms for baseline, end-of treatment and follow-up time points; (b) Testing the baseline rCCA model on each time point’s hold-out set; (c) Testing the end-of-treatment model on each time point’s hold-out set. Dots represent the hold-out correlations for each of the 100 repetitions. Black horizontal lines indicate the mean hold-out correlation coefficients. (*) Holm–Bonferroni corrected significant differences (p-value <0.05).
At baseline, insomnia symptoms were positively associated with depressive symptoms, indicating that greater insomnia symptom severity was linked to more severe depressive symptoms. Higher canonical loadings were observed for general dissatisfaction with the sleep pattern (r = 0.88), the interference of sleep problems with daily functioning (r = 0.82), and difficulty staying asleep (r = 0.81) (Figure 3a). Higher canonical loadings were observed mainly for anxiety-related depressive symptoms, such as worrying about the future (r = 0.52), excessive rumination (r = 0.46), fear of the occurrence of future bad events (r = 0.43) or general feelings of restlessness (r = 0.4). In contrast, decreased involvement and interest in daily life (r = −0.31), increased hopelessness about the future (r = −0.28), negative mood (r = −0.26), psychomotor agitation or retardation (r = −0.26), and increased memory problems (r = −0.22) showed negative canonical loadings, indicating associations with less insomnia symptoms (Figure 3c).
Figure 3. Canonical loadings before and after psychotherapeutic treatment: (a) Canonical loadings of insomnia symptoms at baseline; (b) Canonical loadings of insomnia symptoms at the end of treatment; (c) Canonical loadings of depressive symptoms at baseline; (d) Canonical loadings of depressive symptoms at the end of treatment. Canonical loadings correspond to Pearson correlations between the input variables and the canonical variates. Negative loadings of depressive symptoms indicate more insomnia symptoms, and less depressive symptoms, while positive loadings indicate more insomnia symptoms and more depressive symptoms. Black lines represent error bars (one standard deviation below or above the mean).
At the end of treatment, the canonical loadings for all insomnia symptoms were positively associated with depressive symptoms. The strongest associations were observed for sleep problems being noticeable to others (r = 0.89), sleep problems interfering with daily functioning (r = 0.82), general dissatisfaction with sleep pattern (r = 0.75), and difficulty falling asleep (r = 0.74) (Figure 3b). All depressive symptoms following psychotherapy intervention showed a positive association with insomnia symptoms. The highest loadings were seen in negative affective symptoms, such as overall depressive mood (r = 0.71), restlessness (r = 0.59), perceiving life as empty (r = 0.58), feeling hopeless about the future (r = 0.57) and having a low basic life satisfaction (r = 0.57). Additionally, somatic symptoms, such as low energy and high fatigue, were also strongly and positively associated with insomnia symptoms (r = 0.63) (Figure 3d).
At follow-up, the loadings for insomnia symptoms remained consistent with the canonical loadings at the end of treatment. Similarly, all depressive symptoms at follow-up remained positively associated with insomnia symptoms (Supplementary Material).
Changes in canonical loadings and cross-loadings after psychotherapy
The canonical loadings of most depressive items changed significantly after psychotherapy compared to baseline (corrected p-values between 0.002 and 0.005) (Figure 4a). The calculated difference (delta) scores for the “canonical loadings” revealed the most significant changes in items related to reduced activities (delta = 0.501, p < 0.001), interests, and involvement in daily life (delta = 0.546, p < 0.001), psychomotor agitation or retardation (delta = 0.54, p < 0.001), memory problems (delta = 0.543, p < 0.001), difficulty concentrating and decisiveness (delta = 0.504, p < 0.001) (see Figures 4a and 5 for all items). In contrast, the canonical loadings of insomnia symptoms showed no significant changes after treatment for any items (Figure 4b).
