Globally in 2021, depression ranked second among causes of years lived with disability(1). It substantially impairs academic performance and social interactions(Reference Kovacs2,Reference Birmaher, Arbelaez and Brent3) and reduces overall quality of life(Reference Keenan-Miller, Hammen and Brennan4). It is also linked to increased morbidity and mortality(Reference Markkula, Härkänen and Perälä5,Reference Osby, Brandt and Correia6) . Approximately 50 % of all primary diagnoses of depression are reported to occur before the age of 14 years, and about 75 % emerge before the age of 24 years(Reference Kessler, Berglund and Demler7), which emphasises the importance of this early stage of life for this disorder.
Following psychotherapy, the administration of psychotropic drugs is considered as second-line treatment(8). Currently, the selective serotonin reuptake inhibitor (SSRI) fluoxetine is the only medication approved in Germany for treating children and adolescents with depression. Other SSRI should be used only in case of contraindications(8). However, the use of SSRI is often associated with various side effects, including insomnia, agitation and gastrointestinal problems(Reference Emslie, Kratochvil and Vitiello9). These side effects, along with the small effect sizes(Reference Barth, Kriston and Klostermann10) and frequent non-response to SSRI(Reference Lundorff, Holmgren and Zachariae11,Reference Johnston, Powell and Anderson12) , have led to an increasing pursuit for additional safe and effective treatment options.
In this context, the normalisation of vitamin D status has attracted attention as a potential treatment approach. In Germany, the representative KiGGS study revealed that, according to the vitamin D cut-offs of the Institute of Medicine (IOM)(Reference Ross, Taylor and Yaktine13), 33·1 % of children and adolescents in Germany were ‘potentially at risk for inadequacy’ (25-hydroxy-cholecalciferol (25(OH)D) levels 12 to < 20 ng/ml (30 to < 50 nmol/l)), and additional 12·5 % were ‘at risk for deficiency’ (< 12 ng/ml (< 30 nmol/l))(Reference Rabenberg, Scheidt-Nave and Busch14). Vitamin D comprises secosterols, primarily vitamin D2 and D3 (Reference Charoenngam, Shirvani and Holick15). Unlike other fat-soluble nutrients, its status is mainly determined by endogenous synthesis from cholesterol precursors in the skin during the exposure to UV-B radiation (280–315 nm), rather than dietary intake(Reference Holick, Binkley and Bischoff-Ferrari16,Reference Braegger, Campoy and Colomb17) . Individuals with depression have lower vitamin D levels, with the lowest levels associated with the greatest risk(Reference Hoogendijk, Lips and Dik18–Reference Eskandari, Martinez and Torvik25) and severity of depressive symptoms(Reference Hoogendijk, Lips and Dik18,Reference Wilkins, Birge and Sheline20,Reference Wilkins, Sheline and Roe21,Reference May, Bair and Lappé23,Reference Zhou, Shao and Gan26,Reference Ganji, Milone and Cody27) . Using data from the above-mentioned KiGGS study, we reported earlier that low vitamin D concentrations were associated with emotional problems in children and adolescents in Germany(Reference Husmann, Frank and Schmidt28). Furthermore, meta-analyses indicate therapeutic benefits of vitamin D3 supplementation for depression, particularly at doses > 2800 IU/day for ≥ 8 weeks, with larger effects in individuals with low 25(OH)D(Reference Vellekkatt and Menon29,Reference Xie, Huang and Lou30) . However, findings in children and adolescents are inconsistent(Reference Libuda, Timmesfeld and Antel31). The biological mechanisms underlying the association between vitamin D deficiency and depression remain to be identified, since mechanistic studies, especially in this age group, are limited(Reference Al-Sabah, Al-Taiar and Shaban32).
The hypothesised mechanisms are predominately based on research in cell and animal models. These studies indicate increased region-specific expression of vitamin D receptors in brain areas pivotal for mood regulation(Reference Prüfer, Veenstra and Jirikowski33), along with the homeostatic and trophic effects of vitamin D(Reference Menon, Kar and Suthar34). Recently, vitamin D has been recognised for its antioxidant, pro-neurogenic, neuromodulatory, and especially anti-inflammatory properties, which are increasingly considered essential contributors to its antidepressant effects(Reference Fedotova, Zarembo and Dragasek35–Reference Morello, Landel and Lacassagne38). In this context, neuroinflammation has emerged as a key factor in the onset and progression of depression(Reference Haan, Klein and Hart39). The role of inflammatory markers such as C-reactive protein (CRP), as well as pro- and anti-inflammatory cytokines, including tumor necrosis factor-alpha (TNF-α), interferon-gamma (IFN-γ) and interleukin (IL)-1β, IL-6, IL-8 and IL-10, is frequently discussed regarding the pathophysiology of depression(Reference Harsanyi, Kupcova and Danisovic40–Reference Osimo, Pillinger and Rodriguez45). Elevated levels of CRP, TNF-α, IFN-γ, IL-1β, IL-6 and IL-8 were observed to be linked with greater severity of depressive symptoms(Reference Harsanyi, Kupcova and Danisovic40–Reference Colasanto, Madigan and Korczak42,Reference Kelly, Smith and Mezuk44) . Conversely, higher levels of anti-inflammatory cytokines such as IL-10 have been observed alongside reduced depressive symptoms, highlighting the complex interplay between immune responses and mood disorders(Reference Perry, Upthegrove and Kappelmann43,Reference Osimo, Pillinger and Rodriguez45) . Thus, further investigations into this association are important as they could provide a foundation for promising therapeutic approaches. Accordingly, Vellekkatt and Menon recommended that studies on vitamin D and depression should concurrently examine inflammatory markers as putative pathways linking vitamin D status to symptom severity(Reference Vellekkatt and Menon29).
The few existing studies on the association between depression, vitamin D and inflammation parameters with a concurrent assessment of all three factors were all conducted in adults(Reference Kaviani, Nikooyeh and Etesam46–Reference Wang, Liu and Lian52). Overall, these studies present inconsistent findings and have not yet provided a clear answer whether vitamin D exerts its potential antidepressant effects through its anti-inflammatory properties. Thus, the association and extent of the interaction remain unclear, especially in children and adolescents. Based on these considerations, we aimed to test two a priori hypotheses in cross-sectional analyses: (1) lower vitamin D status is associated with greater depression severity in children and adolescents and (2) this association is moderated by inflammation (CRP and pro-/anti-inflammatory cytokines).
