Machine Learning Prediction of Stress-Induced D-dimer Reactivity in Male Physicians with and without Burnout

Abstract

Acute emotional stress can trigger acute coronary syndrome (ACS), potentially via hypercoagulable states. Circulating D-dimer is an established marker of fibrin turnover and stress-related coagulation activation, yet predictors of D-dimer stress reactivity remain unclear, especially in high-risk groups such as male physicians with burnout. We examined predictors of D-dimer changes during acute stress and recovery in 60 male physicians with and without burnout. Participants underwent the Trier Social Stress Test, with D-dimer and other biomarkers assessed across four time points over 1 hour. The area under the curve (AUC) for D-dimer was calculated to capture overall reactivity. We applied the least absolute shrinkage and selection operator (LASSO) regression to identify relevant predictors among demographic, behavioral, psychosocial, and physiological variables, followed by traditional linear regression to estimate effect sizes. LASSO regression identified five key predictors of D-dimer stress reactivity: Prestress D-dimer, habitual alcohol consumption, prestress cortisol, stress-induced epinephrine (EPI) surge, and adverse childhood experiences (ACEs). In linear regression, all but prestress cortisol remained significant independent predictors, collectively explaining 50.4% of the variance in D-dimer AUC. Specifically, higher alcohol consumption (ΔR ² = 0.117, p < 0.001), larger EPI surge (ΔR ² = 0.081, p = 0.003), and more ACEs (ΔR ² = 0.044, p = 0.026) were associated with heightened D-dimer responses, while higher prestress D-dimer was associated with attenuated reactivity (ΔR ² = 0.208, p < 0.001). Our findings highlight the role of early adversity, alcohol consumption, and sympathoadrenal activation in stress-induced coagulation activation, as reflected by D-dimer reactivity. If validated, these predictors may help identify individuals at elevated risk for stress-triggered ACS and inform targeted prevention strategies.

Keywords

hypercoagulability - D-dimer - psychology - cardiology - atherothrombosis - thrombosis

Acute emotional stress can trigger acute coronary syndrome (ACS) within 1 to 2 hours of stress onset, especially in individuals with subclinical atherosclerosis.[1] About 25% of ACS patients recall a specific trigger, with nearly half citing emotional upset.[2] For example, the risk rises 5.5-fold after an episode of acute anger, an effect attenuated by aspirin and β-blockers.[3] In patients with coronary artery disease (CAD), stress-induced cardiovascular reactivity, including reduction in flow-mediated vasodilation and peripheral vasoconstriction, predicts adverse cardiovascular outcomes independently of traditional risk factors.[4] These findings highlight acute emotional stress as a clinically relevant cardiovascular risk factor, involving mechanisms such as sympathetic overactivation, endothelial dysfunction, and platelet activation, all promoting enhanced coagulation.[5]

Consistent with these suggested mechanisms, experimentally induced acute mental or psychosocial stress has been shown to transiently elevate D-dimer, providing direct evidence of stress-induced activation of coagulation and fibrinolysis.[6] D-dimer, a degradation product of cross-linked fibrin, is a key marker of fibrin turnover and thrombotic risk.[7] Elevated D-dimer levels have independently been associated with increased all-cause mortality and major adverse cardiovascular events in patients with ACS,[7] [8] including those undergoing percutaneous coronary intervention.[9] Stress-induced D-dimer reactivity varies by individual characteristics, with heightened or prolonged responses linked to older age,[10] diastolic hypertension,[11] anxiety,[12] depression,[13] and work stress.[13] However, prior studies typically focused on a narrow set of predictors, limiting comparisons of their relative contributions within a biopsychosocial framework.

Recent advances in machine learning offer new tools to assess multifactorial influences on biomarker responses under stress. Penalized regression methods like least absolute shrinkage and selection operator (LASSO) allow simultaneous evaluation of multiple predictors by shrinking uninformative coefficients to zero, improving model generalizability and interpretability.[14] Applying this method to D-dimer stress reactivity may uncover modifiable biological, behavioral, and psychosocial risk factors, with relevance for cardiovascular outcomes. Identifying such predictors could enhance risk stratification for emotionally triggered ACS, which often occur without traditional clinical warning signs. This integrative approach is particularly relevant for individuals with chronic psychosocial stress, characterized by neuroendocrine, autonomic, and immune dysregulation, along with adverse health behaviors that contribute to CAD.[15] For instance, up to 50% of physicians report clinical symptoms of burnout, a work stress-related syndrome.[16] Chronic work stress[17] and burnout[18] are established CAD risk factors, especially in men,[19] [20] who also show greater stress-induced hypercoagulability than women.[6] However, whether burnout affects acute stress-induced coagulation responses, such as D-dimer reactivity in men, remains unknown.

This study aimed to identify key predictors of D-dimer increases during acute psychosocial stress and recovery over 1 hour in male physicians with and without burnout. Using LASSO regression, we integrated biological, psychosocial, and behavioral variables to identify factors associated with stress-induced coagulation activation in this high-risk group.

Methods

Study Participants and Recruitment

The study was approved by the local ethics committee in Zurich (BASEC-Nr. 2018–01974), and all participants provided written informed consent. Recruitment and data collection took place from September 2019 to December 2021. Male physicians practicing in Switzerland were invited to participate in a study on burnout and cardiovascular health via hospitals, clinics, professional associations, and direct email. Participants received written study information and could contact the research team with questions. The primary outcome, myocardial blood flow assessed by positron emission tomography, is reported elsewhere.[21] Based on power calculation for the main outcome,[21] 60 physicians were enrolled, comprising two matched groups (burnout vs. controls, n = 30 per group). Matching criteria included age (±5 years), body mass index (BMI; ±5 kg/m²), and first-degree family history of premature cardiovascular disease.

