Unravelling the Complexity Gap: A Mechanistic Investigation of Machine Learning  Classification in Panic Disorder 

Unravelling the Complexity Gap: A Mechanistic Investigation of Machine Learning Classification in Panic Disorder | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Unravelling the Complexity Gap: A Mechanistic Investigation of Machine Learning Classification in Panic Disorder Filipe Ricardo Carvalho, Ana Teresa Martins This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-8398524/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract Background Machine learning (ML) models trained on socioeconomic, physiological, and behavioral markers can classify panic disorder (PD) with high accuracy. Yet the mechanisms underlying these predictions remain poorly understood, limiting clinical translation and theoretical integration. Objective To investigate why ML models achieve strong PD classification performance by examining feature interactions, individual contributions, model complexity requirements, and socioeconomic risk gradients. Methods Using complete-case NHANES 1999–2004 data (N = 3,144; 115 PD cases), we applied a multi-method framework including distributional analysis, dimensionality reduction (UMAP, t-SNE), decision trees, SHAP interaction analysis, and socioeconomic stratification. The primary classifier was Gradient Boosting using 11 biopsychosocial predictors. Results Individual features showed modest discriminative power ( Cohen’s d = .13–.70). SHAP identified 10 meaningful interactions, particularly body fat × age and poverty-income ratio × BMI. A shallow decision tree reached only 40.97% accuracy, indicating reliance on multidimensional interactions. Socioeconomic analysis showed a strong gradient (poorest quartile: 6.42% PD prevalence; wealthiest: 1.16%), with highest risk among low-income women. Conclusion High PD classification accuracy emerges from synergistic biopsychosocial patterns. These results clarify why ML-based classification outperforms traditional screening, identify mechanistic pathways consistent with PD models, and highlight high-risk groups for targeted intervention. Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 1. INTRODUCTION Machine learning (ML) has become increasingly prominent in psychiatry, demonstrating high predictive accuracy across a range of disorders (Chekroud et al., 2021 ). However, a critical gap remains between showing that ML works and understanding why it works. Accuracy alone provides limited clinical utility without insight into which features drive classification, how they interact, and which populations are at greatest risk (Doshi-Velez and Kim, 2017 ; Hassija et al., 2024 ). In prior work (Martins et al., 2025 ), we showed that ML models using routinely collected socioeconomic, physiological, and behavioral markers from NHANES achieved robust panic disorder (PD) classification (Precision 0.96, Recall 0.81, ROC-AUC 0.905). These results suggested that health survey data capture meaningful patterns relevant to PD diagnosis, supporting potential use in population screening. Yet several mechanistic questions remained: Do individual features strongly discriminate cases, or is integration across domains necessary? Can simple clinical rules replicate ensemble performance, or is genuine complexity required? Which socioeconomic subgroups are at highest risk, and why? Do identified patterns align with known PD pathophysiology? The current study addresses these questions through a multi-method mechanistic framework, transforming ML from a purely predictive tool into a hypothesis generator. Ensemble methods such as random forests and gradient boosting often achieve superior accuracy but lack transparency, creating challenges in clinical settings where interpretability is crucial for decision-making, patient explanation, regulatory validation, and intervention planning (Doshi-Velez and Kim, 2017 ; Guidotti et al., 2018 ; Kim et al., 2012 ). Modern interpretability methods — including SHAP, LIME, and integrated gradients — allow researchers to quantify feature contributions, explore interactions, and assess the plausibility of model predictions (Lundberg and Lee, 2017 ). Despite these advances, few psychiatric studies have systematically applied such methods to understand the mechanisms underlying high predictive performance. Panic disorder affects 2–3% of adults globally and is characterized by recurrent, unexpected panic attacks accompanied by persistent worry about future episodes (Javaid et al., 2023 ; Kessler and Merikangas, 2006 ). Contemporary biopsychosocial models posit that PD arises from interactions among biological factors (e.g., autonomic dysregulation, HPA-axis abnormalities), cognitive processes (e.g., catastrophic interpretation of bodily sensations, anxiety sensitivity), and social determinants (e.g., socioeconomic stress, childhood adversity, low social support) (Reiss et al., 1986 ; Schmidt et al., 1997 ). This multidomain complexity suggests that accurate classification likely requires integration across domains rather than reliance on single biomarkers (Chekroud et al., 2021 ). To ensure methodological rigor and reduce potential artifacts, we employed a conservative analytic approach using complete-case NHANES data, removed variables with coding irregularities, and focused on 11 core features with optimal data quality. This strategy prioritizes robustness over sample size and tests whether observed patterns reflect genuine biopsychosocial signatures rather than methodological anomalies. Our primary objective was to elucidate mechanisms through which sociodemographic and physiological features enable accurate panic disorder classification. Specifically, we sought to: (1) quantify discriminative power of individual features, (2) identify synergistic feature interactions, (3) test whether simple decision rules can replicate ensemble performance, (4) characterize socioeconomic patterns and high-risk subgroups, and (5) assess biological plausibility of identified patterns. 2. METHODS 2.1. Study Design and Data Source This mechanistic investigation used data from the National Health and Nutrition Examination Survey (NHANES) cycles 1999–2004, a nationally representative cross-sectional survey conducted by the Centers for Disease Control and Prevention to assess the health and nutritional status of adults and children in the United States. Study protocols were approved by the NCHS Research Ethics Review Board, and all participants provided written informed consent. 2.2. Study Population The initial merged dataset included 6,581 participants from the three NHANES cycles. To ensure data quality and avoid potential biases from imputation, we conducted a complete-case analysis, deliberately avoiding synthetic oversampling or multiple imputation methods. After excluding participants with missing values, the final analytical sample consisted of 3,144 individuals, including 115 cases of panic disorder (3.5%) and 3,029 controls (96.5%), consistent with epidemiological estimates from community samples. 2.3. Panic Disorder Classification Panic disorder cases were identified using the CIDPSCOR variable from the NHANES Composite International Diagnostic Interview (CIDI), a validated instrument based on DSM-IV criteria. Code 1 indicates a positive diagnosis, while Code 5 indicates absence of PD. Classification integrates symptom counts, attack characteristics, temporal patterns, associated concerns, and potential alternative aetiologies, ensuring that identified cases represent clinically relevant PD rather than isolated panic events or secondary symptoms. The study focused on a refined set of 11 features spanning sociodemographic, physiological, and behavioral domains, selected from a previous set of 16 to capture the multidimensional nature of panic disorder. 2.4. Feature Set The biological and physiological features included body fat mass, body mass index, and diastolic blood pressure, capturing aspects of cardiovascular and metabolic health. Sociodemographic features comprised age, gender, marital status, household size, and poverty-to-income ratio, reflecting social determinants of health. Behavioral and psychological features included alcohol consumption, family history of hypertension or stroke, and recent lower back pain, serving as indicators of psychological and somatic vulnerability. Detailed descriptions of all features, including NHANES variable codes, measurement protocols, and clinical interpretation, are provided in Supplementary Table S1. 2.5. Machine Learning Model We employed a Gradient Boosting Classifier implemented in scikit-learn, with hyperparameters optimized through cross-validation and preliminary experimentation. The final configuration included 200 boosting stages, a maximum tree depth of 8, a learning rate of 0.1, and balanced class weighting to address the minority prevalence of PD. Features were standardized using the StandardScaler, and any remaining missing numerical values were imputed using the median. The dataset was split into training (75%) and testing (25%) sets using stratified sampling to preserve the natural class distribution, and five-fold cross-validation was applied to optimize hyperparameters and assess model generalization. 2.6. Statistical and Machine Learning Analyses To investigate the mechanisms underlying classification performance, we first evaluated the discriminative power of individual features using descriptive statistics, effect sizes (Cohen’s d), and distributional separation tests (t-tests, Mann-Whitney U, Kolmogorov-Smirnov). SHapley Additive exPlanations (SHAP) values were computed with TreeExplainer to quantify feature importance and pairwise interactions, using the training set only to avoid data leakage. Model complexity was assessed by comparing a simple interpretable decision tree with the full gradient boosting classifier. Socioeconomic stratification involved dividing participants into poverty-to-income quartiles and further by age and gender to identify high-risk subgroups. Dimensionality reduction (UMAP, t-SNE) and clustering metrics (silhouette scores, centroid distances, separation ratios) visualized class separability. Human-readable rules were extracted from the decision tree to enhance interpretability. Sensitivity analyses confirmed robustness by repeating analyses with multiple random seeds, comparing balanced vs unweighted class weighting, and computing SHAP values across all cross-validation folds (Supplementary Table S3). 2.7. Ethical Considerations, Software, and Data Availability This study utilized publicly available, de-identified NHANES data. Secondary analysis was exempt from additional institutional review board approval, as the data contained no identifiable information and posed minimal risk. Analyses were conducted in Python 3.10 using scikit-learn, SHAP, pandas, NumPy, SciPy, and matplotlib. Random operations were seeded for reproducibility. The analytical code is publicly available at GitHub. Table 1 Sample Characteristics. Sample characteristics stratified by panic disorder status (N = 3,144; 115 PD, 3,029 controls). Values shown as mean ± SD or n(%). Statistical comparisons performed using t-tests for continuous variables and chi-square tests for categorical variables. Characteristic Total (N = 3,144) Normal (n = 3,029) Panic Disorder (n = 115) p-value DEMOGRAPHICS Age (years), mean ± SD 29.5 ± 5.9 29.5 ± 5.9 30.6 ± 4.8 0.044 Gender (% Female) 55.7% 54.6% 80.0% < 0.001 SOCIOECONOMIC Poverty-Income Ratio, mean ± SD 2.87 ± 1.61 2.91 ± 1.61 2.06 ± 1.47 < 0.001 Household Size, mean ± SD 3.22 ± 1.56 3.24 ± 1.57 3.05 ± 1.32 0.215 PHYSIOLOGICAL Body Fat Mass (kg), mean ± SD 23.05 ± 11.15 22.91 ± 11.04 26.25 ± 13.68 0.002 Body Mass Index, mean ± SD 26.80 ± 5.77 26.77 ± 5.79 27.26 ± 5.49 0.369 Diastolic BP (mmHg), mean ± SD 69.38 ± 11.38 69.42 ± 11.54 68.85 ± 5.61 0.601 3. RESULTS 3.1 Sample Characteristics and Individual Feature Discriminative Power The conservative analysis included 3,144 participants with complete data on all 11 features (Table 1 ), representing 47.8% of the original NHANES 1999–2004 panic disorder sample. This subset comprised 115 panic disorder (PD) cases (3.66%) and 3,029 normal controls (96.34%), maintaining a prevalence ratio consistent with epidemiological estimates. Females comprised 80.0% of the PD group versus 54.6% of the normal group (p < .001, Cohen's d = .700), the strongest individual discriminator. Age differences were minimal (PD: 30.6 ± 4.8 years vs Normal: 29.5 ± 5.9 years, p = .044, d = .191). Socioeconomic features demonstrated moderate effects. Poverty-income ratio was lower in PD (2.06 ± 1.47 vs 2.91 ± 1.61, p < .001, d=-.531), whereas household size showed negligible difference (d=-.118, p = .215). Physiological measurements revealed subtle differences. Body fat mass was modestly elevated in PD (26.25 ± 13.68 kg vs 22.91 ± 11.04 kg, p = .002, d = .300), while BMI (p = .369) and diastolic blood pressure (p = .601) showed minimal differences. Alcohol consumption showed no association (p = .997, d = 0.000). Table 2 Individual Feature Discriminative Power. Discriminative analysis of 11 features showing effect sizes (Cohen's d), separation percentages, and statistical significance. Features ranked by absolute effect size with domain classification (Biological, Social, Demographic, Psychological). Feature PD Mean ± SD Normal Mean ± SD Cohen's d Separation % p-value Gender 1.80 ± 0.40 1.45 ± 0.50 + 0.700 31.0% < 0.001 Family History HTN/Stroke 1.33 ± 0.47 1.61 ± 0.49 -0.566 22.7% < 0.001 Lower Back Pain 1.37 ± 0.48 1.63 ± 0.48 -0.559 22.0% < 0.001 Poverty-Income Ratio 2.06 ± 1.47 2.91 ± 1.61 -0.531 22.0% < 0.001 Body Fat Mass (kg) 26.25 ± 13.68 22.91 ± 11.04 + 0.300 14.2% 0.002 Marital Status 2.89 ± 1.62 3.35 ± 2.00 -0.232 13.6% 0.015 Age (months) 367.1 ± 58.1 353.8 ± 70.2 + 0.191 11.8% 0.044 Household Size 3.05 ± 1.32 3.24 ± 1.57 -0.118 9.3% 0.215 BMI (kg/m²) 27.26 ± 5.49 26.77 ± 5.79 + 0.085 4.1% 0.369 Diastolic BP (mmHg) 68.85 ± 5.61 69.42 ± 11.54 -0.050 2% 0.601 Alcohol (drinks/day) 3.26 ± 1.99 3.26 ± 3.96 0.000 0% 0.997 Mean |d| = 0.284 Mean = 19.4% Behavioral and clinical features showed mixed patterns. Lower back pain was paradoxically less common in PD ( d =-.559, p < .001), as was family history of hypertension/stroke ( d =-.566, p 0.8), with maximum separation 31.0% for gender and mean separation across all features 19.4%. Four features had medium effects (|d| = 0.5–0.8): gender, family history, lower back pain, and poverty-income ratio, while five features showed negligible effects (|d| < 0.2): age, household size, BMI, blood pressure, and alcohol consumption. The mean absolute Cohen’s d across all 11 features was 0.284 (small effect). Domain-level analysis revealed that univariately, the psychological domain had the strongest effects (mean |d| = 0.563), followed by social (0.295) and biological (0.126). However, multivariate analysis (Section 3.3 ) reversed this pattern, emphasizing the difference between marginal and conditional effects. Full distributional analyses and visualizations are provided in Supplementary Figures S2–S3. These univariate findings established a fundamental paradox: if individual features show weak-to-moderate discrimination at best (maximum 31% separation), how does the gradient boosting ensemble achieve 99.75% accuracy? This question framed the subsequent analyses examining multidimensional complexity and synergistic interactions. 3.2 Multidimensional Complexity and the Necessity of Ensemble Methods Dimensionality reduction analysis (Fig. 1 ) revealed the core mechanism underlying model performance. When the 11-dimensional feature space was projected onto two dimensions using PCA and t-SNE, substantial visual overlap emerged between panic disorder and normal groups despite the gradient boosting model achieving 99.75% test set accuracy. PCA projection yielded Silhouette score of 0.32 and Davies-Bouldin index of 2.14; t-SNE projection showed Silhouette score of 0.38 and Davies-Bouldin index of 1.89. These moderate cluster quality metrics in 2D space stood in stark contrast to near-perfect classification in 11D space. Table 3 Model Performance Comparison - The Complexity Gap. Comparison of decision tree (max_depth = 5) vs gradient boosting classifier performance on test set (n = 786; 29 PD, 757 controls). Performance metrics include accuracy, precision, recall, F1-score, ROC-AUC, and confusion matrix components. *** p < 0.001 (McNemar's test). Metric Decision Tree (Interpretable) Gradient Boosting (Ensemble) Difference (GB - DT) Accuracy (%) 40.97 99.75 + 58.78*** Precision (%) 5.88 100.00 + 94.12*** Recall (%) 100.00 93.10 -6.90 F1-Score 0.111 0.964 + 0.853*** True Negatives 293 756 + 463 False Positives 464 0 -464 Total Errors 464 2 -461 Accurate discrimination required integration across all 11 features, as the separating hyperplane existed only in 11-dimensional space, collapsing into overlapping distributions in 2D projections. The model misclassified only 2 cases out of 786: 2 false negatives (6.9% of PD cases), scattered rather than clustered near decision boundaries. To quantify the complexity gap between interpretable and high-performing models, we compared a simple decision tree (max_depth = 5) against the gradient boosting ensemble (Table 3 , Fig. 3 ). The decision tree achieved only 40.97% accuracy with catastrophic 464 false positives, misclassifying 61.3% of normal controls as panic disorder. Importantly, despite the dominance of poverty-income ratio within the decision tree, the full gradient boosting model did not rely on any single predictor. Body fat mass and poverty-income ratio each contributed only ~ 16% to overall importance, indicating that high performance emerged from integration across multiple domains rather than from dominant individual features. Table 4 SHAP Feature Importance and Class-Stratified Analysis. SHAP-derived feature importance showing global mean absolute SHAP values, class-stratified contributions (PD vs Normal), and domain-level aggregation. Features ranked by importance with theoretical domain classification. Rank Feature Mean SHAP (PD Group) Mean SHAP (Normal) % Total Importance 1 Body Fat Mass + 1.915 -3.019 16.1% 2 Poverty-Income Ratio + 1.720 -3.207 16.1% 3 Age + 1.200 -2.590 12.4% 4 BMI + 0.803 -2.616 11.2% 5 Gender + 1.789 -1.047 9.3% 6 Diastolic BP + 0.655 -2.017 8.7% 7 Lower Back Pain + 1.550 -0.424 6.4% 8 Alcohol + 0.217 -1.482 5.5% 9 Family History + 0.733 -0.483 4.0% 10 Household Size -0.049 -0.880 2.7% 11 Marital Status -0.174 -0.911 2.4% In contrast, the gradient boosting ensemble achieved 99.75% accuracy with 100% precision (zero false positives) and 93.1% recall. This represented a dramatic 58.78 percentage point accuracy gap, which we term the "complexity gap"—quantifying the genuine multidimensional patterns that cannot be captured by simple, human-interpretable rules. The ROC analysis confirmed superior discriminative ability: decision tree AUC = .544 versus gradient boosting AUC = .922. The “complexity gap” highlighted that simple decision rules fail for panic disorder, sometimes performing worse than chance. Interpretability and accuracy are inherently trade-offs: a decision tree achieved 40,97% accuracy, whereas the gradient boosting ensemble reached 99.75%, confirming genuine multidimensional complexity (Hypothesis 3) and invalidating threshold-based approaches. UMAP visualizations further illustrate this complex structure. 