Skip to main content
NCBI home page
International Journal of Methods in Psychiatric Research logoLink to International Journal of Methods in Psychiatric Research
. 2023 Apr 10;32(4):e1968. doi: 10.1002/mpr.1968

Networks of C‐reactive protein and depression symptoms in patients with stable coronary heart disease: Findings from the Heart and Soul Study

Chiyoung Lee 1,2,✉, Mary A Whooley 3
PMCID: PMC10698828  PMID: 37035901

Abstract

Objective

Research addressing the associations between C‐reactive protein (CRP) and depression among patients with stable coronary heart disease (CHD) has produced inconsistent results. This might be attributable to varying associations of CRP with specific depression symptom profiles. We responded to this challenge using various network analysis techniques.

Methods

A total of 967 outpatients with documented CHD were drawn from the baseline cross‐sectional data of the Heart and Soul Study. We first estimated mixed graphical models that included CRP and individual depression symptoms, before and after adjusting for relevant covariates, to explore whether CRP is correlated with specific facets of depression. We also investigated whether CRP levels moderated the associations between specific depression symptoms using moderated network models. Finally, we performed a network comparison test and compared the symptom network properties between non‐elevated and elevated CRP groups.

Results

In the network model without covariates, CRP was positively associated with fatigue, appetite changes, and psychomotor problems. CRP maintained its negative association with concentration difficulty regardless of covariate adjustment. Few symptom‐symptom associations, especially those involving appetite changes, were moderated by CRP. Further, the elevated CRP group showed greater overall symptom connectivity as compared to the non‐elevated group.

Conclusion

This study segues into CRP‐depression relationship with sophisticated methodology.

Keywords: C‐reactive protein, coronary heart disease, depression, network analysis

1. INTRODUCTION

Depression and coronary heart disease (CHD) are both highly prevalent and frequently co‐occur. Identifying the mechanisms linking these two disorders remains a critical challenge. Although much needs to be clarified, an inflammatory pathway has been proposed to be a biological link in the pathophysiology of depression and CHD, since elevations of inflammatory markers can regularly be detected in both.

One of the most studied inflammatory markers is serum C‐reactive protein (CRP), a protein that hepatocytes produce when induced by cytokines, including interleukin‐6 and tumor necrosis factor‐alpha (Khuseyinova & Koenig, 2006). CRP holds promise as a reliable and clinically valuable inflammatory marker that helps to identify depression even among those CHD patients, who already possess chronic, low‐grade baseline inflammation. Bankier et al. (2009) found a significant association between CRP and major depressive disorder (MDD) in stable CHD patients, after accounting for age. One prospective study showed that high levels of CRP at baseline are positively correlated with depression severity in CHD patients (Sforzini et al., 2019). In a study by Nikkheslat et al. (2015), CHD patients with a depression diagnosis had higher levels of CRP compared with CHD patients without depression. However, not all prior research has given consistent results. One well‐designed cross‐sectional study reported no relationship between elevated CRP and MDD in stable CHD patients (Whooley et al., 2007). Two other studies found no association between CRP and depression in the 17 months following a myocardial infarction or coronary revascularization (Annique et al., 2005; Janszky et al., 2005).

At the current stage of knowledge, such incongruent results could be attributable to heterogeneity in study designs or various unaccounted covariates. Yet, an explanation that recently received widespread acceptance in this field proposed that CRP is differentially associated with specific depression symptoms or symptom profiles and any significant versus null associations would be driven by these profiles (Fried et al., 2014). Initial research efforts demonstrating that individual depression symptoms differ in terms of their association with inflammatory markers supported this argument (Felger et al., 2016; Jokela et al., 2016; Majd et al., 2020; White et al., 2017). This inspired calls for further studies of psychopathology at the symptom level (Fried & Nesse, 2015).

Network analysis, a well‐established conceptual technique derived from network science, is a powerful tool that mathematically estimates the unique, pairwise connections between all variables in a single model (Borsboom & Cramer, 2013). In a network comprised of CRP and depression symptoms, network analysis elucidates the interplays between variables to determine which specific symptom in the network has the largest association with CRP, after controlling for other symptoms. Recently, studies by Moriarity, Horn, et al. (2021), as well as other researchers (Fried et al., 2020; Kappelmann et al., 2021; Lee et al., 2023a; Lee & Min, 2023b; Manfro et al., 2022), applied this technique to extend this line of inquiry. For instance, Moriarity, Horn, et al. (2021) constructed a cross‐sectional network using the National Health and Nutrition Examination Survey and found that CRP only correlated with “fatigue” and “appetite changes” among a representative US population.

Simultaneously, since network analysis has rapidly progressed in the past few years, psychiatry researchers have been able to explore complex psychopathology networks using various statistical methods, including moderated network analysis and network comparison test (NCT), all of which can provide an integral perspective in CRP‐depression symptom association. Introduced by Haslbeck et al. (2021), the moderated network models (MNMs) based on moderated network analysis allows for each pair wise interaction between two variables to be moderated by (a subset of) all other variables in a given model. A pioneering study by Moriarity, van Borkulo, et al. (2021) adopted this approach to further explore the moderation effects of CRP on depression symptom structure and found that several symptom‐to‐symptom relations (particularly those including appetite changes) were moderated by CRP. Further, NCT, an exploratory test comparing the networks, enables researchers to test group differences in hypotheses within the comprehensive structural characteristics of the network (van Borkulo et al., 2022). Using this technique, Moriarity, van Borkulo, et al. (2021) found that the elevated CRP group (≥ 3.0 mg/L) showed greater overall depression symptom connectivity (i.e., stronger total associations between individual symptoms in the network) compared to the non‐elevated group. However, all of the aforementioned studies were based on population‐based samples, not exclusively on patients with stable CHD.

1.1. Study aims

We aimed to explain the inconsistencies found in the clinical literature concerning the relationship between CRP and depression among patients with stable CHD by adopting a symptom‐level analysis based on various network approaches. Our first focus was to estimate the network models to include CRP and the individual symptoms of depression with and without relevant covariates with the aim to explore whether CRP is associated with specific facets of depression (Aim 1). Next, we conducted a moderated network analysis to investigate whether the associations between specific depression symptoms can be moderated by the CRP level (Aim 2). Finally, we performed the NCT to assess whether the specific properties of depression symptom network between non‐elevated and elevated CRP groups would differ (Aim 3). No hypotheses were made in view of the exploratory nature of the current study.

2. METHODS

2.1. Study design, database, and settings

This cross‐sectional study used baseline data from the Heart and Soul Study, a prospective cohort study of psychosocial factors and health outcomes conducted on patients who have stable CHD. Details on eligibility and enrollment in the Heart and Soul Study are available in the literature (Ruo et al., 2003).

In brief, outpatients with documented CHD were recruited from two Veterans Affairs Medical Centers (San Francisco and Palo Alto, CA, USA), one university medical center (University of California at San Francisco), and nine community health clinics in Northern California. Criteria for inclusion included a history of any of the following conditions: myocardial infarction, angiographic evidence of ≥50.0% stenosis in one or more major coronary vessels, evidence of exercise‐induced ischemia in treadmill or nuclear testing, coronary artery revascularization, or a diagnosis of coronary disease by an internist or cardiologist. Patients were not eligible if they had an acute coronary syndrome within the past 6 months, could not walk 1 block, or were planning to relocate within 3 years. Between September 2000 and December 2002, 1,024 participants were enrolled and completed a day‐long study appointment at the Veterans Affairs Medical Center in San Francisco; they were screened with a comprehensive assessment. From this baseline sample, we excluded 57 participants with missing information on CRP and depression symptoms (listwise deletion) as network estimation algorithms cannot handle missing values. This process gave 967 participants, whose data were primarily used for analysis.