Screening included telephone interviews using the Maslach Burnout Inventory–Human Services Survey (MBI-HSS)[22] and Patient Health Questionnaire-9 (PHQ-9).[23] Of 143 physicians screened, 83 were excluded due to ineligibility or withdrawal of interest. Burnout classification was based on cutoff scores derived from a comprehensive systematic review of physician burnout, which identified the MBI-HSS with its three subscales of emotional exhaustion (EE), depersonalization (DP), and personal accomplishment (PA) as the most widely used assessment tool.[16] Specifically, participants were allocated to the burnout group if they had an EE score of ≥27 and/or a DP score of ≥10 (with a minimum EE score of 20). Control group participants had to score below 16 for EE and below 7 for DP, reflecting symptom occurrence less than once a month. The PA subscale was not used in the classification process, as it is considered relatively independent of EE and DP.[24] To reduce potential confounding by overlapping biological mechanisms, burnout group participants were required to have a PHQ-9 score of ≤14, and controls ≤10, restricting depressive symptoms to moderate and mild levels, respectively.[23] This criterion ensured conceptual clarity and methodological rigor, considering the ongoing debate and inconclusive evidence regarding the overlap between burnout and depression.[25] Burnout group participants also had to report ≥6 months of work-related stress and prolonged exhaustion.[26]

All participants were aged 28 to 65 years, reflecting typical entry into and retirement from medical practice in Switzerland. They were non-smokers for at least 5 years and had no history of clinical depression or burnout, cardiovascular disease, type I or II diabetes, familial hypercholesterolemia, stage II hypertension, renal insufficiency, other serious medical conditions, or use of medications affecting biomarkers. Additional exclusion criteria included BMI ≥35 kg/m², chronic risky alcohol use, and contraindications to cardiac imaging procedures. Participants who declined disclosure of incidental cardiac imaging findings were excluded. All criteria were assessed via structured interviews; self-reported medical history was accepted due to the participants' medical expertise.

Procedures on the Examination Day

Cardiac imaging was conducted in the Department of Nuclear Medicine and described elsewhere.[21] Participants were then escorted to the Stress and Behavior Research Laboratory for the Trier Social Stress Test (TSST) and further assessments. At 9:40 a.m., they received a standardized light breakfast and were equipped with a hemodynamic monitor. Blood samples were collected by trained laboratory staff via an 18-gauge catheter placed earlier for imaging. During TSST recovery, participants completed questionnaires on demographics, health behaviors, and psychometrics. A debriefing clarified that the test aimed to induce stress rather than assess individual performance.

Psychosocial Stress Protocol

We applied the TSST to induce psychosocial stress through social evaluative threat and loss of control in a standardized laboratory setting. It is considered an ecologically valid stressor that activates the sympathetic nervous system,[27] the hypothalamic–pituitary–adrenal (HPA) axis,[27] as well as hemodynamic[27] and coagulation responses, including changes in plasma D-dimer levels.[10] [11] [13] After a 2-minute introduction, participants had 3 minutes to prepare for a 5-minute mock job interview and a 5-minute mental arithmetic task, performed before a panel of three trained evaluators (two during COVID-19), and a video camera.

Saliva and blood samples for D-dimer and biomarkers were collected at five time points: Before TSST instructions (resting sample), immediately after, and at 15, 45, and 90 minutes after the TSST. Blood pressure (BP) was recorded at these same time points and during the speech and arithmetic tasks. To assess potential anticipatory effects on D-dimer before the TSST began, an additional sample was collected about 3 hours earlier, prior to cardiac imaging. Heart rate (HR) recordings were considered at four time points: Before TSST, immediately after, and at 15 and 45 minutes after TSST. This a priori decision was based on evidence that HR typically returns to prestress levels within 45 minutes in healthy individuals, whereas BP recovery is often more prolonged.[28]

Assessment of Cardiovascular Reactivity

Cardiovascular reactivity during the TSST was assessed using either the Finapres® Nova (Finapres Medical System, Enschede, the Netherlands) or the Omron Evolv (Omron Healthcare Co., Kyoto, Japan). Due to technical issues, the first 13 participants were assessed with Finapres®, and all subsequent participants with the Omron device. To adjust for potential device-related differences, a dummy variable was included in the repeated-measures analysis of variance for HR and BP. Given these limitations, cardiovascular reactivity measures were used solely as a manipulation check to confirm the hemodynamic stress response and were not included as predictors in the LASSO regression for D-dimer reactivity.

Assessment of Plasma D-dimer Levels

D-dimer levels were quantified from ethylenediaminetetraacetic acid (EDTA) plasma using a solid-phase sandwich enzyme-linked immunosorbent assay (ELISA; human D-DIMER ELISA Kit, Thermo Fisher Scientific, Cat. No. EHDDIMER), according to the manufacturer's instructions. A standard curve was generated using a four-parameter logistic fit, with a detection limit of 0.08 pg/mL. Intra- and interassay coefficients of variation (CVs) were 3.4% and 12.8%, respectively.

Assessment of Potential Predictors of D-dimer Stress Reactivity

Demographic and Cardiometabolic Variables

Age was self-reported by participants. BMI was calculated as weight (kg) divided by height (m²), based on measured values. The Systematic Coronary Risk Evaluation (SCORE)2 algorithm, incorporating age, sex, smoking status, total cholesterol, HDL cholesterol, and systolic BP, was used to estimate 10-year cardiovascular disease risk.[29] Systolic BP was assessed as the mean of two resting measurements using the Arteriograph device (Arteriomed GmbH, Grevenbroich, Germany). High-sensitivity C-reactive protein (CRP) and fasting lipid levels were assessed at the Institute of Clinical Chemistry, University Hospital Zurich. Serum CRP was determined via turbidimetric immunoassay using a Cobas c 501® analyzer (Roche Diagnostics, Mannheim, Germany).

