Association of depressive symptoms with systemic immune-inflammation index and platelet parameters among survivors of myocardial infarction: a cross-sectional NHANES study enhanced by machine learning and SHAP analysis

Abstract

Background

Inflammation is implicated in the elevated risk of depressive disorder following myocardial infarction (MI), with platelets serving as a key link between thrombosis, inflammation, and depression. Although the systemic immune-inflammation index (SII), platelet count (PLT), and mean platelet volume (MPV) are readily accessible hematological parameters, their associations with post-MI depressive symptoms remain underexplored. This study investigates these relationships in MI survivors, augmented by machine learning (ML) and SHapley Additive exPlanations (SHAP) analysis for enhanced predictive insights.

Methods

This cross-sectional study utilized data from 1,352 adults with self-reported MI history in the National Health and Nutrition Examination Survey (NHANES) 2009–2020. Multivariable logistic regression, subgroup analyses, dose-response curves, and sensitivity analyses were conducted to assess independent associations between depressive symptoms (Patient Health Questionnaire-9 score ≥ 5) and log₂-transformed SII, PLT, and MPV. Additionally, 14 supervised ML algorithms were benchmarked using 5-fold cross-validation to predict depressive symptoms, with SHAP applied to the top-performing model for feature interpretability.

Results

In multivariable regression, log₂SII (OR = 1.22, 95% CI = 1.06–1.41, P = 0.0069) and log₂PLT (OR = 1.78, 95% CI = 1.30–2.43, P = 0.0003) showed positive associations with depressive symptoms, while log₂MPV did not (OR = 0.53, 95% CI = 0.24–1.17, P = 0.1150). Subgroup analyses revealed robust associations for log₂SII except in females and BMI < 25 kg/m² groups, with no significant interactions for log₂PLT or log₂MPV across demographics or comorbidities. Dose-response curves indicated positive correlations with log₂SII and log₂PLT, and an inverted U-shaped relationship with log₂MPV (inflection point: 3.04). Sensitivity analysis confirmed that the results of this study were robust either after using the methods of multiple imputation or excluding stroke participants. We also observed a strong association of log₂SII and log₂PLT with trouble sleeping, feeling tired and poor appetite, of log₂MPV with poor appetite. ML benchmarking identified Random Forest as optimal (AUC = 0.779, R² = 0.229, RMSE = 0.424), outperforming other models. SHAP analysis ranked age (15.8% impact) and log₂PLT as the top predictors, with higher values of these factors being associated with an increased likelihood of depressive symptoms, thereby reinforcing the interactions between inflammation and platelets.

Conclusions

In a nationally representative U.S. sample, elevated log₂SII and log₂PLT are independently associated with depressive symptoms in MI survivors, with an inverted U-shaped curve for log₂MPV. ML and SHAP integration corroborates and refines the predictive insights gleaned from traditional regression, highlighting age and platelet dynamics as key drivers, and supports targeted screening in vulnerable subgroups like obese or middle aged MI survivors.

Clinical trial number

Not applicable.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12888-025-07763-7.

Introduction

In recent 20 years, a number of studies have evaluated the epidemiology of myocardial infarction with depression, but the prevalence rates vary greatly (13.6%-79.5%) [1–3]. A meta-analysis of 12,315 patients reported the prevalence of depression with myocardial infarction (MI) ranging from 22.39 to 35.46% by a random effects model [4]. The frequent occurrence of depression with MI not only exerts adverse and long-term effects on their quality of life, but also leads to the poor prognosis and mortality. A meta-analysis of 29 studies found that depression might increase cardiovascular adverse outcomes by 1.6 to 2.7 times within 24 months, and it was associated with an increased risk of all-cause mortality, cardiac death, and cardiovascular events [5]. Therefore, it’s important to identify the prevalence and related factors of depression in MI survivors.

Several plausible potential mechanisms of the interaction between cardiovascular diseases and emotional disorders have been proposed including inflammatory response, oxidative stress, autonomic dysregulation, activation of the hypothalamic–pituitary–adrenal (HPA)-axis, serotonin metabolism disorder, neurotrophins pathway dysregulation, endothelial dysfunction and platelet activation. Among these, platelet activation and inflammation play a crucial role for psycho-cardiology diseases. Our previous research found that macrophage/microglia inflammation acted as pivotal pathogenic in depression after MI [6, 7]. Platelets, identified as the main source of chemokines and cytokines at the inflammatory sites, also could interact with innate and acquired immunity to promote inflammatory responses [8]. More than the function in primary haemostasis and thrombo-inflammation, platelets are increasingly recognized as a bridge between psychiatric and clot-related disorders. Platelets are the main transporters of serum serotonin, as well as the main source and storage cells of peripheral brain-derived neurotrophic factor (BDNF) [9], providing evidence for a platelet pathophysiological involvement in depression. Therefore, some studies have regarded platelet activation and inflammation as significant pathomechanism by which coronary-clot diseases promote depressive disorders.

The systemic immune-inflammation index (SII), an available and novel inflammatory biomarker, has prognostic clues for cardiovascular diseases [10], and is also demonstrated the strong link with post-stroke depression [11], diabetic depression [12], depression in corona virus disease 2019 (COVID-19) survivors [13], and bipolar disorder [14]. The mean platelet volume (MPV), known as a marker of platelet activity, as well as a prognostic factor in many inflammatory diseases, is connected with an increased risk of clot formation [15]. Some studies have shown an positively independent association with depression [16], post-stroke depression [17], and autoimmune depression [16]. In contrast, there has been relatively little research on depression and platelet count (PLT). Indeed, there have been many assessments of platelet and inflammatory markers in depressed patients without cardiovascular diseases. In clinical, depression is a common problem in patients with cardiovascular disease. But it is regrettable that there is little published data on platelet and inflammatory indicators in depressed cardiac patients. We examined platelet parameters and inflammatory parameters in a group of MI patients from representative samples of adults in US, attempting to define the relationship of PLT, MPV and SII with depressive symptoms in patients with known cardiovascular disease.

