Nonlinear Relationship Between C-Reactive Protein and Depression Among Obese Middle-Aged Adults : Nursing Research

Secondary Logo

Journal Logo


Nonlinear Relationship Between C-Reactive Protein and Depression Among Obese Middle-Aged Adults

Lee, Chiyoung; Min, Se Hee; Niitsu, Kosuke

Author Information
Nursing Research 72(3):p 236-245, 5/6 2023. | DOI: 10.1097/NNR.0000000000000646


C-reactive protein (CRP)—a nonspecific biomarker of inflammation produced mainly by the liver in response to interleukin-6—has been shown to strongly correlate with depression in preclinical and clinical studies (Osimo et al., 2019). Studies have reported an increased level of circulating CRP in patients with self-reported depressive symptomatology or diagnosed clinical depression after controlling for age, race/ethnicity, gender, smoking status, and body mass index (BMI; Horn et al., 2018). Elevated CRP has also been shown to predict future depression risk and refractoriness to antidepressants (Mac Giollabhui et al., 2021; Yang et al., 2019). Crucially, understanding such association can help elucidate a possible “psychoneuroimmune link” between negative affectivity, inflammatory states, and progression of coronary atherosclerosis and other heart diseases (Khandaker et al., 2020).

Despite the universal agreement on the significance of the relationship between CRP level and depression, the results have been discrepant between genders. Vetter et al. (2013) demonstrated a significant relationship between CRP and major depressive symptoms only in men after accounting for confounders, such as age, metabolic factors, and medications that influence inflammation. Elovainio et al. (2009) used the Whitehall II study to observe that higher CRP levels are related to higher scores on the depression questionnaire in men and women; however, this correlation remained in men alone after controlling for several known risk factors. Similar findings were noted in five large population-based studies (Danner et al., 2003; Ernst et al., 2021; Ford & Erlinger, 2004; Lee et al., 2019; Tayefi et al., 2017). By contrast, Ma et al. (2011) investigated approximately 500 healthy individuals from central Massachusetts and observed an independent relationship between CRP levels and depression scores in women only. Other studies did not detect a relationship between CRP concentrations and depression for both genders (Bremmer et al., 2008; Chocano-Bedoya et al., 2014). Indeed, differences in participant characteristics (population-based vs. clinical), the index of depression (symptom report or clinical diagnosis), or the selection of controlled covariates may explain some inconsistent findings. However, it may be worthwhile to think about these inconsistencies from a more rigorous methodological perspective and apply different analytic approaches to understand the associations between CRP and depression. Until now, parametric models such as generalized linear models (GLMs) in almost all previous studies have assumed the linearity of CRP and depression (Horn et al., 2018); whether the assumed linearity is the correct functional form for the analyzed data is yet unknown. A threshold level of CRP associated with changes in depression risks may exist; parametric models imposing an a priori form for the CRP–depression relationship may potentially bias the results. Nonparametric models are more suitable for examining the association, as these models are flexible for allowing the data to reveal the pattern of association (Hastie & Tibshirani, 2017). Generalized additive model (GAM) is a nonparametric extension of GLM where the modeling of the mean functions relaxes the linearity assumption. It allows the data to determine the functional form of the association, which can be very flexible and curvilinear (nonlinear). In the context of our study, GLM may restrict the pattern of association between CRP and depression to be the same over the entire range of CRP, whereas GAMs might better model heterogeneous associations across the range. However, despite the wealth of literature addressing the effects of CRP on depression, the extent of this relationship has not been comprehensively evaluated beyond the linear pattern.

Study Objective

This exploratory study aimed to analyze the shape of the CRP–depression relationship more carefully using GAMs. Given previous findings on a different relationship between CRP and depression in men and women, we primarily explored the effect of gender (in association with increased inflammation) on this relationship. In particular, the study was conducted on a sample of obese individuals. Obese populations are a unique group within which the interactions between CRP and depression can be assessed in high-risk individuals (Delgado et al., 2018; Moazzami et al., 2019); they present with high levels of inflammatory and metabolic disturbances, as well as high levels of depressive symptoms associated with obesity.


Design, Setting, and Sample

We used publicly available data from the National Health and Nutrition Examination Survey (NHANES). NHANES is a series of cross-sectional health examinations performed by the National Center for Health Statistics to monitor health and nutritional status trends among community-dwelling, nationally representative U.S. populations. About 5,000 individuals participate in the NHANES every year, and the data are reported in 2-year cycles. The NHANES database serves as a unique platform to examine the relationship between CRP and depression by gender while controlling for various sociodemographic, lifestyle, or clinical factors. Based on the usage of gender in the codebook of NHANES, the term gender was used in this study; we acknowledge that this variable captures biological differences between men and women (i.e., sex). Further information on NHANES methodology and data collection are presented on their website (

Our study population comprised middle-aged adults between 40 and 70 years old. A total of 4,905 people within this age range (50.8% women) participated in prepandemic NHANES from 2017 to 2020. From this entire group, we selected participants whose BMI was over 30 kg/m2 (n = 2,460). Next, participants with serum CRP levels higher than 10 mg/L were excluded (n = 302), as such elevations may indicate ongoing acute inflammation (Chandrashekara, 2014). In addition, those with missing CRP values and those who had not replied to the depression screening questionnaires were further excluded (n = 577), leading to a final analysis sample of 1,581. We did not include any women documented as currently pregnant.