Health Behavior Variables

Physical activity was assessed by asking participants how many days per week (0–7) they engaged in exercise intense enough to cause sweating. Weekly alcohol consumption was recorded as the number of standard drinks. Sleep quality and disturbances over the past 4 weeks were measured using the Pittsburgh Sleep Quality Index (PSQI), which covers seven components: Subjective sleep quality, sleep latency, sleep duration, habitual sleep efficiency, sleep disturbances, use of sleep medication, and daytime dysfunction.[30] The global PSQI score ranges from 0 to 21, with higher scores indicating poorer sleep quality. In this sample, internal consistency (Cronbach's α) was 0.61.

Salivary Biomarkers

Saliva samples were collected using salivettes (Sarstedt, Rommelsdorf, Germany) and stored at −80°C. After thawing and centrifugation (3,000 rpm, 10 min), salivary α-amylase (sAA, U/mL; intra-/interassay CV: 2.0%/8.7%) and cortisol (ng/mL; CV: 1.4%/5.9%), markers of sympathetic activation[31] and HPA axis activity,[32] respectively, were determined via enzymatic assays in accordance with manufacturers' instructions; standard values were fitted using four parameter logistics for cortisol.

Plasma Catecholamines

Epinephrine (EPI) and norepinephrine (NE) were measured by high-performance liquid chromatography (HPLC; detection limit: 12 pg/mL; Laboratory for Stress Monitoring, Hardegsen, Germany[33]) from blood drawn into EDTA-coated monovettes (Sarstedt, Numbrecht, Germany), centrifuged at 2,000× g for 10 min, and stored at −80°C. An internal standard (dihydroxybenzylamine, 1,500 pg/mL) was used for correction. Samples were analyzed in a single HPLC run across 17 extraction batches (maximum 23 samples each plus control; RECIPE, Munich); CVs were 7.2% for EPI and 6.0% for NE.

Psychometric Variables

Burnout was assessed with the German 22-item MBI-HSS, comprising three subscales: EE (9 items), DP (5 items), and low PA (8 items).[22] Items were rated from 0 (“never”) to 6 (“daily”). EE reflects energy depletion, DP a detached attitude toward patients, and PA a sense of competence. Burnout was analyzed both categorically (yes/no) and dimensionally via subscales. Cronbach's α was 0.95 (EE), 0.87 (DP), and 0.77 (PA).

Effort (3 items), reward (7 items), and overcommitment (6 items) were assessed using the German short version of the Effort-Reward Imbalance and Overcommitment Questionnaire, measuring extrinsic (effort–reward ratio) and intrinsic (overcommitment) work stress, respectively.[34] Items were rated on a 4-point Likert scale from 1 (“strongly disagree”) to 4 (“strongly agree”). In our sample, Cronbach's α was 0.76 for effort, 0.77 for reward, and 0.81 for overcommitment.

Depressive symptom severity over the past 2 weeks was assessed using the German version of the PHQ-9,[35] with items rated from 0 (“not at all”) to 3 (“nearly every day”); total scores range from 0 to 27, with higher scores indicating greater severity. Cronbach's α in our sample was 0.79.

Anxiety symptoms over the past 2 weeks were assessed using the German version of the Generalized Anxiety Disorder-7 (GAD-7) scale,[36] with items rated from 0 (“not at all”) to 3 (“nearly every day”) and a total score range of 0 to 21. Higher scores reflect greater symptom severity. Cronbach's α was 0.87 in our sample.

Perceived stress over the past month was assessed with the 4-item version of the Perceived Stress Scale-4 (PSS-4). Each item is rated on a 5-point Likert scale, ranging from 0 (“never”) to 5 (“very often”), yielding a total score range of 0 to 16.[37] Cronbach's α was 0.86 in our sample.

A 10-item self-reported Adverse Childhood Experiences Questionnaire (ACE-Q) assessed exposure to trauma before age 18 across abuse, neglect, and household dysfunction subdomains.[38] Items were binary-coded (0 = “no,” 1 = “yes”), yielding a total adverse childhood experience (ACE) score from 0 to 10. In our study, Cronbach's α was 0.70 for the 10 ACE-Q items, supporting the use of the cumulative score to reflect overall exposure burden.

Data Analysis

Data were analyzed using IBM SPSS Statistics for Windows, Version 29.0 (IBM Corp., Armonk, NY) and Python. A two-tailed significance level was set at p < 0.05. To address missing data, we applied multiple imputation (k = 5) in SPSS across the full dataset, which comprised all previously described predictor variables, hemodynamic, and D-dimer measures. To optimize estimation accuracy, the imputation model included all available variables: Demographic and cardiometabolic measures (no missing data), health behaviors (no missing data), biomarkers (5.1% missing), cardiovascular reactivity measures (2.9% missing), and psychometric questionnaire sum scores (0.8% missing). For a detailed overview of the missing data distribution, see [Supplementary Table S1] (available in the online version only). Little's Missing Completely at Random (MCAR) test yielded a p-value of 0.63, indicating no significant evidence of systematic missingness.

