This study utilized data from the NHANES, a nationally representative program administered by the Centers for Disease Control and Prevention (CDC), which employs a complex, multistage probability sampling design. The initial analytic sample comprised 3,632 adults aged 60 years and older from the 2011-2014 NHANES cycles, which provided comprehensive and validated data on depressive symptoms, cognitive function, and anthropometric indicators, including the BRI. Participants were excluded if data on key covariates, depression status, or cognitive performance were missing. Following the application of these exclusion criteria, the final analysis included 2,580 participants (Figure 1).

All NHANES protocols were approved by the National Center for Health Statistics (NCHS) Research Ethics Review Board, and written informed consent was obtained from all participants prior to data collection. The present study was approved by the Ethics Committee of Taizhou Municipal Hospital (Approval No. LWYJ2025297). As the analysis was conducted using publicly available data, the requirement for informed consent was waived by the ethics committee.

The NHANES 2011-2014 modules comprise the Consortium to Establish a Registry for Alzheimer’s Disease (CERAD) battery for immediate word list learning and delayed word recall (CERAD Word Learning [CERAD-WL] and CERAD Delayed Recall [CERAD-DR]) [13], the Animal Fluency test (AFT) [14], and the DSST [15], which represent key cognitive domains, including memory (both immediate and delayed recall), verbal fluency (semantic memory and word retrieval), and processing speed/executive function (information processing and cognitive flexibility), respectively. These tests were chosen because they are the only standardized cognitive evaluations available for older adults in the NHANES 2011-2014 dataset.

Focusing on both immediate and delayed recall, the CERAD-WL subtest evaluates verbal learning capacity. Included in the assessment are three consecutive trials for immediate recall and one for delayed recall. Participants are given a list of ten unrelated words during each immediate trial and required to recall as many as they can right after the words are read aloud. Participants perform a delayed recall task around 10 minutes after the final immediate recall trial, recalling the words they were shown earlier without any visual prompts. In CERAD-WL, the total score can be between 0 and 40, with each of the three immediate trials permitting up to 10 points (30 points in total) and the delayed recall trial providing 10 points. For this analysis, the CERAD test was made up of the CERAD-WL and CERAD-DR cognitive performance evaluations.

In the fields of epidemiological and cognitive screening studies, the AFT is commonly used to assess executive functioning, focusing on verbal fluency and cognitive flexibility. Participants are tasked with naming as many different animals as they can within a 1-minute timeframe. Each animal correctly named earns a point, and the total score demonstrates the individual’s proficiency in semantic memory retrieval and executive control. In a variety of populations, the AFT has proven to be effective in detecting cognitive decline due to aging.

Through the evaluation of short-term memory, sustained attention, processing speed, and visual scanning, the DSST offers a detailed assessment of brain health. The DSST was administered in paper format during the examination session. Nine digit-symbol pairs were displayed in a reference key at the top of the test sheet. This key required participants to match each number in a random sequence to its corresponding symbol within a set time limit, assessing processing speed, attention, and executive function. In a span of 2 minutes, participants needed to connect 133 numbers to the appropriate symbols, receiving a point for every accurate match.

Presently, no universally accepted standard exists for the CERAD test (CERAD-WL and CERAD-DR), AFT, and DSST tests to evaluate cognitive decline. Accordingly, we utilized the 25th percentile of the score, the lowest quartile, as the cutoff, consistent with approaches in prior published studies [16,17]. To assess the robustness of our findings and address potential concerns regarding the dichotomization of data, we will also conduct a sensitivity analysis using continuous test scores, which will allow us to examine the associations without relying on the arbitrary cutoff. By summing the individual scores, a composite score was generated to reflect overall cognitive performance, where higher scores suggest improved cognitive function. Poor cognitive performance was identified by a composite score exceeding 1 standard deviation (SD) below the analytic population’s mean [18], with scores below 47 marking poor performance.

The Patient Health Questionnaire-9 (PHQ-9) represents a well-validated, 9-item self-report instrument that operationalizes diagnostic criteria for major depressive disorder as outlined in the Diagnostic and Statistical Manual of Mental Disorders, Fifth Edition. Made up of nine items, the PHQ-9 is scored from 0 (not at all) to 3 (nearly every day) to measure how often depressive symptoms were experienced in the past 2 weeks. With scores ranging from 0 to 27, higher scores denote more severe symptoms. Participants with a PHQ-9 score of 10 or higher were considered to have moderate to severe depression [19]. Kroenke et al. [20] reported that this cutoff has an 88% sensitivity and an 88% specificity for diagnosing major depression. Its concise format and dual functionality for screening and severity evaluation have established it as the most frequently used depression metric in research and clinical settings internationally.