2.2. Measures

2.2.1. Depression symptoms

Depression symptoms were measured with the nine‐item Patient Health Questionnaire (PHQ‐9), a self‐report inventory of depression (Supplemental Table S1). The PHQ‐9 items are based on the symptom criteria for MDD in the Diagnostic and Statistical Manual of Mental Disorders, Fifth Edition (DSM‐5; Uher et al., 2014), and include the following: loss of interest, feelings of sadness, sleep disturbance, fatigue, appetite changes, feelings of worthlessness, concentration difficulty, psychomotor problems, and suicide ideation. For each symptom, respondents indicate whether, during the previous 2 weeks, the symptom has bothered them: 0 = “not at all”, 1 = “several days”, 2 = “more than half of the days” or 3 = “nearly every day.” The items can be summed to yield a total score, representing the severity of depression.

2.2.2. C‐reactive protein

Serum CRP was quantified by the Roche Integra high‐sensitivity assay (Roche, Indianapolis, IN, USA) for around 25% of the participants and (owing to a change in the laboratory) the Beckman Extended Range high‐sensitivity assay (Beckman Coulter Inc., Fullerton, CA, USA) for the remaining participants. The results from the two assays were strongly correlated (r = 0.99 for 185 participants). Patients were asked to fast for 12 h, to not smoke for 5 h, and to withhold aspirin for 1 week. Obtained serum was frozen at −70°C until the time of the CRP assay.

2.2.3. Covariates

Relevant covariates were selected either because the variables had been previously demonstrated to be important contributors to the CRP–depression link or they had been found to generally affect either CRP or depression among CHD patients. We included demographic, lifestyle, and clinical variables that fulfilled the above criteria based on the empirical evidence (O’Connor et al., 2009; Smith et al., 2018). We also endeavored to align the covariates with the latest network investigations in immunopsychiatry (Fried et al., 2020; Lee et al., 2023a; Moriarity, Horn, et al., 2021) to make the findings comparable.

Demographic variables included age (continuous) and sex (male or female). Lifestyle variables included alcohol use (continuous), current smoking status (yes or no), and physical activity (continuous). Alcohol use was measured using the Alcohol Use Disorders Identification Test (AUDIT‐C; Bush et al., 1998), a validated 3‐item screening tool for identifying individuals with alcohol abuse or dependence. Scores range from 0 to 12, with a higher score signifying a greater severity of problem drinking. Physical activity was determined by the self‐reported question, “which of the following statements best summarizes how physically active you have been during the past month; that is, done activities such as 15–20 min of brisk walking, swimming, general conditioning, or recreational sports?” Responses were scored using a 6‐point, continuous scale with 0 equal to “not at all active” (0 times/month) and 6 being equal to “extremely active” (≥5 times/week). Clinical‐related variables included fibrinogen (RU/mL), white blood cell (WBC; k/cmm), body mass index (BMI; kg/m2), high density lipoprotein (HDL; mg/dL), and triglycerides (mg/dL), and the use of antidepressants including selective serotonin reuptake inhibitors, tricyclics, or other classes of antidepressants (yes or no). Plasma fibrinogen concentrations were measured with the Clauss assay. Fasting serum was analyzed for WBC, HDL, and triglycerides.

2.3. Statistical analysis

All analyses were carried out with the RStudio software. Descriptive statistics were used to analyze the baseline characteristics. Researchers often exclude participants with a CRP >10.0 mg/L from analyses due to the possibility of an active infection or injurious stimuli. However, removing these individuals may ignore those at higher risk of psychopathology secondary to inflammation, as theories linking inflammation with depression primarily concentrate on individuals with elevated inflammatory states (Mac Giollabhui et al., 2020). Moriarity, Horn, et al. (2021) also demonstrated that including individuals with CRP >10.0 mg/L yields more stable and replicable networks. Thus, we did not exclude data based on the CRP level. For the reader's information, we present the models excluding CRP >10 mg/L in the Supplement section, wherein we found the results supporting the argument of Moriarity, Horn, et al. (2021).

2.3.1. Mixed graphical model estimation (Aim 1)

Two network models were constructed to examine the unique associations between CRP and the nine PHQ‐9 items, before and after adjusting for covariates. Because our data contained a mixture of categorical and numeric variables, we estimated mixed graphical models (MGMs) using the R‐package mgm (Haslbeck & Waldorp, 2015). In network models, variables of interest are represented by “nodes” in the network, and the pairwise relations among nodes controlling the effects of all other nodes are represented by “edges”.

We adopted the cross‐validation procedure to specify the LASSO tuning parameter. Cross‐validation is a less conservative model (compared with extended Bayesian information criterion [EBIC]) preferable with small samples as it has greater sensitivity in revealing results (Epskamp & Fried, 2018). Following the recent discussion on nonregularized symptom networks for psychopathology data (Moriarity, Horn, et al., 2021; Williams & Rast, 2020), the models were not regularized. The k parameter was set to 2 to indicate pairwise relations. Edges in MGMs are combined estimates from neighborhood regression (i.e., estimates from regressing node A on B, and vice versa). The ruleReg argument was set to “AND”, which computes the mean of all k parameter estimates if all estimates are non‐zero.

Node predictability, or how well a node is predicted by all neighboring nodes, was also calculated using the mgm's predict() function. Here, the proportion of explained variance (“R2”) was specified as an error function for continuous variables, and the normalized proportion of correct classification (“nCC”) was specified for categorical variables. Estimates are reported on a 0 to 1 scale, in which 1 reflects full predictability. Network models were graphed with the R‐package qgraph, using an average layout estimated with the Fruchterman‐Reingold algorithm. In this layout, nodes that are more strongly correlated are located closer together. Edges that represent positive or negative associations are shown in green and red, respectively. Gray edges represent pairwise interactions wherein no sign is defined (i.e., interactions including categorical variables). The thickness of an edge reflects the strength of the association. Predictability was visually depicted by a pie chart around the node.

We examined the accuracy of edge estimates using the mgm's resample() function, which obtains empirical sampling distributions based on the nonparametric bootstrap; in our study, we ran 1000 bootstrap samples. The mgm's plotRes() function was used to plot the resulting sampling distribution of all edges and the proportion of estimates whose absolute values were non‐zero.

2.3.2. Moderated network analysis (Aim 2)

We fitted a moderated MGM (Haslbeck et al., 2021) that included CRP level as the moderator and nine variables corresponding to PHQ‐9 items to evaluate how the depression symptom network structure changes at different levels of CRP. CRP was introduced as both a categorical and continuous moderator. Following the studies of Horn et al. (2018) and Fried et al. (2020), as a continuous moderator, raw CRP was log‐transformed since it was not normally distributed.