To identify the optimal window for capturing D-dimer changes, we examined its trajectory across all time points ([Supplementary Fig. S1] [available in the online version only]). D-dimer levels before cardiac imaging and TSST instructions were similar, indicating minimal anticipatory effects. Likewise, levels at 45 and 90 minutes poststress plateaued. Repeated-measures analysis of variance confirmed a significant cubic trend across the first four time points (p = 0.044), which became non-significant when including the 90-minute point (p = 0.16), suggesting that D-dimer levels had stabilized beyond 45 minutes. This pattern aligns with prior evidence that coagulation responses to acute psychosocial stress, including changes in D-dimer, typically unfold within the first hour after stress onset (6). Accordingly, we calculated the area under the curve (AUC) for D-dimer and other biomarkers (sAA, cortisol, EPI, NE) over the first four time points using the trapezoidal method, which accounts for unequal time intervals between measurements and reflects total output irrespective of baseline.

Our primary statistical approach was LASSO regression, a machine learning method that performs variable selection through regularization to identify the most relevant predictors while reducing overfitting.[14] Predictors of D-dimer AUC were age, BMI, SCORE2, and hs-CRP; physical activity, alcohol consumption, and the PSQI sum score; burnout group along with EE, DP, and PA subscale scores; sum scores of the PHQ-9, GAD-7, effort–reward ratio, overcommitment, PSS-4, and ACE-Q; AUC measures for sAA, cortisol, EPI, and NE, as well as their corresponding preinstruction values. Preinstruction values were included to assess their potential contribution to individual differences in stress reactivity, including floor or ceiling effects.

We performed a LASSO stability analysis by performing 20 iterations of training the model, using a different subsample each time. This approach was chosen to account for the instability of the LASSO results.[39] Selection probabilities were calculated for each variable by dividing the number of times a variable was included in a model by the total number of models trained. Variables with selection probabilities >75% were considered relevant. Subsampling was performed by randomly splitting the data, with 80% of the data used to train the model and 20% of the samples set aside as a test set. Each model was fitted with 10-fold cross-validation to optimize λ. The performance of the models was assessed by calculating the root mean square error (RMSE) using the test data.

As LASSO does not provide p-values or effect sizes, we conducted follow-up linear regressions to quantify the strength and significance of associations between LASSO-selected predictors and D-dimer AUC. Both univariate (zero-order) and multivariable models were presented, as LASSO may select predictors with minimal bivariate but meaningful joint contributions. D-dimer AUC was normalized using a two-step transformation,[40] and Mahalanobis distance was used to confirm the absence of multivariate outliers. Effect sizes are reported as changes in the coefficient of determination (ΔR ²), interpreted as small (0.02), medium (0.13), and large (0.26). Significant predictors were visualized using SPSS partial regression plots, showing their unique association with D-dimer AUC adjusted for other variables. Because the burnout group was younger, we additionally ran a sensitivity analysis with additional adjustment for age; multivariable linear regression results were unchanged in direction and significance (not shown in detail).

Results

Participant Characteristics

[Table 1] presents the characteristics of the 60 participants stratified by burnout status. Compared with controls, the burnout group was significantly younger but did not differ in BMI, cardiovascular risk score, CRP levels, alcohol consumption, or exercise frequency. As expected, participants with burnout reported significantly higher scores on the three burnout subscales, effort, overcommitment, depressive and anxiety symptoms, perceived stress, and poorer sleep quality, as well as lower reward. ACE scores did not differ between groups.

Table 1 Characteristics of study participants stratified by burnout group

Variable	Burnout group (n = 30)	Control group (n = 30)	p-Value
Age, years	46.77 (10.56)	52.93 (7.48)	0.012
Body mass index, kg/m2	25.63 (3.09)	24.35 (2.72)	0.094
Cardiovascular disease risk score, percentage	3.11 (1.89)	3.47 (1.89)	0.464
C-reactive protein, mg/L	1.13 (0.99)	1.28 (2.37)	0.370
Alcohol, units/week	3.72 (3.22)	2.93 (2.30)	0.280
Exercise, times/week	1.99 (1.62)	2.67 (1.92)	0.147
Pittsburgh Sleep Quality Index, score	5.42 (1.64)	3.29 (1.99)	<0.001
Emotional exhaustion, score	29.17 (7.13)	6.67 (3.99)	<0.001
Depersonalization, score	11.33 (7.00)	3.07 (3.60)	<0.001
Personal accomplishment, score	12.03 (6.74)	5.67 (4.37)	<0.001
Effort, score	10.65 (1.36)	8.07 (2.13)	<0.001
Reward, score	19.58 (4.03)	22.24 (2.96)	0.005
Effort–reward ratio	1.34 (0.41)	0.87 (0.27)	<0.001
Overcommitment, score	16.65 (2.83)	12.43 (3.37)	<0.001
Patient Health Questionnaire-9, score	7.40 (3.13)	2.20 (1.97)	<0.001
General Anxiety Disorder-7, score	6.83 (3.86)	2.23 (2.52)	<0.001
Perceived Stress Scale-4, score	6.37 (2.72)	2.93 (2.39)	<0.001
Adverse childhood experiences, score	1.24 (1.77)	0.73 (1.14)	0.194

Note: Values are presented as mean with standard deviation (in parentheses). Group differences between burnout and control participants were assessed using independent samples t-tests.

Stress-Induced Changes in Cardiovascular Measures and Biomarkers

Confirming effective stress induction, the TSST produced significant cardiovascular responses, with within-subject time effects for HR, systolic, and diastolic BP from prestress to 45 minutes poststress (all p < 0.001; [Supplementary Fig. S2] [available in the online version only]). [Figure 1] presents changes in D-dimer and accompanying biomarkers across the four time points for AUC calculation. Except for sAA, all biomarkers showed significant time effects, including D-dimer (p = 0.044 for within-subjects contrasts), cortisol, EPI, and NE (all p < 0.001 for within-subjects effects), confirming transient coagulation activation and robust neuroendocrine stress responses.