Recent advances in machine learning (ML) offer promising avenues for enhancing the predictive utility of these inflammatory and platelet-related biomarkers in identifying post-MI depression. Ensemble algorithms, such as Random Forest (RF), have demonstrated superior performance in capturing nonlinear interactions among complex biomarker profiles and clinical covariates, achieving high discriminatory accuracy in psychiatric or cardiovascular prediction [18, 19]. When integrated with explainable AI techniques like SHAP (SHapley Additive exPlanations), these models provide interpretable insights into feature importance and directional impacts, facilitating clinical translation by highlighting key drivers like age, platelet activation, and systemic inflammation [20, 21]. Despite these advancements, the application of ML frameworks to evaluate SII, PLT, and MPV in post-MI depression remains underexplored, emphasizing the need for data-driven approaches to refine risk stratification in this vulnerable population. Here, we integrate ML with traditional epidemiological analyses to complement logistic regression by capturing potential nonlinear interactions among biomarkers and covariates, providing a more holistic view of the associations with depressive symptoms.

Methods

Study population

Data were downloaded from the National Health and Nutrition Examination Survey (NHANES) at https://www.cdc.gov/nchs/nhanes, a nationally representative cross-sectional survey. The NHANES data were collected by the National Center for Health Statistics (NCHS), which carried out standardized in-home interviews, physical examinations, and laboratory tests at mobile examination centers to record the general health and nutritional status on the non-institutionalized civilian population in the United States. The data from six cycles of NHANES (2009–2010, 2011–2012, 2013–2014, 2015–2016, 2017–2018, 2017 March–2020 Pre-pandemic) were extracted with independent samples to accumulate an appropriate sample size. The total number of participants enrolled in each cycle was as follows: 2009–2010 (n = 10,537), 2011–2012 (n = 9,756), 2013–2014 (n = 10,175), 2015–2016 (n = 9,971), 2017–2018 (n = 9,254), and 2017-March 2020 Pre-pandemic (n = 15,560). Sample selection flow chart is shown in Fig. 1. We included adult participants who reported a prior MI history, and the outcomes were determined based on the medical condition questionnaire. When the participant answered “yes” to the question “has a doctor ever told you that you had a heart attack”, we considered this individual had MI. The self-reported measures of MI have been proven to be accurate and reliable in prior epidemiological studies. Then we excluded participants with missing data on either SII, depressive symptoms or the platelet counts. Thus, our final sample consisted of 1,352 participants.

Fig. 1.

Flowchart of participant selection. NHANES, National Health and Nutrition Examination Survey; SII, systemic immune-inflammatory index; PHQ-9,Patient Health Questionnaire-9

Outcome variable

Participants were administered the Patient Health Questionnaire-9 (PHQ-9), an effective self-report questionnaire and a valid criteria instrument based on the Diagnostic and Statistical Manual of Mental Disorder (DSM)-V, which was used to measure the frequency of depressive symptoms experienced during the last two weeks. The PHQ-9 questionnaire has been previously published and validated elsewhere [22]. It scores the emotional, somatic or cognitive symptoms in 9 items by using a four-point scale: 0 (not at all), 1 (several days), 2 (more than half the days), 3 (nearly every day), and the sum of all answers to the questionnaire range from 0 to 27. The cut-off point score of ≥ 5 has been recommended to identify depressive symptoms, and previous studies have reported that depressive symptoms are associated with the incidence of cardiovascular diseases, even at symptom levels below the threshold that indicates a depressive disorder [23, 24]. For further sensitivity analysis, every item in PHQ-9 was dichotomized. As previous studies reported, values 0 and 1 were coded as not having the depressive symptom, and values 2 and 3 were coded as having the symptom [25].

Exposure variable

The laboratory methodology of the PLT, lymphocyte and neutrophil is provided on the NHANES website. These hematological samples were measured by complete blood count using automated hematology analyzing devices (Coulter DxH 800 analyzer) and reported as 10³ cells/ml. MPV is also included in complete blood count tests and presented in femtoliter. The SII was calculated by platelet count × neutrophil count/lymphocyte count and expressed as × 10^®9 cells/µl. As shown in Supplementary Table 1, the PLT, MPV and SII were nonnormally distributed and therefore log-transformed. Then they were converted to a categorical variable and divided into quartiles from lowest (Q1) to highest (Q4) to assess the independent effects with and without depression. The details about the quartiles after log-transformation were shown in Supplementary Table 2.

Covariates

Sociodemographic characteristics in NHANES were recorded in household interviews as follows: age, gender (male, female), ethnicity (Mexican American, Other Hispanic, non-Hispanic White, non-Hispanic Black, other Race-including Multi-Racial), education level (less than 9th grade, 9–11th grade/includes 12th grade with no diploma, high school graduate or equivalent, some college or AA degree, college graduate or above) and marital status (married, widowed, divorced, separated, never married, living with partner). Body mass index (BMI) was calculated as weight in kilograms divided by height in meters squared. According to the standard BMI cutoff thresholds, participants were assigned to four groups, with < 18.4 kg/m² for underweight, 18.5–24.9 kg/m² for normal, 25–29.9 kg/m² for overweight, and ≥ 30 kg/m² for obesity. Poverty-income ratio (PIR) was calculated by dividing family income by poverty guideline based on the consumer price index. PIR was categorized as ≤ 1.3, 1.31–3.50, and > 3.50 to estimate socioeconomic status [26]. A higher PIR reflects a relatively higher socioeconomic status. Physical activity (PA) was assessed by summing minutes of activity per week multiplied by the metabolic equivalents of task (MET) score, and classified into high and low level by the cut-off values of 600 min. The presence of diseases was defined by self-reported diagnosis, in other words, asking participants if a doctor or healthcare provider had ever diagnosed them with hypertension, diabetes, stroke, congestive heart failure, high cholesterol level, or trouble sleeping, and the answer “yes” to the questions was ascertained for the diseases. Existing smokers were defined as those who smoked 100 or more cigarettes in life. Heavy alcohol consumption was defined as those who had had at least 12 drinks of any type of alcoholic beverage in any one year. The NHANES team collected the information of prescription medication taken in the prior 30 days as reported by participants at the time of the in-home interview, to get the medicine name from the pill containers of products used. As data collected from the 2009 to 2012 cycle in NHANES, we respectively included participants who were taking antiplatelet agents (clopidogrel, aspirin, cilostazol, prasugrel, dipyridamole) and antidepressants (Amitriptyline, Buspirone, Risperidone, Mirtazapine, Paroxetine, Trazodone, Sertraline). On the basis of the NHANES data from the 2013 to 2020 cycle, “receiving antidepressant treatment” was defined as the self-report use of antidepressant medications according to the primary diagnosis of “major depressive disorder, single episode, unspecified” (F32.9) or “major depressive disorder, recurrent, unspecified” (F33.9), while “receiving antiplatelet treatment” was defined as “long term (current) use of anticoagulants and antithrombotics/antiplatelets” (Z79.02 from 2013 to 2014 or Z79.0 from 2015 to 2020).