To perform moderation analyses, we used mgm to estimate the pairwise and 3‐way interactions (moderation interaction) between each node. Again, the cross‐validation procedure was employed as moderation effects are typically small and might be difficult to detect (Haslbeck et al., 2021). Given that more conservative estimates, such as the EBIC, may result in both important pairwise and interaction estimates being left out, using the less‐conservative estimates can allow the observation of more estimates, thus providing valuable information for interpretation. After all of the pairwise and 3‐way interactions were estimated, we computed and visualized the networks conditioned on both CRP group (<3.0 mg/L, n = 588; ≥3.0 mg/L, n = 379) and the CRP level (one standard deviation [SD] below the mean, n = 156; mean, n = 656; one SD above the mean, n = 155), using the mgm's condition() function. The models were displayed using the “circle” layout.

The stability of the estimated parameters in the MNMs was tested via the resample() function. We applied 1,000 bootstrapped samples. The plotRes() function returned a plot of the bootstrapped sampling distribution of each pairwise and 3‐way interaction. This approach allowed us to assess the strength of moderation for each edge and to calculate confidence intervals (CIs) around the estimate and the number of non‐zero moderation effects arising across the resamples. Non‐zero values with 95.0% CIs that do not include zero suggest the likelihood of moderation effects in the model.

2.3.3. Network comparison test (Aim 3)

We performed NCT on different CRP groups (<3.0 mg/L, ≥3.0 mg/L) using the R‐package NetworkComparisonTest. NCT examines the invariance of global strength (assumes that overall connectivity in both networks is identical), network structure (assumes that the structure of both networks is identical), one‐to‐one edge strength (assumes that each edge between the two networks is of similar strength), and centrality strength (assumes that both networks do not vary in measures of node centrality strength) using two‐tailed permutation tests. For this analysis, the sampling procedure was repeated 1000 times and gamma was set to 0. This process results in a reference distribution (a distribution under the null hypothesis wherein both groups are equal), which can be used to analyze the significance of observed differences between two groups. The observed difference is deemed significant at a p < 0.05 level. NCT generates a p‐value for S (difference in global strength), M (difference in network structure), E (difference in edge strength), and C (difference in centrality strength). Edge strength invariance test was only conducted when a lack of network structure invariance was upheld.

2.3.4. Sensitivity analysis

Considering the possible impact of listwise deletion of missing data on our results, we performed multiple imputation as a sensitivity analysis. The major analysis was concerning the listwise deletion methods of missing values with results being reported in this study. Multiple imputation was performed as a sensitivity analysis and the results are demonstrated in the Supplement section.

3. RESULTS

Out of 967 patients, 82.0% were males and 18.0% were females. The mean age was 66.7 years (SD = 10.99). The mean CRP level was 4.6 mg/L (SD = 8.32). Supplemental Table S2 further presents the background characteristics of the participants.

We only report edges (or moderated edges) that were non‐zero in at least half of the bootstrapped models. In the network model that did not include covariates (Figure 1a, Table 1), CRP was associated with fatigue (edge weight = 0.042), appetite changes (edge weight = 0.047), and psychomotor problems (edge weight = 0.017). The CRP–fatigue edge, CRP–appetite changes edge, and CRP–psychomotor problems edge were non‐zero in 89.0%, 85.0%, and 71.0% of the bootstrapped analyses. The predictability for CRP was 1.3%, which explains only small variance shared with other nodes in the model. Of note, a negative edge was found between CRP and concentration difficulty (edge weight = −0.056; non‐zero in 97.0% of bootstraps).

FIGURE 1.

FIGURE 1

Open in a new tab

Network model of pairwise interactions with (a) and without adjustment for covariates (b). Green edges represent positive associations between the nodes and red edges represent negative edges between the nodes with thicker edges representing stronger association. Gray edges signify pairwise interactions wherein no sign is defined (i.e., interactions involving categorical variables). Alc, alcohol use; Antidepre, antidepressants use; App, appetite changes; BMI, body mass index; Conc, concentration difficulty; CRP, C‐reactive protein; Fati, fatigue; Fgn, fibrinogen; HDL, high density lipoprotein; Lost_int, loss of interest; Motor, psychomotor problems; Phys, physical activity; Sad, feelings of sadness; Slp, sleep disturbance; Worth, feelings of worthlessness; Smok, current smoking status; Suic, suicidal ideation; TG, triglycerides; WBC, white blood cell.

TABLE 1.

CRP predictability and CRP‐symptom edge weights/stability in nonregularized models.

CRP predictability Loss of interest Feelings of sadness Sleep disturbance Fatigue Appetite changes Feelings of worthlessness Concentration difficulty Psychomotor problems Suicidal ideation
Model w/o covariates 1.3% X X X 0.042/89.0% 0.047/85.0% X −0.056/97.0% 0.017/71.0% X
Model w/covariates 27.7% X X X X X X −0.038/63.0% X X
Open in a new tab

Note: X, no edge between CRP and particular symptom criteria, % of 100 bootstraps for which the edge weight was non‐zero. Model w/covariates: adjusting for age, sex, alcohol use, current smoking status, physical activity, fibrinogen, white blood cell, body mass index, high density lipoprotein, triglycerides, and antidepressants use.

Abbreviation: CPR, C‐reactive protein.

After controlling for all covariates (Figure 1b, Table 1), CRP maintained its association only with concentration difficulty (edge weight = −0.038; non‐zero in 63.0% of bootstraps). The predictability for CRP increased to 27.7%, as it shares more variance with other nodes. Although other depression symptoms were not associated with CRP, interesting associations with certain covariates were observed. Smoking status was proximal to several symptom criteria including sleep disturbance, appetite changes, and psychomotor problems (edge weights = 0.080, 0.070, and 0.059; non‐zero in 62.0%, 65.0%, and 73.0% of bootstraps, respectively). Antidepressant use was associated with feelings of sadness, fatigue, appetite changes, concentration difficulty, psychomotor problems, and suicidal ideation (edge weights = 0.079, 0.101, 0.07, 0.054, 0.048, and 0.035; non‐zero in 72.0%, 82.0%, 75.0%, 68.0%, 61.0%, and 59.0% of bootstraps, respectively). Further, CRP was connected to BMI (edge weight = 0.041; non‐zero in 56.0% of bootstraps).

The visualizations of the MNMs (one featuring CRP group as a moderator, the other featuring continuous CRP) are presented in Figure 2 and Figure 3. Supplemental Figures S1 and S2 present the bootstrapped sampling distributions of the MNMs. The moderated edge weights that are illustrated can be interpreted as standardized β coefficients, since the mgm z‐standardizes and centers all continuous nodes before computing edge weights. When conditioning the network by CRP group (<3.0 mg/L, ≥3.0 mg/L; Figure 2), the appetite changes–psychomotor problems edge was stronger (i.e., more positive) in the elevated CRP group (≥3.0 mg/L) (moderated weight = 0.147). The sleep disturbance–appetite changes edge was also stronger in the elevated CRP group (moderated weight = 0.046). Stability analyses found that these moderated edges were non‐zero in 86.0% and 53.0% of bootstraps, respectively. When conditioning the network at three levels of CRP (−1SD, mean, and +1SD; Figure 3), a positive progressive change in the edge between appetite changes and psychomotor problems was clearly evident: at low CRP levels, this relationship was weak and at high levels of CRP it was strong (moderated weight = 0.089; non‐zero in 79.0% of bootstraps). A positive progressive change also emerged for the edge between sleep disturbance and psychomotor problems (moderated weight = 0.033; non‐zero in 54.0% of bootstraps).