Materials and methods
Study design and participants
For the current examination, we analysed data from two studies. Both studies involved psychiatric patients receiving care at the Department of Child and Adolescent Psychiatry, Psychosomatics and Psychotherapy Essen, Germany (LVR-Klinikum Essen). The ‘Vitamin D study’ (June 2016–May 2018; registration number: DRKS00009758) focused on effects of vitamin D supplementation by a two-armed parallel-group, double-blind randomised controlled design. For the presented analysis, however, we only used baseline data(Reference Föcker, Antel and Grasemann53). Second, we included data from the cross-sectional ‘Nutrition and Mental Health study’ (September 2018–May 2020)(Reference Hirtz, Libuda and Hinney54), designed to examine the relationship between diet, nutrient supply, metabolism and mental health in children and adolescents. Both studies were conducted in accordance with the Declaration of Helsinki and approved by the local Ethics Committee (No. 15-6363-BO). Given the congruence in study settings and assessment tools, as well as identical inclusion and exclusion criteria, both datasets were pooled for analysis, resulting in a sample of 512 participants (Figure 1). In both studies, inclusion required inpatient or daycare status, age 11·3–18·9 years and written informed consent. Participants with current severe somatic disease and/or intellectual disabilities (IQ < 70) were excluded. In cases where participants were enrolled in both studies, only data from the ‘Nutrition and Mental Health Study’ were used. For the current analyses, participants without serum 25(OH)D and/or CRP measurement data were excluded. Moreover, data from four participants had to be discarded due to pre-analytical issues or technical problems during biochemical analyses, resulting in a ‘total sample’ of 465 participants (Figure 1). In the total sample (n 465), depression was diagnosed in 293 participants based on clinical psychiatric evaluation according to the International Statistical Classification of Diseases and Related Health Problems Revision 10 (ICD-10), whereas the remaining participants (n 172) predominantly presented with other psychiatric disorders, including conduct-related and emotional disorders, anxiety disorders, eating disorders and substance-related disorders. Depressive symptoms were assessed in all participants of the total sample, with the Beck Depression Inventory II (BDI-II)(Reference Hautzinger, Keller and Kühner55) data available for 450 participants, Diagnostic System for Mental Disorders in Childhood and Adolescence (DISYPS)(Reference Döpfner, Görtz-Dorten and Lehmkuhl56) based on self-reports (DISYPS Self) data available for 441 participants, as well as data for DISYPS based on external assessment (DISYPS Proxy) available for 422 participants. In further analyses, we excluded those participants with missing information on TNF-α, IFN-γ, IL-1β, IL-6, IL-8, and IL-10 which resulted in three subsamples of 166 participants regarding BDI-II, 162 participants regarding DISYPS Self and 150 participants regarding DISYPS Proxy.
Flow chart of participant selection and exclusion in a cohort of patients from a child and adolescents psychiatry (n 512). 25(OH)D, 25-hydroxy-cholecalciferol; BDI-II, Beck Depression Inventory II; CRP, C-reactive protein; DISYPS Proxy, Diagnostic System for Mental Disorders in Childhood and Adolescence based on external assessment; DISYPS Self, DISYPS based on self-assessment; IFN-γ, interferon-gamma; IL, interleukin; TNF-α, tumor necrosis factor-alpha.
Depression severity assessments
At admission, the BDI-II(Reference Hautzinger, Keller and Kühner55) and the DISYPS(Reference Döpfner, Görtz-Dorten and Lehmkuhl56) were used to assess the severity of depressive symptoms. The BDI-II is a self-reporting tool that evaluates twenty-one aspects of symptoms for major depressive disorder based on the Diagnostic and Statistical Manual of Mental Disorders IV (DSM-IV) criteria(Reference Hautzinger, Keller and Kühner55). It measures symptom severity on a four-point Likert scale (0–3 points) based on how well specific statements describe the participant’s symptoms over the past 2 weeks. Higher scores reflect greater severity, with a total score of ≥ 14 indicating at least mild depression(Reference Kumar, Steer and Teitelman57). The DISYPS is based on the diagnostic criteria according to DSM-IV and ICD-10(Reference Döpfner, Görtz-Dorten and Lehmkuhl56). It was conducted both as a self-assessment (DISYPS Self) and as an external assessment by guardians (DISYPS Proxy). Both questionnaires have twenty-nine items, each with a score of 0–3 points, with higher scores reflecting greater severity of symptoms. The total score was converted into stanine values. A stanine score ≥ 8 was considered clinically relevant(Reference Döpfner, Görtz-Dorten and Lehmkuhl56).
Laboratory measures
Blood samples were collected in the early morning after an overnight fast and at rest (no experimental stimulus), from an antecubital vein into monovettes (Sarstedt, Nümbrecht, Germany). The samples were transferred to the Central Laboratory of the University Hospital Essen for analysis within 1 h of collection. Serum 25(OH)D concentrations (ng/ml) were measured using the Siemens ADVIA Centaur® Immunoassay System (Siemens Healthineers, Erlangen, Germany) and CRP concentrations (mg/l) in EDTA plasma using the Siemens Atellica® CH Analyzer (Siemens Healthineers, Erlangen, Germany). The assays’ analytical sensitivities were 4·2 ng/ml for 25(OH)D and 0·04 mg/l for CRP. Serum and plasma aliquots were stored at –80°C. As part of a sub-project of the ‘Nutrition and Mental Health Study’, the aliquots of a subsample of 177 participants (Figure 1) were used to analyse a representative panel of serum pro- and anti-inflammatory cytokines using a multiplex assay (V-plex® human proinflammatory panel 1 kit) on a MESO QuickPlex® SQ 120MM instrument (Meso Scale Diagnostics, Rockville, USA). The sensitivity of the assay was 0·04 pg/ml for TNF-α, 0·37 pg/ml for IFN-γ, 0·05 pg/ml for IL-1β, 0·06 pg/ml for IL-6, 0·07 pg/ml for IL-8 and 0·04 pg/ml for IL-10. Mean intra- and interassay coefficients of variation (CV) were < 5 % and < 10 %, respectively.