3.3 Feature Importance and Synergistic Interaction Networks SHAP analysis (Fig. 2 , Table 4 ) revealed distributed importance across all 11 features, with no dominant predictor. Fat mass and poverty-income ratio were the strongest individual contributors (~ 16% each), yet their effects were modest and depended on interactions with demographic and biological factors. The top 5 features collectively represented only 65.1% of importance, requiring integration across all features for accurate prediction. Feature ranking by mean absolute SHAP value yielded: body fat mass (16.1%), poverty-income ratio (16.1%), age (12.4%), BMI (11.2%), gender (9.3%), diastolic blood pressure (8.7%), lower back pain (6.4%), alcohol consumption (5.5%), family history (4.0%), household size (2.7%), and marital status (2.4%). Notably, several features with weak univariate effects (e.g., blood pressure: Cohen's d =-.050) demonstrated substantial multivariate importance (8.7%), while features with strong univariate effects (e.g., gender: d = .700) showed only moderate multivariate importance (9.3%). Class-stratified SHAP analysis (Table 4 ) revealed asymmetric contributions. Body fat showed strongly positive mean SHAP in PD cases (+ 1.915) and strongly negative in normal controls (-3.019), with difference |Δ|=4.934 indicating high discriminative power. Poverty-income ratio displayed similar magnitude (+ 1.720 vs -3.207, |Δ|=4.927). Age, BMI, and blood pressure showed moderate differences. The beeswarm distribution (Fig. 2 ) illustrated heterogeneity: substantial horizontal spread within each feature row indicated that identical feature values could push predictions in opposite directions depending on other features present—direct visual evidence of interaction effects. At the domain level, biological variables showed (Fig. 6 , Panel B) the largest aggregate importance (~ 36%), but this difference was modest, and all domains contributed meaningfully to prediction. These findings reinforce that the model’s performance reflects distributed, cross-domain integration rather than reliance on a single risk dimension. All domains contributed meaningfully (15.9–36%) with no statistical difference between domains (p > 0.05), validating Hypothesis 2 (distributed biopsychosocial integration). The biological domain's emergence as most important multivariate despite weakest univariate effects exemplified synergistic amplification. Pairwise interaction analysis (Table 5 , Fig. 4 ) identified systematic synergies invisible in univariate examination. Body fat mass functioned as a frequent interaction hub, participating in 6 of the top 10 interactions (60%): with age (strength = 1.047, strongest interaction observed), gender (0.690), alcohol consumption (.670), BMI (.646), poverty-income ratio (.491), and diastolic blood pressure. Age functioned as a universal moderator, showing strong interactions with four features (40%): body fat (1.047), blood pressure (.807), poverty ratio (.661), and lower back pain (.473). A critical pattern emerged: the strongest interactions systematically SPANNED theoretical domains rather than occurring within domains. The top 3 interactions were biological×demographic (body fat × age, 1.047), social×biological (poverty ratio × BMI, 1.016), and demographic×biological (age × blood pressure, 0.807). This cross-domain synergy validated Hypothesis 2 and explained the apparent paradox of weak individual effects producing high ensemble accuracy—predictive power emerged from complex conditional relationships across biopsychosocial domains rather than simple additive contributions. The interaction heatmap (Fig. 4 ) visualized the complete 11×11 matrix of pairwise synergies, with body fat's hub status evident in its row/column showing 6 strong interactions (marked in dark red, strength > 0.47). Age's moderator function appeared in its 4 strong interactions spanning multiple domains. These patterns confirmed Hypothesis 1 (synergistic feature interactions) and demonstrated why computational methods were necessary: the relevant patterns existed not in individual features but in their multivariate interactions. Complete SHAP interaction matrix for all 55 feature pairs is provided in Supplementary Table S2. SHAP importance bar chart (Supplementary Figure S10), dependence plots (Supplementary Figure S11), and detailed interaction visualizations (Supplementary Figure S12) provide comprehensive interpretation. 3.4 Socioeconomic Stratification and Identification of Highest-Risk Subgroups Quartile analysis of poverty-income ratio (Table 6 , Part A) revealed a robust dose-response relationship with panic disorder prevalence. Moving from wealthiest quartile Q4 (PIR > 4.51, prevalence 1.16%) to poorest Q1 (PIR < 1.44, prevalence 6.42%) represented a dramatic 5.5-fold increase in risk. Each quartile increment was associated with approximately 40% reduction in PD risk (Q1→Q2: -36%, Q2→Q3: -28%, Q3→Q4: -60%). Chi-square test for linear trend was highly significant (χ²=32.60, p = 3.9×10⁻⁷), and odds ratios using Q4 as reference showed progressive elevation: Q3 OR = 2.58 (95% CI: 1.24–5.37), Q2 OR = 3.66 (1.80–7.44), Q1 OR = 5.89 (2.95–11.76), all p < 0.05. Stratified analysis revealed substantial effect modification by demographics (Fig. 5 ). Age×Poverty interaction (Table 6 , Part B, p = .002) showed poverty's effect was AMPLIFIED in older adults. Among older participants (≥ median age ~ 30 years), prevalence ranged from 10.40% in Q1 (poorest) to 2.03% in Q4 (wealthiest), representing 5.1-fold difference. Among younger participants, the gradient was less pronounced: Q1 showed 3.64% versus Q4 showing 0.00% (complete protection through wealth in youth, though based on small cell size n = 15). Poor older adults displayed the HIGHEST age-stratified prevalence observed (10.40%, approximately 1 in 10). Gender×Poverty interaction (Table 6 , Part C, p < 0.001) revealed even more extreme disparities. Poor women (Q1) demonstrated HIGHEST OVERALL prevalence of 11.82%—nearly 1 in 8 women living below 150% of federal poverty line met criteria for panic disorder. This represented 4.6-fold increase versus wealthy women (Q4: 2.62%) and 4.6-fold increase versus poor men (Q1: 2.59%). The poverty gradient was strongly linear in women (11.82%→7.87%→3.74%→2.62% across quartiles, p < .001) but inconsistent in men (2.59%→0.74%→2.06%→0.00%, p = .132), suggesting gender-specific vulnerability mechanisms. Table 5 Top 10 Feature Interactions from SHAP Analysis. Top 10 pairwise feature interactions from SHAP interaction analysis, ranked by interaction strength. Domain pairs indicate cross-domain (biological×demographic) vs within-domain interactions. Rank Feature Pair Interaction Strength Interpretation 1 Body Fat × Age 1.047 Adiposity effect increases with age 2 PIR × BMI 1.016 SES modulates body composition effects 3 Age × BP 0.807 Age moderates BP-PD relationship 4 Body Fat × Gender 0.690 Sex-specific adiposity effects 5 Body Fat × Alcohol 0.670 Metabolic-behavioral interaction 6 PIR × Age 0.661 Cumulative adversity pathway 7 Body Fat × BMI 0.646 Body composition synergy 8 BMI × Pain 0.564 Weight-pain-anxiety link 9 Body Fat × PIR 0.491 SES-metabolic pathway 10 Age × Pain 0.473 Chronic pain relevance increases with age Table 6 Panic Disorder Prevalence by Poverty-Income Ratio Quartiles. PART A. PRIMARY QUARTILE ANALYSIS. Panic disorder prevalence by poverty-income ratio quartiles with chi-square test for linear trend and odds ratios (Q4 reference). PART A: PRIMARY QUARTILE ANALYSIS Sample: N = 3,144 (115 PD, 3,029 Normal) | NHANES 1999–2004 Quartile PIR Range N PD Cases PD Prevalence Odds Ratio 95% CI p-value Q1 0.00-1.44 794 51 6.42% 5.89 [2.95–11.76] <0.001 (Poorest) (Below FPL) (ref: Q4) Q2 1.44–2.77 785 32 4.08% 3.66 [1.80–7.44] 4×FPL) (ref) PART B. AGE × POVERTY STRATIFICATION. Panic disorder prevalence stratified by age (median split) and poverty quartile. Interaction p-value from logistic regression. Age Group Q1 (Poorest) Q2 Q3 Q4 (Wealthiest) Trend p Prevalence Prevalence Prevalence Prevalence Younger 3.64% 1.79% 3.53% 0.00% 0.046 (< median age) (n = 11/302) (n = 7/392) (n = 14/397) (n = 0/364) Older 10.40% 6.37% 2.40% 2.03% <0.001 (≥ median age) (n = 40/492) (n = 25/393) (n = 9/392) (n = 9/412) Interaction p = 0.002 Intersection analysis (Table 6 , Part D) identified multiplicative risk: poor older women showed estimated 15–18% PD prevalence (based on multiplicative model combining age and gender effects), representing 4.1–4.9× the population average. This intersection of age, gender, and poverty created a "perfect storm" of vulnerability, identifying the highest-priority population for screening and intervention. Poor older adults combined (gender-aggregated) showed 10.40% prevalence (2.8× average), and poor women across all ages showed 11.82% (3.2× average). In contrast, wealthy young men represented a protected subgroup with 0.00% prevalence. Phenotypic heterogeneity analysis (Table 6 , Part E) revealed distinct PD profiles by socioeconomic status. Among PD cases only (n = 115), comparing low-PIR (below median, n = 83, 72.2%) versus high-PIR (above median, n = 32, 27.8%) showed wealthier cases were significantly older (+ 31.5 months, p = .009), had more body fat (+ 7.1 kg, p = .012), reported more lower back pain (+ 31.6 percentage points, p = .001), but had less family history of hypertension/stroke (-19.8 percentage points, p = .043). This pattern suggested different etiological pathways: poorer PD characterized by younger onset, lower adiposity, and higher genetic loading; wealthier PD characterized by later onset, metabolic components, and chronic pain comorbidity. PART C. GENDER × POVERTY STRATIFICATION. Panic disorder prevalence stratified by gender and poverty quartile. Interaction p-value from logistic regression. Gender Q1 (Poorest) Q2 Q3 Q4 (Wealthiest) Trend p Prevalence Prevalence Prevalence Prevalence Male 2.59% 0.74% 2.06% 0.00% 0.132 (n = 9/347) (n = 3/404) (n = 8/388) (n = 0/395) Female 11.82% 7.87% 3.74% 2.62% < 0.001 (n = 42/447) (n = 29/381) (n = 15/401) (n = 9/381) PART D: HIGHEST RISK SUBGROUPS. Highest-risk subgroups identified through demographic intersection, with prevalence estimates and fold-increase relative to population average (3.66%). Risk Group PD Prevalence Sample Size vs Population Average Poor Older Women (Q1 × Older × Female) ~15–18%* ~250 4.1–4.9× higher Poor Older Adults (Q1 × Older, combined) 10.40% 492 2.8× higher Poor Women (Q1 × Female, all ages) 11.82% 447 3.2× higher Wealthy Young Men (Q4 × Younger × Male) 0.00% ~180* Protected PART E: PD PHENOTYPE BY SOCIOECONOMIC STATUS. Feature comparison among PD cases only (n = 115), stratified by median poverty-income ratio. T-tests and chi-square tests for group differences. Feature Low PIR PD High PIR PD Difference p-value Interpretation Mean ± SD Mean ± SD Age (months) 358.3 ± 56.5 389.8 ± 54.8 +31.5 0.009 Wealthier PD: OLDER Body Fat Mass (kg) 24.3 ± 13.1 31.4 ± 13.7 +7.1 0.012 Wealthier PD: MORE fat Lower Back Pain (% yes) 27.7% 59.4% +31.6% 0.001 Wealthier PD: MORE pain Family History (% yes) 38.6% 18.8% -19.8% 0.043 Poorer PD: MORE FH These socioeconomic patterns validated Hypothesis 4 (socioeconomic stratification creates distinct risk profiles) and carried profound public health implications. The extreme disparities observed—with poor women all ages facing nearly 12% prevalence while wealthy young men showed 0%—indicated that universal screening protocols would be inefficient. Targeted, community-based approaches focusing on socioeconomically disadvantaged populations, particularly older women, would maximize public health yield with number needed to screen of approximately 8 to detect one case in highest-risk subgroups. Poverty distributions (Supplementary Figure S15), prevalence visualization (Supplementary Figure S16), and correlations (Supplementary Figure S17) detail socioeconomic effects. 4. DISCUSSION This multi-method mechanistic investigation revealed that panic disorder classification with near-perfect accuracy (99.75%) emerges from complex, synergistic interactions across 11 biopsychosocial features rather than from simple decision rules or dominant individual predictors. Across all analytic approaches, no single feature emerged as a dominant driver of panic disorder classification. Even the strongest individual predictors—body fat mass and poverty-income ratio—accounted for only ~ 16% of total importance, underscoring that accurate prediction depended on synergistic interactions across biological, demographic, and social factors. The necessity of model complexity was quantified through direct comparison of a simple interpretable decision tree (maximum depth 5) with the full gradient boosting classifier. The simple model achieved only 40.97% accuracy, whereas the complex model reached 99.75%, yielding a "complexity gap" of 58.78 percentage points. This substantial gap provides empirical evidence that panic disorder patterns are inherently multidimensional and require nonlinear feature interactions for accurate classification. These findings align with theoretical conceptualizations of panic disorder as a complex biopsychosocial phenomenon (Clark, 1986 ; Kyriakoulis and Kyrios, 2023 ) and extend recent work demonstrating that machine learning approaches can capture psychiatric disorder patterns that traditional linear models cannot adequately represent (Dwyer et al., 2018 ). The magnitude of this complexity gap suggests that efforts to develop simple screening tools based on single biomarkers or brief questionnaires may be fundamentally limited in their ability to achieve the sensitivity and specificity needed for effective population-level screening. Although body fat mass appeared as a central interaction hub, it did not dominate the model on its own. Its contribution remained modest and gained predictive relevance only through interactions with age, BMI, poverty-income ratio, and gender. This pattern illustrates the model’s reliance on distributed, multi-feature integration. This pattern suggests that body composition may serve as a critical mediator linking biological vulnerability with psychosocial risk factors, consistent with research on metabolic-inflammatory hypotheses in anxiety disorders(Gariepy et al., 2010 ; Strine et al., 2008 ) Age functioned as a universal moderator, appearing in five of the top ten interactions, suggesting that vulnerability profiles change across the lifespan. These findings support dimensional models of psychopathology that emphasize transdiagnostic risk factors operating through complex, context-dependent mechanisms rather than simple additive effects (Kotov et al., 2017 ). Socioeconomic stratification revealed a linear gradient in panic disorder prevalence across poverty quartiles, with the poorest quartile showing 6.42% prevalence compared to 1.16% in the wealthiest quartile. Women living in poverty exhibited the highest observed prevalence at 11.82%. When further stratified by older age, poor older women reached an estimated prevalence of approximately 15–18%. This finding identifies a critical high-risk subgroup that may benefit from targeted screening and early intervention, consistent with social determinants of health frameworks. Our investigation also resolved an apparent paradox wherein SHAP analysis suggested that higher income increased panic disorder risk while raw prevalence data showed the opposite pattern. This contradiction exemplifies Simpson's Paradox, a statistical phenomenon where associations observed in aggregate data can reverse when stratified by confounding variables (Pearl, 2014 ). In our data, poverty correlates negatively with age, body fat mass, and female gender, all of which are themselves risk factors for panic disorder. When the model controls for these confounders through SHAP analysis, the marginal effect of income conditional on all other features can appear paradoxical. The distributed nature of predictive power across biological, psychological, and social domains supports integrative biopsychosocial models of panic disorder etiology (Gorman et al., 2000 ). Contrary to expectations, psychological domain features (family history of cardiovascular disease, lower back pain) showed stronger mean