Ethical Considerations

The publicly available NHANES data do not include personal information. The institutional review board of the U.S. Centers for Disease Control and Prevention (CDC) approved the study protocol, and the participants were required to sign informed consent forms before the NHANES was conducted. The university’s institutional review board, where the corresponding author is affiliated, determined this study as exempt.


C-Reactive Protein

Serum hsCRP concentrations were quantified via a Roche cobas 6000 chemistry analyzer (Roche Diagnostics, Indianapolis, IN) using reagents and calibrators from Roche for the 2017–2020 NHANES cycle. The results were converted to mg/L to conform with the American College of Cardiology/American Heart Association risk guidelines. The lower limit of detection (LLOD) for hsCRP is 0.15 mg/L in the NHANES data set. The hsCRP values below LLOD were replaced by the imputed value based upon an established method (LLOD divided by the square root of 2); all other values had detectable concentrations. Further details on collection, storage, calibration, and quality control procedures for blood samples to determine CRP were published in the NHANES Laboratory Procedures Manual (CDC, 2020, 2021).


Depression was measured using the 9-item Patient Health Questionnaire (PHQ-9), a valid measurement tool for depression and its severity based on the Diagnostic and Statistical Manual of Mental Disorders, Fourth Edition criteria (Zimmerman, 2019). In our analysis, we operationalized depression as a continuous variable that indexed the severity of symptoms. PHQ-9 comprises nine symptom items, where respondents indicate whether each item had disturbed them “not at all,” “several days,” “more than half the days,” or “nearly every day” during the preceding two weeks. Each item was rated with a score of 0, 1, 2, or 3, respectively. The items were summed for the total score, which allowed the classification of symptomatology as follows: none to minimal (0–4), mild (5–9), moderate (10–14), moderately severe (15–19), and severe (20–27). In the present study, the scale’s reliability measured by Cronbach’s alpha was 0.79.


Potentially relevant sociodemographic, lifestyle, and clinical contributors in the association between inflammation and depression were chosen as covariates based on previous immunopsychiatry research (Horn et al., 2018; O’Connor, 2009; Osimo et al., 2019). Sociodemographic covariates included race/ethnicity, marital status, education level, and the ratio of family income to poverty. Lifestyle covariates included recent smoking, moderate work activity, vigorous work activity, walking or cycling, moderate recreational activity, and vigorous recreational activity.

Clinical covariates included white blood cell count, triglycerides, direct high-density lipoprotein cholesterol, and hemoglobin. Other clinical covariates included the presence of arthritis, liver disease, lung disease, asthma, malignancy of any kind, diabetes mellitus, hypertension, coronary heart disease, angina pectoris, and stroke. Participants were considered prevalent lung disease cases if a physician ever told them they had any of the following conditions: chronic obstructive pulmonary disease, chronic bronchitis, or emphysema.

Statistical Analysis

The nonlinear relationship between CRP and depression severity in men and women was estimated using the GAMs. All GAMs were performed using the R-package mgcv v.1.8-31 (Wood, 2022). GAM has two parts to the estimation method: parametric estimation, which is used for those predictors that enter the model parametrically, and nonparametric estimation for smoothing predictors. In this article, only CRP is the smoothing predictor, and the rest are estimated parametrically. GAM can accommodate the interaction between two or more predictors in a way that is conceptually comparable to interactions in GLMs. We used this interaction to map the depression outcome by gender. This model is written as follows:


where Yi is the response of participant i (depression severity), g is the link function, β0 is the intercept, fCRP denotes the nonparametric smooth function of the covariate CRP (indicating a nonlinear association), βgenderxgender represents the main effect of the predictor variable of gender, βxiT indicates the linear predictor of the other categorical covariates, and εi denotes the error term in the model. The f(x) functions can be constructed with various spline smoothing functions, and the mgcv package for GAM utilizes regression splines as smoothers. Models were either unadjusted or were adjusted for multiple covariates known to influence CRP levels and depression in the following way: Model 1, no covariates; Model 2, adjusted for sociodemographic data; Model 3, Model 2 + lifestyle and clinical covariates.