Covariate assessments
At admission, covariates known to affect depressive symptoms were assessed, including smoking status, the use of antidepressants and hormonal contraception(Reference Dolle and Schulte-Körne58). Socio-economic status was calculated based on occupation, income and parents’ level of education and classified as low, middle, or high(Reference Lampert, Müters and Stolzenberg59). The physical activity level was assessed using the International Physical Activity Questionnaire (IPAQ). The results are categorised as low, moderate or high according to the evaluation guidelines(Reference Craig, Marshall and Sjöström60). Information on vitamin D supplement usage was derived from medication lists as well as from an food frequency questionnaire, which included questions on regular supplement usage(Reference Stiegler, Sausenthaler and Buyken61). Furthermore, calendar season was derived from the date of blood sampling.
Anthropometric measures
At admission, participants underwent a physical examination. Body weight was determined to the nearest 0·1 kg in underwear using an electronic scale. Height was determined to 0·1 cm using a stadiometer attached to the wall, barefoot and in an upright position. The BMI was calculated as the quotient of weight in kilograms and the height in metres squared (kg/m2). For participants aged 18·25 years and over (n 8), BMI was categorised as underweight (< 18·5), normal range (18·5–24·9), overweight (25·0–29·9) or obese (> 30·0) according to WHO(62). For participants under 18·25 years, the BMI standard deviation scores (SDS, kg/m2) were calculated using German national reference data according to the Lambda-Mu-Sigma (LMS) method(Reference Kromeyer-Hauschild, Wabitsch and Kunze63). The BMI-SDS was categorised as underweight (< −1·282 sd), normal range (−1·282 to < 1·282 sd), overweight (≥ 1·282 to < 1·881 sd), or obese (≥ 1·881 sd) according to KiGGS(64). The above-described categories for all participants over and under 18·25 years were summarised in the variable ‘weight class’.
Pubertal stage was assessed during the physical examination using Tanner stages based on pubic hair development. Tanner stage information was missing, or assessment was refused in forty-six participants. For female participants older than 13·4 years and male participants older than 14·1 years, Tanner stage 5 was assumed (n 33), based on population-based data on sexual maturation in German children and adolescents on sexual maturation(Reference Kahl, Schaffrath Rosario and Schlaud65). Participants below these age thresholds with missing Tanner stage information were excluded from sensitivity analyses (n 13).
Statistical analysis
Data are presented as frequencies and percentages for categorical variables and median with the percentiles 25th and 75th for continuous variables, which were all not normally distributed (Table 1). Group differences between ‘depressive’ (BDI-II ≥ 14) and ‘non-depressive’ participants (BDI-II < 14) were analysed using χ 2 tests for categorical variables, and Mann–Whitney U tests were applied for continuous variables. Spearman’s rank correlation analysis was conducted to calculate the correlation coefficients between 25(OH)D, inflammatory markers, as well as depression severity according to BDI-II, DISYPS Self and DISYPS Proxy (Table 1). To analyse the relationship between 25(OH)D and depression outcomes, regression analyses were performed for BDI-II, DISYPS Self and DISYPS Proxy. For BDI-II score as outcome variable, a linear regression analysis was performed (Table 2). To fulfil the assumptions of the linear regression analysis, a cubic transformation was applied to 25(OH)D as exposition variable. For the outcome variables DISYPS Self and DISYPS Proxy, binomial logistic regressions were conducted with the untransformed 25(OH)D variable as exposition variable (Tables 3 and 4). In this analysis, participants with a stanine value of ≥ 8 for DISYPS Self and DISYPS Proxy were classified as ‘depressed’.
Characteristics of patients from a child and adolescent psychiatry (n 450) at admission, stratified by depression status based on BDI-II
25(OH)D, 25-hydroxy-cholecalciferol; BDI-II, Beck Depression Inventory II; CRP, C-reactive protein; IFN-γ, interferon-gamma; IL, interleukin; SES, socio-economic status; TNF-α, tumor necrosis factor-alpha.
Provided are the number of participants, median, P25 and P75 (in round brackets) for interval scaled variables, and for categorical variables, the number of participants and percentages. Bold indicates P < 0·05.
* BDI-II score ≥ 14 was classified ‘depressed’, and BDI-II score < 14 was classified ‘not depressed’.
† For comparison of continuous variables, Mann–Whitney U test, and for categorial variables, χ 2 was performed.
‡ Categories for 25(OH)D status according to the Institute of Medicine (‘at risk for deficiency’ (serum 25(OH)D levels < 12 ng/ml (30 nmol/l)); ‘potentially at risk for inadequacy’ (serum 25(OH)D levels 12–< 20 ng/ml (30–< 50 nmol/l)); ‘adequate’ (serum 25(OH)D levels 20–< 30 ng/ml (50–< 75 nmol/l)).
§ Recommendation according to the Endocrine Society (serum 25(OH)D levels ≥ 30 ng/ml (≥ 75 nmol/l)).
|| Measurements only in subsample (n 166).
¶ Socio-economic status was calculated based on occupation, income and parents’ level of education and classified as low, middle or high(Reference Lampert, Müters and Stolzenberg59).
Linear regression models predicting BDI-II scores based on 25(OH)D levels with inclusion of inflammatory markers in full and subsample analyses (full BDI-II sample: n 450; BDI-II subsample*: n 166)
25(OH)D, 25-hydroxy-cholecalciferol; BDI-II, Beck Depression Inventory II; CRP, C-reactive protein; IL, Interleukin.
Bold indicates P < 0·05. Full BDI-II sample – crude model: included only 25(OH)D as exposition variable. Full BDI-II sample – model 2: included 25(OH)D and CRP as exposition variables. Full BDI-II sample – model 3: model 2 further adjusted for sex, smoking, antidepressant use and age. BDI-II subsample – crude model: included only 25(OH)D as exposition variable. BDI-II subsample – model 2a: 25(OH)D and CRP as exposition variables. BDI-II subsample – model 2b: model 2a plus additional inflammatory markers as exposition variables. BDI-II subsample – model 3: model 2b further adjusted for sex, smoking, antidepressant use and age.