Statistical analysis

In accordance with the analysis guidelines by NCHS, the appropriate sampling weights (WTMEC2YR) were used to merge more cycles and ensure the estimates national representation. The multiple interpolation method was employed in IBM SPSS Statistics 27.0 to deal with outliers. The Bonferroni correction was used for multiple testing. Multivariable conditional logistic regression was employed to quantify the association of various factors with depressive symptoms and adjusted for all covariates that were significant in the univariate analysis. The multivariate tests were constructed in 4 models: no covariates in model 1; age and gender were adjusted in model 2; age, gender, race, marital status, education level were adjusted in model 3; model 3 plus adjustment of income level, smoking, alcohol, BMI, hypertension, stroke, diabetes, sleeping disorder, anti-platelet drugs in model 4. Subgroups analysis was conducted with stratified factors including gender (male/female), age (≤ 44/45–59/60–74/75–89 years), BMI (underweight/normal/overweight/obesity), hypertension (yes/no), stroke (yes/no), sleep disorders (yes/no), anti-platelet drugs (yes/no), and antidepressants (yes/no). Then we conducted interaction test to examine the heterogeneity of associations between the subgroups. For sensitivity analyses, the methods of multiple imputation and excluding the stroke population were performed, and logistic regression was used to analyze the associations of exposure measurements with each of the depressive symptoms. The dose–response associations were assessed on a continuous scale by restricted cubic spline curves. A stratified analysis of the dose-response relationship by age, gender, and BMI was conducted. The statistical analysis was performed in the software R (version 4.3.1) and EmpowerStats (version 4.0), with a P < 0.05 considered statistically significant. To assess multicollinearity, pairwise correlation coefficients among the markers were examine, confirming moderate correlations but none exceeding 0.80, which supported their inclusion without undue inflation of variance.

Machine learning predictive modeling and interpretability analysis

To assess the predictive utility of log₂-transformed inflammatory and platelet-related biomarkers (SII, PLT, and MPV) alongside covariates for depressive symptoms (binary outcome: PHQ-9 ≥ 5) in MI survivors, a machine learning pipeline was implemented in Python (version 3.13) using scikit-learn, XGBoost, LightGBM, CatBoost, and SHAP libraries. The dataset was first divided into an 80% training set and a 20% held-out testing set. Hyperparameter optimization and benchmarking of the 14 supervised classification algorithms were conducted using 5-fold stratified cross-validation solely on the training set. Ensemble methods (Random Forest [RF], XGBoost, LightGBM, CatBoost), support vector classifiers (SVC and SVR with RBF kernel), neural networks (Multi-Layer Perceptron [MLP] with two hidden layers of 100 neurons each and ReLU activation), and baseline models (Decision Tree [DT], Gradient Boosting Regressor [GBR], K-Nearest Neighbors [KNN with k = 5], Logistic Regression, AdaBoost, Extreme Learning Machine [ELM], and Bayesian Ridge [Bay]). All models were trained on balanced classes via Synthetic Minority Over-sampling Technique (SMOTE) to mitigate class imbalance. Although the core task is binary classification of depressive symptoms (PHQ-9 ≥ 5 vs. <5), regression metrics were incorporated to evaluate performance on continuous predicted probabilities output by classifiers (probability of positive class), enabling assessment of model calibration and nuance beyond discrete thresholds.

Model performance was evaluated using Mean Absolute Error (MAE), Mean Squared Error (MSE), Root Mean Squared Error (RMSE), coefficient of determination (R²), and Area Under the Receiver Operating Characteristic Curve (AUC), with the optimal model selected based on the highest AUC and balanced regression metrics. The final selected model (RF) was then evaluated on the held-out test set to provide unbiased performance estimates. For interpretability, SHAP (SHapley Additive exPlanations) analysis was conducted solely on the top-performing RF model, utilizing a Kernel SHAP explainer with 100 background samples from the training set to compute SHAP values for the entire test cohort. Global feature contributions were visualized through a beeswarm plot, which ranked features by mean absolute SHAP value and depicted the distributions of positive/negative impacts across instances, stratified by feature values.

Results

Baseline characteristics of participants

Descriptive statistics are shown in Table 1. This study included 1,352 participants with a self-reported history of MI. Among these participants, 53.8% were non-Hispanic white, 20.3% were non-Hispanic black, 9.4% were Mexican American, 8.5% were other Hispanic, and 7.7% were from other race. The average age of the adults with the history of MI was 66.29 ± 11.67 years, and more than half of the sample were male (67.2%). These participants with with PHQ ≥ 5 accounted for 37.06% of all participants with a prior MI. Compared with the non-depressive MI population, participants screening positive for depressive symptoms in survivors of heart attack were more likely to be younger; more often female; living in a lower-income family (i.e., PIR ≤ 1.30); never married, widowed, divorced or separated; underweight or obesity; other Hispanic; lower education degree (i.e., Less than 9th grade); people with sleep disorder, previous hypertension or stroke, with higher log₂PLT or log₂SII.

Table 1.

Survey-weighted baseline characteristics of the study population based on depressive symptoms from NHANES 2009 to 2020