Apart from the equation, there are two other important arguments in GAM, including “family” and “method.” Family determines the distribution and the link function for fitting. GAM allows adoption of a broad range of distributions for the response variable and link functions for assessing the effects of the predictors on the dependent regressors. We used a default Gaussian (i.e., normal) distribution with an identity link. An identity link function means the relationships are predicted directly, without transformation (Garson, 2012). Method is employed to specify the procedure for estimating the smoothing parameter. The choice of the smoothing parameter or the extent of smoothing applied to the data can strongly affect model fit. The smoothness of f(x) is computed to achieve an optimal balance between the fit to the data versus a penalty for excessive wiggliness of the functions. In the present study, we estimated the smoothing parameter based on a restricted maximum likelihood approach. The use of the restricted maximum likelihood approach is preferred for finite sample sizes to avoid overfitting (Stevenson & Woods, 2006). The basis dimensions used to represent the smoothing term (k) were set to 6 for interaction, after verification using the “gam.check” function implemented in the mgcv that this basis function was sufficient to capture the wiggliness of the data (p > .05; see Supplemental Digital Content, This optimal number of k was chosen based on Akaike information criterion values via iterative processes. We compared k = 4, 5, 6, 7, and 8, as most of the research demonstrates that k = 3 is sufficient, and there is often no remarkable improvement with larger than 8 knots. We checked the model assumptions with basic residual diagnostic plots, as well as ensured model convergence was reached and that the smoothing parameter was not overfit based on the k-index test (Wood, 2022). The predicted smooth functions and confidence intervals were plotted for both unadjusted and adjusted GAMs.

In the GAM, the reported values of the effective degree of freedom (EDF) output reflect the degree of curvature of the smooth, with a value equal to 1 signifying a linear association. A value of EDF > 1 represents a more complex relationship between CRP and depression severity. The threshold effect was investigated using a two-piecewise linear regression model in R (segmented package; Muggeo & Muggeo, 2017) in cases of pronounced non-linearity. The recursive method automatically determined the inflection point, where the maximum model likelihood was used. Piecewise linear regression also obtained multiple slope parameters separated by the inflection point, where the slope significantly changed direction (i.e., where the smoothed plots exhibited a clear transition point). All statistical analyses were conducted with R (Version 4.0.5;


Sample Characteristics

Our study sample included 1,581 participants, with a mean age of 55.3 years. Among them, 48.6% were men and 51.4% were women. The highest proportions of participants were non-Hispanic Black (33.9%) and non-Hispanic White (31.8%), followed by Mexican American (13.7%), other Hispanic (11.6%), other (4.9%), and non-Hispanic Asian (4.1%). More than half (61.5%) were married or living with a partner. Approximately a quarter or more of the participants received high school graduate/GED-level education (24.1%), about a third completed some college or an associate of art degree (35.5%), and 22.6% at least graduated from college (22.6%). The mean BMI was 36.1 kg/m2. Further details can be found in Table 1.

TABLE 1 - Baseline Characteristics of the Study Sample (N = 1,581)
Variables n (%) or mean ± SD
C-reactive protein (mg/L) 3.6 ± 2.39
Body mass index (kg/m2) 36.1 ± 5.71
Sociodemographic factors Clinical factors
Age 55.3 ± 8.78 WBC count (1000 cells/uL) 7.4 ± 2.14
Race/ethnicity Triglycerides (mmol/L) 158.5 ± 123.19
 Mexican American 217 (13.7) HDL cholesterol (mmol/L) 49.9 ± 14.24
 Other Hispanic 183 (11.6) Hemoglobin (g/dL) 14.1 ± 1.58
 Non-Hispanic White 502 (31.8) Arthritis
 Non-Hispanic Black 536 (33.9)  Yes 654 (41.5)
 Non-Hispanic Asian 65 (4.1)  No 922 (58.5)
 Other 78 (4.9) Liver disease
Gender  Yes 118 (7.5)
 Men 768 (48.6)  No 1,461 (92.5)
 Women 813 (51.4) Lung disease
Marital status  Yes 171 (10.8)
 Married/living with partner 971 (61.5)  No 1,406 (89.2)
Widowed/divorced/separated 412 (26.1) Asthma
 Never married 196 (12.4)  Yes 286 (18.1)
Education level  No 1,293 (81.9)
 Less than 9th grade 124 (7.9) Malignancy of any kind
 9th–11th grade 157 (9.9)  Yes 137 (8.7)
 High school graduate/GED 380 (24.1)  No 1,444 (91.3)
 Some college or AA degree 560 (35.5) Diabetes mellitus
 College graduate or above 358 (22.6)  Yes 366 (23.2)
The ratio of family income to poverty 2.7 ± 1.66  No 1,150 (72.7)
Lifestyle factors  Prediabetes 65 (4.1)
Recent smoking Hypertension
 Yes 288 (18.2)  Yes 849 (53.7)
 No 1,293 (81.8)  No 731 (46.3)
Moderate work activity Coronary heart disease
 Yes 714 (45.2)  Yes 74 (4.7)
 No 867 (54.8)  No 1,505 (95.3)
Vigorous work activity Angina pectoris
 Yes 425 (26.9)  Yes 39 (2.5)
 No 1155 (73.1)  No 1,538 (97.5)
Walking or cycling Stroke
 Yes 305 (19.3)  Yes 86 (5.4)
 No 1,275 (80.7)  No 1,494 (94.6)
Moderate recreational activity
 Yes 586 (37.1)
 No 994 (62.9)
Vigorous recreational activity
 Yes 271 (17.1)
 No 1,310 (82.9)
Note. HDL = high-density lipoprotein; SD = standard deviation; WBC = white blood cell.