Figure 4. Changes in canonical loadings and cross-loadings before and after psychotherapeutic treatment: (a) Fisher z-scores of canonical loadings at baseline and after treatment for depressive symptoms; (b) Fisher z-scores of canonical loadings at baseline and after treatment for insomnia symptoms; (c) Fisher z-scores of cross-loadings at baseline and after treatment for depressive symptoms; (d) Fisher z-scores of cross-loadings at baseline and after treatment for insomnia symptoms. (*) Holm–Bonferroni corrected significant differences between time points (p-value <0.05).
Figure 5. Schematic representation of canonical loadings and cross-loadings shifts in the multivariate association between depressive symptoms and insomnia symptoms following psychotherapeutic intervention: Some insomnia-related depressive symptom loadings altered after treatment (dark orange – canonical loadings; bold names in legend – canonical cross-loadings) based on Holm-Bonferroni corrected significant differences between time points (p-value <0.05), while the others remained stable (light orange). Interestingly, depressive-related insomnia items remained fully unchanged. The thickness of the two-sided arrow shows the multivariate association between insomnia and depressive symptoms before (r = 0.24) and after (r = 0.42) treatment. This figure was created with Biorender.com.
With regard to the cross-loadings of depressive symptoms (as the association with the insomnia canonical variate), significant changes after psychotherapy were observed for lower energy and higher fatigue (delta = 0.429, p = 0.001), decrease in involvement and interests (delta = 0.41, p = 0.003), increased negative mood (delta = 0.368, p = 0.015), increased depressive mood (delta = 0.363, p = 0.018), and avoiding social gatherings (delta = 0.334, p = 0.049) (Figure 4c). Conversely, the cross-loadings of insomnia symptoms (as the association with the depressive symptoms canonical variate) did not show any significant changes (Figure 4d), which emphasized that the insomnia symptoms pattern remained stable and independent from changes in depressive symptoms following CBT/SUI psychotherapy.
Discussion
In the present study, we found a moderate association between insomnia and depressive symptoms at baseline, providing further evidence of their bidirectional relationship in LLD. This association increased at the end of treatment and at the six-month follow-up. Interestingly, while we observed improvement in depressive symptoms, insomnia symptoms did not change after psychotherapy. We observed a trend of increasing multivariate associations between insomnia and depressive symptoms after psychotherapy. Additionally, by projecting the end of the treatment model onto the baseline test set, we observed a decrease in the observed correlation, highlighting the emergence of new or increased patterns of association after psychotherapy that could not be captured by the baseline model. Since the applied interventions primarily addressed depressive symptoms rather than insomnia, unresolved sleep problems could have exacerbated specific depressive symptoms, thereby strengthening their interrelationship. Of note, changes between the baseline and the end of the treatment model were not significant after correction for multiple comparisons, probably due to our small sample size and high data variability. However, changes between the baseline and the follow-up sessions were significant. Several existing studies performed CCA analyses to identify biotypes of depression using the brain’s functional connectivity data in patients with MDD [Reference Drysdale, Grosenick and Downar44–Reference Dinga, Schmaal and Marquand46]. Some studies also applied CCA analysis between depression and anxiety symptoms among college students [Reference Li, Wang and Li47] or between sleep quality, depressive symptoms, brain structure, and motor performance in the general population samples [Reference Küppers, Bi, Nicolaisen-Sobesky, Hoffstaedter, Yeo and Drzezga31]. However, to the best of our knowledge, no existing study has assessed the role of psychotherapy intervention on the interplay between insomnia and depressive symptoms in LLD using such multivariate analysis.