Characteristic	Not depressed (N = 844)	Depressed (N = 498)	P-value
Gender			< 0.0001
Male	621(73.29%)	281(48.44%)
Female	223(26.71%)	217(51.56%)
Age			0.0031
≤ 44	43(6.04%)	37(9.31%)
45–59	133(20.29%)	118(30.41%)
60–74	386(47.75%)	226(39.02%)
75–89	282(25.92%)	117(21.26%)
Race			0.0015
Mexican American	66(3.18%)	62(5.95%)
Other Hispanic	59(2.97%)	57(6.44%)
Non-Hispanic White	467(77.15%)	252(71.80%)
Non-Hispanic Black	181(9.32%)	93(10.28%)
Other Race - Including Multi-Racial	71(7.39%)	34(5.54%)
Education Level			0.002
Less than 9th grade	97(6.63%)	74(8.90%)
9-11th grade (Includes 12th grade with no diploma)	129(13.62%)	92(13.36%)
High school graduate/GED or equivalent	218(29.07%)	128(30.17%)
Some college or AA degree	251(28.72%)	153(36.98%)
College graduate or above	149(21.96%)	51(10.58%)
Marital Status			0.016
Married	381(49.91%)	174(41.67%)
Widowed	158(15.42%)	122(23.68%)
Divorced	178(18.58%)	119(19.59%)
Separated	39(4.21%)	28(4.16%)
Never married	51(5.67%)	37(7.89%)
Living with partner	37(6.21%)	18(3.02%)
Poverty-income ratio			< 0.0001
≤ 1.3	275(21.46%)	249(39.68%)
1.31–3.50	362(44.12%)	181(35.70%)
> 3.50	207(34.43%)	68(24.62%)
Body Mass Index			0.0321
Normal (25-29.9)	168(17.51%)	83(14.93%)
Underweight (≤ 18.4)	11(1.16%)	11(2.94%)
Overweight (25-29.9)	279(34.08%)	141(27.31%)
Obesity (≥ 30)	386(47.25%)	263(54.82%)
Smoke			0.1775
Yes	547(65.54%)	343(70.59%)
No	297(34.46%)	155(29.41%)
Drink			0.0784
Yes	642(79.33%)	357(73.12%)
No	202(20.67%)	141(26.88%)
Diabetes			0.0695
Yes	310(33.47%)	202(40.86%)
No	509(62.48%)	282(56.92%)
Borderline	25(4.05%)	14(2.23%)
High Blood Pressure			0.036
Yes	621(70.04%)	395(77.61%)
No	223(29.96%)	103(22.39%)
High Cholesterol			0.5942
Yes	571(69.37%)	342(71.12%)
No	273(30.63%)	156(28.88%)
Stroke			< 0.0001
Yes	141(13.36%)	130(24.44%)
No	703(86.64%)	368(75.56%)
Sleep Disorders			< 0.0001
Yes	252(31.54%)	313(66.91%)
No	592(68.46%)	185(33.09%)
Physical Activity			0.3278
≤ 600	767(87.94%)	424(85.63%)
> 600	77(12.06%)	74(14.37%)
Antiplatelet Drugs			0.1044
Yes	125(15.36%)	61(10.83%)
No	719(84.64%)	437(89.17%)
Antidepressants			< 0.0001
Yes	51(6.54%)	147(32.89%)
No	793(93.46%)	351(67.11%)

Frequency (percentage), unless otherwise stated

Univariate and multivariate logistic regression models

The results demonstrated that the depressive symptoms were significantly associated with log₂SII (OR = 1.32, 95%CI = 1.16–1.50, P = 0.0001) and log₂PLT (OR = 2.32, 95%CI = 1.76–3.08, P < 0.0001), but there was no clear-cut correlation between log₂MPV and depressive symptoms in the crude model (OR = 0.72, 95%CI = 0.37–1.44, P = 0.3550). There were respectively 122% and 178% increased risk of depressive symptoms per one-unit increment in log₂SII and log₂PLT after adjusting for most confounders. To determine whether the associations changed on a gradient, we stratified levels of the indicators into quartiles. As shown in Table 2, the significant difference was found in the second quartile of log₂MPV (OR = 1.40, 95%CI = 1.02–1.94, P = 0.0403) in the non-adjusted model, however, after adjusting for all covariates, this correlation became insignificant in model 3 (OR = 1.24, 95%CI = 0.86–1.77, P = 0.2458). Moreover, log₂SII and log₂PLT manifested stronger relationships with increased likelihood of depressive symptoms in Q3 and Q4 groups than the reference group. After controlling for a number of confounding variables, the results remained robust and statistically significant. Compared with the first quartile of log₂SII, multivariate-adjusted ORs for patients in the forth quartile tend to be higher, with model 3 (OR = 1.55, 95%CI = 1.08–2.22, P = 0.0168). In addition, the multivariate demonstrated that individuals in log₂PLT group (Q4) was correlated with a higher risk for depressive symptoms in model 3 (OR = 1.77, 95%CI = 1.23–2.54, P = 0.0020).

Table 2.

The association of depressive symptoms with log₂SII, log₂PLT and log₂MPV

Characteristics	Unadjusted		Adjust Model 1		Adjust Model 2		Adjust Model 3
	OR(95%Cl)	P-value	OR(95%Cl)	P-value	OR(95%Cl)	P-value	OR(95%Cl)	P-value
log 2 SII	1.32 (1.16, 1.50)	< 0.0001	1.30 (1.14, 1.49)	< 0.0001	1.27 (1.11, 1.46)	0.0005	1.22 (1.06, 1.41)	0.0069
log 2 SII quartile
Q1	1		1		1		1
Q2	0.96 (0.69, 1.33)	0.8023	0.91 (0.65, 1.28)	0.6004	0.88 (0.62, 1.24)	0.4569	0.79 (0.55, 1.13)	0.1983
Q3	1.61 (1.17, 2.20)	0.0032	1.60 (1.16, 2.21)	0.0043	1.53 (1.10, 2.14)	0.0114	1.56 (1.09, 2.22)	0.0148
Q4	1.81 (1.32, 2.48)	0.0002	1.75 (1.26, 2.41)	0.0007	1.68 (1.20, 2.34)	0.0023	1.55 (1.08, 2.22)	0.0168
log 2 PLT	2.32 (1.76, 3.08)	< 0.0001	1.82 (1.36, 2.43)	< 0.0001	1.81 (1.35, 2.44)	< 0.0001	1.78 (1.30, 2.43)	0.0003
log 2 PLT quartile
Q1	1		1		1		1
Q2	1.23 (0.89, 1.71)	0.2162	1.11 (0.79, 1.55)	0.5568	1.05 (0.75, 1.49)	0.7629	1.03 (0.72, 1.49)	0.8546
Q3	1.87 (1.35, 2.58)	0.0001	1.58 (1.13, 2.20)	0.0073	1.61 (1.15, 2.25)	0.006	1.54 (1.08, 2.21)	0.0174
Q4	2.32 (1.69, 3.19)	< 0.0001	1.75 (1.25, 2.44)	0.001	1.74 (1.24, 2.45)	0.0015	1.77 (1.23, 2.54)	0.002
log 2 MPV	0.72 (0.37, 1.44)	0.355	0.57 (0.28, 1.17)	0.1255	0.56 (0.27, 1.16)	0.1211	0.53 (0.24, 1.17)	0.115
log 2 MPV quartile
Q1	1		1		1		1
Q2	1.40 (1.02, 1.94)	0.0403	1.44 (1.04, 2.01)	0.0297	1.44 (1.03, 2.01)	0.0352	1.24 (0.86, 1.77)	0.2458
Q3	1.12 (0.82, 1.53)	0.4632	1.05 (0.76, 1.45)	0.7608	1.05 (0.76, 1.46)	0.7583	0.91 (0.64, 1.30)	0.6173
Q4	0.87 (0.63, 1.19)	0.3817	0.79 (0.57, 1.09)	0.153	0.79 (0.57, 1.10)	0.1688	0.74 (0.52, 1.06)	0.1007