The BRI, an innovative anthropometric measure, calculates body type based on height and WC, assessing central adiposity, where a higher score indicates more fat accumulation. Data on height and WC were obtained from the participants’ examination records. Trained health technicians, with the help of a recorder, measured body dimensions at Mobile Examination Centers to ensure accuracy. Before taking measurements, participants had to remove their clothing and shoes. In an upright position, participants had their standing height measured directly, and their WC was measured at the midpoint between the lower rib and the top of the hips. BRI was computed by 364.2-365.5*(1-[WC(m)/2π]2/[0.5*height(m)]²)^½ [4].

Potential confounding variables adjusted for in the analysis included demographic characteristics (age [continuous, years, mean±SD], sex [male/female], race/ethnicity [Mexican American/ non-Hispanic Black/non-Hispanic White/other], marital status [married or living with partner/non-married]), socioeconomic factors (education level [less than high school/high school/more than high school], poverty income ratio [PIR: <1/1-3/>3]), lifestyle behaviors (smoking status [never/former/ current], alcohol use [never/former/current]), anthropometric measures (BMI [continuous, kg/m², mean±SD]), and health conditions (hypertension [yes/no], stroke [yes/no]). Health conditions, including hypertension and stroke, were identified through self-reported physician diagnoses obtained during the household interview. According to NHANES protocol, hypertension is diagnosed when the systolic blood pressure is 130 mm Hg or higher, the diastolic blood pressure is 80 mm Hg or higher, or there is ongoing use of prescribed antihypertensive medication. In a face-to-face interview, stroke was defined by a previous diagnosis reported by the individual and confirmed by a physician. Those who replied “yes” to the question, “Has a medical professional ever told you that you suffered a stroke?” were deemed to have had a stroke. According to U.S. weight classification standards (CDC, 2021), BMI is computed by dividing a person’s weight in kilograms by their height in meters squared.

We applied the appropriate NHANES weights to account for the complex survey design, ensuring the results were representative of the U.S. adult population. Specifically, the sampling weights prescribed for the 2011-2014 combined cycles were used, combining the weights from individual survey years to account for the multi-year design. In the analysis, the strata and primary sampling units (PSU) supplied by NHANES were used to adjust for the clustering effects of the survey. The design features were vital for securing unbiased estimates and maintaining the reproducibility of our outcomes. The analyses were performed using logistic regression models that were survey-weighted, adjusting for the complex survey design by including weights, strata, and PSU. For continuous variables, weighted means with SEs were used, and categorical variables were expressed in terms of weighted proportions. We formulated a series of multivariable survey-weighted logistic regression models for each cognitive domain (CERAD-WL/DR, AFT, and DSST) to examine the relationship between depression severity and cognitive performance. Both continuous variables and categorical indicators representing clinically relevant severity levels were used to analyze PHQ-9 scores. By employing Schoenfeld residuals, we tested the proportional hazards assumption, which was satisfied for all covariates in the Cox models (all p>0.05), ensuring the models were appropriate. To minimize multicollinearity and overadjustment, we conducted a variance inflation factor (VIF) analysis before including covariates in the model, ensuring that only variables with a VIF less than 3 were included, with no variable exceeding this threshold in the final model, thereby minimizing the risk of multicollinearity and overadjustment (Supplementary Table 1). Confounders were identified using prior research and expert knowledge to understand potential influences on the relationship between depressive symptoms and cognitive performance. In order to assess the association between depressive symptoms and cognitive performance, three hierarchical regression models were constructed, with progressively adjusted covariates. The unadjusted crude model, referred to as Model 1, provided an initial estimate of the associations without controlling for potential confounders. To mitigate basic confounding factors, Model 2 was adjusted for demographic variables including age, sex, race, and marital status. To fully account for potential confounders, Model 3 added more lifestyle and health-related factors, such as drinking and smoking status, education level, PIR, BMI, hypertension, and stroke. Odds ratios (ORs) with their respective 95% confidence intervals (CIs) were used to report all estimates. To analyze potential non-linear connections between depressive symptoms and cognitive outcomes, survey-weighted restricted cubic spline (RCS) models with four knots at the 5th, 35th, 65th, and 95th percentiles of the PHQ-9 distribution were developed. To assess the robustness of our findings and mitigate potential biases from the dichotomization of data, we performed a sensitivity analysis using continuous test scores. Subgroup analyses were executed across demographic and clinical layers, and interaction terms were tested to explore effect heterogeneity. For analyzing mediation, logistic regression models were employed to estimate the mediator (BRI) and the outcome (cognitive performance). Using mediation analysis, we assessed the extent to which BRI acted as a mediator in the relationship between depressive symptoms and cognitive performance. Specifically, we evaluated the indirect effect of depression on cognitive performance through BRI, while controlling for relevant covariates such as age, sex, race/ethnicity, and health conditions. By employing bootstrap resampling with 1,000 iterations, we estimated the CIs for the indirect effects, achieving robust estimates that take into account their non-normal distribution. Chained equations were used for multivariate multiple imputation to deal with missing data, generating 20 imputed datasets with the assumption that the data were missing at random. The robustness of the primary findings was assessed through sensitivity analyses using the imputed datasets. The statistical tests were two-sided, with p-values less than 0.05 indicating statistical significance. Analyses were performed with R (version 4.3.2; R Foundation for Statistical Computing).