FIGURE 2.

FIGURE 2

Open in a new tab

Moderated network models conditioned on CRP group (<3.0 mg/L, left panel; ≥3.0 mg/L, right panel). As CRP is the moderator being conditioned on in the network, there are no edges connected to CRP and it is simply included in the visualization of the models to indicate its role as the moderator. Green edges represent positive linear relations. App, appetite changes; Conc, concentration difficulty; CRP, C‐reactive protein; Fati, fatigue; Lost_int, loss of interest; Motor, psychomotor problems; Sad, feelings of sadness; Slp, sleep disturbance; Worth, feelings of worthlessness; Suic, suicidal ideatio.

FIGURE 3.

FIGURE 3

Open in a new tab

Moderated network models conditioned on three CRP levels. Network is depicted at −1SD, mean, and +1SD of CRP level. As CRP is the moderator being conditioned on in the network, there are no edges connected to CRP and it is simply included in the visualization of the models to indicate its role as the moderator. Green edges represent positive linear relations. App, appetite changes; Conc, concentration difficulty; CRP, C‐reactive protein; Fati, fatigue; Lost_int, loss of interest; Motor, psychomotor problems; Sad, feelings of sadness; Slp, sleep disturbance; Suic, suicidal ideation; Worth, feelings of worthlessness.

The NCT indicated a significant difference in global strength between the non‐elevated and elevated CRP group: S = 0.65 (<3.0 mg/L, S = 3.75; ≥3.0 mg/L, S = 4.40), p = .029. The global strength invariance test result is plotted in Supplemental Figure S3. However, both networks were invariant in terms of network structure (M = 0.22, p = .338). Considering this, differences between edge strengths were not tested. Further, centrality strength did not differ between the two groups (C = 0.18, p = .479).

All the results from the analysis regarding the listwise deletion methods (main analysis) elaborated above largely mirror the analysis concerning multiple imputation, except for the slightly varying associations between covariates found in MGMs (refer Supplement). This ensures the robustness of the data.

4. DISCUSSION

Using various network approaches, we identified specific depression symptoms that are uniquely influenced by CRP and demarcated CRP‐specific moderation effects on several symptom‐to‐symptom relations among patients with stable CHD. Further, we observed differences in overall depression symptom connectivity between non‐elevated and elevated CRP groups. Such CRP‐related variability in depression symptom profiles may explain the inconsistencies that have been previously reported on the CRP–depression relationship in this patient group.

In the network model without covariates, CRP was correlated with fatigue and appetite changes. Prior network research has also found these two symptoms to be particularly characterized by an inflammatory response relative to other depression criteria (Fried et al., 2020; Kappelmann et al., 2021; Lee et al., 2023a; Manfro et al., 2022; Moriarity, Horn, et al., 2021). Other studies using medical populations also underscored the role of CRP on fatigue (Booker et al., 2009; Orre et al., 2011). Indeed, basic science research of neural‐immune signaling has proposed that inflammatory processes induce “sickness behavior”—a constellation of behavioral changes that include fatigue, appetite changes, or depression‐like symptoms—via the effects of proinflammatory cytokines on the central nervous system (Dantzer et al., 2008; Eisenberger et al., 2010; Irwin & Cole, 2011).

The edge between CRP and psychomotor problems was also noticeable, although the edge weight was somewhat weak. These findings extend the previous work by (Moriarity, van Borkulo, et al., 2021), which reported that psychomotor problems had a significantly higher expected influence (i.e., a measure of overall connectivity in networks with both positive and negative edges) in an elevated CRP group (≥3.0 mg/L) among a representative adult sample. A review of the evidence (Majd et al., 2020) also provided support for a link between CRP and psychomotor problems, specifically psychomotor retardation which has both motor and cognitive components. Further, prior research has consistently shown that CRP is uniquely associated with a change in the attention‐executive‐psychomotor domain relative to other cognitive domains in patients with pre‐existing, stable CHD (Hoth et al., 2008; Weinstein et al., 2017).

Yet it is important to note that CRP–fatigue, CRP–appetite changes, and CRP–psychomotor problems edge disappeared after adjusting for covariates, indicating that some covariates contribute to these associations. Specifically, current smoking status and antidepressant use showed connections with appetite changes, fatigue, or psychomotor problems. Although it remains a challenge to draw any causal or mediating inferences regarding these covariates in the CRP–depression relationship due to the nature of network methodology, this issue can be addressed as an extension of the presented work.

Interestingly, CRP maintained its negative association with concentration difficulty regardless of covariate adjustment. This result is somewhat counterintuitive to prior evidence, which underscores the role of CRP in cognitive performance among patients with stable CHD. Generally, people with pre‐existing CHD (compared with the general population) have a greater likelihood of future cognitive impairment (Muller et al., 2007; Singh‐Manoux et al., 2008) that can be partially explained by the presence of chronic, systematic inflammation related to the underlying atherosclerotic process (Casserly & Topol, 2004). One possible mechanism proposes that circulating CRP can negatively impact the vasculature by altering vascular homeostasis and promoting coagulation (Grad & Danenberg, 2013). Studies have also found that CRP may reflect localized central nervous system inflammation, including cerebral small vessel disease (Hilal et al., 2018), reduced cerebral blood flow (Warren et al., 2018), and vascular dementia (Miralbell et al., 2013), all of which can affect cognitive function (Redwine et al., 2022). Yet, discussion around our findings should be treated with caution; the present study is unable to determine if concentration difficulty in depression reflects an independent dysfunction related to cognitive impairment among patients with stable CHD. Further accumulation of evidence is needed to expand this interpretation.

The MNMs prompted us to reanalyze our data. Especially, in both group comparisons and continuous statistical approaches, the appetite changes–psychomotor problems edge was particularly stronger in the higher CRP groups. It is possible that, among CHD patients with a high inflammatory state, psychomotor problems may reflect cognitive slowing/dysfunction, and these changes may have been influenced by a loss of appetite—an association that has been underscored by previous studies (i.e., a strong association between appetite loss symptoms and neurocognitive performance; Potter et al., 2015; Saha et al., 2016). Yet, a precise conclusion cannot be put forth at present, because “appetite changes” and “psychomotor problems” items on the PHQ‐9 are double‐barreled, (i.e., they combined two different questions). Many argue that CRP is more specific to increased, rather than decreased, appetite because it is released in adipose tissue (Hickman et al., 2014; Lamers et al., 2018; Simmons et al., 2020). Indeed, in our study, CRP was positively correlated with BMI.

The sleep disturbance–appetite changes edge was also stronger (i.e., more positive) in the elevated CRP group (≥3.0 mg/L). A positive, progressive change also emerged for the edge between sleep disturbance and psychomotor problems when CRP level was a continuous moderator. Again, we do not know what aspects of sleep disturbance (sleep deprivation or oversleep) are driving these associations, limiting our interpretation. Yet, the finding that CRP moderates some symptom–symptom associations may provide researchers with valuable information on differential symptom‐level risk/maintenance mechanisms among people with stable CHD who have varying levels of CRP (Moriarity, van Borkulo, et al., 2021).