* Subsample includes only those with measurements of additional pro- and anti-inflammatory markers.
† For 25(OH)D, post-cubic transformation values are presented.
Binomial regression models predicting depression according to DISYPS Self based on 25(OH)D levels with inclusion of inflammatory markers (full DISYPS Self sample: n 441; DISYPS Self subsample*: n 162)
25(OH)D, 25-hydroxy-cholecalciferol; CRP, C-reactive protein; DISYPS Self, Diagnostic System for Mental Disorders in Childhood and Adolescence based on self-assessment; IL, interleukin; OR, Odds Ratio.
For R2, max-rescaled R2 was used. Bold indicates P < 0·05. Full DISYPS Self sample – crude model: included only 25(OH)D as exposition variable. Full DISYPS Self sample – model 2: included 25(OH)D and CRP as exposition variables. Full DISYPS Self sample – model 3: model 2 further adjusted for sex, antidepressant use and age. DISYPS Self subsample – crude model: included only 25(OH)D as exposition variable. DISYPS Self subsample – model 2a: 25(OH)D and CRP as exposition variables. DISYPS Self subsample – model 2b: model 2a plus additional inflammatory markers as exposition variables. DISYPS Self subsample – model 3: model 2b further adjusted for sex, antidepressant use and age.
* Subsample includes only those with measurements of additional pro- and anti-inflammatory markers.
Binomial regression models predicting depression according to DISYPS Proxy based on 25(OH)D levels with inclusion of inflammatory markers (full DISYPS Proxy sample: n 422; DISYPS Proxy subsample*: n 150)
25(OH)D, 25-hydroxy-cholecalciferol; CRP, C-reactive protein; DISYPS Proxy, Diagnostic System for Mental Disorders in Childhood and Adolescence based on external assessment; IL, interleukin; OR, Odds Ratio.
For R2, max-rescaled R2 was used. Bold indicates P < 0·05. Full DISYPS Proxy sample – crude model: included only 25(OH)D as exposition variable. Full DISYPS Proxy sample – model 2: included 25(OH)D and CRP as exposition variables. Full DISYPS Proxy sample – model 3: model 2 further adjusted for antidepressant use and age. DISYPS Proxy subsample – crude model: included only 25(OH)D as exposition variable. DISYPS Proxy subsample – model 2a: 25(OH)D and CRP as exposition variables. DISYPS Proxy subsample – model 2b: model 2a plus additional inflammatory markers as exposition variables. DISYPS Proxy subsample – model 3: model 2b further adjusted for antidepressant use and age.
* Subsample includes only those with measurements of additional pro- and anti-inflammatory markers.
For each outcome, analyses were separately conducted using either the full sample or only the subsample data with additional inflammatory markers (Tables 2, 3 and 4). For both samples, models were built sequentially, starting with crude models with 25(OH)D as the sole predictor, followed by models that additionally included CRP (models 2 full sample and 2a in the subsample). In the subsample, additional inflammatory markers that significantly correlated with 25(OH)D were included in the next step (models 2b). Finally, models were adjusted for relevant covariates (models 3). To this end, potential covariates (e.g. sex assigned at birth, age, antidepressant use and smoking status) were identified from the literature(Reference Dolle and Schulte-Körne58). Each potential covariate was tested individually as the sole independent variable and afterwards with 25(OH)D as an additional independent variable in the regression model. Only variables that were significantly associated with the respective outcome in both steps were included in the final regression models (Table 2, 3 and 4).
Finally, each independent variable, i.e. inflammatory parameters and relevant covariates (e.g. sex), was tested for interactions with 25(OH)D. For this purpose, the respective interaction term as well as 25(OH)D and the respective covariate were included in the models for the different outcomes. Since none of these tested interactions were robust, no stratified analyses were conducted.
As the cubic transformation of 25(OH)D as an explanatory variable hampered the interpretation of effects on BDI-II, we decided to additionally visualise this association. To this end, adjusted Least Squares Means (lsmeans) were calculated using the ‘Full BDI-II sample’, with the untransformed 25(OH)D variable converted into a class variable. The categories were derived from literature-based thresholds for 25(OH)D status and defined as follows: category 1, ‘at risk for deficiency’ according to the IOM (serum 25(OH)D < 12 ng/ml (< 30 nmol/l)); category 2, ‘potentially at risk for inadequacy’ according to the IOM (serum 25(OH)D 12–< 20 ng/ml (30–< 50 nmol/l)); category 3, ‘adequate’ according to the IOM (serum 25(OH)D 20–< 30 ng/ml (50–< 75 nmol/l))(Reference Ross, Taylor and Yaktine13); category 4, serum 25(OH)D ≥ 30 ng/ml (≥ 75 nmol/l) as recommended by the Endocrine Society (serum 25(OH)D ≥ 30 ng/ml (≥ 75 nmol/l)))(Reference Holick, Binkley and Bischoff-Ferrari16). The model was adjusted for sex, smoking status, antidepressant use and age (Figure 2).
Relationship between 25(OH)D status thresholds and BDI-II scorea in a cohort of patients from a child and adolescents psychiatry (n 450). a Adjusted for sex, smoking, antidepressant use and age. 1, ‘at risk for deficiency’ according to the Institute of Medicine (serum 25(OH)D levels < 12 ng/ml (< 30 nmol/l))(Reference Ross, Taylor and Yaktine13), n 195. 2, ‘potentially at risk for inadequacy’ according to the Institute of Medicine (serum 25(OH)D levels 12–< 20 ng/ml (30–< 50 nmol/l))(Reference Ross, Taylor and Yaktine13), n 154. 3, ‘adequate’ according to the Institute of Medicine (serum 25(OH)D levels 20–< 30 ng/ml (50–< 75 nmol/l))(Reference Ross, Taylor and Yaktine13), n 80. 4, Recommendation according to the Endocrine Society (serum 25(OH)D levels ≥ 30 ng/ml (≥ 75 nmol/l))(Reference Holick, Binkley and Bischoff-Ferrari16), n 21. 25(OH)D, 25-hydroxy-cholecalciferol; BDI-II, Beck Depression Inventory II.