It has been shown that pharmacotherapy treatment and commonly used psychotherapy interventions alleviate certain types of depressive symptoms but fail to fully resolve insomnia symptoms in depressed patients [Reference Danon, Poupon, Courtet and Gorwood48, Reference Staner49]. For example, clusters of negative emotional and cognitive bias and neurovegetative symptoms remained after 6 weeks of antidepressant treatment in patients with MDD [Reference Danon, Poupon, Courtet and Gorwood48]. These residual insomnia symptoms increase vulnerability to depressive relapse (in the same episode) or recurrence (in the new episode) and a lower likelihood for patients to have full benefit from treatment [Reference Kwasny, Wlodarczyk, Dywel, Szarmach, Strandberg and Cubala50, Reference Lustberg and Reynolds51]. Residual insomnia symptoms are then often overlooked or treated with hypnotic medications or particular antidepressants despite the risks of side effects (e.g., daytime impairment, amnesia, falling down, and dependency), particularly in older adults, who are nearly five times more likely to receive prescriptions of medications than younger individuals [Reference Abad and Guilleminault52–Reference Hauglund, Andersen, Tokarska, Radovanovic, Kjaerby and Sørensen54]. Given these concerns, a non-pharmacological approach, such as cognitive behavioral therapy for insomnia (CBT-I)—which includes cognitive therapy, stimulus control, sleep restriction, sleep hygiene, and relaxation techniques—may be an additional and effective first-line treatment and could be integrated into clinical practice for older adults with LLD to prevent residual insomnia [Reference Irwin, Carrillo, Sadeghi, Bjurstrom, Breen and Olmstead21, Reference Riemann, Espie, Altena, Arnardottir, Baglioni and Bassetti55]. CBT-I is suggested as an early effective intervention not only for insomnia disorder but also in mood disorders [Reference Palagini, Aquino, Alfi, Massoni, Gambini and Miniati56]. The stable association observed between insomnia and depressive symptoms after a six-month follow-up in the present study suggests a persistent relationship and symptoms’ interaction pattern. Interestingly, a study based on the Growth Mixture Modeling in patients with MDD demonstrated that characteristics of insomnia can predict long-term depression outcomes. In particular, patients with lower insomnia severity were the optimal responders to psychotherapy and exhibited more favorable treatment responses, even at a two-year follow-up, compared to patients with higher insomnia severity [Reference Bei, Asarnow, Krystal, Edinger, Buysse and Manber57]. These findings support the need to consider careful management of insomnia symptoms during the treatment of clinical depression.
At baseline, our canonical loadings indicated a positive relationship between insomnia symptoms and anxiety-related depressive symptoms, particularly excessive worry, rumination, and feelings of restlessness (Figure 4a). Excessive uncontrollable worry, rumination, emotion dysregulation, and intrusive thoughts have been associated with increased insomnia symptoms in previous studies due to hyperarousal conditions [Reference Kirwan, Pickett and Jarrett58–Reference Samea, Mortazavi, Reimann, Ebneabbasi, Zarei and Khazaie61], also for older patients [Reference Zhou, Birditt, Leger and Fingermann62]. In this study, after psychotherapy and at follow-up, this association shifted, with insomnia symptoms becoming more strongly linked to negative affective symptoms, such as low mood, negative outlook, and reduced life satisfaction, as well as somatic symptoms, including low energy and heightened fatigue (Figure 4). Figure 5 provides a schematic representation of significant shifts in the association of canonical loadings. This further suggests that while CBT/SUI treatment may have focused more on cognitive aspects, sleep-related and somatic symptoms remained unchanged, contributing to persistent negative mood. It has been shown that older MDD patients with insomnia, compared to those without sleep problems, were more likely to experience somatic symptoms such as low energy, which were, in turn, associated with persistent negative affect [Reference Zhao, Liu, Wu, Wang, Cao and Lyu63].
Individuals with the same severity of depression can exhibit substantial differences in how their symptoms interact [Reference Ebrahimi, Borsboom, Hoekstra, Epskamp, Ostinelli and Bastiaansen64]. A novel “person-specific symptom network” model suggests that examining individual symptom relationships could help identify maladaptive patterns, enabling personalized treatment approaches to weaken these persistent associations [Reference Schumacher, Klein, Hautzinger, Härter, Schramm and Kriston65]. We observed that psychomotor agitation or retardation, memory and concentration difficulties, and decreased engagement in daily activities were negatively associated with insomnia at baseline but became positively linked by the end of treatment and follow-up. Items such as decreased involvement and interests and somatic symptoms such as higher fatigue showed the highest cross-loading delta scores, reflecting significant shifts in their relationship with insomnia over time (Figure 4c-d). Psychomotor agitation has been linked to depression in older adults, often alongside increased insomnia symptoms [Reference Olgiati, Serreti and Colombo66]. Agitated depression may represent a distinct LLD subtype, where heightened agitation initially masks perceived sleep difficulties, explaining the negative correlation at baseline. After therapy, patients may have gained a greater awareness of their sleep disturbances, leading to the observed change in the association. Current models of insomnia emphasize its role in hyperarousal driven by central and peripheral activation of the stress system and its involvement in GABAergic transmission [Reference Palagini and Bianchini67]. This dysregulation may contribute to symptoms such as psychomotor agitation or retardation.