NCT revealed that the elevated CRP group had more strengthened associations between individual symptoms than the non‐elevated CRP group, which is consistent with the findings of the study of Moriarity, van Borkulo, et al. (2021). Viewed from the network perspective, a more strongly and densely connected network can signify that an individual feels “trapped” in the disordered state (i.e., depression). More densely connected networks will feature strong feedback among their symptoms and, therefore, can be associated with a higher vulnerability to depression and a less positive prognosis (van Borkulo et al., 2015). However, network structure (accordingly one‐to‐one edge strength) did not statistically significantly differ between the groups, although we noted certain strong symptom‐to‐symptom connections in the elevated CRP group while using the moderated network analysis.

Yet, the results from the moderated network analysis and NCT should be interpreted with extreme caution as we did not control covariates that may contribute to the complex interplay between CRP and depression. Potential ways to include covariates in MNMs have been briefly discussed by Haslbeck (2019) (i.e., specifying a particular set of moderation effects in the model). However, identifying very minimal moderation effects requires a large sample (i.e., significant power; Haslbeck et al., 2021), which can be evidenced by the relatively small and less stable moderation effects found in this study. The MNMs include many more parameters, thereby decreasing the power (Moriarity, van Borkulo, et al., 2021). Given our relatively small sample, it was difficult to consider the number of covariates in the MNM. In this regard, the methods we applied should be seen as a humble contribution to the literature. Additionally, NCT is a relativity new method that has not been validated for non‐continuous variables such as binary, which partially limited our ability to include certain types of covariates in the model (van Borkulo et al., 2022). More importantly, leveraging the covariates as nodes would compare CRP–group differences in symptom and covariate networks, instead of only symptom networks (Moriarity, van Borkulo, et al., 2021). One solution is that researchers may use the propensity score matching technique (Rosenbaum & Rubin, 1983) to account for baseline differences and create comparable groups, and subsequently perform the NCT, as in the study of Lee and Hu (2022).

4.1. Limitations

This study has several limitations. First, as a cross‐sectional study using the undirected network models, we are limited in our ability to test the directionality of the associations among variables. In the future, longitudinal designs of network analysis (e.g., cross‐lagged panel network models; Rhemtulla et al., in press) can be used to examine how the relationship between CRP and depression symptoms unfolds over time. Second, the generalizability of the findings is limited, as the recruitment was from only one site and the patients comprised of mostly males. Third, we could not incorporate other important covariates into analyses due to data limitations. For instance, other than antidepressant use, we failed to consider other essential clinical or treatment covariates. Especially for CHD patients, controlling the use of statin, a frontline therapy for regulating hyperlipidemia and CHD risk that is known to demonstrate anti‐inflammatory effects independent of lipid‐lowering action (Li et al., 2019), may be critical. Further, CHD is regarded as a complex multifactorial disease that is closely associated with environmental, genetic, and other metabolic risk factors, all of which should be closely inspected as potential covariates in future investigations. Fourth, we used an established CRP cut‐off value (3.0 mg/L) recommended by the American Heart Association for the group differences analyses. However, CHD is itself a pro‐inflammatory state, and participants in the Heart and Soul Study are older and have more comorbid illnesses than those in other studies, which likely contributed to a more chronically elevated state of inflammation (mean CRP in the current study = 4.6 mg/L). As such, another cut‐off value might have better demarcated the level at which CRP moderated depression symptoms. Finally, even though it is not uncommon for depression scales to have minimal content overlap, having different types of content in scales may produce different psychopathology networks. Research leveraging beyond DSM‐5 dimensions will be also crucial to more accurately reflect the heterogeneity of depression and its underlying etiology.

Some methodological limitations should be noted. First, we used cross‐validation to select the regularization parameter. Yet, there are numerous methods for choosing network models, and the sensitivity and specificity of each approach remain the subject of ongoing research. A cross‐validation approach increases the possibility that non‐zero estimated edges approximate the true network population, but at the risk of lower specificity (Epskamp & Fried, 2018). Contrastingly, one simulation study demonstrated that cross‐validation yielded both the highest sensitivity and specificity under various conditions as compared to other regularization‐based parameter selection criteria (Wysocki & Rhemtulla, 2021). Thus, interpretation should be cautious. Second, the edge weights observed between CRP and depression symptoms were generally small, as observed in other studies (Fried et al., 2020; Kappelmann et al., 2021; Lee et al., 2023a; Moriarity, Horn, et al., 2021, Moriarity, van Borkulo, et al., 2021). One reason may be that the blood assay and self‐report estimates are prone to downward bias due to measurement–domain‐specific variance, thereby, attenuating power (Moriarity & Alloy, 2021). Moreover, network analysis is affected by the sample size. Although there are no “rules of thumb” for sample size requirements for network analysis, generally, larger sample sizes lead to easier recovery of edges in the network (Epskamp et al., 2018).

In conclusion, the current study encourages researchers to approach the subject on CRP–depression relationship among patients with stable CHD with sophisticated research methodology. Yet, we reemphasize that this study is an exploratory, preliminary investigation. More research is warranted to recognize specific profiles for depression wherein CRP confers risk among this patient group and address the limitations faced by the current study.

AUTHOR CONTRIBUTIONS

Chiyoung Lee: Conceptualization, Formal analysis, Methodology, Writing – original draft. Mary A. Whooley: Data curation, Resources, Supervision, Writing – review & editing.

CONFLICT OF INTEREST

The authors declare there are no conflicts of interest.

ETHICS STATEMENT

For the present secondary data analysis, the first author received the data from the principal investigator of the Heart and Soul Study after receiving approval from University of Washington (approval no. STUDY00015032). All methods and procedures were conducted in accordance with the relevant guidelines and regulations. The Heart and Soul Study was approved by the following institutional review boards: the Committee on Human Research at the University of California, San Francisco; the Research and Development Committee at the San Francisco Veterans Affairs Medical Center; the Medical Human Subjects Committee at Stanford University; the Human Subjects Committee at the Veterans Affairs Palo Alto Health Care System; and the Data Governance Board of the Community Health Network of San Francisco.

CONSENT TO PARTICIPATE

All participants provided written informed consent.

Supporting information

Supporting Information S1

Click here for additional data file. (2.5MB, docx)

ACKNOWLEDGMENTS

None.

Lee, C. , & Whooley, M. A. (2023). Networks of C‐reactive protein and depression symptoms in patients with stable coronary heart disease: Findings from the Heart and Soul Study. International Journal of Methods in Psychiatric Research, 32(4), e1968. 10.1002/mpr.1968

DATA AVAILABILITY STATEMENT

The datasets used and analyzed during the current study are available by submitting a proposal request to the Heart and Soul Study team (https://heartandsoulstudy.ucsf.edu/).