Model 1: no covariates were adjusted

Model 2: adjusted for age and gender

Model 3: adjusted for age, gender, race, marital status and education level

Model 4: adjusted for age, gender, race, marital status, education level, income level, smoking, alcohol, BMI, hypertension, stroke, diabetes, sleeping disorder, anti-platelet drugs

P < 0.05 presents significant difference. CI, confidence interval; OR, odds ratio; SII, systemic immune-inflammatory index; MPV, mean platelet volume; PLT, platelet count

Dose-response relationship of depressive symptoms with log₂SII, log₂MPV and log₂PLT

As shown in Fig. 2, we discovered an inverted U-shaped curve between log₂MPV and depressive symptoms using restricted cubic spline curves (P-nonlinearity < 0.0001). The results revealed that the risk of depressive symptoms first increased and reached the maximum at the log₂MPV value of 3.04 and then decreased as MPV increased. The positive association of log₂SII and log₂PLT with depressive symptoms was statistically significant (P-nonlinearity < 0.0001). Subsequently, we conducted a stratified analysis of the dose-response relationship by age, gender, and BMI. The results have been presented in Supplementary Fig. 1 (Fig. S1). After stratified analysis by gender, a positive dose-response relationship between depressive symptoms and log₂SII was evident in males (P < 0.05), as well as was maintained between depressive symptoms and log₂PLT in females (P = 0.0083). By contrast, no significant correlation between log₂MPV and depressive symptoms was observed in either males or females. When stratified by age, positive dose-response relationships between depressive symptoms and both log₂SII and log₂PLT were apparent in the age groups of 45–59 years and 60–74 years, and the direct correlation also appeared between depressive symptoms and log₂MPV in age group from 45 to 59 (P = 0.0017). After stratification by BMI, positive correlations between depressive symptoms and both log₂SII and log₂PLT were evident in the overweight and obese groups, which was also observed between depressive symptoms and log₂MPV in the obese group (P = 0.0311).

Fig. 2.

Restricted cubic spline fitting for the association between exposure variables with depressive symptoms. Association of depressive symptoms with log₂SII (A), log₂PLT (B) and log₂MPV (C). The shaded portion represent the 95% confidence interval from the fit. CI, confidence interval; OR, odds ratio; SII, systemic immune-inflammatory index; MPV, mean platelet volume; PLT, platelet count

Subgroup analyses

The results of interaction analysis in Fig. 3 were consistent except for the gender and BMI group. In subgroup analyses stratified by gender, the positive association was independently significantly positive in male but not statistically significant in female (P for interaction = 0.0009). Besides, a significant interaction was also found between log₂SII and BMI (P for interaction = 0.0361). The depressive symptoms was more strongly correlated with log₂SII in overweight and obesity populations, with an increased OR ranging from 124% to 159%. But the risk of depressive symptoms prevalence was not increased with higher log₂SII in underweight and normal populations. As for log₂PLT and log₂MPV, consistent results were observed by the interaction test when analyses were stratified by age, sex, race, BMI, smoking status, hypertension, stroke, insomnia and antiplatelet agents (P for interaction > 0.05).

Fig. 3.

Forest plot for subgroup analysis on the association of depressive symptoms with log₂SII, log₂PLT and log₂MPV. CI, confidence interval; OR, odds ratio; SII, systemic immune-inflammatory index; MPV, mean platelet volume; PLT, platelet count

Sensitivity analyses

The sensitivity analyses demonstrated a similar result after multiple imputations (Supplementary Table 3) or excluding participants with the history of stroke (Supplementary Table 4) in the fully-adjusted model. It’s worth that the correlation was significantly closer between log₂SII and depressive symptoms after excluding patients with stroke, suggesting that the relationship was stable and independent of a history of stroke. Depressive symptoms is a heterogeneous disorder covering somatic, affective, and cognitive symptoms. So we dichotomized every item of PHQ9 to examine whether SII was associated with individual depressive symptom. As shown in Supplementary Table 5, we observed a strong association of log₂SII with trouble sleeping (OR = 1.29, 95%CI = 1.10–1.52, P = 0.0018), feeling tired (OR = 1.25, 95%CI = 1.07–1.44, P = 0.0039), and poor appetite (OR = 1.21, 95%CI = 1.00-1.47, P = 0.0450), of log₂PLT with poor appetite (OR = 2.08, 95%CI = 1.38–3.14, P = 0.0005), feeling tired (OR = 1.49, 95%CI = 1.08–2.05, P = 0.0149), and trouble sleeping (OR = 1.85, 95%CI = 1.31–2.62, P = 0.0005), of log₂MPV with poor appetite (OR = 0.15, 95%CI = 0.05–0.42, P = 0.0004).

Machine learning predictive modeling

To further explore the predictive performance of inflammatory and platelet-related features for depressive symptoms in MI survivors, we employed a comprehensive machine learning framework. Fourteen supervised learning algorithms were evaluated using 5-fold cross-validation on the dataset comprising log-transformed SII, PLT, MPV, and covariates (age, gender, neutrophil, LYMNO, PLR, MLR) (Fig. 4). The models included tree-based ensembles (RF, XGBoost, LightGBM, CatBoost), support vector machines (SVR, SVC), neural networks (MLP), and others (DT, GBR, KNN, Logistic Regression, AdaBoost, ELM, Bay). Performance metrics included MAE, MSE, RMSE, R², and AUC for binary depression classification (PHQ-9 ≥ 5). Table 3 summarizes comparative performance across models. RF showed superior accuracy, with the highest AUC (0.779), followed by CatBoost (0.762) and XGBoost (0.758). Tree-based models outperformed linear and kernel-based alternatives, with RF also achieving the best R² (0.229) and lowest RMSE (0.424). Simpler models like Logistic Regression and ELM had lower AUCs (< 0.64), indicating limited discriminatory power.

Fig. 4.

Illustrates the Receiver Operating Characteristic (ROC) curves for all models, confirming the leading position of RF with the curve closest to the ideal (AUC = 0.779), surpassing the random classifier (AUC = 0.5). This robustness across metrics positioned RF as the optimal model for subsequent interpretability analysis

Table 3.