Table 1 details baseline characteristics based on depressive symptom severity, as measured by the PHQ-9. In the group of 2,580 participants, 75.3% reported minimal symptoms with a PHQ-9 score of 0-4, 15.5% had mild symptoms with scores from 5 to 9, and 9.1% experienced moderate to severe symptoms with scores ranging from 10 to 27. Differences in demographic, socioeconomic, behavioral, and clinical profiles were evident among the various severity groups. The burden of depressive symptoms was greater among younger individuals (mean age: 67.48 vs. 69.09 years; p=0.020), those with higher BMI (31.71 vs. 28.68 kg/m²; p<0.001), and a higher proportion of females (65.07% vs. 50.93%; p<0.001). More severe groups had a higher proportion of Non-Hispanic Black individuals (p<0.001). People experiencing more severe symptoms were also more likely to be unmarried, have lower levels of education, and belong to lower-income brackets (all p<0.001). The behavior analysis indicated that former drinking increased with symptom severity (p<0.001), but smoking status showed no significant difference (p=0.88). In clinical terms, an increased prevalence of hypertension and stroke was observed in individuals with moderate to severe depressive symptoms, with p-values of 0.040 and less than 0.001, respectively. Cognitive function, evaluated through CERAD, AFT, and DSST, showed a progressive decline with increasing symptom severity (all p<0.01).

Increased severity of depressive symptoms was consistently related to higher chances of lower cognitive performance in every domain assessed. In fully adjusted models (Model 3), each one-point increase in PHQ-9 score was significantly associated with poorer performance on the CERAD test score (OR=1.034, 95% CI: 1.000-1.070, p=0.049) (Table 2), AFT score (OR=1.046, 95% CI: 1.014-1.079, p=0.007) (Table 3), and DSST score (OR=1.080, 95% CI: 1.040-1.123, p<0.001) (Table 4).

Individuals with moderate to severe depression, as measured by a PHQ-9 score between 10 and 27, had increased odds of experiencing lower cognitive performance compared to those with minimal symptoms, who scored between 0 and 4. The most robust associations were with DSST scores (OR=2.562, 95% CI: 1.432-4.584, p=0.004) (Table 4), followed by AFT (OR=1.834, 95% CI: 1.239-2.714, p=0.005) (Table 3). Although the connection with CERAD test score in this group wasn’t statistically significant (OR=1.454, 95% CI: 0.936-2.259, p=0.090) (Table 2), the linear trend across the severity of depression categories was significant (p for trend=0.026).