effect sizes (Cohen's d = 0.56) than biological features (Cohen's d = 0.13), challenging views that panic disorder is primarily a disorder of physiological dysregulation. These findings align with the neurovisceral integration model (Thayer et al., 2012 ) and suggest that panic disorder emerges from dysregulation of bidirectional brain-body communication rather than isolated autonomic dysfunction. From a clinical perspective, our findings have several important implications. First, the near-perfect classification achieved with readily available sociodemographic, physiological, and behavioral markers suggests potential for developing scalable screening tools that do not require specialized psychiatric assessment or expensive biomarker assays. The model's precision ensures that individuals flagged as high-risk genuinely warrant further evaluation, avoiding unnecessary follow-up assessments. However, the recall indicates that approximately 6.9% panic disorder cases would not be detected, highlighting the continued importance of clinical judgment and comprehensive psychiatric evaluation. Second, the identification of poor older women as a high-risk group suggests that screening efforts may be most efficient when targeted toward this vulnerable population. Third, the distributed pattern of risk factors across multiple domains underscores the need for multidisciplinary treatment approaches addressing biological, psychological, and social contributors to panic disorder. Several methodological strengths enhance confidence in our findings. Our conservative analytical approach, employing complete-case analysis without synthetic oversampling, ensures that performance metrics reflect real-world classification scenarios rather than artificially inflated estimates. The removal of data artifacts demonstrates our commitment to identifying genuine clinical signals rather than capitalizing on data quality issues. The use of multiple convergent analytical approaches, including distributional analysis, SHAP interpretation, decision tree comparison, and dimensionality reduction, provides triangulated evidence supporting our conclusions. Nevertheless, several limitations warrant consideration. First, the cross-sectional design precludes causal inference regarding the directionality of observed associations. Second, the relatively small number of panic disorder cases (n = 115) limits statistical power. Third, our feature set does not include genetic markers, direct measures of autonomic function, or neuroimaging data. Fourth, the NHANES sample may not generalize to other cultural contexts. Fifth, reliance on the CIDI diagnostic algorithm may not perfectly correspond to clinical diagnoses made through comprehensive psychiatric evaluation. Future research should address these limitations through validation in independent longitudinal samples, incorporation of additional biological, genetic, and neurophysiological measures (Costello et al., 2019 ; Fonzo and Etkin, 2017 ; Hettema et al., 2001 ; Thayer et al., 2012 ), and exploration of whether similar patterns of distributed discriminative power and synergistic interactions characterize other anxiety disorders. Prospective studies could evaluate whether machine learning-based screening improves clinical outcomes, and qualitative research may elucidate barriers in high-risk subgroups. In conclusion, this mechanistic investigation demonstrates that near-perfect panic disorder classification emerges from complex, synergistic interactions across biopsychosocial features rather than from simple decision rules or dominant individual predictors. The substantial complexity gap between simple and sophisticated models quantifies the genuinely multidimensional nature of panic disorder patterns, validating the use of machine learning approaches for psychiatric screening and risk stratification. The discovery that women living in poverty exhibit high prevalence, rising further with age, identifies a critical high-risk population for targeted intervention. These findings advance theoretical understanding of panic disorder as a complex biopsychosocial phenomenon while demonstrating practical potential for scalable screening tools. Declarations Author Contributions Ana Teresa Martins: Investigation, Data Curation, Software, Writing – Original Draft, Visualization. Filipe Ricardo Carvalho: Formal analysis, Conceptualization, Methodology, Validation, Writing – Review & Editing, Supervision, Project administration. DATA AVAILABILITY STATEMENT The data used in this study are publicly available from the National Health and Nutrition Examination Survey (NHANES) 2017-2020, accessible at https://www.cdc.gov/nchs/nhanes/index.htm. All analysis code is available on GitHub at https://github.com/frpcarvalho/panic-mechanisms. Ethics Statement This secondary analysis used de-identified public NHANES data. Original NHANES data collection was approved by the National Center for Health Statistics Research Ethics Review Board with informed consent from all participants. No additional ethics approval was required for this study. Funding This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors. Declaration of Competing Interests The author declares no competing interests. References Chekroud AM, Bondar J, Delgadillo J, Doherty G, Wasil A, Fokkema M, Cohen Z, Belgrave D, DeRubeis R, Iniesta R, Dwyer D, Choi K (2021) The promise of machine learning in predicting treatment outcomes in psychiatry. World Psychiatry 20:154–170. https://doi.org/10.1002/WPS.20882 Clark DM (1986) A cognitive approach to panic. Behav Res Ther 24:461–470. https://doi.org/10.1016/0005-7967(86)90011-2 Costello H, Gould RL, Abrol E, Howard R (2019) Systematic review and meta-analysis of the association between peripheral inflammatory cytokines and generalised anxiety disorder. BMJ Open 9:e027925. https://doi.org/10.1136/BMJOPEN-2018-027925 Doshi-Velez F, Kim B (2017) Towards A Rigorous Science of Interpretable Machine Learning Dwyer DB, Falkai P, Koutsouleris N (2018) Machine Learning Approaches for Clinical Psychology and Psychiatry. Annu Rev Clin Psychol 14:91–118. https://doi.org/10.1146/ANNUREV-CLINPSY-032816-045037 Fonzo GA, Etkin A (2017) Affective neuroimaging in generalized anxiety disorder: an integrated review. Dialogues Clin Neurosci 19:169–179. https://doi.org/10.31887/DCNS.2017.19.2/GFONZO Gariepy G, Nitka D, Schmitz N (2010) The association between obesity and anxiety disorders in the population: a systematic review and meta-analysis. Int J Obes (Lond) 34:407–419. https://doi.org/10.1038/IJO.2009.252 Gorman JM, Kent JM, Sullivan GM, Coplan JD (2000) Neuroanatomical hypothesis of panic disorder, revised. Am J Psychiatry 157:493–505. https://doi.org/10.1176/APPI.AJP.157.4.493 Guidotti R, Monreale A, Ruggieri S, Turini F, Giannotti F, Pedreschi D (2018) A Survey Of Methods For Explaining Black Box Models. ACM Comput Surv 51. https://doi.org/10.1145/3236009 Hassija V, Chamola V, Mahapatra A, Singal A, Goel D, Huang K, Scardapane S, Spinelli I, Mahmud M, Hussain A (2024) Interpreting Black-Box Models: A Review on Explainable Artificial Intelligence. Cognit Comput 16:45–74. https://doi.org/10.1007/s12559-023-10179-8 Hettema JM, Neale MC, Kendler KS (2001) A review and meta-analysis of the genetic epidemiology of anxiety disorders. Am J Psychiatry 158:1568–1578. https://doi.org/10.1176/APPI.AJP.158.10.1568 Javaid SF, Hashim IJ, Hashim MJ, Stip E, Samad MA, Ahbabi A, Al (2023) Epidemiology of anxiety disorders: global burden and sociodemographic associations. Middle East Curr Psychiatry 2023 30(1 30):44. https://doi.org/10.1186/S43045-023-00315-3 Kessler RC, Merikangas KR (2006) The National Comorbidity Survey Replication (NCS-R): background and aims. International Journal of Methods in Psychiatric Research Kim JE, Dager SR, Lyoo IK (2012) The role of the amygdala in the pathophysiology of panic disorder: evidence from neuroimaging studies. Biol Mood Anxiety Disord 2. https://doi.org/10.1186/2045-5380-2-20 Kotov R, Waszczuk MA, Krueger RF, Forbes MK, Watson D, Clark LA, Achenbach TM, Althoff RR, Ivanova MY, Bagby M, Brown R, Carpenter TA, Caspi WT, Moffitt A, Eaton TE, Forbush NR, Goldberg KT, Hasin D, Hyman D, Lynam SE, Samuel DR, South DB, Markon SC, Miller K, Morey JD, Mullins-Sweatt LC, Ormel SN, Patrick J, Regier CJ, Rescorla DA, Ruggero L, Sellbom CJ, Simms M, Skodol LJ, Slade AE, Tackett T, Waldman JL, Widiger ID, Wright TA, Zimmerman AGC, M (2017) The Hierarchical Taxonomy of Psychopathology (HiTOP): A dimensional alternative to traditional nosologies. J Abnorm Psychol 126:454–477. https://doi.org/10.1037/ABN0000258 Kyriakoulis P, Kyrios M (2023) Biological and cognitive theories explaining panic disorder: A narrative review. Front Psychiatry Volume. https://doi.org/10.3389/fpsyt.2023.957515 . 14-2023 Lundberg SM, Lee SI (2017) A Unified Approach to Interpreting Model Predictions. Adv Neural Inf Process Syst 2017-December, 4766–4775 Martins AT, Gomes A, Ros A, Santos J, Carvalho FR (2025) Predicting Panic Disorder with Socioeconomic, Physiological, and Behavioral Markers: A Machine Learning Study. PsyArXiv Preprints Pearl J (2014) Comment: Understanding Simpson’s paradox. Am Stat 68:8–13. https://doi.org/10.1080/00031305.2014.876829 ;CTYPE:STRING:JOURNAL Reiss S, Peterson RA, Gursky DM, McNally RJ (1986) Anxiety sensitivity, anxiety frequency and the prediction of fearfulness. Behav Res Ther 24:1–8. https://doi.org/10.1016/0005-7967(86)90143-9 Schmidt NB, Lerew DR, Jackson RJ (1997) The role of anxiety sensitivity in the pathogenesis of panic: prospective evaluation of spontaneous panic attacks during acute stress. J Abnorm Psychol 106:355–364. https://doi.org/10.1037//0021-843X.106.3.355 Strine TW, Chapman DP, Balluz L, Mokdad AH (2008) Health-related quality of life and health behaviors by social and emotional support. Their relevance to psychiatry and medicine. Soc Psychiatry Psychiatr Epidemiol 43:151–159. https://doi.org/10.1007/S00127-007-0277-X Thayer JF, Åhs F, Fredrikson M, Sollers JJ, Wager TD (2012) A meta-analysis of heart rate variability and neuroimaging studies: Implications for heart rate variability as a marker of stress and health. Neurosci Biobehav Rev 36:747–756. https://doi.org/10.1016/j.neubiorev.2011.11.009 Additional Declarations No competing interests reported. Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-8398524","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":569803080,"identity":"ca0162f9-3d25-4a9e-bc14-74c971a33b94","order_by":0,"name":"Filipe Ricardo Carvalho","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAAz0lEQVRIie3RMQuCQBTA8QfCuTyaOxT7Cg+ERIL6KkLQVNBoVHYQOPl5nAVBF91zlKCpLYiChpTaPbeI+/Hg3fLnDQegKD9IF59tEmgJwFoiweS7CZgHQM2TySfNyCVGWt824RQdPboPH7QHZyQ6ksHC5mU4RzcqYx5RDm7YcWWGCFwUGtJpFQ+RMmim4wqi9hTFoUmWF/6STBgXftomzEDaySRs7Ao/Ryoye2JSgm7odSXauRK0tSg/1tXVDyyn/dI+0vZ3+gmgd6IoivL33lhqOg8nM6raAAAAAElFTkSuQmCC","orcid":"","institution":"University of Algarve","correspondingAuthor":true,"prefix":"","firstName":"Filipe","middleName":"Ricardo","lastName":"Carvalho","suffix":""},{"id":569803082,"identity":"da4ee1e1-415a-4c7b-8d1a-7651180d03da","order_by":1,"name":"Ana Teresa Martins","email":"","orcid":"","institution":"University of Algarve","correspondingAuthor":false,"prefix":"","firstName":"Ana","middleName":"Teresa","lastName":"Martins","suffix":""}],"badges":[],"createdAt":"2025-12-18 19:38:12","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-8398524/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-8398524/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":99814961,"identity":"1c190453-a7f3-4b3f-ae73-df4ece0cc712","added_by":"auto","created_at":"2026-01-08 14:43:13","extension":"docx","order_by":0,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":1452405,"visible":true,"origin":"","legend":"","description":"","filename":"manuscriptpanic.docx","url":"https://assets-eu.researchsquare.com/files/rs-8398524/v1/c91f1c3cfce9ee447aa83b2b.docx"},{"id":99814703,"identity":"f71ffe8d-38ab-4c52-9cab-17769e8cf34b","added_by":"auto","created_at":"2026-01-08 14:42:37","extension":"json","order_by":1,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":4622,"visible":true,"origin":"","legend":"","description":"","filename":"0f5b73208bdf4c3b8272708d73ab6433.json","url":"https://assets-eu.researchsquare.com/files/rs-8398524/v1/027146dcfbf74e137b0d39bc.json"},{"id":99814875,"identity":"c8eb832a-9911-4978-9f54-47fd1501f812","added_by":"auto","created_at":"2026-01-08 14:43:02","extension":"xml","order_by":2,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":130392,"visible":true,"origin":"","legend":"","description":"","filename":"0f5b73208bdf4c3b8272708d73ab64331enriched.xml","url":"https://assets-eu.researchsquare.com/files/rs-8398524/v1/a563b6e85de5c0992fcc64e8.xml"},{"id":99814599,"identity":"662cc418-a612-410c-aa07-dc3d34f4874a","added_by":"auto","created_at":"2026-01-08 14:42:07","extension":"png","order_by":9,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":96971,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-8398524/v1/ae6a3bd427450c219e609eaf.png"},{"id":99814901,"identity":"3ffc441c-8760-42fb-a683-cb768f4585ed","added_by":"auto","created_at":"2026-01-08 14:43:06","extension":"png","order_by":10,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":70004,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-8398524/v1/edddb842ccca4750a1649520.png"},{"id":99814595,"identity":"d963673b-cdbf-4213-bec3-ae9b0b8e722b","added_by":"auto","created_at":"2026-01-08 14:42:07","extension":"png","order_by":11,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":22264,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-8398524/v1/74b87126e2b599028c06b59a.png"},{"id":99814737,"identity":"e1209a8b-68a7-49e2-9521-c09600c8b7ab","added_by":"auto","created_at":"2026-01-08 14:42:41","extension":"png","order_by":12,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":26689,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage4.png","url":"https://assets-eu.researchsquare.com/files/rs-8398524/v1/ad84e63fe850e62048811277.png"},{"id":99814866,"identity":"b2868513-3c25-4268-b75a-e3d1a7195701","added_by":"auto","created_at":"2026-01-08 14:43:00","extension":"png","order_by":13,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":21704,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage5.png","url":"https://assets-eu.researchsquare.com/files/rs-8398524/v1/d94c36e56d412baa3999384b.png"},{"id":99814614,"identity":"82f84773-a6d9-4f06-af80-68b0df00bad1","added_by":"auto","created_at":"2026-01-08 14:42:18","extension":"png","order_by":14,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":25060,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage6.png","url":"https://assets-eu.researchsquare.com/files/rs-8398524/v1/a0a9effc165938990481e911.png"},{"id":99814758,"identity":"2e816df3-5443-498c-a162-c213c97be54b","added_by":"auto","created_at":"2026-01-08 14:42:44","extension":"xml","order_by":15,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":133176,"visible":true,"origin":"","legend":"","description":"","filename":"0f5b73208bdf4c3b8272708d73ab64331structuring.xml","url":"https://assets-eu.researchsquare.com/files/rs-8398524/v1/cd48b2dc06092ea45abbacbe.xml"},{"id":99814674,"identity":"32b540d6-361a-4533-b133-d9b2f5b64730","added_by":"auto","created_at":"2026-01-08 14:42:32","extension":"html","order_by":16,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":142019,"visible":true,"origin":"","legend":"","description":"","filename":"earlyproof.html","url":"https://assets-eu.researchsquare.com/files/rs-8398524/v1/a456a1680d17b8e5e1a3d339.html"},{"id":99814768,"identity":"ab1c5088-51a6-4b3c-913c-237753d72757","added_by":"auto","created_at":"2026-01-08 14:42:46","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":689624,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eMultidimensional Visualization Reveals Separation Paradox.\u003c/strong\u003e Two-dimensional projections (PCA left, t-SNE right) of 11-dimensional feature space. PD cases (red, n=115) and controls (teal, n=3,029) show substantial overlap despite high classification accuracy in full dimensional space. Misclassified cases marked with yellow crosses.\u003c/p\u003e","description":"","filename":"floatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-8398524/v1/39560df5caa1a97a24b4cf35.png"},{"id":99814999,"identity":"a55963b8-1425-4104-ba86-397d97318c52","added_by":"auto","created_at":"2026-01-08 14:43:18","extension":"jpeg","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":317880,"visible":true,"origin":"","legend":"\u003cp\u003eSHAP beeswarm plot showing feature contributions for all participants (n=3,144). Each point represents one participant; horizontal position indicates SHAP value magnitude and direction, color indicates feature value (red=high, blue=low). Features ranked by mean absolute SHAP value.\u003c/p\u003e","description":"","filename":"floatimage2.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-8398524/v1/6f02fcb9602f9b0bff75a2ad.jpeg"},{"id":99814706,"identity":"9ec240f6-f9d9-4875-88e2-4436fef71585","added_by":"auto","created_at":"2026-01-08 14:42:37","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":96921,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eThe Complexity Gap - Why Simple Models Fail.\u003c/strong\u003e Confusion matrices comparing decision tree (left, max_depth=5) vs gradient boosting (right) on test set (n=786; 29 PD, 757 controls). Color intensity represents case count.\u003c/p\u003e","description":"","filename":"floatimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-8398524/v1/74e7ea1595e74012423433fb.png"},{"id":99814760,"identity":"0834ae9a-8ac1-49c1-bbf5-4f73ad01fc87","added_by":"auto","created_at":"2026-01-08 14:42:44","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":114224,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eSHAP Interaction Network Reveals Body Fat Hub and Cross-Domain Synergy.