To assess the robustness of results, the above-mentioned analyses were conducted using five different subsets of participants in separate sensitivity analyses: (1) only participants with full information on all depression questionnaires, i.e. BDI-II, DISYPS Self, plus DISYPS Proxy (n 417); (2) only participants classified as ‘depressed’ according to all of these questionnaires (BDI-II ≥ 14), DISYPS Self and DISYPS Proxy (DISYPS ≥ 8) (n 272); (3) equal to subset 2, but with a lower cut-off score for DISYPS (≥ 7) (n 308), considering that stanine scores ≥ 7 are considered ‘above average’(Reference Schmidt-Atzert, Krumm and Amelang66); (4) only ‘depressed’ participants according to BDI-II (BDI-II ≥ 14) as well as a diagnosis of depression verified by clinical assessment and the Schedule for Affective Disorders and Schizophrenia for School-Aged Children – Present and Lifetime Version (K-SADS-PL)(Reference Kaufman, Birmaher and Brent67) (n 217); and (5) only participants without diagnoses for substance use or eating disorder (n 336) due to potential effects on vitamin D metabolism(Reference Tardelli, Lago and Da Silveira68–Reference Modan-Moses, Levy-Shraga and Pinhas-Hamiel71). In addition, sensitivity analyses were performed by including calendar season, pubertal stage (Tanner status) and vitamin D mono-supplementation as additional covariates.
Additionally, we evaluated whether inflammatory markers mediate the association between serum 25(OH)D and depressive symptoms (Table 5). Mediation was examined in fully adjusted models: in the full sample with CRP as mediator (full sample – model 3) and in the subsample with CRP, IL-1β and IL-6 as parallel mediators (subsample – model 3) (Table 5). For BDI-II (continuous outcome), we fit a structural equation model (SEM) following Preacher and Hayes to estimate total, direct and indirect effects of serum 25(OH)D(Reference Preacher and Hayes72). For the DISYPS Self and DISYPS Proxy (binary outcomes), we used conditional process analysis with the SAS PROCESS macro (model 4; parallel mediators) and non-parametric percentile bootstrapping (5000 resamples) to obtain 95 % CI for indirect effects(Reference Hayes73).
Direct, indirect and total effects of serum 25(OH)D on depressive symptoms with inflammatory markers as a potential mediator in full and subsample* analyses
25(OH)D, 25-hydroxy-cholecalciferol; BDI-II, Beck Depression Inventory II; CRP, C-reactive protein; DISYPS Proxy, Diagnostic System for Mental Disorders in Childhood and Adolescence based on external assessment; DISYPS Self, DISYPS based on self-assessment; IL, interleukin.
Bold indicates P < 0·05. Full BDI-II sample – model 3: Effects of 25(OH)D on BDI-II; CRP as mediator; sex, smoking, antidepressant use and age as covariates. BDI-II subsample – model 3: Effects of 25(OH)D on BDI-II; CRP, IL-1β and IL-6 as mediators; sex, smoking, antidepressant use and age as covariates. Full DISYPS Self sample – model 3: Effects of 25(OH)D on DISYPS Self; CRP as mediator; sex, antidepressant use and age as covariates. DISYPS Self subsample – model 3: Effects of 25(OH)D on DISYPS Self; CRP, IL-1β and IL-6 as mediators; sex, smoking, antidepressant use and age as covariates. Full DISYPS Proxy sample – model 3: Effects of 25(OH)D on DISYPS Proxy; CRP as mediator; antidepressant use and age as covariates. DISYPS Proxy subsample – model 3: effects of 25(OH)D on DISYPS Proxy; CRP, IL-1β and IL-6 as mediators; antidepressant use and age as covariates.
* Subsample includes only those with measurements of additional pro- and anti-inflammatory markers.
† Structural equation model following Preacher and Hayes to estimate effects. 95 % CI for total, direct and indirect effects were computed post hoc as Wald intervals (β ± 1·96 × se) from PROC CALIS output.
‡ Conditional process analysis with the SAS PROCESS macro to estimate effects; and non-parametric percentile bootstrapping to obtain 95 % CI for indirect effects.
Data processing and analyses were conducted using SAS 9.4 (SAS Institute Inc., Cary, USA). Two-sided significances were calculated, and the α level was set to 0·05.
Results
Sample characteristics
A total of 465 participants (64·7 % ♀) aged 11·3–18·9 years were included for analysis (‘total sample’). Of these, 14·4 % (n 67) were taking antidepressants (data not shown). In the total sample, 34·6 % (n 161) had 25(OH)D levels in the ‘potentially at risk for inadequacy’ range (12 to < 20 ng/ml), and additional 43·2 % (n 201) were categorised as ‘at risk for deficiency‘ (< 12 ng/ml). Based on the DISYPS Self, 73·5 % (n 324 from 441) and, according to the DISYPS Proxy, 83·2 % (n 351 from 422) experienced at least mild depression (data not shown). The BDI-II indicated a prevalence of 77·8 % (n 350 from 450) for at least mild depression (Table 1). While 25(OH)D levels as a continuous variable did not differ between depressive and non-depressive participants according to the BDI-II (n 450), the division into vitamin D categories revealed significant differences, with a lower proportion of depressive participants meeting the Endocrine Society recommendations compared with non-depressive participants. No differences in inflammatory markers were observed between depressive and non-depressive participants. Depressive participants, however, were older and included a higher proportion of females compared with non-depressive participants (Table 1).
Spearman’s correlations revealed moderate correlations between DISYPS Proxy and the self-assessment questionnaires (BDI-II: r = 0·40, P < 0·0001, n 420; DISYPS Self: r = 0·43, P 0·0001, n 417), while the self-assessment questionnaires correlated strongly with each other (r = 0·75; P < 0·0001, n 441) (Appendix Table 1). A significant, albeit weak correlation between 25(OH)D levels and depression was found only for DISYPS Proxy (r = –0·11; P = 0·02; n 422). Additionally, the 25(OH)D levels had a weak negative correlation with CRP (r = –0·21, P < 0·0001, n 465), IL-1β (r = –0·20, P = 0·01, n 177) and IL-6 (r = –0·19, P = 0·01, n 177). None of the measured inflammatory markers correlated with depression according to all three depression questionnaires.