Performance comparison of machine learning models for predicting depressive symptoms in MI survivors

Model	MAE	MSE	RMSE	R 2	AUC
RF	0.366134	0.180072	0.424349	0.228984	0.778935
CatBoost	0.375321	0.187809	0.433369	0.195855	0.762189
XGBoost	0.319181	0.201808	0.449231	0.135913	0.758166
LightGBM	0.337983	0.197442	0.444345	0.154609	0.754793
MLP	0.379667	0.210177	0.45845	0.100083	0.703905
GBR	0.408048	0.213593	0.462162	0.085453	0.686746
SVR	0.427623	0.212977	0.461494	0.088094	0.68287
SVC	0.427623	0.212977	0.461494	0.088094	0.68287
AdaBoost	0.461348	0.223304	0.472551	0.043874	0.662219
KNN	0.389591	0.232268	0.481942	0.005496	0.656331
DT	0.327138	0.327138	0.571959	-0.40071	0.647781
LogisticRegression	0.438951	0.224162	0.473457	0.040203	0.637278
ELM	0.508934	0.27105	0.520624	-0.16056	0.628047
Bay	0.421725	0.241954	0.491888	-0.03598	0.626864

Bold values highlight the top-performing model (RF) across key metrics

SHAP interpretability analysis

To elucidate feature contributions underlying RF predictions, we conducted SHAP analysis on the trained RF model. The beeswarm plot (Fig. 5) reveals global feature impacts on depressive symptoms risk. Age was the strongest predictor (15.8% of total impact), with younger ages (low values, blue points) consistently increasing SHAP values and elevating post-MI depressive symptoms probability, likely due to greater psychological distress or socioeconomic factors. Log₂PLT achieved the second-highest rank, where higher values (red points) positively influenced outputs, consistent with logistic regression results. Gender showed bidirectional effects, with females exhibiting comparable risk to males but subtle instance variations. Neutrophil count and log₂SII demonstrated positive correlations, with higher values driving predictions toward depressive symptoms and emphasizing systemic inflammation. Other features had moderate influences, highlighting interplay among platelet dynamics, immune dysregulation, and demographics. These findings reinforce mechanistic links between younger age, platelet activation, and inflammation in post-MI depressive symptoms from prior analyses.

Fig. 5.

SHAP beeswarm plot for the Random Forest model, showing the distribution of SHAP values for each feature across the dataset. Features are ranked by mean absolute SHAP value (right y-axis), with age having the highest impact (15.8%). Points are colored by feature value (red: high, blue: low), and positioned horizontally by SHAP value (impact on model output). Positive SHAP values indicate features pushing predictions toward depressive symptoms, while negative values push away. The bottom band represents the sum of SHAP values from the three least influential features

Discussion

In this nationally representative cross-sectional study, we observed a positive correlation of log₂PLT and log₂SII with the risk of clinically relevant depressive symptoms. Specifically, the third and fourth quantile in log₂SII and log₂PLT both indicated a significantly higher prevalence of depressive symptoms compared with the first quartile. The consequence was consistent after various sensitivity and stratified analyses. A dose-response relationship also shown that higher levels of log₂SII or log₂PLT were associated with depressive symptoms, particularly when log₂SII level was higher than 9 or log₂PLT was greater than 6.5. Moreover, restricted cubic spline analyses suggested an inverted U-shaped curve with an inflection point of 3.04 between log₂MPV and depressive symptoms. We also employed subgroup analysis to delve into potential moderating variables. However, after subgroup analysis by sex, age, and BMI, no inverted U-shaped relationship between log₂MPV and depressive symptoms was observed in any subgroup, further highlighting the need for replication in independent cohorts to confirm its robustness. Interestingly, positive associations of log₂SII, log₂PLT, and log₂MPV with depressive symptoms were discerned both in the 45–59 year age subgroup and in the obesity subgroup, indicating that these three biological indicators might be correlated with for depressive symptoms in middle-aged and obese populations. According to SHAP analysis, age (45-59 years) and log₂PLT emerged as the leading predictors. This study provided a reference for biological indicators in identifying associations with depressive symptoms among individuals with MI, and also offered evidence for platelet activation and inflammation in the pathomechanism of depression.

To enhance the preliminary utility of these inflammatory and platelet-related biomarkers for post-MI depressive symptoms, we further implemented a comprehensive machine learning framework benchmarking 14 supervised classification algorithms. The RF model emerged as the top performer, achieving the highest AUC of 0.779, alongside superior regression metrics such as R²=0.229 and RMSE = 0.424, outperforming other tree-based ensembles and linear/kernel-based alternatives. Although the RF model exhibits robust discriminatory capacity in classifying depressive symptoms based on current biomarker levels, its cross-sectional data structure precludes prospective predictive utility. This ML approach serves as a complementary tool to epidemiological analyses by uncovering nonlinear associations that traditional logistic regression may fail to detect, yet its generalizability remains contingent upon external validation. Consistent with the findings of Abdulla et al. [27], who demonstrated that XGBoost attained an area under the curve (AUC) of 0.95 for predicting depressive symptoms progression using inflammatory biomarkers (e.g., IL-1β and MCP-1), our results further validate the superiority of ensemble learning approaches in modeling complex nonlinear interactions among inflammatory features. Similarly, in the context of psychosocial maladjustment in AMI patients, Wang et al. [28] demonstrated the effectiveness of a stacked generalization model (LDS-R, combining Support Vector Classification, Logistic Regression, Decision Tree, and RF) with SHAP integration, achieving an AUC of 0.909 for predicting psychosocial maladjustment, highlighting the applicability of advanced ensemble methods to systemic inflammation and psychosocial factors in acute cardiac settings. This emphasizes the nonlinear interplay of features like log₂SII, log₂PLT, and log2MPV with covariates (age, gender, neutrophil count, lymphocyte count, PLR, MLR) in discriminating depressive symptoms risk (PHQ-9 ≥ 5). Subsequent SHAP-based interpretability analysis of the RF model identified age as the dominant predictor, with younger individuals exhibiting consistently higher predicted probabilities of depressive symptoms. This finding corroborates our stratified analysis, which demonstrated stronger associations between the predictor variables and depressive symptoms in the 45–59 years age group compared to older cohorts. Platelet count (log₂PLT) ranked second, reinforcing its direct positive influence, while gender showed bidirectional effects, with nuanced variations potentially reflecting sex-specific vulnerabilities. These ML-derived insights largely reinforce our logistic regression results rather than providing independent validation, but they add value by elucidating directional impacts and feature interactions. These insights suggest that integrating these biomarkers into preliminary predictive tools could aid further exploration of post-MI emotional disorders, particularly in young female AMI patients where associations were strongest, thereby extending the potential clinical promise of interpretable AI in psycho-cardiology pending external validation.