\u003c/strong\u003e Heatmap of pairwise SHAP interaction strengths for all 55 feature combinations (11×11 symmetric matrix). Color intensity represents interaction magnitude; values shown for interactions \u0026gt;0.30.\u003c/p\u003e","description":"","filename":"floatimage4.png","url":"https://assets-eu.researchsquare.com/files/rs-8398524/v1/b14e420b90189ba229d90a17.png"},{"id":99814827,"identity":"92904de1-4fe4-4338-8d1b-012ee9faf2dc","added_by":"auto","created_at":"2026-01-08 14:42:56","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":112668,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eSocioeconomic Stratification Reveals Extreme Disparities and Highest-Risk Subgroups.\u003c/strong\u003e Heatmaps showing panic disorder prevalence (%) stratified by (A) age and poverty quartile, and (B) gender and poverty quartile. Color intensity represents prevalence; cell values show percentage and sample size.\u003c/p\u003e","description":"","filename":"floatimage5.png","url":"https://assets-eu.researchsquare.com/files/rs-8398524/v1/bbed1c2d6a018c48bc2acb49.png"},{"id":100356274,"identity":"4dc06c58-9458-452d-92cd-3f971082281e","added_by":"auto","created_at":"2026-01-16 06:59:23","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":107864,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eBalanced Biopsychosocial Domain Contributions.\u003c/strong\u003e Comparison of (A) univariate effect sizes (mean Cohen's d) vs (B) multivariate importance (SHAP %) across four theoretical domains. Error bars represent 95% confidence intervals.\u003c/p\u003e","description":"","filename":"floatimage6.png","url":"https://assets-eu.researchsquare.com/files/rs-8398524/v1/15aa87f32cedb8ba2bdd66f4.png"},{"id":102296217,"identity":"f3e60ceb-f435-4e0f-ad85-d32eefa7e3e3","added_by":"auto","created_at":"2026-02-10 10:18:14","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":2793004,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-8398524/v1/a405eee2-5ca6-4d93-9633-450dbeb85efe.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"Unravelling the Complexity Gap: A Mechanistic Investigation of Machine Learning Classification in Panic Disorder ","fulltext":[{"header":"1. INTRODUCTION","content":"\u003cp\u003eMachine learning (ML) has become increasingly prominent in psychiatry, demonstrating high predictive accuracy across a range of disorders (Chekroud et al., \u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e2021\u003c/span\u003e). However, a critical gap remains between showing that ML works and understanding why it works. Accuracy alone provides limited clinical utility without insight into which features drive classification, how they interact, and which populations are at greatest risk (Doshi-Velez and Kim, \u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e2017\u003c/span\u003e; Hassija et al., \u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e2024\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eIn prior work (Martins et al., \u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e2025\u003c/span\u003e), we showed that ML models using routinely collected socioeconomic, physiological, and behavioral markers from NHANES achieved robust panic disorder (PD) classification (Precision 0.96, Recall 0.81, ROC-AUC 0.905). These results suggested that health survey data capture meaningful patterns relevant to PD diagnosis, supporting potential use in population screening. Yet several mechanistic questions remained: Do individual features strongly discriminate cases, or is integration across domains necessary? Can simple clinical rules replicate ensemble performance, or is genuine complexity required? Which socioeconomic subgroups are at highest risk, and why? Do identified patterns align with known PD pathophysiology?\u003c/p\u003e \u003cp\u003eThe current study addresses these questions through a multi-method mechanistic framework, transforming ML from a purely predictive tool into a hypothesis generator. Ensemble methods such as random forests and gradient boosting often achieve superior accuracy but lack transparency, creating challenges in clinical settings where interpretability is crucial for decision-making, patient explanation, regulatory validation, and intervention planning (Doshi-Velez and Kim, \u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e2017\u003c/span\u003e; Guidotti et al., \u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e2018\u003c/span\u003e; Kim et al., \u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e2012\u003c/span\u003e). Modern interpretability methods \u0026mdash; including SHAP, LIME, and integrated gradients \u0026mdash; allow researchers to quantify feature contributions, explore interactions, and assess the plausibility of model predictions (Lundberg and Lee, \u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e2017\u003c/span\u003e). Despite these advances, few psychiatric studies have systematically applied such methods to understand the mechanisms underlying high predictive performance.\u003c/p\u003e \u003cp\u003ePanic disorder affects 2\u0026ndash;3% of adults globally and is characterized by recurrent, unexpected panic attacks accompanied by persistent worry about future episodes (Javaid et al., \u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Kessler and Merikangas, \u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e2006\u003c/span\u003e). Contemporary biopsychosocial models posit that PD arises from interactions among biological factors (e.g., autonomic dysregulation, HPA-axis abnormalities), cognitive processes (e.g., catastrophic interpretation of bodily sensations, anxiety sensitivity), and social determinants (e.g., socioeconomic stress, childhood adversity, low social support) (Reiss et al., \u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e1986\u003c/span\u003e; Schmidt et al., \u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e1997\u003c/span\u003e). This multidomain complexity suggests that accurate classification likely requires integration across domains rather than reliance on single biomarkers (Chekroud et al., \u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e2021\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eTo ensure methodological rigor and reduce potential artifacts, we employed a conservative analytic approach using complete-case NHANES data, removed variables with coding irregularities, and focused on 11 core features with optimal data quality. This strategy prioritizes robustness over sample size and tests whether observed patterns reflect genuine biopsychosocial signatures rather than methodological anomalies.\u003c/p\u003e \u003cp\u003eOur primary objective was to elucidate mechanisms through which sociodemographic and physiological features enable accurate panic disorder classification. Specifically, we sought to: \u003cem\u003e(1)\u003c/em\u003e quantify discriminative power of individual features, \u003cem\u003e(2)\u003c/em\u003e identify synergistic feature interactions, \u003cem\u003e(3)\u003c/em\u003e test whether simple decision rules can replicate ensemble performance, \u003cem\u003e(4)\u003c/em\u003e characterize socioeconomic patterns and high-risk subgroups, and (5) assess biological plausibility of identified patterns.\u003c/p\u003e"},{"header":"2. METHODS","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003e2.1. Study Design and Data Source\u003c/h2\u003e \u003cp\u003eThis mechanistic investigation used data from the National Health and Nutrition Examination Survey (NHANES) cycles 1999\u0026ndash;2004, a nationally representative cross-sectional survey conducted by the Centers for Disease Control and Prevention to assess the health and nutritional status of adults and children in the United States. Study protocols were approved by the NCHS Research Ethics Review Board, and all participants provided written informed consent.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec4\" class=\"Section2\"\u003e \u003ch2\u003e2.2. Study Population\u003c/h2\u003e \u003cp\u003eThe initial merged dataset included 6,581 participants from the three NHANES cycles. To ensure data quality and avoid potential biases from imputation, we conducted a complete-case analysis, deliberately avoiding synthetic oversampling or multiple imputation methods. After excluding participants with missing values, the final analytical sample consisted of 3,144 individuals, including 115 cases of panic disorder (3.5%) and 3,029 controls (96.5%), consistent with epidemiological estimates from community samples.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec5\" class=\"Section2\"\u003e \u003ch2\u003e2.3. Panic Disorder Classification\u003c/h2\u003e \u003cp\u003ePanic disorder cases were identified using the CIDPSCOR variable from the NHANES Composite International Diagnostic Interview (CIDI), a validated instrument based on DSM-IV criteria. Code 1 indicates a positive diagnosis, while Code 5 indicates absence of PD. Classification integrates symptom counts, attack characteristics, temporal patterns, associated concerns, and potential alternative aetiologies, ensuring that identified cases represent clinically relevant PD rather than isolated panic events or secondary symptoms. The study focused on a refined set of 11 features spanning sociodemographic, physiological, and behavioral domains, selected from a previous set of 16 to capture the multidimensional nature of panic disorder.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec6\" class=\"Section2\"\u003e \u003ch2\u003e2.4. Feature Set\u003c/h2\u003e \u003cp\u003eThe biological and physiological features included body fat mass, body mass index, and diastolic blood pressure, capturing aspects of cardiovascular and metabolic health. Sociodemographic features comprised age, gender, marital status, household size, and poverty-to-income ratio, reflecting social determinants of health. Behavioral and psychological features included alcohol consumption, family history of hypertension or stroke, and recent lower back pain, serving as indicators of psychological and somatic vulnerability. Detailed descriptions of all features, including NHANES variable codes, measurement protocols, and clinical interpretation, are provided in Supplementary Table S1.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec7\" class=\"Section2\"\u003e \u003ch2\u003e2.5. Machine Learning Model\u003c/h2\u003e \u003cp\u003eWe employed a Gradient Boosting Classifier implemented in scikit-learn, with hyperparameters optimized through cross-validation and preliminary experimentation. The final configuration included 200 boosting stages, a maximum tree depth of 8, a learning rate of 0.1, and balanced class weighting to address the minority prevalence of PD. Features were standardized using the StandardScaler, and any remaining missing numerical values were imputed using the median. The dataset was split into training (75%) and testing (25%) sets using stratified sampling to preserve the natural class distribution, and five-fold cross-validation was applied to optimize hyperparameters and assess model generalization.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec8\" class=\"Section2\"\u003e \u003ch2\u003e2.6. Statistical and Machine Learning Analyses\u003c/h2\u003e \u003cp\u003eTo investigate the mechanisms underlying classification performance, we first evaluated the discriminative power of individual features using descriptive statistics, effect sizes (Cohen\u0026rsquo;s d), and distributional separation tests (t-tests, Mann-Whitney U, Kolmogorov-Smirnov). SHapley Additive exPlanations (SHAP) values were computed with TreeExplainer to quantify feature importance and pairwise interactions, using the training set only to avoid data leakage. Model complexity was assessed by comparing a simple interpretable decision tree with the full gradient boosting classifier. Socioeconomic stratification involved dividing participants into poverty-to-income quartiles and further by age and gender to identify high-risk subgroups. Dimensionality reduction (UMAP, t-SNE) and clustering metrics (silhouette scores, centroid distances, separation ratios) visualized class separability. Human-readable rules were extracted from the decision tree to enhance interpretability.\u003c/p\u003e \u003cp\u003eSensitivity analyses confirmed robustness by repeating analyses with multiple random seeds, comparing balanced vs unweighted class weighting, and computing SHAP values across all cross-validation folds (Supplementary Table S3).\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec9\" class=\"Section2\"\u003e \u003ch2\u003e2.7. Ethical Considerations, Software, and Data Availability\u003c/h2\u003e \u003cp\u003eThis study utilized publicly available, de-identified NHANES data. Secondary analysis was exempt from additional institutional review board approval, as the data contained no identifiable information and posed minimal risk. Analyses were conducted in Python 3.10 using scikit-learn, SHAP, pandas, NumPy, SciPy, and matplotlib. Random operations were seeded for reproducibility. The analytical code is publicly available at GitHub.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003e\u003cb\u003eSample Characteristics.\u003c/b\u003e Sample characteristics stratified by panic disorder status (N\u0026thinsp;=\u0026thinsp;3,144; 115 PD, 3,029 controls). Values shown as mean\u0026thinsp;\u0026plusmn;\u0026thinsp;SD or n(%). Statistical comparisons performed using t-tests for continuous variables and chi-square tests for categorical variables.\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"5\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCharacteristic\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eTotal\u003c/p\u003e \u003cp\u003e(N\u0026thinsp;=\u0026thinsp;3,144)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eNormal\u003c/p\u003e \u003cp\u003e(n\u0026thinsp;=\u0026thinsp;3,029)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003ePanic Disorder\u003c/p\u003e \u003cp\u003e(n\u0026thinsp;=\u0026thinsp;115)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003ep-value\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDEMOGRAPHICS\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAge (years), mean\u0026thinsp;\u0026plusmn;\u0026thinsp;SD\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e29.5\u0026thinsp;\u0026plusmn;\u0026thinsp;5.9\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e29.5\u0026thinsp;\u0026plusmn;\u0026thinsp;5.9\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e30.6\u0026thinsp;\u0026plusmn;\u0026thinsp;4.8\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.044\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eGender (% Female)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e55.7%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e54.6%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e80.0%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSOCIOECONOMIC\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePoverty-Income Ratio, mean\u0026thinsp;\u0026plusmn;\u0026thinsp;SD\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e2.87\u0026thinsp;\u0026plusmn;\u0026thinsp;1.61\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e2.91\u0026thinsp;\u0026plusmn;\u0026thinsp;1.61\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e2.06\u0026thinsp;\u0026plusmn;\u0026thinsp;1.47\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eHousehold Size, mean\u0026thinsp;\u0026plusmn;\u0026thinsp;SD\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e3.22\u0026thinsp;\u0026plusmn;\u0026thinsp;1.56\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e3.24\u0026thinsp;\u0026plusmn;\u0026thinsp;1.57\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e3.05\u0026thinsp;\u0026plusmn;\u0026thinsp;1.32\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.215\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePHYSIOLOGICAL\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eBody Fat Mass (kg), mean\u0026thinsp;\u0026plusmn;\u0026thinsp;SD\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e23.05\u0026thinsp;\u0026plusmn;\u0026thinsp;11.15\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e22.91\u0026thinsp;\u0026plusmn;\u0026thinsp;11.04\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e26.25\u0026thinsp;\u0026plusmn;\u0026thinsp;13.68\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.002\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eBody Mass Index, mean\u0026thinsp;\u0026plusmn;\u0026thinsp;SD\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e26.80\u0026thinsp;\u0026plusmn;\u0026thinsp;5.77\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e26.77\u0026thinsp;\u0026plusmn;\u0026thinsp;5.79\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e27.26\u0026thinsp;\u0026plusmn;\u0026thinsp;5.49\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.369\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDiastolic BP (mmHg), mean\u0026thinsp;\u0026plusmn;\u0026thinsp;SD\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e69.38\u0026thinsp;\u0026plusmn;\u0026thinsp;11.38\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e69.42\u0026thinsp;\u0026plusmn;\u0026thinsp;11.54\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e68.85\u0026thinsp;\u0026plusmn;\u0026thinsp;5.61\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.601\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003c/div\u003e"},{"header":"3. RESULTS","content":"\u003cdiv id=\"Sec11\" class=\"Section2\"\u003e \u003ch2\u003e3.1 Sample Characteristics and Individual Feature Discriminative Power\u003c/h2\u003e \u003cp\u003eThe conservative analysis included 3,144 participants with complete data on all 11 features (Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e), representing 47.8% of the original NHANES 1999\u0026ndash;2004 panic disorder sample. This subset comprised 115 panic disorder (PD) cases (3.66%) and 3,029 normal controls (96.34%), maintaining a prevalence ratio consistent with epidemiological estimates.