GAM Comparison

Three models were established to estimate the difference in the smooth functions by gender. Table 2 presents the estimated regression coefficients of the overall model with the factor-smooth interaction Gender × CRP.

TABLE 2 - Estimated Regression Coefficients of the Overall Model With a Factor Smooth Interaction
Model 1 (N = 1,581) Model 2 (n = 1,386) Model 3 (n = 1,368)
Estimate p Estimate p Estimate p
Intercept 3.69 <.001 3.99 <.001 9.57 <.001
Gender × CRP (smooth) Men 1.01 [EDF] .008 1.00 [EDF] .057 1.00 [EDF] .278
Women 2.42 [EDF] .095 2.48 [EDF] .106 2.34 [EDF] .211
Race/ethnicity Mexican American Ref Ref
Other Hispanic 0.59 .215 0.16 .737
Non-Hispanic White 0.99 .015 0.27 .510
Non-Hispanic Black −0.48 .246 −1.08 .011
Non-Hispanic Asian 0.52 .453 0.13 .849
Other 1.47 .019 0.78 .202
Marital status Married/living with partner Ref Ref
Widowed/divorced/separated 1.08 <.001 0.89 .001
Never married 1.31 <.001 0.97 .009
Education level Less than 9th grade Ref Ref
9th–11th grade 0.66 .267 0.55 .342
High school graduate/GED or equivalent 0.28 .602 0.24 .657
Some college or AA degree 0.57 .281 0.41 .429
College graduate or above 0.06 .911 0.29 .603
Ratio of family income to poverty Continuous −0.53 <.001 −0.38 <.001
Recent smoking No (ref = yes) −1.23 <.001
Moderate work activity No (ref = yes) −0.15 .584
Vigorous work activity No (ref = yes) −0.16 .579
Walking or cycling No (ref = yes) 0.15 .619
Moderate recreational activity No (ref = yes) 0.44 .083
Vigorous recreational activity No (ref = yes) 0.14 .676
WBC count Continuous 0.05 0.393
Triglycerides Continuous 0.0002 .785
HDL cholesterol Continuous 0.03 .001
Hemoglobin Continuous −0.09 .278
Arthritis No (ref = yes) −0.97 <.001
Liver disease No (ref = yes) −0.65 .139
Lung disease No (ref = yes) −1.70 <.001
Asthma No (ref = yes) −0.58 .058
Malignancy No (ref = yes) −0.12 .772
Diabetes mellitus Yes Ref
No −0.39 .180
Prediabetes −0.49 .424
Hypertension No (ref = yes) −0.48 .052
Coronary heart disease No (ref = yes) −0.10 .860
Angina pectoris No (ref = yes) −1.16 .137
Stroke No (ref = yes) −0.20 .692
Fit statistics AIC = 9275.98, R 2 adj = 0.007, DE = 0.962%, REML = 3919.60 AIC = 7997.84
R 2 adj = 0.088,
DE = 9.84%, REML = 3862.40
AIC = 7795.12
R 2 adj = 0.155, DE = 17.8%, REML = 3815.00
Note. CRP = C-reactive protein; EDF = effective degree of freedom; Ref = reference; WBC = white blood cell; HDL = high-density lipoprotein; AIC = Akaike information criterion; DE = deviance explained; REML = restricted maximum likelihood.

Model 1 used depression severity as an outcome with a smoothing spline function of CRP as a univariable predictor in men and women. Given the smoothing function applied, the EDF from Model 1 showed a statistically significant linear relationship between CRP and depression severity in men with EDF equivalent to 1 (p = .008). A highly nonlinear relationship was found in women with EDF > 2 but was not significant (p = .095). The plots of estimated smoothing spline functions in Figure 1a demonstrate the linear relationship between CRP and depression severity in men and an inverse U-shaped relationship in women. Model 2 was built upon Model 1 but was adjusted for sociodemographic variables, such as race/ethnicity, marital status, educational level, and the ratio of family income to poverty in men and women. Similarly, a linear relationship between CRP and severity of depression was identified in men (EDF = 1.00), and a highly nonlinear relationship was identified in women (EDF = 2.48); however, both relationships were not significant (p = .057 and p = .106, respectively; Figure 1b). Model 3 was further adjusted for other lifestyle and clinical covariates in addition to sociodemographic variables in men and women. The estimated smooth functions in Figure 1c exhibit the linear relationship between CRP and depression severity in men (EDF = 1.00) and an inverse U-shaped relationship in women (EDF = 2.34). These relationships were not statistically significant (p = .278 and p = .210, respectively).