As far as we know, at present, there is very few study to examine the association of the platelet biochemical index and SII with depressive symptoms in those who suffered from heart attacks. Study by Öztürk et al. demonstrated that the basal levels of MPV were slightly higher in depressed patients than healthy controls [29]. A significant correlation of depression with MPV (OR = 1.720, P = 0.012) in autoimmune disorders has been reported as well [30]. Likewise, a prospective stroke cohort by Lassale et al. showed that high MPV levels (≥ 9.1 fl.) were independently correlated with post-stroke depression (OR = 2.762, 95% CI = 1.138–6.702, P = 0.025). Moreover, a cross-sectional study involving 14,007 Chinese affective disorder patients indicated that, compared with the healthy control, both the first-episode and recurrent major depressive disorder group possessed increased MPV and decreased PLT (P < 0.05) [31]. A case control study which included 61 depressed patients and 30 healthy subjects in China suggested that MPV was positively correlated with HAMD scores for work and interest, gastrointestinal symptoms, and hopelessness [16]. MPV is considered as a biological indicator of platelet activity, and is also used as biomarkers of inflammation in psychiatric disorders in some studies [32, 33]. The present state significantly deviates from the prior results. We found that log₂PLT was positively associated with an elevated risk of depressive symptoms, while the statistical association of depressive symptoms with log₂MPV was only observed at the second quartile than the reference in crude and partially corrected model. In sensitivity analysis, our study indicated that log₂PLT was significantly associated with poor appetite, feeling tired and trouble sleeping, while log₂MPV was only associated with poor appetite. Our data thus provide evidence for associations between platelet parameters and specific depressive symptoms in MI survivors, though the underlying reasons for these deviations warrant further investigation. Several mechanisms might explain the problem. Firstly, previous reports have reported the association with MPV and PLT in varying epidemiological methods and target population, and our study included a different population of depressive comorbidities than previous studies. Secondly, MPV is derived from dividing plateletcrit (PCT) by PLT, in other words, MPV and PLT are inversely proportional, and PCT also acts as an important factor affecting the MPV value. Unfortunately, the PCT was not included in the NHANES database for further analysis. But at least, it could partially account for the increased vulnerability of depressed patients to acute thrombotic event and the increased mortality post-MI.

SII, characterized by accessible, non-invasive and low cost, has promising prospect for depression in clinical application. Among individuals from the 2005–2018 NHANES, a cross-sectional study discovered the association of each 100-unit increase in SII with a 2% increase in the risk of depression [34]. There are some classical inflammatory markers with widespread therapeutic practice, such as monocyte-to-lymphocyte ratio (MLR). Ding et al. reported that MLR was an independent risk factor for depression by 3 months after stroke [11]. A cross-sectional retrospective study consisting of 84 patients with coronary heart disease demonstrated that SII and cluster of differentiation (CD)4/CD8 T-cell ratios were higher in patients with coronary heart disease (CHD) with depression than in those without depressive disorder (Z=-3.249, P = 0.001) [35]. Consistent with most studies, our study claimed that log₂SII had a strong association with increased risk of depressive symptoms with heart attacks. It’s now well established that dysregulation of inflammatory processes occurred in pathophysiology of depression. However, the pathological mechanism of post-MI depression remains to be elucidated, and the identification of peripheral biomarkers might provide new avenues and selective targets. Our study also provided a robust association of log₂SII with trouble sleeping, feeling tired, and poor appetite. The link of systemic inflammation with specific symptoms of depression had been previously reported. Alessandro Gialluisi also demonstrated a strong association of the altered appetite with E-selectin and c-reactive Protein (CRP), and white blood cells count, and of tiredness with granulocyte-to-lymphocyte ratio [36].

Research on the bidirectional interaction and underlying mechanisms between depression and cardiovascular diseases has surged. Currently, studies on the mechanisms of disease comorbidity primarily focus on neuroendocrine factors, immune-inflammatory responses, and platelet dysfunction. Among these, our research provides clinical data references for the mechanisms of immune-inflammatory responses and platelet dysfunction in psycho-cardiological diseases. Our previous animal experiments have observed that inflammatory factors can lead to microglial activation and alterations in neurotransmitters, thereby exacerbating depressive symptoms [6, 7]. Furthermore, inflammation has been implicated as a risk factor for the development of suicidal behaviors, extending the potential clinical relevance of our findings in post-MI depression management, particularly in vulnerable populations like MI survivors [37]. Potential mechanisms may include Cytokine activation promotes the upregulation of adhesion molecules on endothelial cells, which then drives the infiltration of monocytes and lymphocytes into the vascular wall, triggering a local inflammatory response in the inner layer of the blood vessels, ultimately promoting the emergence and progression of cardiovascular diseases [38, 39]. Previous studies suggest that, Compared to a control group without depressive symptoms, patients affected by depression exhibit increased platelet activity, assessed by MPV [9]. Platelet dysfunction, in turn, leads to sustained enhancement of platelet activation and aggregation, increased adhesion to the vascular endothelium, and promotion of thrombosis and vascular stenosis [40]. Platelets share some similarities with serotonergic neurons in terms of 5-hydroxytryptamine (5-HT) uptake, storage, metabolism, and release mechanisms [41]. Platelets from patients with depression exhibit greater aggregation in response to 5-HT [42]. As an important neurotransmitter, 5-HT can bind to 5-HT receptors located on the platelet surface, enhancing serotonin uptake by platelets and thereby augmenting their aggregation process. The serotonin-mediated platelet activation might represent a common pathophysiological process in both depression and cardiovascular diseases.

Previous studies have reported the gender differences in the risk of depression disorder post-cardiovascular diseases, as well as the influence of gender on the relationship between inflammatory markers and depression. A Deswal et al. found a linear increase in circulating levels of tumor necrosis factor (TNF) with advancing age in men, while plasma TNF-α level in women at an age of 50 years or less with heart diseases was lower, but disproportionately higher in women > 50 years of age [43]. There was also evidence from in vivo experiment that male rats with heart failure exhibited depression-like behaviors and an increase of cytokine levels in the prefrontal cortex, which were not observed in female rats [44]. After ovariectomy, the similar pattern of manifestation was shown in female rats, which was prevented by 17β-estradiol replacement. It possibly reflects an inhibitory action of estrogens for neuroinflammation to protect against depression. Moreover, the heightened platelet aggregation responses to serotonin and parasympathetic autonomic activity have been demonstrated in women [45]. Thus, the hormonal status alone is not sufficient to explain the different impact of inflammation on depression between women and men. The exact mechanisms might be not confined to hormone theory and worthy of further investigation.