\u003c/p\u003e \u003cp\u003eFemales comprised 80.0% of the PD group versus 54.6% of the normal group (p\u0026thinsp;\u0026lt;\u0026thinsp;.001, Cohen's d\u0026thinsp;=\u0026thinsp;.700), the strongest individual discriminator. Age differences were minimal (PD: 30.6\u0026thinsp;\u0026plusmn;\u0026thinsp;4.8 years vs Normal: 29.5\u0026thinsp;\u0026plusmn;\u0026thinsp;5.9 years, p\u0026thinsp;=\u0026thinsp;.044, d\u0026thinsp;=\u0026thinsp;.191).\u003c/p\u003e \u003cp\u003eSocioeconomic features demonstrated moderate effects. Poverty-income ratio was lower in PD (2.06\u0026thinsp;\u0026plusmn;\u0026thinsp;1.47 vs 2.91\u0026thinsp;\u0026plusmn;\u0026thinsp;1.61, p\u0026thinsp;\u0026lt;\u0026thinsp;.001, d=-.531), whereas household size showed negligible difference (d=-.118, p\u0026thinsp;=\u0026thinsp;.215).\u003c/p\u003e \u003cp\u003ePhysiological measurements revealed subtle differences. Body fat mass was modestly elevated in PD (26.25\u0026thinsp;\u0026plusmn;\u0026thinsp;13.68 kg vs 22.91\u0026thinsp;\u0026plusmn;\u0026thinsp;11.04 kg, p\u0026thinsp;=\u0026thinsp;.002, d\u0026thinsp;=\u0026thinsp;.300), while BMI (p\u0026thinsp;=\u0026thinsp;.369) and diastolic blood pressure (p\u0026thinsp;=\u0026thinsp;.601) showed minimal differences. Alcohol consumption showed no association (p\u0026thinsp;=\u0026thinsp;.997, d\u0026thinsp;=\u0026thinsp;0.000).\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab2\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003e\u003cb\u003eIndividual Feature Discriminative Power.\u003c/b\u003e Discriminative analysis of 11 features showing effect sizes (Cohen's d), separation percentages, and statistical significance. Features ranked by absolute effect size with domain classification (Biological, Social, Demographic, Psychological).\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"6\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\"\u0026plusmn;\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eFeature\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003ePD Mean\u0026thinsp;\u0026plusmn;\u0026thinsp;SD\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eNormal Mean\u0026thinsp;\u0026plusmn;\u0026thinsp;SD\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eCohen's d\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eSeparation %\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003ep-value\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eGender\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c2\"\u003e \u003cp\u003e1.80\u0026thinsp;\u0026plusmn;\u0026thinsp;0.40\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1.45\u0026thinsp;\u0026plusmn;\u0026thinsp;0.50\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e+\u0026thinsp;0.700\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e31.0%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eFamily History HTN/Stroke\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c2\"\u003e \u003cp\u003e1.33\u0026thinsp;\u0026plusmn;\u0026thinsp;0.47\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1.61\u0026thinsp;\u0026plusmn;\u0026thinsp;0.49\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e-0.566\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e22.7%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLower Back Pain\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c2\"\u003e \u003cp\u003e1.37\u0026thinsp;\u0026plusmn;\u0026thinsp;0.48\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1.63\u0026thinsp;\u0026plusmn;\u0026thinsp;0.48\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e-0.559\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e22.0%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePoverty-Income Ratio\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c2\"\u003e \u003cp\u003e2.06\u0026thinsp;\u0026plusmn;\u0026thinsp;1.47\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e2.91\u0026thinsp;\u0026plusmn;\u0026thinsp;1.61\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e-0.531\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e22.0%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eBody Fat Mass (kg)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c2\"\u003e \u003cp\u003e26.25\u0026thinsp;\u0026plusmn;\u0026thinsp;13.68\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e22.91\u0026thinsp;\u0026plusmn;\u0026thinsp;11.04\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e+\u0026thinsp;0.300\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e14.2%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.002\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMarital Status\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c2\"\u003e \u003cp\u003e2.89\u0026thinsp;\u0026plusmn;\u0026thinsp;1.62\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e3.35\u0026thinsp;\u0026plusmn;\u0026thinsp;2.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e-0.232\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e13.6%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.015\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAge (months)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c2\"\u003e \u003cp\u003e367.1\u0026thinsp;\u0026plusmn;\u0026thinsp;58.1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e353.8\u0026thinsp;\u0026plusmn;\u0026thinsp;70.2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e+\u0026thinsp;0.191\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e11.8%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.044\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eHousehold Size\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c2\"\u003e \u003cp\u003e3.05\u0026thinsp;\u0026plusmn;\u0026thinsp;1.32\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e3.24\u0026thinsp;\u0026plusmn;\u0026thinsp;1.57\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e-0.118\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e9.3%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.215\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eBMI (kg/m\u0026sup2;)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c2\"\u003e \u003cp\u003e27.26\u0026thinsp;\u0026plusmn;\u0026thinsp;5.49\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e26.77\u0026thinsp;\u0026plusmn;\u0026thinsp;5.79\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e+\u0026thinsp;0.085\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e4.1%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.369\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDiastolic BP (mmHg)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c2\"\u003e \u003cp\u003e68.85\u0026thinsp;\u0026plusmn;\u0026thinsp;5.61\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e69.42\u0026thinsp;\u0026plusmn;\u0026thinsp;11.54\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e-0.050\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e2%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.601\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAlcohol (drinks/day)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c2\"\u003e \u003cp\u003e3.26\u0026thinsp;\u0026plusmn;\u0026thinsp;1.99\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e3.26\u0026thinsp;\u0026plusmn;\u0026thinsp;3.96\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.000\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.997\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eMean |d| = 0.284\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eMean\u0026thinsp;=\u0026thinsp;19.4%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eBehavioral and clinical features showed mixed patterns. Lower back pain was paradoxically less common in PD (\u003cem\u003ed\u003c/em\u003e=-.559, \u003cem\u003ep\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;.001), as was family history of hypertension/stroke (\u003cem\u003ed\u003c/em\u003e=-.566, \u003cem\u003ep\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;.001).\u003c/p\u003e \u003cp\u003eComprehensive discriminative analysis (Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e) showed that no single feature achieved strong discrimination (Cohen\u0026rsquo;s d\u0026thinsp;\u0026gt;\u0026thinsp;0.8), with maximum separation 31.0% for gender and mean separation across all features 19.4%. Four features had medium effects (|d| = 0.5\u0026ndash;0.8): gender, family history, lower back pain, and poverty-income ratio, while five features showed negligible effects (|d| \u0026lt; 0.2): age, household size, BMI, blood pressure, and alcohol consumption. The mean absolute Cohen\u0026rsquo;s d across all 11 features was 0.284 (small effect).\u003c/p\u003e \u003cp\u003eDomain-level analysis revealed that univariately, the psychological domain had the strongest effects (mean |d| = 0.563), followed by social (0.295) and biological (0.126). However, multivariate analysis (Section \u003cspan refid=\"Sec13\" class=\"InternalRef\"\u003e3.3\u003c/span\u003e) reversed this pattern, emphasizing the difference between marginal and conditional effects. Full distributional analyses and visualizations are provided in Supplementary Figures S2\u0026ndash;S3.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eThese univariate findings established a fundamental paradox: if individual features show weak-to-moderate discrimination at best (maximum 31% separation), how does the gradient boosting ensemble achieve 99.75% accuracy? This question framed the subsequent analyses examining multidimensional complexity and synergistic interactions.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec12\" class=\"Section2\"\u003e \u003ch2\u003e3.2 Multidimensional Complexity and the Necessity of Ensemble Methods\u003c/h2\u003e \u003cp\u003eDimensionality reduction analysis (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e) revealed the core mechanism underlying model performance. When the 11-dimensional feature space was projected onto two dimensions using PCA and t-SNE, substantial visual overlap emerged between panic disorder and normal groups despite the gradient boosting model achieving 99.75% test set accuracy. PCA projection yielded Silhouette score of 0.32 and Davies-Bouldin index of 2.14; t-SNE projection showed Silhouette score of 0.38 and Davies-Bouldin index of 1.89. These moderate cluster quality metrics in 2D space stood in stark contrast to near-perfect classification in 11D space.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab3\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 3\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003e\u003cb\u003eModel Performance Comparison - The Complexity Gap.\u003c/b\u003e Comparison of decision tree (max_depth\u0026thinsp;=\u0026thinsp;5) vs gradient boosting classifier performance on test set (n\u0026thinsp;=\u0026thinsp;786; 29 PD, 757 controls). Performance metrics include accuracy, precision, recall, F1-score, ROC-AUC, and confusion matrix components. *** p\u0026thinsp;\u0026lt;\u0026thinsp;0.001 (McNemar's test).\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"4\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMetric\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eDecision Tree\u003c/p\u003e \u003cp\u003e(Interpretable)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eGradient Boosting\u003c/p\u003e \u003cp\u003e(Ensemble)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eDifference\u003c/p\u003e \u003cp\u003e(GB - DT)\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAccuracy (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e40.97\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e99.75\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e+\u0026thinsp;58.78***\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePrecision (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e5.88\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e100.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e+\u0026thinsp;94.12***\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eRecall (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e100.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e93.10\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e-6.90\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eF1-Score\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0.111\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.964\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e+\u0026thinsp;0.853***\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eTrue Negatives\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e293\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e756\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e+\u0026thinsp;463\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eFalse Positives\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e464\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e-464\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eTotal Errors\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e464\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e-461\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eAccurate discrimination required integration across all 11 features, as the separating hyperplane existed only in 11-dimensional space, collapsing into overlapping distributions in 2D projections. The model misclassified only 2 cases out of 786: 2 false negatives (6.9% of PD cases), scattered rather than clustered near decision boundaries.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eTo quantify the complexity gap between interpretable and high-performing models, we compared a simple decision tree (max_depth\u0026thinsp;=\u0026thinsp;5) against the gradient boosting ensemble (Table\u0026nbsp;\u003cspan refid=\"Tab3\" class=\"InternalRef\"\u003e3\u003c/span\u003e, Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e). The decision tree achieved only 40.97% accuracy with catastrophic 464 false positives, misclassifying 61.3% of normal controls as panic disorder. Importantly, despite the dominance of poverty-income ratio within the decision tree, the full gradient boosting model did not rely on any single predictor. Body fat mass and poverty-income ratio each contributed only\u0026thinsp;~\u0026thinsp;16% to overall importance, indicating that high performance emerged from integration across multiple domains rather than from dominant individual features.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab4\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 4\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003e\u003cb\u003eSHAP Feature Importance and Class-Stratified Analysis.\u003c/b\u003e SHAP-derived feature importance showing global mean absolute SHAP values, class-stratified contributions (PD vs Normal), and domain-level aggregation. Features ranked by importance with theoretical domain classification.\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"5\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eRank\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eFeature\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eMean SHAP\u003c/p\u003e \u003cp\u003e(PD Group)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eMean SHAP\u003c/p\u003e \u003cp\u003e(Normal)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003e% Total\u003c/p\u003e \u003cp\u003eImportance\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eBody Fat Mass\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e+\u0026thinsp;1.915\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e-3.019\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e16.1%\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003ePoverty-Income Ratio\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e+\u0026thinsp;1.720\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e-3.207\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e16.1%\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eAge\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e+\u0026thinsp;1.200\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e-2.590\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e12.4%\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e4\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eBMI\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e+\u0026thinsp;0.803\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e-2.616\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e11.2%\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e5\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eGender\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e+\u0026thinsp;1.789\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e-1.047\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e9.3%\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e6\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eDiastolic BP\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e+\u0026thinsp;0.655\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e-2.017\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e8.7%\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e7\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eLower Back Pain\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e+\u0026thinsp;1.550\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e-0.424\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e6.4%\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e8\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eAlcohol\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e+\u0026thinsp;0.217\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e-1.482\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e5.5%\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e9\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eFamily History\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e+\u0026thinsp;0.733\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e-0.483\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e4.0%\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e10\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eHousehold Size\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e-0.049\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e-0.880\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e2.7%\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e11\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eMarital Status\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e-0.174\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e-0.911\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e2.4%\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eIn contrast, the gradient boosting ensemble achieved 99.75% accuracy with 100% precision (zero false positives) and 93.1% recall. This represented a dramatic 58.78 percentage point accuracy gap, which we term the \"complexity gap\"\u0026mdash;quantifying the genuine multidimensional patterns that cannot be captured by simple, human-interpretable rules. The ROC analysis confirmed superior discriminative ability: decision tree AUC\u0026thinsp;=\u0026thinsp;.544 versus gradient boosting AUC\u0026thinsp;=\u0026thinsp;.922.