Regression analyses
Linear regression in the full BDI-II sample (crude model) revealed a negative association between the cubic-transformed 25(OH)D variable and BDI-II scores (β = –0·000119; P = 0·006; R 2 = 0·017) (full BDI-II sample – models, Table 2). Regarding the role of inflammation, the association between 25(OH)D and BDI-II score remained unchanged after adjusting for CRP (model 2) as well as covariates (model 3). Equally, in the subsample models, which included only participants with additional pro- and anti-inflammatory marker data (n 166), the β values for 25(OH)D remained similar to the crude model, regardless of the number of inflammatory markers included (BDI-II subsample models, Table 2). However, in the subsample, the association between 25(OH)D levels and BDI-II scores did not reach significance. Also, none of the associations between the inflammatory markers and BDI-II scores were significant.
The association between 25(OH)D levels and BDI-II scores is visualised in Figure 2, to illustrate effect estimates. BDI-II scores slightly increased in the first three vitamin D categories, i.e. up to the IOM threshold for ‘adequate’ 25(OH)D levels. 25(OH)D levels in the fourth category, i.e. within the range categorised as ‘recommended’ by the Endocrine Society, were accompanied by a decrease of 8–9 BDI-II points compared with the other categories (Figure 2).
Binomial logistic regression based on DISYPS Self showed no significant association between 25(OH)D levels and depression in the crude models, which remained robust after adjustment for inflammatory markers (models 2, 2a and 2b) and covariates (model 3, Table 3). Among all inflammatory parameters, only IL-6 reached statistical significance (OR = 2·274 (1·024; 5·048)), and only in the adjusted subsample model for DISYPS Self (DISYPS Self subsample – models, Table 3). In contrast, a significant inverse association between 25(OH)D levels and depression was observed in the crude models for DISYPS Proxy (full sample: OR = 0·950 (0·922; 0·979); R 2 = 0·043; subsample: OR = 0·931 (0·880; 0·986); R 2 = 0·067) (Table 4). This association remained robust, and the OR did not change after adjustment for inflammatory markers (models 2, 2a and 2b, Table 4) and after the inclusion of covariates (model 3, Table 4). None of the associations between inflammatory markers and DISYPS Proxy were significant (Table 4).
Mediation analyses
We examined whether inflammatory markers mediate the association between serum 25(OH)D and depressive symptoms (Table 5). In the SEM for BDI-II (full BDI-II sample – model 3, Table 5), the total (β = −0·088 (−0·168, −0·009)) and direct effects (β = −0·090 (−0·170, −0·011)) were statistically significant, whereas the indirect effect via inflammation was not. In the subsample (BDI-II subsample – model 3, Table 5), neither total, direct, nor indirect effects reached significance.
For DISYPS Self, conditional process analyses showed no significant total or direct effects in either the full sample or the subsample (Table 5). For DISYPS Proxy, both the total and direct effects were inverse and significant in the full sample (total β = –0·048 (–0·079, –0·016); direct β = –0·047 (–0·079, –0·016)) and the subsample (total β = –0·073 (–0·133, –0·013); direct β = –0·080 (–0·141, –0·019)). For both DISYPS outcomes, indirect effects via CRP (full sample) and via CRP, IL-1β and IL-6 (subsample) were not significant (Table 5).
Discussion
To the best of our knowledge, this is the first study to investigate the association between depression, vitamin D status and inflammation in children and adolescents. Our findings indicate an inverse but – as reflected by the β, OR and R2 values – statistically modest association between vitamin D status and depression. Additionally, weak negative correlations were found between vitamin D and certain inflammatory markers. However, the measured inflammatory parameters appear to play no significant role in depression during early life stages, as we found no associations between inflammatory markers and depression. Since the inclusion of these inflammatory markers did not alter the relationship between vitamin D and depression, and formal mediation analyses likewise showed no significant indirect effects via inflammation, our study does not support the hypothesis that potential antidepressant effects of vitamin D are mediated by anti-inflammatory mechanisms in this age group.
Vitamin D and depression
Although not confirmed for DISYPS Self, our findings overall suggest an inverse association between vitamin D levels and depression, which aligns with existing research on both adult(Reference Musazadeh, Keramati and Ghalichi74–Reference Ju, Lee and Jeong77) and – to a lesser extent – younger populations. While research on children and adolescents is limited, some studies indicate an inverse relationship. Lower levels of vitamin D have been associated with an increased risk of depression(Reference Tarikere Satyanarayana, Suryanarayana and Theophilus Yesupatham78) as well as more severe symptoms of depression in adolescents(Reference Huang, Gong and Bao79). However, a cross-sectional study in this age group found no clear association between vitamin D levels and depressive symptoms(Reference Al-Sabah, Al-Taiar and Shaban32).
While our overall findings support the notion that higher vitamin D levels are linked to fewer symptoms and a lower risk of depression, this association was not consistent across all outcomes. For example, the association was not confirmed by DISYPS Self, and the association observed for BDI-II was not robust, losing significance in the subsample analysis. In contrast, associations were stronger when depression outcomes were based on parent-reported data. This discrepancy in reporters’ evaluation seems to be of relevance, especially in youth mental health assessments(Reference Los Reyes, Augenstein and Wang80,Reference Martel, Markon and Smith81) . A meta-analysis by De Los Reyes et al. (2015) demonstrated that different informants capture complementary aspects of youth psychopathology, with parent reports tending to be more sensitive to observable behavioural impairments, whereas adolescent self-reports may more accurately reflect internalising distress(Reference Los Reyes, Augenstein and Wang80). This may partly explain why associations were more pronounced for parent-reported outcomes in the present study, although further investigation is warranted. Additionally, in clinically referred children, parents often emphasise symptom severity, whereas adolescents tend to under-report their symptoms and, thus, do not match with parents evaluation anymore(Reference Salbach-Andrae, Klinkowski and Lenz82). This can lead to misclassification, particularly when categorical outcomes (e.g. depressive vs. non-depressive) are used. Consistently, in our sample, depressive scores were higher in DISYPS Proxy compared with DISYPS Self, making the latter prone to misclassification. Notably, studies examining the relationship between vitamin D and depression in children and adolescents have shown more consistent results when relying on external evaluations(Reference Libuda, Timmesfeld and Antel31,Reference Tolppanen, Sayers and Fraser83,Reference Mohamed, Khalil and El Melegy84) , while self-reports often reveal a less clear connection(Reference Al-Sabah, Al-Taiar and Shaban32,Reference Tarikere Satyanarayana, Suryanarayana and Theophilus Yesupatham78,Reference Huang, Gong and Bao79,Reference Bahrami, Mazloum and Maghsoudi85,Reference Högberg, Gustafsson and Hällström86) . Similarly, the first randomised controlled trial investigating vitamin D supplementation in depressive adolescents by Libuda et al. (2020) found that vitamin D supplementation reduced parent-reported, but not self-reported symptoms(Reference Libuda, Timmesfeld and Antel31).