REFERENCES

  1. Annique, S. , Dorien, T. , Richel, L. , Gunter, K. , Joris, D. , Harry, J. C. , Gert, G. , Frank, S. , Michael, M. , & Adriaan, H. (2005). Inflammatory markers in depressed post‐myocardial infarction patients. Journal of Psychiatric Research, 39(2), 137–144. 10.1016/j.jpsychires.2004.05.009 [DOI] [PubMed] [Google Scholar]
  2. Bankier, B. , Barajas, J. , Martinez‐Rumayor, A. , & Januzzi, J. L. (2009). Association between major depressive disorder and C‐reactive protein levels in stable coronary heart disease patients. Journal of Psychosomatic Research, 66(3), 189–194. 10.1016/j.jpsychores.2008.09.010 [DOI] [PubMed] [Google Scholar]
  3. Booker, R. , Olson, K. , Pilarski, L. M. , Noon, J. P. , & Bahlis, N. J. (2009). The relationships among physiologic variables, quality of life, and fatigue in patients with multiple myeloma. Oncology Nursing Forum, 36(2), 209–216. 10.1188/09.ONF.209-216 [DOI] [PubMed] [Google Scholar]
  4. Borsboom, D. , & Cramer, A. O. (2013). Network analysis: An integrative approach to the structure of psychopathology. Annual Review of Clinical Psychology, 9(1), 91–121. 10.1146/annurev-clinpsy-050212-185608 [DOI] [PubMed] [Google Scholar]
  5. Bush, K. , Kivlahan, D. R. , McDonell, M. B. , Fihn, S. D. , Bradley, K. A. , & Ambulatory Care Quality Improvement Project . (1998). The AUDIT alcohol consumption questions (AUDIT‐C): An effective brief screening test for problem drinking. Archives of Internal Medicine, 158(16), 1789–1795. 10.1001/archinte.158.16.1789 [DOI] [PubMed] [Google Scholar]
  6. Casserly, I. , & Topol, E. J. (2004). Convergence of atherosclerosis and Alzheimer's disease: Inflammation, cholesterol, and misfolded proteins. The Lancet, 363(9415), 1139–1146. 10.1016/S0140-6736(04)15900-X [DOI] [PubMed] [Google Scholar]
  7. Dantzer, R. , O'connor, J. C. , Freund, G. G. , Johnson, R. W. , & Kelley, K. W. (2008). From inflammation to sickness and depression: When the immune system subjugates the brain. Nature Reviews Neuroscience, 9(1), 46–56. 10.1038/nrn2297 [DOI] [PMC free article] [PubMed] [Google Scholar]
  8. Eisenberger, N. I. , Inagaki, T. K. , Mashal, N. M. , & Irwin, M. R. (2010). Inflammation and social experience: An inflammatory challenge induces feelings of social disconnection in addition to depressed mood. Brain, Behavior, and Immunity, 24(4), 558–563. 10.1016/j.bbi.2009.12.009 [DOI] [PMC free article] [PubMed] [Google Scholar]
  9. Epskamp, S. , & Fried, E. I. (2018). A tutorial on regularized partial correlation networks. Psychological Methods, 23(4), 617–634. 10.31234/OSF.IO/236W8 [DOI] [PubMed] [Google Scholar]
  10. Epskamp, S. , Borsboom, D. , & Fried, E. I. (2018). Estimating psychological networks and their accuracy: A tutorial paper. Behavior Research Methods, 50(1), 195–212. 10.3758/s13428-017-0862-1 [DOI] [PMC free article] [PubMed] [Google Scholar]
  11. Felger, J. C. , Li, Z. , Haroon, E. , Woolwine, B. J. , Jung, M. Y. , Hu, X. , & Miller, A. H. (2016). Inflammation is associated with decreased functional connectivity within corticostriatal reward circuitry in depression. Molecular Psychiatry, 21(10), 1358–1365. 10.1038/mp.2015.168 [DOI] [PMC free article] [PubMed] [Google Scholar]
  12. Fried, E. I. , & Nesse, R. M. (2015). Depression sum‐scores don’t add up: Why analyzing specific depression symptoms is essential. BMC Medicine, 13(1), 1–11. 10.1186/s12916-015-0325-4 [DOI] [PMC free article] [PubMed] [Google Scholar]
  13. Fried, E. I. , Nesse, R. M. , Zivin, K. , Guille, C. , & Sen, S. (2014). Depression is more than the sum score of its parts: Individual DSM symptoms have different risk factors. Psychological Medicine, 44(10), 2067–2076. 10.1017/S0033291713002900 [DOI] [PMC free article] [PubMed] [Google Scholar]
  14. Fried, E. I. , Von Stockert, S. , Haslbeck, J. M. B. , Lamers, F. , Schoevers, R. A. , & Penninx, B. W. J. H. (2020). Using network analysis to examine links between individual depressive symptoms, inflammatory markers, and covariates. Psychological Medicine, 50(16), 2682–2690. 10.1017/S0033291719002770 [DOI] [PubMed] [Google Scholar]
  15. Grad, E. , & Danenberg, H. D. (2013). C‐Reactive protein and atherothrombosis: Cause or effect? Blood Reviews, 27(1), 23–29. 10.1016/j.blre.2012.12.001 [DOI] [PubMed] [Google Scholar]
  16. Haslbeck (2019). Moderated network models for continuous data. https://jonashaslbeck.com/Moderated‐Network‐Models‐for‐continuous‐data/
  17. Haslbeck, J. M. , & Waldorp, L. J. (2015). mgm: Structure estimation for time‐varying mixed graphical models in high‐dimensional data. 10.48550/arXiv.1510.06871 [DOI]
  18. Haslbeck, J. M. , Borsboom, D. , & Waldorp, L. J. (2021). Moderated network models. Multivariate Behavioral Research, 56(2), 256–287. 10.1080/00273171.2019.1677207 [DOI] [PubMed] [Google Scholar]
  19. Hickman, R. J. , Khambaty, T. , & Stewart, J. C. (2014). C‐Reactive protein is elevated in atypical but not nonatypical depression: Data from the National Health and Nutrition Examination Survey (NHANES) 1999–2004. Journal of Behavioral Medicine, 37(4), 621–629. 10.1007/s10865-013-9510-0 [DOI] [PubMed] [Google Scholar]
  20. Hilal, S. , Ikram, M. A. , Verbeek, M. M. , Franco, O. H. , Stoops, E. , Vanderstichele, H. , Niessen, W. J. , & Vernooij, M. W. (2018). C‐reactive protein, plasma amyloid‐β levels, and their interaction with magnetic resonance imaging markers. Stroke, 49(11), 2692–2698. 10.1161/STROKEAHA.118.022317 [DOI] [PubMed] [Google Scholar]
  21. 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 [DOI] [PMC free article] [PubMed] [Google Scholar]
  22. Hoth, K. F. , Haley, A. P. , Gunstad, J. , Paul, R. H. , Poppas, A. , Jefferson, A. L. , Tate, D. F. , Ono, M. , Jerskey, B. A. , & Cohen, R. A. (2008). Elevated C‐reactive protein is related to cognitive decline in older adults with cardiovascular disease. Journal of the American Geriatrics Society, 56(10), 1898–1903. 10.1111/j.1532-5415.2008.01930.x [DOI] [PMC free article] [PubMed] [Google Scholar]