\u003c/p\u003e \u003cp\u003eThe \u0026ldquo;complexity gap\u0026rdquo; highlighted that simple decision rules fail for panic disorder, sometimes performing worse than chance. Interpretability and accuracy are inherently trade-offs: a decision tree achieved 40,97% accuracy, whereas the gradient boosting ensemble reached 99.75%, confirming genuine multidimensional complexity (Hypothesis 3) and invalidating threshold-based approaches. UMAP visualizations further illustrate this complex structure.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec13\" class=\"Section2\"\u003e \u003ch2\u003e3.3 Feature Importance and Synergistic Interaction Networks\u003c/h2\u003e \u003cp\u003eSHAP analysis (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e, Table\u0026nbsp;\u003cspan refid=\"Tab4\" class=\"InternalRef\"\u003e4\u003c/span\u003e) revealed distributed importance across all 11 features, with no dominant predictor. Fat mass and poverty-income ratio were the strongest individual contributors (~\u0026thinsp;16% each), yet their effects were modest and depended on interactions with demographic and biological factors. The top 5 features collectively represented only 65.1% of importance, requiring integration across all features for accurate prediction.\u003c/p\u003e \u003cp\u003eFeature ranking by mean absolute SHAP value yielded: body fat mass (16.1%), poverty-income ratio (16.1%), age (12.4%), BMI (11.2%), gender (9.3%), diastolic blood pressure (8.7%), lower back pain (6.4%), alcohol consumption (5.5%), family history (4.0%), household size (2.7%), and marital status (2.4%). Notably, several features with weak univariate effects (e.g., blood pressure: \u003cem\u003eCohen's d\u003c/em\u003e=-.050) demonstrated substantial multivariate importance (8.7%), while features with strong univariate effects (e.g., gender: \u003cem\u003ed\u003c/em\u003e\u0026thinsp;=\u0026thinsp;.700) showed only moderate multivariate importance (9.3%).\u003c/p\u003e \u003cp\u003eClass-stratified SHAP analysis (Table\u0026nbsp;\u003cspan refid=\"Tab4\" class=\"InternalRef\"\u003e4\u003c/span\u003e) revealed asymmetric contributions. Body fat showed strongly positive mean SHAP in PD cases (+\u0026thinsp;1.915) and strongly negative in normal controls (-3.019), with difference |Δ|=4.934 indicating high discriminative power. Poverty-income ratio displayed similar magnitude (+\u0026thinsp;1.720 vs -3.207, |Δ|=4.927). Age, BMI, and blood pressure showed moderate differences. The beeswarm distribution (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e) illustrated heterogeneity: substantial horizontal spread within each feature row indicated that identical feature values could push predictions in opposite directions depending on other features present\u0026mdash;direct visual evidence of interaction effects.\u003c/p\u003e \u003cp\u003eAt the domain level, biological variables showed (Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003e, Panel B) the largest aggregate importance (~\u0026thinsp;36%), but this difference was modest, and all domains contributed meaningfully to prediction. These findings reinforce that the model\u0026rsquo;s performance reflects distributed, cross-domain integration rather than reliance on a single risk dimension. All domains contributed meaningfully (15.9\u0026ndash;36%) with no statistical difference between domains (p\u0026thinsp;\u0026gt;\u0026thinsp;0.05), validating Hypothesis 2 (distributed biopsychosocial integration). The biological domain's emergence as most important multivariate despite weakest univariate effects exemplified synergistic amplification.\u003c/p\u003e \u003cp\u003ePairwise interaction analysis (Table\u0026nbsp;\u003cspan refid=\"Tab5\" class=\"InternalRef\"\u003e5\u003c/span\u003e, Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e) identified systematic synergies invisible in univariate examination. Body fat mass functioned as a frequent interaction hub, participating in 6 of the top 10 interactions (60%): with age (strength\u0026thinsp;=\u0026thinsp;1.047, strongest interaction observed), gender (0.690), alcohol consumption (.670), BMI (.646), poverty-income ratio (.491), and diastolic blood pressure. Age functioned as a universal moderator, showing strong interactions with \u003cem\u003efour features\u003c/em\u003e (40%): body fat (1.047), blood pressure (.807), poverty ratio (.661), and lower back pain (.473).\u003c/p\u003e \u003cp\u003eA critical pattern emerged: the strongest interactions systematically SPANNED theoretical domains rather than occurring within domains. The top 3 interactions were biological\u0026times;demographic (body fat \u0026times; age, 1.047), social\u0026times;biological (poverty ratio \u0026times; BMI, 1.016), and demographic\u0026times;biological (age \u0026times; blood pressure, 0.807). This cross-domain synergy validated Hypothesis 2 and explained the apparent paradox of weak individual effects producing high ensemble accuracy\u0026mdash;predictive power emerged from complex conditional relationships across biopsychosocial domains rather than simple additive contributions.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eThe interaction heatmap (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e) visualized the complete 11\u0026times;11 matrix of pairwise synergies, with body fat's hub status evident in its row/column showing 6 strong interactions (marked in dark red, strength\u0026thinsp;\u0026gt;\u0026thinsp;0.47). Age's moderator function appeared in its 4 strong interactions spanning multiple domains. These patterns confirmed Hypothesis 1 (synergistic feature interactions) and demonstrated why computational methods were necessary: the relevant patterns existed not in individual features but in their multivariate interactions. Complete SHAP interaction matrix for all 55 feature pairs is provided in Supplementary Table S2. SHAP importance bar chart (Supplementary Figure S10), dependence plots (Supplementary Figure S11), and detailed interaction visualizations (Supplementary Figure S12) provide comprehensive interpretation.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec14\" class=\"Section2\"\u003e \u003ch2\u003e3.4 Socioeconomic Stratification and Identification of Highest-Risk Subgroups\u003c/h2\u003e \u003cp\u003eQuartile analysis of poverty-income ratio (Table\u0026nbsp;\u003cspan refid=\"Tab6\" class=\"InternalRef\"\u003e6\u003c/span\u003e, Part A) revealed a robust dose-response relationship with panic disorder prevalence. Moving from wealthiest quartile Q4 (PIR\u0026thinsp;\u0026gt;\u0026thinsp;4.51, prevalence 1.16%) to poorest Q1 (PIR\u0026thinsp;\u0026lt;\u0026thinsp;1.44, prevalence 6.42%) represented a dramatic 5.5-fold increase in risk. Each quartile increment was associated with approximately 40% reduction in PD risk (Q1\u0026rarr;Q2: -36%, Q2\u0026rarr;Q3: -28%, Q3\u0026rarr;Q4: -60%). Chi-square test for linear trend was highly significant (χ\u0026sup2;=32.60, p\u0026thinsp;=\u0026thinsp;3.9\u0026times;10⁻⁷), and odds ratios using Q4 as reference showed progressive elevation: Q3 OR\u0026thinsp;=\u0026thinsp;2.58 (95% CI: 1.24\u0026ndash;5.37), Q2 OR\u0026thinsp;=\u0026thinsp;3.66 (1.80\u0026ndash;7.44), Q1 OR\u0026thinsp;=\u0026thinsp;5.89 (2.95\u0026ndash;11.76), all p\u0026thinsp;\u0026lt;\u0026thinsp;0.05.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eStratified analysis revealed substantial effect modification by demographics (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003e). Age\u0026times;Poverty interaction (Table\u0026nbsp;\u003cspan refid=\"Tab6\" class=\"InternalRef\"\u003e6\u003c/span\u003e, Part B, \u003cem\u003ep\u003c/em\u003e\u0026thinsp;=\u0026thinsp;.002) showed poverty's effect was AMPLIFIED in older adults. Among older participants (\u0026ge;\u0026thinsp;median age\u0026thinsp;~\u0026thinsp;30 years), prevalence ranged from 10.40% in Q1 (poorest) to 2.03% in Q4 (wealthiest), representing 5.1-fold difference. Among younger participants, the gradient was less pronounced: Q1 showed 3.64% versus Q4 showing 0.00% (complete protection through wealth in youth, though based on small cell size n\u0026thinsp;=\u0026thinsp;15). Poor older adults displayed the HIGHEST age-stratified prevalence observed (10.40%, approximately 1 in 10).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eGender\u0026times;Poverty interaction (Table\u0026nbsp;\u003cspan refid=\"Tab6\" class=\"InternalRef\"\u003e6\u003c/span\u003e, Part C, p\u0026thinsp;\u0026lt;\u0026thinsp;0.001) revealed even more extreme disparities. Poor women (Q1) demonstrated HIGHEST OVERALL prevalence of 11.82%\u0026mdash;nearly 1 in 8 women living below 150% of federal poverty line met criteria for panic disorder. This represented 4.6-fold increase versus wealthy women (Q4: 2.62%) and 4.6-fold increase versus poor men (Q1: 2.59%). The poverty gradient was strongly linear in women (11.82%\u0026rarr;7.87%\u0026rarr;3.74%\u0026rarr;2.62% across quartiles, \u003cem\u003ep\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;.001) but inconsistent in men (2.59%\u0026rarr;0.74%\u0026rarr;2.06%\u0026rarr;0.00%, \u003cem\u003ep\u003c/em\u003e\u0026thinsp;=\u0026thinsp;.132), suggesting gender-specific vulnerability mechanisms.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab5\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 5\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003e\u003cb\u003eTop 10 Feature Interactions from SHAP Analysis.\u003c/b\u003e Top 10 pairwise feature interactions from SHAP interaction analysis, ranked by interaction strength. Domain pairs indicate cross-domain (biological\u0026times;demographic) vs within-domain interactions.\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"4\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eRank\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eFeature Pair\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eInteraction Strength\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eInterpretation\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eBody Fat \u0026times; Age\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e1.047\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eAdiposity effect increases with age\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003ePIR \u0026times; BMI\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e1.016\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eSES modulates body composition effects\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eAge \u0026times; BP\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.807\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eAge moderates BP-PD relationship\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e4\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eBody Fat \u0026times; Gender\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.690\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eSex-specific adiposity effects\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e5\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eBody Fat \u0026times; Alcohol\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.670\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eMetabolic-behavioral interaction\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e6\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003ePIR \u0026times; Age\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.661\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eCumulative adversity pathway\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e7\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eBody Fat \u0026times; BMI\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.646\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eBody composition synergy\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e8\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eBMI \u0026times; Pain\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.564\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eWeight-pain-anxiety link\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e9\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eBody Fat \u0026times; PIR\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.491\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eSES-metabolic pathway\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e10\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eAge \u0026times; Pain\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.473\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eChronic pain relevance increases with age\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab6\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 6\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003e\u003cb\u003ePanic Disorder Prevalence by Poverty-Income Ratio Quartiles. PART A.\u003c/b\u003e PRIMARY QUARTILE ANALYSIS. Panic disorder prevalence by poverty-income ratio quartiles with chi-square test for linear trend and odds ratios (Q4 reference).\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"8\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c7\" colnum=\"7\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c8\" colnum=\"8\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colspan=\"8\" nameend=\"c8\" namest=\"c1\"\u003e \u003cp\u003ePART A: PRIMARY QUARTILE ANALYSIS Sample: N\u0026thinsp;=\u0026thinsp;3,144 (115 PD, 3,029 Normal) | NHANES 1999\u0026ndash;2004\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eQuartile\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003ePIR Range\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eN\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003ePD Cases\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003ePD Prevalence\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eOdds Ratio\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e95% CI\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003ep-value\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eQ1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0.00-1.44\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e794\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e51\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e6.42%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e5.89\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e[2.95\u0026ndash;11.76]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e\u0026lt;0.001\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e(Poorest)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e(Below FPL)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e(ref: Q4)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eQ2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e1.44\u0026ndash;2.77\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e785\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e32\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e4.08%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e3.66\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e[1.80\u0026ndash;7.44]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e\u0026lt;0.001\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e(~\u0026thinsp;FPL-2\u0026times;FPL)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e(ref: Q4)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eQ3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e2.77\u0026ndash;4.51\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e789\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e23\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e2.92%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e2.58\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e[1.24\u0026ndash;5.37]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e0.011\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e(2\u0026ndash;4\u0026times;FPL)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e(ref: Q4)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eQ4\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e4.51-5.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e776\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e9\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e1.16%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e1.