Fig. 1 (A–E) Biomarkers (D-dimer, salivary α-amylase, salivary cortisol, epinephrine, and norepinephrine) were assessed at four time points used to calculate the area under the curve (AUC): Before Trier Social Stress Test (TSST) instructions (baseline), immediately after the TSST, and at 15 and 45 minutes poststress. Values are presented as means with 95% confidence intervals.

Data-Guided Selection of Relevant Predictors for D-dimer Stress Reactivity

In the 20 LASSO models, between 3 and 15 variables were selected. The selection probabilities for all variables that were selected at least once are shown in [Fig. 2]. Based on our cutoff value of 75% for the importance of a variable, five variables met this inclusion criterion: Alcohol consumption and prestress D-dimer were selected in every model, ACEs were selected in 19 of 20 models, AUC for EPI was selected in 18 models, and prestress cortisol was selected 16 times. The average RMSE across all models was 3,742.75, suggesting that the model's power to predict the outcome on new data is limited.

Fig. 2 Selection possibilities in the LASSO stability analysis. Quotient of the number of times a variable was included in one of the models and the total number of models. Only variables that were selected at least once are shown. The cutoff value of 75%, which was used to classify the variables as important or less important, is indicated by the dashed line. ACE-Q, Adverse Childhood Experiences Questionnaire; AUC, area under the curve; BMI, body mass index; CRP, C-reactive protein; DP, depersonalization; EPI, epinephrine; GAD-7, General Anxiety Disorder-7; LASSO, least absolute shrinkage and selection operator; MBI, Maslach Burnout Inventory; NE, norepinephrine; PHQ-9, Patient Health Questionnaire-9; SCORE2, Systematic Coronary Risk Evaluation 2.

Associations of LASSO-Identified Predictors with D-dimer Stress Reactivity

[Table 2] displays the univariate and multivariable associations of the five predictors selected by LASSO regression. Some predictors showed only weak or non-significant zero-order associations but became significant when adjusted for other selected predictors, highlighting the added value of multivariable modeling in revealing predictors whose effects are conditional on other factors. In the multivariable model, the five predictors collectively explained 50.4% of the variance, with prestress D-dimer as the strongest predictor, showing a large effect (ΔR ² = 0.21), indicating that higher baseline levels were associated with smaller stress-induced increases. Small to medium effect sizes were observed for alcohol consumption (ΔR ² = 0.12), AUC of EPI (ΔR ² = 0.08), and ACEs (ΔR ² = 0.04), all positively associated with heightened D-dimer stress responses. [Figure 3] displays the partial regression plots of these significant associations. Prestress cortisol, although selected by LASSO, did not show a significant independent association in the multivariable model.

Table 2 Univariate and multivariable associations of least absolute shrinkage and selection operator-identified predictors with D-dimer stress reactivity

Predictor	Univariate model			Multivariable model
	Unstandardized B (95% confidence interval)	p-Value	R 2 change	Unstandardized B (95% confidence interval)	p-Value	R 2 change
D-dimer before instruction for TSST	−21.8 (−30.1, −13.6)	<0.001	0.325	−18.2 (−25.6, −10.9)	<0.001	0.208
Alcohol consumption	816 (361, 1,272)	<0.001	0.181	706 (326, 1,086)	<0.001	0.117
Adverse childhood experiences	768 (−151, 1,688)	0.100	0.046	802 (100, 1,504)	0.026	0.044
Epinephrine area under the curve	1.41 (−0.47, 3.28)	0.139	0.037	2.14 (0.76, 3.52)	0.003	0.081
Cortisol before instruction for TSST	−4,992 (−10,600, 616)	0.080	0.052	−1,261 (−5,495, 2,972)	0.553	0.003

Abbreviation: TSST, Trier Social Stress Test.

Note: Some predictors showed weak or nonsignificant univariate associations but became significant in the multivariable model, reflecting mutual adjustment and suppression effects. This illustrates the added value of the least absolute shrinkage and selection operator in identifying predictors with conditional contributions when accounting for other variables.

Fig. 3 (A–D) The partial regression plots show the significant multivariable associations of prestress D-dimer (A), habitual alcohol consumption (B), stress-induced epinephrine area under the curve (C), and adverse childhood experiences (D) with D-dimer area under the curve (AUC) across prestress, immediately after stress, and 15 and 45 minutes poststress.

Discussion

This study examined predictors of D-dimer changes in response to acute psychosocial stress over a 1-hour interval. This time window is critical for triggering ACS following emotional upset, especially in male physicians experiencing occupational stress. To our knowledge, this is the first study applying a data-driven approach to identify a multivariate biopsychosocial profile predictive of stress-induced coagulation activation, thereby addressing a critical gap in the literature where traditional methods have limited discovery.[6] Using machine learning-based variable selection, we identified prestress D-dimer and habitual alcohol consumption as the strongest predictors, followed by ACEs, stress-induced EPI surge, and prestress cortisol. Even though the predictive power of the LASSO models was limited, we consider the variables with a high selection frequency to be relevant to the outcome. These results support the view that coagulation activation during emotionally triggered ACS may result from interacting neurobiological, psychosocial, and behavioral factors. The findings may be particularly relevant for male physicians in high-stress settings, such as emergency departments, operating rooms, or during on-call duties, where acute emotional strain could transiently elevate thrombotic risk.

