Association of serum oleic acid level with depression in American adults: a cross-sectional study

The participants in this study were recruited from the NHANES, a major program conducted by the Centers for Disease Control and Prevention (CDC) to assess the health and nutritional status of adults and children in the United States of America. The NHANES comprises demographic, socioeconomic, health-related, and medical information. NHANES survey is approved by the research ethics review board of the Centers for Disease Control and Prevention [18]. All participants provided informed consent. The NHANES data are publicly available. There was no additional authorization or ethical review required for the release of NHANES data for this study. Further information regarding the NHANES is available on the CDC website [19].

Participants in our study were screened on the basis of the following inclusion criteria: 1) age 18 years or older, 2) participation in serum fatty acid testing and availability of fasting laboratory specimens. The exclusion criteria were as follows: 1) incomplete Patient Health Questionnaire-9 (PHQ-9) (n = 452) and 2) missing information on oleic acid level (n = 46).

Esterified fatty acids are hydrolyzed primarily from triglycerides, phospholipids and cholesteryl esters using sequential treatment with mineral acid and base in the presence of heat. Using a modification of [20], total fatty acids are hexane-extracted from the matrix (100uL serum) along with an internal standard solution containing stable isotopically-labeled fatty acids to account for recovery. The extract is derivatized with pentafluorobenzyl bromide (PFBBr) in the presence of triethylamine to form pentafluorobenzyl esters. The reaction mixture is injected onto a capillary gas chromatograph column to resolve oleic acid from other matrix constituents. Oleic acid is detected using electron capture negative-ion mass spectrometry. Oleic acid is measured using selected ion monitoring. Quantitation is accomplished by comparing the peak area of the analyte in the unknown with the peak area of a known amount in a calibrator solution. Calculations are corrected based on the peak area of the internal standard in the unknown compared with the peak area of the internal standard in the calibrator solution. Serum samples were processed, stored, and shipped to the Division of Laboratory Sciences, National Center for Environmental Health, CDC, Atlanta, GA, for testing. More information regarding laboratory techniques and quality assurance has been previously documented [21].

The 9-item PHQ-9 was used to assess depression. The PHQ-9 is well accepted as an accurate and reliable technique for depression screening [22‐24]. The PHQ-9 contains nine items regarding the frequency of depressive symptoms [25]. Each question is scored from “0” (not at all) to “3” (nearly every day). The total PHQ-9 score can vary from 0 to 27, and a score of ≥ 10 was defined as depression in this study. Based on the cutoff point, the PHQ-9 achieved a sensitivity of 85% and a specificity of 89% for detecting MDD [26].

Model covariates were selected based on a priori knowledge [11, 27]. Self-reported sociodemographic characteristics included age, sex (male/female), race/ethnicity (non-Hispanic white, Mexican American, non-Hispanic black, other Hispanic, or other race/multiple races), education level (< high school/completed high school or > high school), and marital status (married/living with partner or never married/widowed/divorced/separated).

Physical activity was assessed by vigorous physical activity (high-intensity activities and fitness and sports such as running or basketball) and moderate physical activity (e.g., brisk walking, swimming, and bicycling at a regular pace) reported by participants. Alcohol status and smoking status were used as categorical variables. Body mass index (BMI) was measured as weight (kg) divided by height (m) squared. Alcohol status was determined by answers to the question, “Have you had at least 12 alcohol drinks a year?” (yes/no). Participants were divided into never smokers (smoked < 100 cigarettes), former smokers (currently not smoking but smoked ≥ 100 cigarettes), and current smokers (≥ 100 cigarettes and currently smoking every day or some days).

Metabolic syndrome (MetS) was defined according to the updated National Cholesterol Education Program/Adult Treatment Panel III criteria for Americans [28].

Omega-3 PUFA levels were detected using electron capture negative-ion mass spectrometry. Total cholesterol levels were measured using the enzymatic method, which is a single-reagent, endpoint reaction specific for cholesterol.

We have collated the above confounding variables and their detailed definitions into Table S1.

First we searched the NHANES database for oleic acid data and PHQ-9 questionnaire data. We found that this information was collected in the 2011–2014 surveys. Next, covariate information was retrieved for participants in the 2011–2014 surveys, and the download websites for the covariates were identified. We downloaded the required data files locally and combined the data according to the unique identification number assigned to each participant by NHANES. Then, we arrange and clean the combined data to obtain the final dataset to be analysed.