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003eReference\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e(Wealthiest)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e(\u0026gt;\u0026thinsp;4\u0026times;FPL)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e(ref)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003cb\u003ePART B. AGE \u0026times; POVERTY STRATIFICATION.\u003c/b\u003e Panic disorder prevalence stratified by age (median split) and poverty quartile. Interaction p-value from logistic regression.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"No\" id=\"Taba\" border=\"1\"\u003e \u003ccolgroup cols=\"6\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAge Group\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eQ1 (Poorest)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eQ2\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eQ3\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eQ4 (Wealthiest)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003eTrend p\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003ePrevalence\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003ePrevalence\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003ePrevalence\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c6\" namest=\"c5\"\u003e \u003cp\u003ePrevalence\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eYounger\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e3.64%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1.79%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e3.53%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.00%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e0.046\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e(\u0026lt;\u0026thinsp;median age)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e(n\u0026thinsp;=\u0026thinsp;11/302)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e(n\u0026thinsp;=\u0026thinsp;7/392)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e(n\u0026thinsp;=\u0026thinsp;14/397)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c6\" namest=\"c5\"\u003e \u003cp\u003e(n\u0026thinsp;=\u0026thinsp;0/364)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eOlder\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e10.40%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e6.37%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e2.40%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e2.03%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e\u0026lt;0.001\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e(\u0026ge;\u0026thinsp;median age)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e(n\u0026thinsp;=\u0026thinsp;40/492)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e(n\u0026thinsp;=\u0026thinsp;25/393)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e(n\u0026thinsp;=\u0026thinsp;9/392)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c6\" namest=\"c5\"\u003e \u003cp\u003e(n\u0026thinsp;=\u0026thinsp;9/412)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eInteraction\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003ep\u0026thinsp;=\u0026thinsp;0.002\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eIntersection analysis (Table\u0026nbsp;\u003cspan refid=\"Tab6\" class=\"InternalRef\"\u003e6\u003c/span\u003e, Part D) identified multiplicative risk: poor older women showed estimated 15\u0026ndash;18% PD prevalence (based on multiplicative model combining age and gender effects), representing 4.1\u0026ndash;4.9\u0026times; the population average. This intersection of age, gender, and poverty created a \"perfect storm\" of vulnerability, identifying the highest-priority population for screening and intervention. Poor older adults combined (gender-aggregated) showed 10.40% prevalence (2.8\u0026times; average), and poor women across all ages showed 11.82% (3.2\u0026times; average). In contrast, wealthy young men represented a protected subgroup with 0.00% prevalence.\u003c/p\u003e \u003cp\u003ePhenotypic heterogeneity analysis (Table\u0026nbsp;\u003cspan refid=\"Tab6\" class=\"InternalRef\"\u003e6\u003c/span\u003e, Part E) revealed distinct PD profiles by socioeconomic status. Among PD cases only (n\u0026thinsp;=\u0026thinsp;115), comparing low-PIR (below median, n\u0026thinsp;=\u0026thinsp;83, 72.2%) versus high-PIR (above median, n\u0026thinsp;=\u0026thinsp;32, 27.8%) showed wealthier cases were significantly older (+\u0026thinsp;31.5 months, \u003cem\u003ep\u003c/em\u003e\u0026thinsp;=\u0026thinsp;.009), had more body fat (+\u0026thinsp;7.1 kg, \u003cem\u003ep\u003c/em\u003e\u0026thinsp;=\u0026thinsp;.012), reported more lower back pain (+\u0026thinsp;31.6 percentage points, \u003cem\u003ep\u003c/em\u003e\u0026thinsp;=\u0026thinsp;.001), but had less family history of hypertension/stroke (-19.8 percentage points, \u003cem\u003ep\u003c/em\u003e\u0026thinsp;=\u0026thinsp;.043). This pattern suggested different etiological pathways: poorer PD characterized by younger onset, lower adiposity, and higher genetic loading; wealthier PD characterized by later onset, metabolic components, and chronic pain comorbidity.\u003c/p\u003e \u003cp\u003e \u003cb\u003ePART C. GENDER \u0026times; POVERTY STRATIFICATION.\u003c/b\u003e Panic disorder prevalence stratified by gender and poverty quartile. Interaction p-value from logistic regression.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"No\" id=\"Tabb\" border=\"1\"\u003e \u003ccolgroup cols=\"6\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eGender\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eQ1 (Poorest)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eQ2\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eQ3\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eQ4 (Wealthiest)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003eTrend p\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003ePrevalence\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003ePrevalence\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003ePrevalence\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c6\" namest=\"c5\"\u003e \u003cp\u003ePrevalence\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMale\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e2.59%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.74%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e2.06%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.00%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e0.132\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e(n\u0026thinsp;=\u0026thinsp;9/347)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e(n\u0026thinsp;=\u0026thinsp;3/404)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e(n\u0026thinsp;=\u0026thinsp;8/388)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c6\" namest=\"c5\"\u003e \u003cp\u003e(n\u0026thinsp;=\u0026thinsp;0/395)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eFemale\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e11.82%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e7.87%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e3.74%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e2.62%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e(n\u0026thinsp;=\u0026thinsp;42/447)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e(n\u0026thinsp;=\u0026thinsp;29/381)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e(n\u0026thinsp;=\u0026thinsp;15/401)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c6\" namest=\"c5\"\u003e \u003cp\u003e(n\u0026thinsp;=\u0026thinsp;9/381)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003cb\u003ePART D: HIGHEST RISK SUBGROUPS.\u003c/b\u003e Highest-risk subgroups identified through demographic intersection, with prevalence estimates and fold-increase relative to population average (3.66%).\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"No\" id=\"Tabc\" border=\"1\"\u003e \u003ccolgroup cols=\"4\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eRisk Group\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003ePD Prevalence\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eSample Size\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003evs Population Average\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePoor Older Women (Q1 \u0026times; Older \u0026times; Female)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e~15\u0026ndash;18%*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e~250\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e4.1\u0026ndash;4.9\u0026times; higher\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePoor Older Adults (Q1 \u0026times; Older, combined)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e10.40%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e492\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e2.8\u0026times; higher\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePoor Women (Q1 \u0026times; Female, all ages)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e11.82%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e447\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e3.2\u0026times; higher\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eWealthy Young Men (Q4 \u0026times; Younger \u0026times; Male)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0.00%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e~180*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eProtected\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003cb\u003ePART E: PD PHENOTYPE BY SOCIOECONOMIC STATUS.\u003c/b\u003e Feature comparison among PD cases only (n\u0026thinsp;=\u0026thinsp;115), stratified by median poverty-income ratio. T-tests and chi-square tests for group differences.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"No\" id=\"Tabd\" border=\"1\"\u003e \u003ccolgroup cols=\"6\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eFeature\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eLow PIR PD\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eHigh PIR PD\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eDifference\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003ep-value\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003eInterpretation\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eMean\u0026thinsp;\u0026plusmn;\u0026thinsp;SD\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eMean\u0026thinsp;\u0026plusmn;\u0026thinsp;SD\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAge (months)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e358.3\u0026thinsp;\u0026plusmn;\u0026thinsp;56.5\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e389.8\u0026thinsp;\u0026plusmn;\u0026thinsp;54.8\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e+31.5\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.009\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eWealthier PD: OLDER\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eBody Fat Mass (kg)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e24.3\u0026thinsp;\u0026plusmn;\u0026thinsp;13.1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e31.4\u0026thinsp;\u0026plusmn;\u0026thinsp;13.7\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e+7.1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.012\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eWealthier PD: MORE fat\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLower Back Pain (% yes)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e27.7%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e59.4%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e+31.6%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.001\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eWealthier PD: MORE pain\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eFamily History (% yes)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e38.6%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e18.8%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e-19.8%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.043\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003ePoorer PD: MORE FH\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eThese socioeconomic patterns validated Hypothesis 4 (socioeconomic stratification creates distinct risk profiles) and carried profound public health implications. The extreme disparities observed\u0026mdash;with poor women all ages facing nearly 12% prevalence while wealthy young men showed 0%\u0026mdash;indicated that universal screening protocols would be inefficient. Targeted, community-based approaches focusing on socioeconomically disadvantaged populations, particularly older women, would maximize public health yield with number needed to screen of approximately 8 to detect one case in highest-risk subgroups. Poverty distributions (Supplementary Figure S15), prevalence visualization (Supplementary Figure S16), and correlations (Supplementary Figure S17) detail socioeconomic effects.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e"},{"header":"4. DISCUSSION","content":"\u003cp\u003eThis multi-method mechanistic investigation revealed that panic disorder classification with near-perfect accuracy (99.75%) emerges from complex, synergistic interactions across 11 biopsychosocial features rather than from simple decision rules or dominant individual predictors. Across all analytic approaches, no single feature emerged as a dominant driver of panic disorder classification. Even the strongest individual predictors\u0026mdash;body fat mass and poverty-income ratio\u0026mdash;accounted for only\u0026thinsp;~\u0026thinsp;16% of total importance, underscoring that accurate prediction depended on synergistic interactions across biological, demographic, and social factors.\u003c/p\u003e \u003cp\u003eThe necessity of model complexity was quantified through direct comparison of a simple interpretable decision tree (maximum depth 5) with the full gradient boosting classifier. The simple model achieved only 40.97% accuracy, whereas the complex model reached 99.75%, yielding a \"complexity gap\" of 58.78 percentage points. This substantial gap provides empirical evidence that panic disorder patterns are inherently multidimensional and require nonlinear feature interactions for accurate classification. These findings align with theoretical conceptualizations of panic disorder as a complex biopsychosocial phenomenon (Clark, \u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e1986\u003c/span\u003e; Kyriakoulis and Kyrios, \u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e2023\u003c/span\u003e) and extend recent work demonstrating that machine learning approaches can capture psychiatric disorder patterns that traditional linear models cannot adequately represent (Dwyer et al., \u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e2018\u003c/span\u003e). The magnitude of this complexity gap suggests that efforts to develop simple screening tools based on single biomarkers or brief questionnaires may be fundamentally limited in their ability to achieve the sensitivity and specificity needed for effective population-level screening.\u003c/p\u003e \u003cp\u003eAlthough body fat mass appeared as a central interaction hub, it did not dominate the model on its own. Its contribution remained modest and gained predictive relevance only through interactions with age, BMI, poverty-income ratio, and gender. This pattern illustrates the model\u0026rsquo;s reliance on distributed, multi-feature integration. This pattern suggests that body composition may serve as a critical mediator linking biological vulnerability with psychosocial risk factors, consistent with research on metabolic-inflammatory hypotheses in anxiety disorders(Gariepy et al., \u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e2010\u003c/span\u003e; Strine et al., \u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e2008\u003c/span\u003e) Age functioned as a universal moderator, appearing in five of the top ten interactions, suggesting that vulnerability profiles change across the lifespan. These findings support dimensional models of psychopathology that emphasize transdiagnostic risk factors operating through complex, context-dependent mechanisms rather than simple additive effects (Kotov et al., \u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e2017\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eSocioeconomic stratification revealed a linear gradient in panic disorder prevalence across poverty quartiles, with the poorest quartile showing 6.42% prevalence compared to 1.16% in the wealthiest quartile. Women living in poverty exhibited the highest observed prevalence at 11.82%. When further stratified by older age, poor older women reached an estimated prevalence of approximately 15\u0026ndash;18%. This finding identifies a critical high-risk subgroup that may benefit from targeted screening and early intervention, consistent with social determinants of health frameworks.\u003c/p\u003e \u003cp\u003eOur investigation also resolved an apparent paradox wherein SHAP analysis suggested that higher income increased panic disorder risk while raw prevalence data showed the opposite pattern. This contradiction exemplifies Simpson's Paradox, a statistical phenomenon where associations observed in aggregate data can reverse when stratified by confounding variables (Pearl, \u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e2014\u003c/span\u003e). In our data, poverty correlates negatively with age, body fat mass, and female gender, all of which are themselves risk factors for panic disorder. When the model controls for these confounders through SHAP analysis, the marginal effect of income conditional on all other features can appear paradoxical.