As penalized regression methods like LASSO are optimized for variable selection and prediction, we conducted follow-up linear regression to estimate effect sizes and statistical significance for the retained variables. The five predictors collectively explained 50.4% of the variance, with four showing significant independent associations in the multivariable model. D-dimer stress reactivity was higher in participants with more ACEs, greater alcohol consumption, and a stronger stress-induced EPI surge, each showing small-to-medium effect sizes. In contrast, higher prestress D-dimer was associated with smaller stress-induced increases, showing a large effect, consistent with a ceiling effect where fibrin formation may be limited at already elevated baseline levels. Prestress cortisol was not independently significant, likely due to shared variance with other predictors. Similarly, several predictors identified by LASSO showed no significant zero-order associations but became significant in the multivariable model, suggesting suppression effects. This illustrates a key advantage of penalized regression methods, which identify predictors based on their joint contributions rather than isolated correlations, underscoring the value of multivariable approaches in revealing predictors that might be overlooked in simpler analyses.

The independent associations of higher ACEs and greater EPI surge with increased D-dimer stress reactivity align with expectations and support the validity of the LASSO results. Prior studies have linked ACEs to elevated cardiovascular risk,[41] including incident ACS.[42] Extending this literature, our study is, to our knowledge, the first to suggest an association between ACEs and heightened stress-induced coagulation activation in adulthood. Individuals with early adversity often show poorer emotion regulation and greater stress sensitivity, contributing to heightened physiological vulnerability, such as increased inflammation implicated in cardiovascular disease.[43] Although not directly assessed, impaired emotion regulation may underlie the stronger D-dimer response among participants with more ACEs. The broad health impacts of ACEs may also explain why ACEs emerged as a key predictor in our LASSO model, while other psychosocial factors, including burnout and its subscales, showed weaker predictive strength. Given that nearly 50% of physicians report ACEs,[44] our findings underscore the need to integrate trauma-informed care and stress interventions into occupational health programs, especially in high-stakes medical settings.

Sympathetic nervous system activity, particularly catecholamine release and adrenergic receptor signaling, is central to hemostatic changes during acute stress.[6] D-dimer levels have been shown to increase following inhalation of a short-acting β2-adrenoreceptor agonist in healthy individuals[45] and in response to NE infusion in healthy middle-aged men, suggesting additional involvement of an α-adrenergic mechanism.[46] In another study with healthy men, NE increases from pre- to poststress were directly associated with D-dimer increases, though no significant association was found between NE AUC and D-dimer over 60 minutes,[47] consistent with our findings. Our study complements these previous findings by emphasizing that EPI plays a critical role in driving the coagulation and fibrinolytic activation steps that lead to fibrin formation and degradation during acute stress.[6]

Light-to-moderate alcohol consumption may reduce CAD risk, potentially by lowering coagulation factor levels.[48] However, its effect on coagulation responses to acute stress may differ. In our sample, none met the criteria for substance use disorder, yet even light-to-moderate habitual drinking was linearly associated with heightened stress-induced coagulation activation. Given that acute emotional stress,[1] [2] [3] including during mass events like football tournaments,[49] can trigger ACS, and alcohol consumption often rises during these occasions, our findings may have public health relevance. They support educational efforts to raise awareness of the cardiovascular risks of alcohol use during acute stress and targeted prevention campaigns to reduce stress-related drinking, particularly in individuals at elevated ACS risk. Notably, we did not have data on specific drinking patterns such as binge or episodic heavy drinking, which may exert distinct effects on D-dimer responses compared with habitual consumption. Moreover, information on the timing of alcohol intake relative to the TSST was unavailable, precluding control for potential acute effects of recent drinking. Future studies should, therefore, distinguish between regular and binge drinking behaviors and account for drinking timing to clarify their respective contributions to stress-induced coagulation activation.

This study leveraged a data-driven machine learning approach to evaluate a broad range of biological, behavioral, and psychosocial predictors of coagulation reactivity in a clinically relevant cohort, providing both mechanistic and translational insights. However, the exclusive focus on male physicians limits the generalizability of our findings to women, whose HPA axis, autonomic, and coagulation stress responses may differ from men prior to menopause,[6] [50] as well as to other professional groups. While D-dimer is a robust marker of fibrin turnover, future studies should include additional coagulation markers to clarify which hemostatic pathways (e.g., thrombin formation, platelet activation) are most affected by acute stress.[6] The identified predictors were not validated in an independent sample, so future studies should confirm their generalizability and robustness, particularly for those significant only in the multivariable model.

In summary, early adversity, habitual alcohol use, and sympathoadrenal activation emerged as predictors of heightened D-dimer responses to acute psychosocial stress, highlighting the need for validation in independent cohorts. Longitudinal studies should clarify whether increased D-dimer stress reactivity predicts ACS, thereby informing strategies to mitigate ACS risk during emotionally stressful situations by targeting these predictors.

Conflict of Interest

The authors declare that they have no conflict of interest.

Contributors' Statement

R.v.K. contributed to conceptualization, formal analysis, funding acquisition, methodology, resources, supervision, and both original drafting and review and editing of the manuscript. M.G. contributed to conceptualization, formal analysis, methodology, validation, and review and editing. C.Z.H. contributed to conceptualization, data curation, investigation, methodology, supervision, and review and editing. S.A.H. contributed to conceptualization, data curation, investigation, methodology, and review and editing. S.S. contributed to data curation, investigation, methodology, and review and editing. A.P.P. contributed to conceptualization, investigation, methodology, project administration, supervision, validation, and review and editing. D.G.V. contributed to data curation, investigation, methodology, validation, and review and editing. M.P. contributed to conceptualization, data curation, investigation, methodology, project administration, supervision, and review and editing.