Interestingly, an examination of the Avon Longitudinal Study for Parents and Children (ALSPAC) cohort revealed that higher concentrations of 25(OH)D3 at an average age of 9·8 years were associated with a lower risk of depressive symptoms at age 13·8 years, but not at age 10·6 years. The authors suggested that longer intervals between the measurement of vitamin D3 and assessment of depressive symptoms lead to clearer results, as the biological pathways linking vitamin D to depression could involve processes requiring more time to manifest(Reference Tolppanen, Sayers and Fraser83). This raises the possibility that the associations we observed between vitamin D and depression might have been more pronounced in a longitudinal design compared with our cross-sectional study. However, the ALSPAC cohort represents a healthier reference population with less variability in depressive symptoms and fewer cases of manifest depression than our sample(Reference Tolppanen, Sayers and Fraser83), potentially limiting the generalisation of their findings to populations with higher rates of depressive disorders, such as ours.
Meta-analyses examining the impact of vitamin D supplementation on depression in adults report a relatively wide range of effect sizes between 0·28 (–0·14, 0·69)(Reference Gowda, Mutowo and Smith87) and 0·58 (0·45, 0·72)(Reference Vellekkatt and Menon29). Notably, these effect sizes were comparable or even higher than those observed in a meta-analysis of SSRI (0·27)(Reference Barth, Kriston and Klostermann10). In our study, regression analyses indicated a small overall association between vitamin D status and depression severity or likelihood, as reflected by β, OR and R2 values. However, the BDI-II model revealed a cubic, non-linear relationship, with an inverse effect emerging at higher vitamin D levels. Notably, Figure 2 highlights a difference of more than eight BDI-II points between vitamin D status categories 1–3 and participants meeting the Endocrine Society’s recommendations (category 4)(Reference Holick, Binkley and Bischoff-Ferrari16). This difference may suggest a clinically relevant effect(Reference Senft, Fischer-Hansal and Schosser88), although cautious interpretation is warranted given the small size of the highest vitamin D subgroup. Future studies on children and adolescents should ensure greater representation of this group.
In this context, the debate about optimal vitamin D levels needs to be considered, particularly regarding mental health outcomes. Recommendations from the IOM classify serum 25(OH)D levels < 12 ng/ml (< 30 nmol/l) as ‘at risk for deficiency’ and levels between 12 and < 20 ng/ml (30 and < 50 nmol/l) as ‘potentially at risk for inadequacy’(Reference Ross, Taylor and Yaktine13). When applying these thresholds to our cohort, where most participants experienced at least mild depression, the prevalence of vitamin D levels ‘at risk for deficiency’ was 30·7 % higher, and the percentage with ‘potentially at risk for inadequacy’ levels was 1·5 % higher compared with the general population for the same age group(Reference Rabenberg, Scheidt-Nave and Busch14). While the IOM considered 25(OH)D concentrations ≥ 20 ng/ml (≥ 50 nmol/l) as ‘sufficient’(Reference Ross, Taylor and Yaktine13), the Endocrine Society recommends levels above 30 ng/ml (75 nmol/l)(Reference Holick, Binkley and Bischoff-Ferrari16). Our results might be an indication that this cut-off may be more relevant for depression prevention, as lower BDI-II scores were observed among participants meeting the Endocrine Society’s recommendations(Reference Holick, Binkley and Bischoff-Ferrari16). As one of the few studies dealing with this topic, Tolppanen et al. used 30 ng/ml (75 nmol/l) as threshold for vitamin D insufficiency and 20 ng/ml (50 nmol/l) as threshold for vitamin D deficiency at the age of 9·8 years. Since both were associated with a 20–30 % increased risk of depressive symptoms at age 13·8 years(Reference Tolppanen, Sayers and Fraser83), relatively high vitamin D levels might be necessary for the prevention of depression. Additionally, Kaviani et al. (2022) found that even at baseline vitamin D levels considered sufficient (≥ 30 ng/ml (≥ 75 nmol/l)), vitamin D supplementation reduced depressive symptoms, suggesting higher levels may benefit those with depression(Reference Kaviani, Nikooyeh and Etesam46). Similarly, our findings showed an inverse association with BDI-II scores only at levels above 30 ng/ml, consistent with the Endocrine Society’s recommendations(Reference Holick, Binkley and Bischoff-Ferrari16).
Role of inflammation
The role of inflammation as a potential pathway linking vitamin D and depression is currently discussed(Reference Vellekkatt and Menon29). To the best of our knowledge, to date, no studies have examined the interplay between vitamin D, inflammation and depression in children and adolescents. However, research on vitamin D and inflammation points to inverse associations between vitamin D status and inflammatory markers, particularly CRP and IL-6(Reference Filgueiras, Rocha and Novaes89). Our data also indicated an inverse, albeit weak statistical association between vitamin D levels and inflammation, reflected in small inverse correlations with CRP, IL-1β and IL-6.
Regarding the role of inflammation in depression, the levels of inflammatory markers did not differ between participants with depression and those without depression in our sample. Additionally, none of the three outcomes showed any correlation with the inflammatory markers we measured. In the regression analyses, only IL-6 showed a significant association and only with one of the three outcomes (DISYPS Self). Therefore, our findings suggest that inflammation, at least as reflected in circulating cytokines and CRP in this cross-sectional study, does not appear to play a role in depression, at least within our cohort of children and adolescents. The missing significance for single inflammatory parameters might also be attributable to the small sample size, low inflammation levels and the cross-sectional nature of our data. Over time, this relationship may become evident, as studies show depression is linked to future inflammation in childhood and adolescence, and vice versa(Reference Colasanto, Madigan and Korczak42).