Participant characteristics are expressed as means (95% CIs) for continuous variables and as proportions and percentages of the total for categorical variables. Continuous data were compared using one-way analysis of variance, and categorical data were compared using the chi-squared test. Restricted cubic spline (RCS) analysis with 5 knots (5th, 28th, 50th, 73th, and 95th percentiles) was used to characterize the shape of the association between oleic acid levels and depression. Multivariable logistic regression analysis was performed to quantify the association between oleic acid levels and depression. We used three levels of adjustment: Model 1 was adjusted for age, sex, and race/ethnicity; Model 2 was further adjusted for education level, marital status, physical activity, BMI, smoking status, and alcohol status; and Model 3 was additionally adjusted for MetS, omega-3 PUFAs, and total cholesterol. We also performed a sensitivity analysis to ensure that the results were robust. Oleic acid was converted into a categorical variable (quartiles), and the p-value for trend was calculated. This method provides a more visual representation of the changes of depression with increasing oleic acid levels [29‐32].

In total, 456 (10.2%) participants were excluded from the analyses because of missing values in their covariates. These participants were not included in the analyses that adjusted for the corresponding missing covariates. Imputations of missing data were conducted for sensitivity analyses using the missForest R package, which is a random forest-based technique that is highly computationally efficient for high-dimensional data consisting of both categorical and continuous predictors [33]. Subgroup analyses were performed using stratified multivariable logistic regression analyses.

In total, 4,459 participants were included for statistical analysis (Fig. S1). Table 1 shows the characteristics of participants according to quartiles of serum oleic acid levels (Q1, ≤ 1.54 mmol/L; Q2, > 1.54 mmol/L to 1.94 mmol/L; Q3, > 1.94 mmol/L to < 2.51 mmol/L; Q4, ≥ 2.51 mmol/L). Participants with the highest oleic acid level (Q4; ≥ 2.51 mmol/L) were more likely to be older, non-Hispanic White, married or living with the partner, and a current or former smoker; have lower physical activity; have higher BMI values, omega-3 PUFA levels, and total cholesterol levels; and were more likely to have MetS and depression than participants in the other groups (all p-value < 0.05).

A comparison of characteristics between participants included and excluded (excluded for the absence of oleic acid level data or incomplete PHQ-9) from the analysis is shown in Table S1 in the Supplementary Material. Missing data are listed in Table S2. To avoid the effect of outliers on the results, we excluded outlier patients (1.05%) from subsequent analysis. Owing to the skewed distribution of serum oleic acid levels, outliers were identified using the Huber method [34].

According to RCS, oleic acid levels were positively associated with depression (Fig. 1). Using multivariable logistic regression analysis, Table 2 quantifies the relationship between oleic acid levels and depression. Oleic acid was positively linked with depression in the crude model (OR = 1.35, 95% confidence interval [CI]: 1.16–1.57, p = 0.003). A significant correlation between oleic acid levels and depression was still detected in models after accounting for various variables (model 1: OR = 1.40, 95% CI: 1.21–1.61; model 2: OR = 1.29, 95% CI: 1.10–1.50; model 3: OR = 1.40, 95% CI: 1.03–1.90).

We also converted oleic acid from a continuous variable to a categorical variable (quartiles). In the fully adjusted Model 3, compared with participants in the first oleic acid quartile, participants in the fourth quartile were associated with an increased prevalence of depression (OR = 2.22, 95% CI: 1.04–4.73). Moreover, the p value for trend was significant in all models. We also performed sensitivity analyses using datasets containing outliers, and the results were stable (Table S4).

Sensitivity analyses using the five data sets generated by multiple imputations for missing covariates yielded results consistent with those of the primary analysis (Table S5). All covariable subgroup analyses revealed a pattern similar to the main analysis pattern (all p for interaction > 0.05) (Table S6).

According to a recent study, a PHQ-9 cutoff score of 10 may overestimate the prevalence of depression [35]. Moreover, a systematic review reports that the overall sensitivity (0.37–0.98) and selectivity (0.42–0.99) of PHQ-9 to be ranged widely in primary care samples [36]. Therefore, we also examined the correlation between serum oleic acid levels and PHQ-9 scores as a sensitivity analysis. Because of the approximate gamma distribution of the PHQ-9 score, the association of oleic acid with the PHQ-9 score was tested by generalized linear models (log-link function and a gamma distribution). Because the outcome variable could not include 0 in the model, an increase of 0.1 in the original PHQ-9 score was used as the outcome variable (0.1–27.1). We also analyzed the association between oleic acid levels and PHQ-9 score using linear regression. Analyses of the PHQ-9 score as a continuous variable were performed as follows. We found that serum oleic acid levels were positively correlated with PHQ-9 score (Table S7).