• References

• 1 Strike PC, Steptoe A. Behavioral and emotional triggers of acute coronary syndromes: A systematic review and critique. Psychosom Med 2005; 67 (02) 179-186
• 2 Tofler GH, Kopel E, Klempfner R, Eldar M, Buckley T, Goldenberg I. National Israel Survey of Acute Coronary Syndrome Investigators. Triggers and timing of acute coronary syndromes. Am J Cardiol 2017; 119 (10) 1560-1565
• 3 Mostofsky E, Penner EA, Mittleman MA. Outbursts of anger as a trigger of acute cardiovascular events: A systematic review and meta-analysis. Eur Heart J 2014; 35 (21) 1404-1410
• 4 Moazzami K, Sullivan S, Wang M. et al. Cardiovascular reactivity to mental stress and adverse cardiovascular outcomes in patients with coronary artery disease. J Am Heart Assoc 2025; 14 (03) e034683
• 5 von Känel R. Acute mental stress and hemostasis: When physiology becomes vascular harm. Thromb Res 2015; 135 (Suppl. 01) S52-S55
• 6 von Känel R. Stress-induced hypercoagulability: Insights from epidemiological and mechanistic studies, and clinical integration. Semin Thromb Hemost 2025; 51 (04) 381-400
• 7 Simes J, Robledo KP, White HD. et al; LIPID Study Investigators. D-dimer predicts long-term cause-specific mortality, cardiovascular events, and cancer in patients with stable coronary heart disease: LIPID Study. Circulation 2018; 138 (07) 712-723
• 8 Brügger-Andersen T, Pönitz V, Staines H, Grundt H, Hetland Ø, Nilsen DW. The prognostic utility of D-dimer and fibrin monomer at long-term follow-up after hospitalization with coronary chest pain. Blood Coagul Fibrinolysis 2008; 19 (07) 701-707
• 9 Chen R, Liu C, Zhou P. et al. Prognostic value of D-dimer in patients with acute coronary syndrome treated by percutaneous coronary intervention: A retrospective cohort study. Thromb J 2021; 19 (01) 30
• 10 Wirtz PH, Redwine LS, Baertschi C, Spillmann M, Ehlert U, von Känel R. Coagulation activity before and after acute psychosocial stress increases with age. Psychosom Med 2008; 70 (04) 476-481
• 11 Wirtz PH, Ehlert U, Emini L, Rüdisüli K, Groessbauer S, von Känel R. Procoagulant stress reactivity and recovery in apparently healthy men with systolic and diastolic hypertension. J Psychosom Res 2007; 63 (01) 51-58
• 12 von Känel R, Dimsdale JE, Adler KA, Patterson TL, Mills PJ, Grant I. Effects of depressive symptoms and anxiety on hemostatic responses to acute mental stress and recovery in the elderly. Psychiatry Res 2004; 126 (03) 253-264
• 13 von Känel R, Bellingrath S, Kudielka BM. Association of vital exhaustion and depressive symptoms with changes in fibrin D-dimer to acute psychosocial stress. J Psychosom Res 2009; 67 (01) 93-101
• 14 Tibshirani R. Regression shrinkage and selection via the lasso. J R Stat Soc Series B Stat Methodol 1996; 58 (01) 267-288
• 15 Schmidt-Trucksäss A, Lichtenstein AH, von Känel R. Lifestyle factors as determinants of atherosclerotic cardiovascular health. Atherosclerosis 2024; 395: 117577
• 16 Rotenstein LS, Torre M, Ramos MA. et al. Prevalence of burnout among physicians: A systematic review. JAMA 2018; 320 (11) 1131-1150
• 17 Kivimäki M, Nyberg ST, Batty GD. et al; IPD-Work Consortium. Job strain as a risk factor for coronary heart disease: A collaborative meta-analysis of individual participant data. Lancet 2012; 380 (9852): 1491-1497
• 18 Toker S, Melamed S, Berliner S, Zeltser D, Shapira I. Burnout and risk of coronary heart disease: A prospective study of 8838 employees. Psychosom Med 2012; 74 (08) 840-847
• 19 Sara JD, Prasad M, Eleid MF, Zhang M, Widmer RJ, Lerman A. Association between work-related stress and coronary heart disease: A review of prospective studies through the job strain, effort-reward balance, and organizational justice models. J Am Heart Assoc 2018; 7 (09) e008073
• 20 Appels A, Schouten E. Burnout as a risk factor for coronary heart disease. Behav Med 1991; 17 (02) 53-59
• 21 von Känel R, Princip M, Holzgang SA. et al. Coronary microvascular function in male physicians with burnout and job stress: An observational study. BMC Med 2023; 21 (01) 477
• 22 Büssing A, Perrar KM. Die Messung von Burnout. Untersuchung einer deutschen Fassung des Maslach Burnout Inventory (MBI-D). [Measuring burnout: A study of a German version of the Maslach Burnout Inventory (MBI-D)]. Diagnostica 1992; 38: 328-353
• 23 Kroenke K, Spitzer RL, Williams JB. The PHQ-9: Validity of a brief depression severity measure. J Gen Intern Med 2001; 16 (09) 606-613
• 24 Lee RT, Ashforth BE. A meta-analytic examination of the correlates of the three dimensions of job burnout. J Appl Psychol 1996; 81 (02) 123-133
• 25 Koutsimani P, Montgomery A, Georganta K. The relationship between burnout, depression, and anxiety: A systematic review and meta-analysis. Front Psychol 2019; 10: 284
• 26 Jönsson P, Österberg K, Wallergård M. et al. Exhaustion-related changes in cardiovascular and cortisol reactivity to acute psychosocial stress. Physiol Behav 2015; 151: 327-337