\u003c/p\u003e \u003cp\u003eThe distributed nature of predictive power across biological, psychological, and social domains supports integrative biopsychosocial models of panic disorder etiology (Gorman et al., \u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e2000\u003c/span\u003e). Contrary to expectations, psychological domain features (family history of cardiovascular disease, lower back pain) showed stronger mean effect sizes (Cohen's d\u0026thinsp;=\u0026thinsp;0.56) than biological features (Cohen's d\u0026thinsp;=\u0026thinsp;0.13), challenging views that panic disorder is primarily a disorder of physiological dysregulation. These findings align with the neurovisceral integration model (Thayer et al., \u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e2012\u003c/span\u003e) and suggest that panic disorder emerges from dysregulation of bidirectional brain-body communication rather than isolated autonomic dysfunction.\u003c/p\u003e \u003cp\u003eFrom a clinical perspective, our findings have several important implications. First, the near-perfect classification achieved with readily available sociodemographic, physiological, and behavioral markers suggests potential for developing scalable screening tools that do not require specialized psychiatric assessment or expensive biomarker assays. The model's precision ensures that individuals flagged as high-risk genuinely warrant further evaluation, avoiding unnecessary follow-up assessments. However, the recall indicates that approximately 6.9% panic disorder cases would not be detected, highlighting the continued importance of clinical judgment and comprehensive psychiatric evaluation. Second, the identification of poor older women as a high-risk group suggests that screening efforts may be most efficient when targeted toward this vulnerable population. Third, the distributed pattern of risk factors across multiple domains underscores the need for multidisciplinary treatment approaches addressing biological, psychological, and social contributors to panic disorder.\u003c/p\u003e \u003cp\u003eSeveral methodological strengths enhance confidence in our findings. Our conservative analytical approach, employing complete-case analysis without synthetic oversampling, ensures that performance metrics reflect real-world classification scenarios rather than artificially inflated estimates. The removal of data artifacts demonstrates our commitment to identifying genuine clinical signals rather than capitalizing on data quality issues. The use of multiple convergent analytical approaches, including distributional analysis, SHAP interpretation, decision tree comparison, and dimensionality reduction, provides triangulated evidence supporting our conclusions.\u003c/p\u003e \u003cp\u003eNevertheless, several limitations warrant consideration. First, the cross-sectional design precludes causal inference regarding the directionality of observed associations. Second, the relatively small number of panic disorder cases (n\u0026thinsp;=\u0026thinsp;115) limits statistical power. Third, our feature set does not include genetic markers, direct measures of autonomic function, or neuroimaging data. Fourth, the NHANES sample may not generalize to other cultural contexts. Fifth, reliance on the CIDI diagnostic algorithm may not perfectly correspond to clinical diagnoses made through comprehensive psychiatric evaluation.\u003c/p\u003e \u003cp\u003eFuture research should address these limitations through validation in independent longitudinal samples, incorporation of additional biological, genetic, and neurophysiological measures (Costello et al., \u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e2019\u003c/span\u003e; Fonzo and Etkin, \u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e2017\u003c/span\u003e; Hettema et al., \u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e2001\u003c/span\u003e; Thayer et al., \u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e2012\u003c/span\u003e), and exploration of whether similar patterns of distributed discriminative power and synergistic interactions characterize other anxiety disorders. Prospective studies could evaluate whether machine learning-based screening improves clinical outcomes, and qualitative research may elucidate barriers in high-risk subgroups.\u003c/p\u003e \u003cp\u003eIn conclusion, this mechanistic investigation demonstrates that near-perfect panic disorder classification emerges from complex, synergistic interactions across biopsychosocial features rather than from simple decision rules or dominant individual predictors. The substantial complexity gap between simple and sophisticated models quantifies the genuinely multidimensional nature of panic disorder patterns, validating the use of machine learning approaches for psychiatric screening and risk stratification. The discovery that women living in poverty exhibit high prevalence, rising further with age, identifies a critical high-risk population for targeted intervention. These findings advance theoretical understanding of panic disorder as a complex biopsychosocial phenomenon while demonstrating practical potential for scalable screening tools.\u003c/p\u003e "},{"header":"Declarations","content":"\u003ch2\u003eAuthor Contributions\u003c/h2\u003e\n\u003cp\u003e\u003cstrong\u003eAna Teresa Martins:\u003c/strong\u003e Investigation, Data Curation, Software, Writing – Original Draft, Visualization. \u003cstrong\u003eFilipe Ricardo Carvalho:\u003c/strong\u003e Formal analysis, Conceptualization, Methodology, Validation, Writing – Review \u0026amp; Editing, Supervision, Project administration.\u003c/p\u003e\n\u003ch2\u003eDATA\u0026nbsp;AVAILABILITY STATEMENT\u003c/h2\u003e\n\u003cp\u003eThe data used in this study are publicly available from the National Health and Nutrition Examination Survey (NHANES) 2017-2020, accessible at https://www.cdc.gov/nchs/nhanes/index.htm. All analysis code is available on GitHub at https://github.com/frpcarvalho/panic-mechanisms.\u003c/p\u003e\n\u003ch2\u003eEthics Statement\u003c/h2\u003e\n\u003cp\u003eThis secondary analysis used de-identified public NHANES data. Original NHANES data collection was approved by the National Center for Health Statistics Research Ethics Review Board with informed consent from all participants. No additional ethics approval was required for this study.\u003c/p\u003e\n\u003ch2\u003eFunding\u003c/h2\u003e\n\u003cp\u003eThis research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors.\u003c/p\u003e\n\u003ch2\u003eDeclaration of Competing Interests\u003c/h2\u003e\n\u003cp\u003eThe author declares no competing interests.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eChekroud AM, Bondar J, Delgadillo J, Doherty G, Wasil A, Fokkema M, Cohen Z, Belgrave D, DeRubeis R, Iniesta R, Dwyer D, Choi K (2021) The promise of machine learning in predicting treatment outcomes in psychiatry. World Psychiatry 20:154\u0026ndash;170. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1002/WPS.20882\u003c/span\u003e\u003cspan address=\"10.1002/WPS.20882\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eClark DM (1986) A cognitive approach to panic. Behav Res Ther 24:461\u0026ndash;470. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/0005-7967(86)90011-2\u003c/span\u003e\u003cspan address=\"10.1016/0005-7967(86)90011-2\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCostello H, Gould RL, Abrol E, Howard R (2019) Systematic review and meta-analysis of the association between peripheral inflammatory cytokines and generalised anxiety disorder. BMJ Open 9:e027925. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1136/BMJOPEN-2018-027925\u003c/span\u003e\u003cspan address=\"10.1136/BMJOPEN-2018-027925\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDoshi-Velez F, Kim B (2017) Towards A Rigorous Science of Interpretable Machine Learning\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDwyer DB, Falkai P, Koutsouleris N (2018) Machine Learning Approaches for Clinical Psychology and Psychiatry. Annu Rev Clin Psychol 14:91\u0026ndash;118. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1146/ANNUREV-CLINPSY-032816-045037\u003c/span\u003e\u003cspan address=\"10.1146/ANNUREV-CLINPSY-032816-045037\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eFonzo GA, Etkin A (2017) Affective neuroimaging in generalized anxiety disorder: an integrated review. Dialogues Clin Neurosci 19:169\u0026ndash;179. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.31887/DCNS.2017.19.2/GFONZO\u003c/span\u003e\u003cspan address=\"10.31887/DCNS.2017.19.2/GFONZO\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGariepy G, Nitka D, Schmitz N (2010) The association between obesity and anxiety disorders in the population: a systematic review and meta-analysis. Int J Obes (Lond) 34:407\u0026ndash;419. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1038/IJO.2009.252\u003c/span\u003e\u003cspan address=\"10.1038/IJO.2009.252\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGorman JM, Kent JM, Sullivan GM, Coplan JD (2000) Neuroanatomical hypothesis of panic disorder, revised. Am J Psychiatry 157:493\u0026ndash;505. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1176/APPI.AJP.157.4.493\u003c/span\u003e\u003cspan address=\"10.1176/APPI.AJP.157.4.493\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGuidotti R, Monreale A, Ruggieri S, Turini F, Giannotti F, Pedreschi D (2018) A Survey Of Methods For Explaining Black Box Models. ACM Comput Surv 51. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1145/3236009\u003c/span\u003e\u003cspan address=\"10.1145/3236009\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHassija V, Chamola V, Mahapatra A, Singal A, Goel D, Huang K, Scardapane S, Spinelli I, Mahmud M, Hussain A (2024) Interpreting Black-Box Models: A Review on Explainable Artificial Intelligence. Cognit Comput 16:45\u0026ndash;74. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/s12559-023-10179-8\u003c/span\u003e\u003cspan address=\"10.1007/s12559-023-10179-8\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHettema JM, Neale MC, Kendler KS (2001) A review and meta-analysis of the genetic epidemiology of anxiety disorders. Am J Psychiatry 158:1568\u0026ndash;1578. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1176/APPI.AJP.158.10.1568\u003c/span\u003e\u003cspan address=\"10.1176/APPI.AJP.158.10.1568\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eJavaid SF, Hashim IJ, Hashim MJ, Stip E, Samad MA, Ahbabi A, Al (2023) Epidemiology of anxiety disorders: global burden and sociodemographic associations. Middle East Curr Psychiatry 2023 30(1 30):44. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1186/S43045-023-00315-3\u003c/span\u003e\u003cspan address=\"10.1186/S43045-023-00315-3\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKessler RC, Merikangas KR (2006) The National Comorbidity Survey Replication (NCS-R): background and aims. International Journal of Methods in Psychiatric Research\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKim JE, Dager SR, Lyoo IK (2012) The role of the amygdala in the pathophysiology of panic disorder: evidence from neuroimaging studies. Biol Mood Anxiety Disord 2. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1186/2045-5380-2-20\u003c/span\u003e\u003cspan address=\"10.1186/2045-5380-2-20\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKotov R, Waszczuk MA, Krueger RF, Forbes MK, Watson D, Clark LA, Achenbach TM, Althoff RR, Ivanova MY, Bagby M, Brown R, Carpenter TA, Caspi WT, Moffitt A, Eaton TE, Forbush NR, Goldberg KT, Hasin D, Hyman D, Lynam SE, Samuel DR, South DB, Markon SC, Miller K, Morey JD, Mullins-Sweatt LC, Ormel SN, Patrick J, Regier CJ, Rescorla DA, Ruggero L, Sellbom CJ, Simms M, Skodol LJ, Slade AE, Tackett T, Waldman JL, Widiger ID, Wright TA, Zimmerman AGC, M (2017) The Hierarchical Taxonomy of Psychopathology (HiTOP): A dimensional alternative to traditional nosologies. J Abnorm Psychol 126:454\u0026ndash;477. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1037/ABN0000258\u003c/span\u003e\u003cspan address=\"10.1037/ABN0000258\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKyriakoulis P, Kyrios M (2023) Biological and cognitive theories explaining panic disorder: A narrative review. Front Psychiatry Volume. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3389/fpsyt.2023.957515\u003c/span\u003e\u003cspan address=\"10.3389/fpsyt.2023.957515\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e. 14-2023\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLundberg SM, Lee SI (2017) A Unified Approach to Interpreting Model Predictions. Adv Neural Inf Process Syst 2017-December, 4766\u0026ndash;4775\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMartins AT, Gomes A, Ros A, Santos J, Carvalho FR (2025) Predicting Panic Disorder with Socioeconomic, Physiological, and Behavioral Markers: A Machine Learning Study. PsyArXiv Preprints\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePearl J (2014) Comment: Understanding Simpson\u0026rsquo;s paradox. Am Stat 68:8\u0026ndash;13. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1080/00031305.2014.876829\u003c/span\u003e\u003cspan address=\"10.1080/00031305.2014.876829\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e;CTYPE:STRING:JOURNAL\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eReiss S, Peterson RA, Gursky DM, McNally RJ (1986) Anxiety sensitivity, anxiety frequency and the prediction of fearfulness. Behav Res Ther 24:1\u0026ndash;8. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/0005-7967(86)90143-9\u003c/span\u003e\u003cspan address=\"10.1016/0005-7967(86)90143-9\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSchmidt NB, Lerew DR, Jackson RJ (1997) The role of anxiety sensitivity in the pathogenesis of panic: prospective evaluation of spontaneous panic attacks during acute stress. J Abnorm Psychol 106:355\u0026ndash;364. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1037//0021-843X.106.3.355\u003c/span\u003e\u003cspan address=\"10.1037//0021-843X.106.3.355\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eStrine TW, Chapman DP, Balluz L, Mokdad AH (2008) Health-related quality of life and health behaviors by social and emotional support. Their relevance to psychiatry and medicine. Soc Psychiatry Psychiatr Epidemiol 43:151\u0026ndash;159. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/S00127-007-0277-X\u003c/span\u003e\u003cspan address=\"10.1007/S00127-007-0277-X\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eThayer JF, \u0026Aring;hs F, Fredrikson M, Sollers JJ, Wager TD (2012) A meta-analysis of heart rate variability and neuroimaging studies: Implications for heart rate variability as a marker of stress and health. Neurosci Biobehav Rev 36:747\u0026ndash;756. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.neubiorev.2011.11.009\u003c/span\u003e\u003cspan address=\"10.1016/j.neubiorev.2011.11.009\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"","lastPublishedDoi":"10.21203/rs.3.rs-8398524/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-8398524/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003ch2\u003eBackground\u003c/h2\u003e \u003cp\u003eMachine learning (ML) models trained on socioeconomic, physiological, and behavioral markers can classify panic disorder (PD) with high accuracy. Yet the mechanisms underlying these predictions remain poorly understood, limiting clinical translation and theoretical integration.\u003c/p\u003e\u003ch2\u003eObjective\u003c/h2\u003e \u003cp\u003eTo investigate why ML models achieve strong PD classification performance by examining feature interactions, individual contributions, model complexity requirements, and socioeconomic risk gradients.\u003c/p\u003e\u003ch2\u003eMethods\u003c/h2\u003e \u003cp\u003eUsing complete-case NHANES 1999\u0026ndash;2004 data (N\u0026thinsp;=\u0026thinsp;3,144; 115 PD cases), we applied a multi-method framework including distributional analysis, dimensionality reduction (UMAP, t-SNE), decision trees, SHAP interaction analysis, and socioeconomic stratification. The primary classifier was Gradient Boosting using 11 biopsychosocial predictors.\u003c/p\u003e\u003ch2\u003eResults\u003c/h2\u003e \u003cp\u003eIndividual features showed modest discriminative power (\u003cem\u003eCohen\u0026rsquo;s d\u003c/em\u003e\u0026thinsp;=\u0026thinsp;.13\u0026ndash;.70). SHAP identified 10 meaningful interactions, particularly body fat \u0026times; age and poverty-income ratio \u0026times; BMI. A shallow decision tree reached only 40.97% accuracy, indicating reliance on multidimensional interactions. Socioeconomic analysis showed a strong gradient (poorest quartile: 6.42% PD prevalence; wealthiest: 1.16%), with highest risk among low-income women.\u003c/p\u003e\u003ch2\u003eConclusion\u003c/h2\u003e \u003cp\u003eHigh PD classification accuracy emerges from synergistic biopsychosocial patterns. These results clarify why ML-based classification outperforms traditional screening, identify mechanistic pathways consistent with PD models, and highlight high-risk groups for targeted intervention.\u003c/p\u003e","manuscriptTitle":"Unravelling the Complexity Gap: A Mechanistic Investigation of Machine Learning Classification in Panic Disorder ","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2026-01-08 14:17:16","doi":"10.21203/rs.3.rs-8398524/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"3f36d73e-0def-44d3-b29d-0ce60586093e","owner":[],"postedDate":"January 8th, 2026","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[],"tags":[],"updatedAt":"2026-02-07T15:54:48+00:00","versionOfRecord":[],"versionCreatedAt":"2026-01-08 14:17:16","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-8398524","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-8398524","identity":"rs-8398524","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}