Although our findings suggest that vitamin D may be associated with some inflammatory markers, our analyses indicate that inflammation does not statistically mediate the association between vitamin D status and depression. Specifically, the associations between vitamin D and depression largely remained unchanged after accounting for inflammatory markers, and formal mediation analyses indicated no significant indirect effects of inflammation. Additionally, inflammatory markers were not associated with depression, indicating that, at least in this cohort of children and adolescents, inflammation does not explain the link between depression and vitamin D.
Strength and limitations
Our cohort consists solely of participants receiving care at a child and adolescent psychiatric clinic. This may limit the generalisability of the findings, as only a small minority (about 20 %) of children with mental disorders, including depression, seek clinical help(Reference Boyle, Duncan and Georgiades90–Reference Merikangas, He and Burstein92). Moreover, the prevalence of low vitamin D status in our clinically referred sample was higher than in population-based cohorts (e.g. the KiGGS study(Reference Rabenberg, Scheidt-Nave and Busch14)). This overrepresentation may limit the transferability of our findings to the general population and should be considered when interpreting the observed associations. However, all measurements were taken on the day of clinic admission, thereby reducing the risk of bias from the clinical stay itself.
Sex distribution differed between depressed and non-depressed participants, with a higher proportion of females in the depressed group. While this imbalance may limit interpretation, formal interaction analyses provided no evidence that associations between vitamin D, inflammatory markers and depression differed by sex.
Due to the cross-sectional nature of our data, we cannot establish a causal relationship between vitamin D status and depression. While some studies suggest a prospective link between vitamin D and the onset and progression of depression(Reference Kaviani, Nikooyeh and Etesam46,Reference Tolppanen, Sayers and Fraser83,Reference Alavi, Khademalhoseini and Vakili93) , the possibility of reverse causality remains. For example, depressed individuals might avoid outdoor activities, leading to less UV-B exposure, or may experience reduced appetite, resulting in malnutrition and consequently in lower vitamin D levels(Reference Menon, Kar and Suthar34). Even though we examined the potential influence of activity level and weight class in our analyses, we were unable to determine the direction of the association between vitamin D and depression. To further assess robustness, we conducted sensitivity analyses including pubertal stage, calendar season and vitamin D mono-supplementation as additional covariates. These additional analyses did not materially change the results for any outcome (data not shown), supporting the stability of the observed associations. Nevertheless, information on some factors such as pigmentation as well as detailed information on sunlight exposure patterns were not available and could not be considered as covariates in the model. Accordingly, residual confounding cannot be ruled out.
A further limitation is that antidepressant use was included only as a categorical variable in our adjusted regression models, without consideration of duration or dosage. However, with only 14·4 % of the full sample reporting antidepressant use, this limitation likely has minimal impact on the overall findings.
Methodologically, our assessments were comprehensive, despite being conducted at a single time point. Depression was assessed using both self-reports and external reports by legal guardians, following the multi-informant approach recommended in the literature for evaluating youth mental health(Reference Los Reyes, Augenstein and Wang80,Reference Martel, Markon and Smith81) . Further assessments included the K-SADS-PL and clinical psychiatric diagnoses by physicians. We also conducted several sensitivity analyses, which confirmed the robustness of our findings (data not shown). However, the overlap of symptoms between depression and anxiety presents a challenge, as it may lead to a misclassification as ‘depressed’ and therefore potential inaccuracies in association estimates. Although misdiagnosis is unlikely due to the use of multiple diagnostic tools and sensitivity analyses, it cannot be completely dismissed. Nonetheless, this concern may be mitigated by research interest in the potential beneficial effects of vitamin D on both disorders(Reference Kouba, Camargo and Gil-Mohapel94).
Finally, the measurement of eight inflammatory markers was particularly comprehensive, allowing for an in-depth analysis of the role of inflammation in the relationship between vitamin D and depression. While we included key circulating inflammatory markers, others, such as those related to B-cell homeostasis(Reference Ahmetspahic, Schwarte and Ambrée95), components of the NLRP3 inflammasome or markers of central neuroimmune activity including microglial activation, which are also discussed in the context of the aetiology of depression(Reference Kouba, Camargo and Gil-Mohapel94), were not available. This limitation indicates that inflammatory pathways beyond peripheral cytokines and CRP cannot be excluded. Additionally, recent infections, which could influence inflammatory marker levels, were not assessed. To assess whether inflammation mediates the association between vitamin D and depression, we applied complementary methods (covariate-adjusted regression and formal mediation analyses) which yielded convergent evidence. However, temporal precedence cannot be established in cross-sectional data; thus, these mediation findings are associational and should be replicated in longitudinal designs(Reference Maxwell, Cole and Mitchell96). Expanding marker selection and accounting for such factors in future studies may provide further insights.
Conclusion
Our results implicate a minor inverse relationship between vitamin D levels and depression in children and adolescents. Although it was confirmed that vitamin D status is related with certain inflammatory markers, our data did not indicate that the relationship between vitamin D status and depression is explained by inflammatory markers. Future studies should not only examine the mechanisms underlying the inverse association between vitamin D and depression, but also the optimal vitamin D levels for depression-related outcomes in youth since the present study indicated beneficial effects only at levels above 30 ng/ml (75 nmol/l). To this end, longitudinal studies starting in childhood and incorporating a multi-informant approach to assess youth mental health are urgently needed.
Supplementary material
For supplementary material/s referred to in this article, please visit https://doi.org/10.1017/S0007114526106928
Acknowledgements
This research received no specific grant from any funding agency in the public, commercial or not-for-profit sectors.
L. S.: conceptualisation, formal analysis, visualisation, writing – original draft and writing – review and editing. NJ: writing – review and editing and supervision. J. B., H. E., R. H., C. G., A. H., J. A., J. H. and T. P.: critical input for manuscript. M. F.: conceptualisation and critical input for manuscript. L. L.: conceptualisation, writing – review and editing and supervision. All authors have read and agreed to the published version of the manuscript.
The authors declare that they have no conflict of interest.
Open access