Plots of estimated smoothing spline function of C-reactive protein (CRP) with 95% confidence band for the generalized additive model with outcome variable as depression severity by gender. EDF = effective degree of freedom. (a) Model 1 represents the univariable smooth function of CRP in men (left plot; EDF = 1.01, p = .008) and women (right plot; EDF = 2.42, p = .095). (b) Model 2 shows the multivariable smooth function of CRP with all sociodemographic variables adjusted in men (left plot; EDF = 1.00, p = .057) and women (right plot; EDF = 2.48, p = .106). (c) Model 3 represents the multivariable smooth function of CRP with all covariates adjusted in men (left plot; EDF = 1.00, p = .278) and women (right plot; EDF = 2.34, p = .211).

Relationship Between CRP and Depression Severity in Women

Although the p-value was not significant, Model 1, Model 2, and Model 3 demonstrated a pronounced nonlinear relationship between CRP and depression severity in women based on the EDF value. In particular, the resulting curve displayed a two-stage change with a turning point (i.e., inflection point). Thus, we used two-piecewise linear regression models to further characterize this nonlinear relationship. When the statistically significant covariates (i.e., marital status, educational level, the ratio of family income to poverty, recent smoking status, direct high-density lipoprotein cholesterol, and the presence of arthritis and lung disease) from Model 3 were added, the recursive method automatically determined the inflection point where the smoothed plots had a significant transition point. In women, the inflection point for CRP was 3.6 mg/L (standard error of 0.83), which suggests that the relationship between CRP and depression severity is different below and above 3.6 mg/L. Especially depression severity related to CRP significantly increased as the CRP level increased to an inflection point of 3.6 mg/L (p = .048) but decreased thereafter. Table 3 and Figure 2 provide detailed information on the relationship between CRP and depression severity in women below or above the inflection point.

TABLE 3 - Adjusted Association Between C-Reactive Protein and Depression Severity in Women < or ≥ Optimum Threshold Level (3.6 mg/L With Standard Error of 0.83)
Estimate Standard error p
Intercept 6.52 1.11 <.001
< Threshold 0.50 0.25 .048
> Threshold −0.25 0.14 NA
Note. Statistically significant covariates from Model 3 were adjusted in a two-piecewise linear regression model.

Segmented regression plot showing the adjusted relationship of C-reactive protein (CRP) and depression severity among women. The threshold level affected the association pattern particularly for women, among whom the depression severity related to CRP significantly increased as the CRP level increased to an inflection point of 3.6 mg/L (i.e., the point indicated by the circle in the figure) but decreased thereafter.


This exploratory study aimed to analyze the relationship between CRP and depression severity in middle-aged obese adults with CRP of ≤10 mg/L using GAMs and examine potential gender differences. There were apparent differences in the relationship between CRP and depression severity between men and women. Among men, an increasing linear pattern was observed (EDF ≈ 1). Contrastingly, among women, the EDF value was greater than 2 in all unadjusted and adjusted models, indicating the smooth (curved) association of CRP and depression scores. The threshold level affected the association pattern particularly for women; the depression severity related to CRP significantly increased as the CRP level increased to a certain threshold (3.6 mg/L) but decreased thereafter.

This study broadly corroborates the extensive literature, which has consistently reported a statistically positive significant relationship between CRP and depression (assuming linearity) only in men (Danner et al., 2003; Elovainio et al., 2009; Ernst et al., 2021; Ford & Erlinger, 2004; Lee et al., 2019; Tayefi et al., 2017; Vetter et al., 2013). We also provide insight into the underlying mechanisms regarding why previous researchers have observed inconsistent relationships between these two variables—specifically in women with CRP of ≤10 mg/L. In particular, we suspect that the existing findings on women might have varied by the range of CRP in the sample population. For instance, in Ma et al.’s (2011) study conducted on a population-based, healthy control group where the CRP level was heavily clustered between 0 and 3 (mean = 1.8 mg/L, median = 1.2 mg/L), CRP was positively associated with a clinical diagnosis of depression—which partially aligns with the findings of the current study. However, interpreting these results to a wider population is challenging because the current study’s depression scores were within a narrow range (those with depression scores between 0 and 9 comprised approximately 90.0%), thereby limiting the capacity to conclude individuals with clinical depression.

Based on the results, in middle-aged obese men with CRP of ≤10 mg/L, inflammation and depression are intertwined. Given that inflammation is a potential mechanism underlying depression as a risk factor for atherosclerotic vascular disease, the results may explain the relatively high susceptibility of men to cardiovascular disease attributable to depression (Ferketich et al., 2000; Ford et al., 1998; Kamphuis et al., 2006). The American Heart Association/CDC joint statement on cardiovascular risk stratification suggests the following cutoff values: low risk, < 1.0 mg/L; medium risk, 1.0–3.0 mg/L; and high risk, > 3.0 mg/L (Pearson et al., 2003). In one study, more than one third of individuals with unipolar depression, bipolar depression, bipolar mania, or schizophrenia had a CRP of > 3 mg/L, suggesting that it could be related to an elevated risk of cardiovascular events (Wysokiński et al., 2015). Contrastingly, for women with CRP of ≤10 mg/L, assuming linearity for the CRP relationship with depression severity may not be appropriate when assessing associated disease risk. It is possible that another cutoff better demarcates the level at which CRP correlates depression severity. In addition, the current clinical risk assessment tools guided by CRP levels may become more clinically useful if CRP measures are combined with gender-specific depressive psychopathology.