  23. Irwin, M. R. , & Cole, S. W. (2011). Reciprocal regulation of the neural and innate immune systems. Nature Reviews Immunology, 11(9), 625–632. 10.1038/nri3042 [DOI] [PMC free article] [PubMed] [Google Scholar]
  24. Janszky, I. , Lekander, M. , Blom, M. , Georgiades, A. , & Ahnve, S. (2005). Self‐rated health and vital exhaustion, but not depression, is related to inflammation in women with coronary heart disease. Brain, Behavior, and Immunity, 19(6), 555–563. 10.1016/j.bbi.2005.01.001 [DOI] [PubMed] [Google Scholar]
  25. Jokela, M. , Virtanen, M. , Batty, G. D. , & Kivimäki, M. (2016). Inflammation and specific symptoms of depression. JAMA Psychiatry, 73(1), 87–88. 10.1001/jamapsychiatry.2015.1977 [DOI] [PubMed] [Google Scholar]
  26. Kappelmann, N. , Czamara, D. , Rost, N. , Moser, S. , Schmoll, V. , Trastulla, L. , Stochl, J. , Lucae, S. , Binder, E. B. , Khandaker, G. M. , Arloth, J. , & CHARGE inflammation working group . (2021). Polygenic risk for immuno‐metabolic markers and specific depressive symptoms: A multi‐sample network analysis study. Brain, Behavior, and Immunity, 95, 256–268. 10.1016/j.bbi.2021.03.024 [DOI] [PubMed] [Google Scholar]
  27. Khuseyinova, N. , & Koenig, W. (2006). Biomarkers of outcome from cardiovascular disease. Current Opinion in Critical Care, 12(5), 412–419. 10.1097/01.ccx.0000244119.16377.75 [DOI] [PubMed] [Google Scholar]
  28. Lamers, F. , Milaneschi, Y. , De Jonge, P. , Giltay, E. J. , & Penninx, B. W. J. H. (2018). Metabolic and inflammatory markers: Associations with individual depressive symptoms. Psychological Medicine, 48(7), 1102–1110. 10.1017/S0033291717002483 [DOI] [PubMed] [Google Scholar]
  29. Lee, C. , & Hu, X. (2022). Sex differences in depressive symptom networks among community‐dwelling older adults. Nursing Research, 71(5), 370–379. 10.1097/NNR.0000000000000601 [DOI] [PubMed] [Google Scholar]
  30. Lee, C. , & Min, S. H. (2023b). Racial differences in C‐reactive protein, depression symptoms, and social relationships in older adults: A moderated network analysis. Biological Research For Nursing, 109980042311577. 10.1177/10998004231157767 [DOI] [PubMed] [Google Scholar]
  31. Lee, C. , Min, S. H. , & Niitsu, K. (2023a). C‐Reactive protein and specific depression symptoms among older adults: An exploratory investigation of multi‐plane networks using cross‐sectional data from NHANES (2017–2020). Biological Research For Nursing, 25(1), 14–23. 10.1177/10998004221110602 [DOI] [PubMed] [Google Scholar]
  32. Li, M. , Wang, X. , Li, X. , Chen, H. , Hu, Y. , Zhang, X. , Tang, X. , Miao, Y. , Tian, G. , & Shang, H. (2019). Statins for the primary prevention of coronary heart disease. BioMed Research International, 2019, 4870350–4870415. 10.1155/2019/4870350 [DOI] [PMC free article] [PubMed] [Google Scholar]
  33. Mac Giollabhui, N. , Ellman, L. M. , Coe, C. L. , Byrne, M. L. , Abramson, L. Y. , & Alloy, L. B. (2020). To exclude or not to exclude: Considerations and recommendations for C‐reactive protein values higher than 10 mg/L. Brain, Behavior, and Immunity, 87, 898–900. 10.1016/j.bbi.2020.01.023 [DOI] [PMC free article] [PubMed] [Google Scholar]
  34. Majd, M. , Saunders, E. F. , & Engeland, C. G. (2020). Inflammation and the dimensions of depression: A review. Frontiers in Neuroendocrinology, 56, 100800. 10.1016/j.yfrne.2019.100800 [DOI] [PubMed] [Google Scholar]
  35. Manfro, P. H. , Anselmi, L. , Barros, F. , Gonçalves, H. , Murray, J. , Oliveira, I. O. , Tovo‐Rodrigues, L. , Wehrmeister, F. C. , Menezes, A. M. , Mondelli, V. , Rohde, L. A. , & Kieling, C. (2022). Youth depression and inflammation: Cross‐sectional network analyses of C‐Reactive protein, interleukin‐6 and symptoms in a population‐based sample. Journal of Psychiatric Research, 150, 197–201. 10.1016/j.jpsychires.2022.03.065 [DOI] [PubMed] [Google Scholar]
  36. Miralbell, J. , López‐Cancio, E. , López‐Oloriz, J. , Arenillas, J. F. , Barrios, M. , Soriano‐Raya, J. J. , Galan, A. , Caceres, C. , Alzamora, M. , Pera, G. , Toran, P. , Davalos, A. , & Mataró, M. (2013). Cognitive patterns in relation to biomarkers of cerebrovascular disease and vascular risk factors. Cerebrovascular Diseases, 36(2), 98–105. 10.1159/000352059 [DOI] [PubMed] [Google Scholar]
  37. Moriarity, D. P. , & Alloy, L. B. (2021). Back to basics: The importance of measurement properties in biological psychiatry. Neuroscience & Biobehavioral Reviews, 123, 72–82. 10.1016/j.neubiorev.2021.01.008 [DOI] [PMC free article] [PubMed] [Google Scholar]
  38. 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 [DOI] [PMC free article] [PubMed] [Google Scholar]
  39. Moriarity, D. P. , van Borkulo, C. , & Alloy, L. B. (2021). Inflammatory phenotype of depression symptom structure: A network perspective. Brain, Behavior, and Immunity, 93, 35–42. 10.1016/j.bbi.2020.12.005 [DOI] [PMC free article] [PubMed] [Google Scholar]
  40. Muller, M. , Grobbee, D. E. , Aleman, A. , Bots, M. , & van der Schouw, Y. T. (2007). Cardiovascular disease and cognitive performance in middle‐aged and elderly men. Atherosclerosis, 190(1), 143–149. 10.1016/j.atherosclerosis.2006.01.005 [DOI] [PubMed] [Google Scholar]
  41. Nikkheslat, N. , Zunszain, P. A. , Horowitz, M. A. , Barbosa, I. G. , Parker, J. A. , Myint, A. M. , Schwarz, M. J. , Tylee, A. T. , Carvalho, L. A. , & Pariante, C. M. (2015). Insufficient glucocorticoid signaling and elevated inflammation in coronary heart disease patients with comorbid depression. Brain, Behavior, and Immunity, 48, 8–18. 10.1016/j.bbi.2015.02.002 [DOI] [PubMed] [Google Scholar]
  42. Orre, I. J. , Reinertsen, K. V. , Aukrust, P. , Dahl, A. A. , Fosså, S. D. , Ueland, T. , & Murison, R. (2011). Higher levels of fatigue are associated with higher CRP levels in disease‐free breast cancer survivors. Journal of Psychosomatic Research, 71(3), 136–141. 10.1016/j.jpsychores.2011.04.003 [DOI] [PubMed] [Google Scholar]
  43. 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(7), 887–897. 10.1016/j.bbi.2009.04.005 [DOI] [PMC free article] [PubMed] [Google Scholar]