Stratification analysis suggested that depressive symptoms were significantly related to the levels of log₂SII in obese adults but not in nonobese adults. The finding of BMI interacting with the relationship between inflammatory biomarkers and depressive symptoms was consistent with previous studies. Bin Zou et al. demonstrated that in participants with BMI ≥ 25, depression was independently associated with pro-inflammatory markers such as intercellular cell adhesion molecule-1 (ICAM-1), CRP, but these associations were not observed in BMI < 25 [46]. Nils Kappelmann et al. reported small but robust genetic correlations of BMI and CRP levels with depressive symptoms [47]. The pooled analysis of 15 population-based cohorts comprising 57,532 individuals found that CRP exerted the greatest attenuating effect (about 20%) on the identified obesity-depressive symptom associations [48]. Our results indicates that more attention should be paid to individuals with cardiovascular diseases with higher SII especially in obese population. Although subgroup analyses show stronger associations in the obese population, these results should be interpreted cautiously, as these analyses are exploratory and still require validation in independent, larger prospective studies.

Our study holds some strengths. First of all, the samples were nationally representative, and the data were selected from demographics, dietary, examination, laboratory and questionnaire modules, ensuring the associations credible. Certainly, sensitivity and subgroup analyses were further conducted to confirm the robustness of our results. Our study aims to determine the correlation between depressive patients and PLT, MPV, as well as SII in the population with MI, providing clinicians with a potential means of identifying patients with depressive symptoms. Furthermore, these biomarkers derived from peripheral blood offer accessibility, cost-effectiveness, and broad applicability. Therefore, the inflammatory biomarkers SII, along with PLT and MPV, discussed in this study, hold significant clinical importance in predicting the risk of depressive symptoms among patients with MI. The incorporation of machine learning and SHAP analysis further strengthens these findings by demonstrating the practical predictive power of these biomarkers in a data-driven context, offering interpretable insights that bridge statistical associations with clinical decision-making.

However, this study also has some limitations. Despite the fact that confounding variables were adjusted in the regression analysis, the effect of other possible residual variables such as inflammatory diseases, long-term use of other cardiovascular medications, specific MI characteristics (STEMI/NSTEMI), time since MI, and psychosocial support on the results cannot be eliminated. It should be noted that excluding inflammatory comorbidities from the final model might limit the generalizability of our findings to populations with different inflammatory comorbidity profiles. In addition, after conducting a sensitivity analysis by excluding participants with a history of stroke, the main results of the multivariable conditional logistic regression did not significantly change. However, it should be noted that the NHANES database does not specify the classification of stroke. Given the significant differences in pathophysiological mechanisms, clinical manifestations, prognosis, and their divergent effects on inflammation and platelet activity between ischemic and hemorrhagic stroke, caution is warranted when interpreting the results. Furthermore, the diagnoses of heart attacks and depressive symptoms were subjectively self-reported, which might exert potential recall bias on the results. Only the adults with heart attacks were included in this study, so it should be carefully exercised when extrapolating this association to general populations or depression with other complications. Additionally, the inverted U-shaped association for log₂MPV was not consistently reproduced in subgroup analyses, and this pattern should be interpreted cautiously as its biological plausibility remains speculative. This observation may be attributable to residual confounding (from unmeasured hematological factors like plateletcrit), measurement variability in MPV (due to pre-analytic factors like sample collection timing or storage), or statistical artifacts (arising from data skewness, model overfitting, or limited power in nonlinear estimations). Likewise, although SMOTE addressed class imbalance (~ 37% prevalence of depressive symptoms), it may introduce synthetic data artifacts, potentially inflating performance metrics and limiting generalizability to unbalanced real-world datasets. Regarding the ML analysis, the sample size (n = 1,352) is modest for training 14 complex algorithms, which increases the risk of overfitting despite the use of cross-validation. Consequently, the superior performance of Random Forest may not generalize perfectly to external cohorts. Moreover, due to the cross-sectional design, we cannot rule out the possibility that depressive symptoms influence inflammatory or platelet markers. This bidirectional relationship between depressive symptoms and inflammation is supported by prior literature [38]. It’s plausible that depressive symptoms in our sample may have contributed to observed variations in inflammatory or platelet markers, rather than these markers driving depression. This two-way association highlights the necessity of longitudinal research to clarify temporal dynamics. Thus, the results should be interpreted with caution for several mentioned limitations.

Conclusion

In a nationally representative sample of U.S. adults with a history of myocardial infarction, elevated log₂SII and log₂PLT were positively associated with depressive symptoms, while log₂MPV suggested an inverted U-shaped relationship with an inflection point of 3.04. Machine learning analysis, particularly the Random Forest model (AUC = 0.779), reinforced these associations and highlighted age and platelet count as key predictors via SHAP interpretability, emphasizing the interplay of inflammation and platelet dynamics in post-MI depressive symptoms. Given the cross-sectional design and the absence of external validation, longitudinal studies are needed to elucidate the causal mechanisms of SII, PLT, and MPV in depression onset and progression and to confirm the generalizability of our ML findings.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

Yize Sun designed the study and drafted the manuscript. Zheyi Wang drafted the machine learning sections of the manuscript and contributed to the data analyses. Shencun Yu and Qiong Xu contributed to the interpretation of the results. Qiong Xu, Jiao Sun and Qiaoyi Huang contributed to the language polishing and formatting. Chen Huang and Jian Lv contributed to proofreading the entire manuscript and correct the grammatical errors. All authors read and approved the final manuscript.

Funding

The project was funded by the Natural Science Foundation of Shandong Province (ZR2025QC973) and Science and Technology Project of Traditional Chinese Medicine in Qingdao (2023-zyyz04).

Data availability

The NHANES Survey is a public open database, the datasets generated from which are available at the website: https://www.cdc.gov/nchs/nhanes/index.htm.

Declarations

Ethics approval and consent to participate

The protocols of NHANES were approved by the institutional review board of the NCHS, CDC (https://www.cdc.gov/nchs/nhanes/irba98.htm). All participants have provided written informed consent.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.