For women, the nonlinear pattern persisted irrespective of any adjustments for potential covariates or mechanisms possibly responsible for the association between CRP and depression, such as socioeconomic, lifestyle, or anthropometric factors or chronic illness conditions. Given this outcome, the complex trend may have been due to other factors not considered in the NHANES, such as critical hormones that fluctuate during the menstrual cycle. In women, estrogen has anti-inflammatory effects, and the menstrual cycle affects adiposity and inflammation (Bjune et al., 2022), which could explain the weak relationship between CRP levels and depression among women. In line with this, estrogen therapy and oral contraceptive use—which have long-term effects on the mood of menopausal women (Chedraui & Pérez-López, 2019)—could have influenced this trend. However, these variables were not included in the analyses because of missing information.

From the methodological perspective, the complex relationship between CRP and depression severity identified in the sample of women can also be partially explained by the sickness syndrome theory (Raison & Miller, 2003; Saper et al., 2012), which argues that chronic low-grade inflammation might be associated with depression subtypes. Depression is a highly heterogeneous disorder, and there is an increasing number of research in immunopsychiatry arguing that psychopathology needs to be investigated on an individual symptom level (Fried & Nesse, 2015). For instance, Köhler-Forsberg et al. (2017) investigated the association between CRP and specific depressive symptomatology by gender, finding significant symptom-specific associations only among women (e.g., higher CRP is mainly associated with increased suicidality severity) after controlling for important factors, including BMI and smoking.

Limitations and Strengths

We acknowledge several limitations that make these results only a first step toward the optimal modeling of the relationship between CRP and depression. First, cross-sectional data were analyzed; therefore, it is not possible to confirm whether inflammation precedes the onset of depressive symptoms. Longitudinal data are needed to better understand the temporal relationship between these complex biological processes. Also, as with most observational studies, residual confounding cannot be ruled out. For instance, other than the discussed gender-specific hormones, we could not assess information on prior diagnoses of depression or treatment history (e.g., psychotropics), which could preclude the actual relationship between CRP and depression.

Other limitations predominantly relate to measurement issues. First, CRP is an acute-phase reactant that may vary considerably according to the underlying inflammatory state at the time of measurement. Therefore, serial measurements of CRP over time provide a more clinically relevant information and may have improved the reliability of the study findings. Second, depression was measured through self-reports rather than structured clinical interviews, leaving room for inaccuracies in the analysis. Furthermore, the estimates of associations between the different measurement domains (i.e., blood assays vs. self-report) are more prone to downward bias because of measurement-domain-specific error variance (Moriarity et al., 2021).

The strengths include the application of GAM, which enabled us to model a nonlinear relationship between an independent variable and an outcome without a priori assumptions regarding the functional form of the relationship. This study also contributes to the growing evidence that gender plays a crucial role in the relationship between CRP and depression. Finally, as mentioned, the data may only represent less severe depression; the NHANES is designed to collect data from noninstitutionalized community populations, possibly limiting the statistical analysis power by not reflecting the depression severity. However, the positive relationship between the CRP and depressive score among men (and women to a certain threshold) who were not clinically depressed on average may also be of interest as it implies that the entire range of depressive symptomatology could be associated with health risks.


Although much remains to be learned from future research on inflammatory signaling pathways as possible underlying mechanisms between CRP and depression, this study is a starting point to explore the CRP–depression relationship before posing any inflexible analytic form. Especially for obese women with CRP of ≤10 mg/L, it might be inadequate to draw implications about this relationship from standard parametric models, as a threshold level of CRP associated with changes in depression score may exist.