Recognizing our limitations can guide future research to strengthen and expand our findings. For example, recall/response bias is common in self-report questionnaires. Moreover, psychotherapy may enhance introspection and self-awareness, enabling patients to better recognize the connection between sleep disturbances and depressive symptoms, making this relationship more apparent after the treatment [Reference Wampold68]. In addition to psychotherapy, 38% of our patients were also receiving antidepressant medication. However, their pharmacological treatment was required to remain stable before and throughout the intervention. A supplementary analysis including the use of psychotropic medication as an additional confound showed no significant difference in the results (see Supplementary Figure S2, S3). Moreover, while individual depressive symptoms showed varying canonical loadings, all depressive symptoms exhibited high loadings post-treatment and at follow-up. Notably, individuals with mood disorders often report excessive sleep disturbances and poor sleep quality, though objective measures like polysomnography may show different results [Reference Rumble, White and Benca69, Reference Scarlett, Nolan, Kenny and O’Connell70]. This study relied solely on the ISI scale, which focuses on insomnia; incorporating additional subjective and objective sleep measures, such as sleep duration and efficiency, could provide a more comprehensive assessment on the sleep-depression relationship. Future research should also compare different treatment modalities, particularly LLD-CBT, versus insomnia-specific interventions (e.g., CBT-I). Neuroimaging data could further reveal underlying neurobiological substrates of interventions. Patients in this study received either LLD-CBT or SUI, but separate rCCA models for each group were not statistically valid due to sample size limitations. Larger datasets are needed to confirm these findings and perform clustering approaches based on the link between sleep and depression, as suggested recently [Reference Solomonov, Lee, Banerjee, Chen, Sirey and Gunning71, Reference Tozzi, Zhang, Pines, Olmsted, Zhai and Anene72].
In conclusion, this study demonstrated increased multivariate association between insomnia and depressive symptoms in LLD patients following psychotherapy, alongside a shift in the pattern of depressive symptoms, while insomnia symptoms remained unchanged. These findings emphasize the need to consider their relationship when assessing treatment effects. Furthermore, the findings underscore the value of multivariate approaches in examining symptom interactions, ultimately aiming to enhance clinical treatment.
Supplementary material
The supplementary material for this article can be found at http://doi.org/10.1192/j.eurpsy.2025.10088.
Data availability statement
The data supporting this study’s findings are not publicly available due to ethics restrictions but are available from Dr. Forugh Salimi Dafsari upon reasonable request with individual permission from the local institutions’ ethics board. The preprint of this work is online in medRxiv (https://doi.org/10.1101/2025.03.28.25324836).
Acknowledgements
The authors would like to thank all participating patients, therapists, and outcome evaluators who participated in the study.
Author contribution
Conception and study design: R.H., V.K., F.D., and M.T; Preprocessing and data analysis: R.H., V.K., F.S.; Interpretation F.S., F.D., D.R., S.B.E., F.J., and M.T. Paper writing and editing: all authors.
Financial support
The initial randomized controlled trial was funded by grant BMBF 01KG1716 from the German Federal Ministry of Education and Research (BMBF). V.K. is supported by the Deutsche Forschungsgemeinschaft (DFG, German Research Foundation) - Project-ID 431549029 - SFB 1451. The current analyses received no specific grant from any funding agency, commercial or not-for-profit sectors.
Competing interests
The authors declare no conflict of interest.
Open access
Comments
No Comments have been published for this article.