RCS analyses provided additional insights into the dose-response relationships between the severity of depressive symptoms and cognitive performance across multiple domains. The CERAD test score revealed a modest linear association, where increased PHQ-9 scores were linked to higher odds of lower cognitive performance (p for overall=0.026), but the non-linear component was not significant (p for non-linear=0.303) (Figure 2A). The AFT score showed a similar pattern, where increased depressive symptoms were linearly related to lower cognitive performance (p for overall=0.002; p for non-linear= 0.483) (Figure 2B). DSST scores showed a significant non-linear relationship (overall p<0.001; non-linear p=0.012), notably indicating that the effect of depressive symptoms on processing speed and executive function may increase past a certain threshold (Figure 2C).

Analyses of subgroups showed that the link between the severity of depression and cognitive performance, as indicated by CERAD, AFT, and DSST scores, differed among various demographic and clinical groups. For CERAD, the association was statistically significant among females (OR=1.051, 95% CI: 1.016-1.087, p=0.005) and individuals with less than a high school education (OR=1.057, 95% CI: 1.001-1.115, p=0.044), with no significant interactions detected for age, sex, race, education, or other covariates (p for interaction>0.05) (Supplementary Table 2). In contrast, depression severity was significantly associated with lower AFT scores across nearly all subgroups, including both younger and older adults, males and females, and individuals of diverse racial backgrounds. Notably, the relationship was especially pronounced in participants with less than a high school education (OR=1.069, 95% CI: 1.018-1.122, p=0.009), current smokers (OR=1.142, 95% CI: 1.066-1.223, p<0.001), and non-drinkers (OR=1.115, 95% CI: 1.046-1.189, p=0.002), though none of the interaction terms reached statistical significance (Supplementary Table 3). For DSST scores, a robust and consistent association with depression severity was observed across all demographic categories, with the effect most prominent in individuals with lower educational attainment (less than high school: OR=1.126, 95% CI: 1.074-1.180, p<0.001) and those married or living with a partner (OR=1.120, 95% CI: 1.079-1.162, p<0.001) (Supplementary Table 4). Importantly, significant interaction effects were noted for race (p for interaction=0.024), education level (p for interaction=0.003), and marital status (p for interaction<0.001), suggesting that the detrimental impact of depressive symptoms on cognitive performance may be moderated by social and educational contexts (Supplementary Table 4).

Mediation analysis revealed heterogeneous effects of BRI on the relationship between depressive symptoms and domain-specific cognitive function. For the DSST, BRI statistically mediated the association between depression severity and cognitive performance, with an indirect effect of β=-0.021 (p=0.006), accounting for 7.23% of the total effect (95% CI: 1.14% to 12.46%) (Figure 3). It implies that BRI is partially responsible for the detrimental impact of depression on processing speed and executive function. In contrast, mediation was not statistically significant for the CERAD (indirect effect β=-0.005, p=0.184; proportion mediated: 6.21%) or AFT (indirect effect β=-0.002, p=0.448; proportion mediated: 4.41%) scores (Figure 3), with wide CIs indicating substantial uncertainty in the mediating role of BRI for memory and verbal fluency outcomes.

Sensitivity analyses bolstered the reliability of the link between depression severity and cognitive performance. The direction and magnitude of the associations remained consistent across all cognitive domains in multiply imputed datasets. Every one-point rise in the PHQ-9 score was consistently linked to a significant increase in the odds of lower cognitive performance, as measured by CERAD, AFT, and DSST scores (Supplementary Tables 5-7). Upon categorizing depression severity, individuals exhibiting moderate to severe symptoms (PHQ-9 score 10-27) demonstrated a notably higher risk of lower cognitive performance than those with minimal symptoms (score 0-4) (Supplementary Tables 5-7) with the strongest associations observed for DSST, followed by AFT; CERAD results remained directionally consistent though slightly attenuated. To confirm the stability of our results and resolve any concerns about categorizing cognitive performance, we executed a sensitivity analysis using continuous test scores. Model 3, which included adjustments for a full range of covariates, found that depression was still significantly associated with diminished cognitive performance. The depression group had an OR of 2.23 (95% CI: 1.33-3.79, p<0.001) compared to the non-depression group (Supplementary Table 8). The findings of this analysis were in line with the main results, revealing a significant connection between depression and lower cognitive function.

This study found that more severe depressive symptoms are associated with lower cognitive performance in older U.S. adults, with BRI, a measure of central obesity, statistically mediating this relationship. The findings suggest a role for metabolic dysfunction in the association between depression and cognitive performance, underscoring the need for further research and interventions that consider both psychological and metabolic factors to support cognitive health in aging populations.