Chiyoung Lee

Se Hee Min

Kosuke Niitsu


Bjune J. I., Strømland P. P., Jersin R. Å., Mellgren G., Dankel S. N. (2022). Metabolic and epigenetic regulation by estrogen in adipocytes. Frontiers in Endocrinology, 153. 10.3389/fendo.2022.828780
Bremmer M. A., Beekman A. T. F., Deeg D. J. H., Penninx B. W. J. H., Dik M. G., Hack C. E., Hoogendijk W. J. G. (2008). Inflammatory markers in late-life depression: Results from a population-based study. Journal of Affective Disorders, 106, 249–255. 10.1016/j.jad.2007.07.002
Centers for Disease Control and Prevention. (2020). Laboratory procedure manual: High sensitivity C-reactive protein (hs-CRP).
Centers for Disease Control and Prevention. (2021). Laboratory procedure manual: High sensitivity C-reactive protein (hs-CRP).
Chandrashekara S. (2014). C-reactive protein: An inflammatory marker with specific role in physiology, pathology, and diagnosis. Internet Journal of Rheumatology and Clinical Immunology, 2.
Chedraui P., Pérez-López F. R. (2019). Metabolic syndrome during female midlife: What are the risks? Climacteric, 22, 127–132. 10.1080/13697137.2018.1561666
Chocano-Bedoya P. O., Mirzaei F., O’Reilly E. J., Lucas M., Okereke O. I., Hu F. B., Rimm E. B., Ascherio A. (2014). C-reactive protein, interleukin-6, soluble tumor necrosis factor α receptor 2 and incident clinical depression. Journal of Affective Disorders, 163, 25–32. 10.1016/j.jad.2014.03.023
Danner M., Kasl S. V., Abramson J. L., Vaccarino V. (2003). Association between depression and elevated C-reactive protein. Psychosomatic Medicine, 65, 347–356. 10.1097/01.PSY.0000041542.29808.01
Delgado I., Huet L., Dexpert S., Beau C., Forestier D., Ledaguenel P., Aubert A., Sauvant J., Aouizerate B., Magne E., Capuron L. (2018). Depressive symptoms in obesity: Relative contribution of low-grade inflammation and metabolic health. Psychoneuroendocrinology, 91, 55–61. 10.1016/j.psyneuen.2018.02.030
Elovainio M., Aalto A.-M., Kivimäki M., Pirkola S., Sundvall J., Lönnqvist J., Reunanen A. (2009). Depression and C-reactive protein: Population-based Health 2000 Study. Psychosomatic Medicine, 71, 423–430. 10.1097/PSY.0b013e31819e333a
Ernst M., Brähler E., Otten D., Werner A. M., Tibubos A. N., Reiner I., Wicke F., Wiltink J., Michal M., Nagler M., Münzel T., Wild P. S., König J., Pfeiffer N., Borta A., Lackner K. J., Beutel M. E. (2021). Inflammation predicts new onset of depression in men, but not in women within a prospective, representative community cohort. Scientific Reports, 11, 2271. 10.1038/s41598-021-81927-9
Ferketich A. K., Schwartzbaum J. A., Frid D. J., Moeschberger M. L. (2000). Depression as an antecedent to heart disease among women and men in the NHANES I study. Archives of Internal Medicine, 160, 1261–1268. 10.1001/archinte.160.9.1261
Ford D. E., Erlinger T. P. (2004). Depression and C-reactive protein in U.S. adults: Data from the third National Health and Nutrition Examination Survey. Archives of Internal Medicine, 164, 1010–1014. 10.1001/archinte.164.9.1010
Ford D. E., Mead L. A., Chang P. P., Cooper-Patrick L., Wang N.-Y., Klag M. J. (1998). Depression is a risk factor for coronary artery disease in men: The precursors study. Archives of Internal Medicine, 158, 1422–1426. 10.1001/archinte.158.13.1422
Fried E. I., Nesse R. M. (2015). Depression sum-scores don’t add up: Why analyzing specific depression symptoms is essential. BMC Medicine, 13, 72. 10.1186/s12916-015-0325-4
Garson G. D. (2012). Testing statistical assumptions. Statistical Associates.
Hastie T. J., Tibshirani R. J. (2017). Generalized additive models (1st ed.). CRC Press.
Horn S. R., Long M. M., Nelson B. W., Allen N. B., Fisher P. A., Byrne M. L. (2018). Replication and reproducibility issues in the relationship between C-reactive protein and depression: A systematic review and focused meta-analysis. Brain, Behavior, and Immunity, 73, 85–114. 10.1016/j.bbi.2018.06.016
Kamphuis M. H., Kalmijn S., Tijhuis M. A. R., Geerlings M. I., Giampaoli S., Nissinen A., Grobbee D. E., Kromhout D. (2006). Depressive symptoms as risk factor of cardiovascular mortality in older European men: The Finland, Italy and Netherlands Elderly (FINE) study. European Journal of Preventive Cardiology, 13, 199–206. 10.1097/01.hjr.0000188242.64590.92
Khandaker G. M., Zuber V., Rees J. M. B., Carvalho L., Mason A. M., Foley C. N., Gkatzionis Apostolos, Jones Peter B., Burgess S. (2020). Shared mechanisms between coronary heart disease and depression: Findings from a large U.K. general population-based cohort. Molecular Psychiatry, 25, 1477–1486. 10.1038/s41380-019-0395-3
Köhler-Forsberg O., Buttenschøn H. N., Tansey K. E., Maier W., Hauser J., Dernovsek M. Z., Henigsberg N., Souery D., Farmer A., Rietschel M., McGuffin P., Aitchison K. J., Uher R., Mors O. (2017). Association between C-reactive protein (CRP) with depression symptom severity and specific depressive symptoms in major depression. Brain, Behavior, and Immunity, 62, 344–350. 10.1016/j.bbi.2017.02.020