  44. Potter, G. G. , McQuoid, D. R. , & Steffens, D. C. (2015). Appetite loss and neurocognitive deficits in late‐life depression. International Journal of Geriatric Psychiatry, 30(6), 647–654. 10.1002/gps.4196 [DOI] [PMC free article] [PubMed] [Google Scholar]
  45. Redwine, L. S. , Hong, S. , Kohn, J. , Martinez, C. , Hurwitz, B. E. , Pung, M. A. , Wilson, K. , Pruitt, C. , Greenberg, B. H. , & Mills, P. J. (2022). Systemic inflammation and cognitive decrements in patients with stage B heart failure. Psychosomatic Medicine, 84(2), 133–140. 10.1097/PSY.0000000000001033 [DOI] [PubMed] [Google Scholar]
  46. Rhemtulla, M ., van Bork, R ., & Cramer, A. O. J . (in press). Cross‐lagged network models. Multivariate Behavioral Research.  [Google Scholar]
  47. Rosenbaum, P. R. , & Rubin, D. B. (1983). The central role of the propensity score in observational studies for causal effects. Biometrika, 70(1), 41–55. 10.1093/biomet/70.1.41 [DOI] [Google Scholar]
  48. Ruo, B. , Rumsfeld, J. S. , Hlatky, M. A. , Liu, H. , Browner, W. S. , & Whooley, M. A. (2003). Depressive symptoms and health‐related quality of life: The Heart and Soul Study. JAMA, 290(2), 215–221. 10.1001/jama.290.2.215 [DOI] [PMC free article] [PubMed] [Google Scholar]
  49. Saha, S. , Hatch, D. J. , Hayden, K. M. , Steffens, D. C. , & Potter, G. G. (2016). Appetite and weight loss symptoms in late‐life depression predict dementia outcomes. American Journal of Geriatric Psychiatry, 24(10), 870–878. 10.1016/j.jagp.2016.05.004 [DOI] [PMC free article] [PubMed] [Google Scholar]
  50. Sforzini, L. , Pariante, C. M. , Palacios, J. E. , Tylee, A. , Carvalho, L. A. , Viganò, C. A. , & Nikkheslat, N. (2019). Inflammation associated with coronary heart disease predicts onset of depression in a three‐year prospective follow‐up: A preliminary study. Brain, Behavior, and Immunity, 81, 659–664. 10.1016/j.bbi.2019.07.023 [DOI] [PubMed] [Google Scholar]
  51. Simmons, W. K. , Burrows, K. , Avery, J. A. , Kerr, K. L. , Taylor, A. , Bodurka, J. , Potter, W. , Teague, T. K. , & Drevets, W. C. (2020). Appetite changes reveal depression subgroups with distinct endocrine, metabolic, and immune states. Molecular Psychiatry, 25(7), 1457–1468. 10.1038/s41380-018-0093-6 [DOI] [PMC free article] [PubMed] [Google Scholar]
  52. Singh‐Manoux, A. , Sabia, S. , Lajnef, M. , Ferrie, J. E. , Nabi, H. , Britton, A. R. , Marmot, M. G. , & Shipley, M. J. (2008). History of coronary heart disease and cognitive performance in midlife: The Whitehall II study. European Heart Journal, 29(17), 2100–2107. 10.1093/eurheartj/ehn298 [DOI] [PMC free article] [PubMed] [Google Scholar]
  53. Smith, K. J. , Au, B. , Ollis, L. , & Schmitz, N. (2018). The association between C‐reactive protein, interleukin‐6 and depression among older adults in the community: A systematic review and meta‐analysis. Experimental Gerontology, 102, 109–132. 10.1016/j.exger.2017.12.005 [DOI] [PubMed] [Google Scholar]
  54. Uher, R. , Payne, J. L. , Pavlova, B. , & Perlis, R. H. (2014). Major depressive disorder in DSM‐5: Implications for clinical practice and research of changes from DSM‐IV. Depression and Anxiety, 31(6), 459–471. 10.1002/da.22217 [DOI] [PubMed] [Google Scholar]
  55. van Borkulo, C. D. , van Bork, R. , Boschloo, L. , Kossakowski, J. J. , Tio, P. , Schoevers, R. A. , Borsboom, D. , & Waldorp, L. J. (2022). Comparing network structures on three aspects: A permutation test. Psychological Methods. Advance online publication. 10.1037/met0000476 [DOI] [PubMed] [Google Scholar]
  56. van Borkulo, C. , Boschloo, L. , Borsboom, D. , Penninx, B. W. , Waldorp, L. J. , & Schoevers, R. A. (2015). Association of symptom network structure with the course of depression. JAMA Psychiatry, 72(12), 1219–1226. 10.1001/jamapsychiatry.2015.2079 [DOI] [PubMed] [Google Scholar]
  57. Warren, K. N. , Beason‐Held, L. L. , Carlson, O. , Egan, J. M. , An, Y. , Doshi, J. , Davatzikos, C. , Ferrucci, L. , & Resnick, S. M. (2018). Elevated markers of inflammation are associated with longitudinal changes in brain function in older adults. The Journals of Gerontology: Series A, 73(6), 770–778. 10.1093/gerona/glx199 [DOI] [PMC free article] [PubMed] [Google Scholar]
  58. Weinstein, G. , Lutski, M. , Goldbourt, U. , & Tanne, D. (2017). C‐reactive protein is related to future cognitive impairment and decline in elderly individuals with cardiovascular disease. Archives of Gerontology and Geriatrics, 69, 31–37. 10.1016/j.archger.2016.11.002 [DOI] [PubMed] [Google Scholar]
  59. White, J. , Kivimäki, M. , Jokela, M. , & Batty, G. D. (2017). Association of inflammation with specific symptoms of depression in a general population of older people: The English Longitudinal Study of Ageing. Brain, Behavior, and Immunity, 61, 27–30. 10.1016/j.bbi.2016.08.012 [DOI] [PubMed] [Google Scholar]
  60. Whooley, M. A. , Caska, C. M. , Hendrickson, B. E. , Rourke, M. A. , Ho, J. , & Ali, S. (2007). Depression and inflammation in patients with coronary heart disease: Findings from the heart and Soul study. Biological Psychiatry, 62(4), 314–320. 10.1016/j.biopsych.2006.10.016 [DOI] [PMC free article] [PubMed] [Google Scholar]
  61. Williams, D. R. , & Rast, P. (2020). Back to the basics: Rethinking partial correlation network methodology. British Journal of Mathematical and Statistical Psychology, 73(2), 187–212. 10.1111/bmsp.12173 [DOI] [PMC free article] [PubMed] [Google Scholar]
  62. Wysocki, A. C. , & Rhemtulla, M. (2021). On penalty parameter selection for estimating network models. Multivariate Behavioral Research, 56(2), 288–302. 10.1080/00273171.2019.1672516 [DOI] [PubMed] [Google Scholar]

Associated Data

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

Supplementary Materials

Supporting Information S1

Click here for additional data file. (2.5MB, docx)

Data Availability Statement

The datasets used and analyzed during the current study are available by submitting a proposal request to the Heart and Soul Study team (https://heartandsoulstudy.ucsf.edu/).

Articles from International Journal of Methods in Psychiatric Research are provided here courtesy of Wiley

RESOURCES