• 27 Schommer NC, Hellhammer DH, Kirschbaum C. Dissociation between reactivity of the hypothalamus-pituitary-adrenal axis and the sympathetic-adrenal-medullary system to repeated psychosocial stress. Psychosom Med 2003; 65 (03) 450-460
• 28 Steptoe A, Marmot M. Psychosocial, hemostatic, and inflammatory correlates of delayed poststress blood pressure recovery. Psychosom Med 2006; 68 (04) 531-537
• 29 SCORE2 Working Group and ESC Cardiovascular risk collaboration. SCORE2 risk prediction algorithms: New models to estimate 10-year risk of cardiovascular disease in Europe. Eur Heart J 2021; 42 (25) 2439-2454
• 30 Buysse DJ, Reynolds III CF, Monk TH, Berman SR, Kupfer DJ. The Pittsburgh Sleep Quality Index: a new instrument for psychiatric practice and research. Psychiatry Res 1989; 28 (02) 193-213
• 31 Nater UM, Rohleder N. Salivary alpha-amylase as a non-invasive biomarker for the sympathetic nervous system: Current state of research. Psychoneuroendocrinology 2009; 34 (04) 486-496
• 32 Kudielka BM, Wüst S. Human models in acute and chronic stress: Assessing determinants of individual hypothalamus-pituitary-adrenal axis activity and reactivity. Stress 2010; 13 (01) 1-14
• 33 Ehrenreich H, Schuck J, Stender N. et al. Endocrine and hemodynamic effects of stress versus systemic CRF in alcoholics during early and medium term abstinence. Alcohol Clin Exp Res 1997; 21 (07) 1285-1293
• 34 Siegrist J, Wege N, Pühlhofer F, Wahrendorf M. A short generic measure of work stress in the era of globalization: Effort-reward imbalance. Int Arch Occup Environ Health 2009; 82 (08) 1005-1013
• 35 Gräfe K, Zipfel S, Herzog W, Löwe B. Screening psychischer Störungen mit dem “Gesundheitsfragebogen für Patienten (PHQ-D)”: Ergebnisse der deutschen Validierungsstudie [Screening for psychiatric disorders with the Patient Health Questionnaire (PHQ). Results from the German validation study]. Diagnostica 2004; 50: 171-181
• 36 Spitzer RL, Kroenke K, Williams JB, Löwe B. A brief measure for assessing generalized anxiety disorder: The GAD-7. Arch Intern Med 2006; 166 (10) 1092-1097
• 37 Cohen S, Kamarck T, Mermelstein R. A global measure of perceived stress. J Health Soc Behav 1983; 24 (04) 385-396
• 38 Anda RF, Dong M, Brown DW. et al. The relationship of adverse childhood experiences to a history of premature death of family members. BMC Public Health 2009; 9: 106
• 39 Meinshausen N, Bühlmann P. Stability selection. J R Stat Soc Series B Stat Methodol 2010; 72 (04) 417-473
• 40 Templeton GF. A two-step approach for transforming continuous variables to normal: Implications and recommendations for IS research. Comm Assoc Inform Syst 2011; 28: 4
• 41 Suglia SF, Koenen KC, Boynton-Jarrett R. et al; American Heart Association Council on Epidemiology and Prevention; Council on Cardiovascular Disease in the Young; Council on Functional Genomics and Translational Biology; Council on Cardiovascular and Stroke Nursing; and Council on Quality of Care and Outcomes Research. Childhood and adolescent adversity and cardiometabolic outcomes: A scientific statement from the American Heart Association. Circulation 2018; 137 (05) e15-e28
• 42 Jacquet-Smailovic M, Brennstuhl MJ, Tarquinio CL, Tarquinio C. Relationship between cumulative adverse childhood experiences and myocardial infarction in adulthood: A systematic review and meta-analysis. J Child Adolesc Trauma 2021; 15 (03) 701-714
• 43 Mathur A, Li JC, Lipitz SR, Graham-Engeland JE. Emotion regulation as a pathway connecting early life adversity and inflammation in adulthood: A conceptual framework. Advers Resil Sci 2022; 3 (01) 1-19
• 44 Stork BR, Akselberg NJ, Qin Y, Miller DC. Adverse childhood experiences (ACEs) and community physicians: What we've learned. Perm J 2020; 24: 19-99
• 45 Ali-Saleh M, Sarig G, Ablin JN, Brenner B, Jacob G. Inhalation of a short-acting β2-adrenoreceptor agonist induces a hypercoagulable state in healthy subjects. PLoS ONE 2016; 11 (07) e0158652
• 46 von Känel R, Heimgartner N, Stutz M. et al. Prothrombotic response to norepinephrine infusion, mimicking norepinephrine stress-reactivity effects, is partly mediated by α-adrenergic mechanisms. Psychoneuroendocrinology 2019; 105: 44-50
• 47 Wirtz PH, Ehlert U, Emini L. et al. Anticipatory cognitive stress appraisal and the acute procoagulant stress response in men. Psychosom Med 2006; 68 (06) 851-858
• 48 Mukamal KJ, Jadhav PP, D'Agostino RB. et al. Alcohol consumption and hemostatic factors: analysis of the Framingham Offspring cohort. Circulation 2001; 104 (12) 1367-1373
• 49 Lin LL, Gu HY, Yao YY. et al. The association between watching football matches and the risk of cardiovascular events: A meta-analysis. J Sports Sci 2019; 37 (24) 2826-2834
• 50 Kajantie E, Phillips DI. The effects of sex and hormonal status on the physiological response to acute psychosocial stress. Psychoneuroendocrinology 2006; 31 (02) 151-178

Publication History

Received: 05 June 2025

Accepted: 27 October 2025

Accepted Manuscript online:
08 December 2025

Article published online:
19 December 2025