Lee S., Oh S. S., Jang S.-I., Park E.-C. (2019). Sex difference in the association between high-sensitivity C-reactive protein and depression: The 2016 Korea National Health and Nutrition Examination Survey. Scientific Reports, 9, 1918. 10.1038/s41598-018-36402-3
Ma Y., Chiriboga D. E., Pagoto S. L., Rosal M. C., Li W., Merriam P. A., Hébert J. R., Whited M. C., Ockene I. S. (2011). Association between depression and C-reactive protein. Cardiology Research and Practice, 2011, 286509. 10.4061/2011/286509
Mac Giollabhui N., Ng T. H., Ellman L. M., Alloy L. B. (2021). The longitudinal associations of inflammatory biomarkers and depression revisited: Systematic review, meta-analysis, and meta-regression. Molecular Psychiatry, 26, 3302–3314 10.1038/s41380-020-00867-4
Moazzami K., Lima B. B., Sullivan S., Shah A., Bremner J. D., Vaccarino V. (2019). Independent and joint association of obesity and metabolic syndrome with depression and inflammation. Health Psychology, 38, 586–595. 10.1037/hea0000764
Moriarity D. P., Horn S. R., Kautz M. M., Haslbeck J. M., Alloy L. B. (2021). How handling extreme C-reactive protein (CRP) values and regularization influences CRP and depression criteria associations in network analyses. Brain, Behavior, and Immunity, 91, 393–403. 10.1016/j.bbi.2020.10.020
Muggeo V. M. R. (2022). Package ‘segmented’: Regression models with break-points/change-points (with possibly random effects) estimation.
O’Connor M.-F., Bower J. E., Cho H. J., Creswell J. D., Dimitrov S., Hamby M. E., Hoyt M. A., Martin J. L., Robles T. F., Sloan E. K., Thomas K. S., Irwin M. R. (2009). To assess, to control, to exclude: Effects of biobehavioral factors on circulating inflammatory markers. Brain, Behavior, and Immunity, 23, 887–897. 10.1016/j.bbi.2009.04.005
Osimo E. F., Baxter L. J., Lewis G., Jones P. B., Khandaker G. M. (2019). Prevalence of low-grade inflammation in depression: A systematic review and meta-analysis of CRP levels. Psychological Medicine, 49, 1958–1970. 10.1017/S0033291719001454
Pearson T. A., Mensah G. A., Alexander R. W., Anderson J. L., Cannon R. O. 3rd, Criqui M., Fadl Y. Y., Fortmann S. P., Hong Y., Myers G. L., Rifai N., Smith S. C. Jr., Taubert K., Tracy R. P., Vinicor F. (2003). Markers of inflammation and cardiovascular disease: Application to clinical and public health practice: A statement for healthcare professionals from the Centers for Disease Control and Prevention and the American Heart Association. Circulation, 107, 499–511. 10.1161/01.CIR.0000052939.59093.45
Raison C. L., Miller A. H. (2003). Depression in cancer: New developments regarding diagnosis and treatment. Biological Psychiatry, 54, 283–294. 10.1016/S0006-3223(03)00413-X
Saper C. B., Romanovsky A. A., Scammell T. E. (2012). Neural circuitry engaged by prostaglandins during the sickness syndrome. Nature Neuroscience, 15, 1088–1095. 10.1038/nn.3159
Stevenson R. D., Woods W. A. Jr. (2006). Condition indices for conservation: New uses for evolving tools. Integrative & Comparative Biology, 46, 1169–1190. 10.1093/icb/icl052
Tayefi M., Shafiee M., Kazemi-Bajestani S. M. R., Esmaeili H., Darroudi S., Khakpouri S., Mohammad M., Ghaneifar Z., Azarpajouh M. R., Moohebati M., Heidari-Bakavoli A., Parizadeh M. R., Nematy M., Safarian M., Ebrahimi M., Ferns G. A., Naghmeh M., Ghayour-Mobarhan M. (2017). Depression and anxiety both associate with serum level of hs-CRP: A gender-stratified analysis in a population-based study. Psychoneuroendocrinology, 81, 63–69. 10.1016/j.psyneuen.2017.02.035
Vetter M. L., Wadden T. A., Vinnard C., Moore R. H., Khan Z., Volger S., Sarwer D. B., Faulconbridge L. F. (2013). Gender differences in the relationship between symptoms of depression and high-sensitivity CRP. International Journal of Obesity, 37, S38–S43. 10.1038/ijo.2013.95
Wood S.. (2022). Package ‘mgcv’: Mixed GAM computation vehicle with automatic smoothness estimation.
Wysokiński A., Margulska A., Strzelecki D., Kłoszewska I. (2015). Levels of C-reactive protein (CRP) in patients with schizophrenia, unipolar depression and bipolar disorder. Nordic Journal of Psychiatry, 69, 346–353. 10.3109/08039488.2014.984755
Yang C., Wardenaar K. J., Bosker F. J., Li J., Schoevers R. A. (2019). Inflammatory markers and treatment outcome in treatment resistant depression: A systematic review. Journal of Affective Disorders, 257, 640–649. 10.1016/j.jad.2019.07.045
Zimmerman M. (2019). Using the 9-item Patient Health Questionnaire to screen for and monitor depression. JAMA, 322, 2125–2126. 10.1001/jama.2019.15883

C-reactive protein; depression; inflammation; obesity

Supplemental Digital Content

Copyright © 2023 Wolters Kluwer Health, Inc. All rights reserved.