Hongwei Liu
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Hongwei Liu has a total of 28 co-authors. Here are the top co-authors by number of shared publications:
- Haixia Fan
First hospital of shanxi medical university, Taiyuan, People’s Republic of China
3 shared publications - Minheng Zhang
Department of Gerontology The First People's Hospital of Jinzhong Yuci Shanxi Province China
3 shared publications - Zhinan Ye
Department of Neurology, Taizhou Municipal Hospital (Taizhou University Affiliated Municipal Hospital), School of Medicine, Taizhou University, China
3 shared publications - Haiyan Wu
Department of Sleep Center, First Hospital of Shanxi Medical University, 85, Jiefang South Road, Taiyuan, 030001, China, 86 13623515703
2 shared publications - Caidi He
Department of Nursing, Taizhou Municipal Hospital (Taizhou University Affiliated Municipal Hospital), School of Medicine, Taizhou University, Taizhou, China
1 shared publication - Gaiying Ma
Department of Neurology, Peking University First Hospital Taiyuan Branch (Taiyuan Central Hospital of Shanxi Medical University), Taiyuan, Shanxi Province, China
1 shared publication
Latest Publications
- Open Access
- 25 October 2026
- Published by Korean Neuropsychiatric Association in Psychiatry Investigation
- Vol. 23 (10) , 1278-1297
Identification of Gene Signatures and Molecular Mechanisms Underlying the Comorbidity of Alzheimer’s Disease and Crohn’s Disease Using Machine Learning
Objective Alzheimer’s disease (AD) and Crohn’s disease (CD) both involve inflammation and immune dysregulation, yet the potential molecular mechanisms underlying their comorbidity remain unclear.Methods We integrated transcriptomic data from AD and CD patients and applied differential expression analysis, weighted gene coexpression network analysis, protein-protein interaction networks, and multiple machine learning approaches to identify key comorbidity genes. Functional enrichment, single-cell sequencing validation, and virtual knockout analyses were used to explore their biological roles. Molecular docking was performed to evaluate the binding affinity of candidate small-molecule drugs to the identified core genes.Results CXCL1 and IGFBP5 were identified as core comorbidity genes. CXCL1 was associated with inflammatory signaling, including cytokine receptor binding, neutrophil migration, and NOD-like receptor signaling. IGFBP5 was linked to growth factor binding, smooth muscle cell proliferation, and extracellular matrix-receptor interactions. Single-cell and virtual knockout analyses indicated that these genes play pivotal roles in inflammation, immune regulation, cell migration, and tissue remodeling, potentially bridging central and peripheral inflammation via the gut-brain axis and IgSF CAM signaling. Candidate drug prediction and molecular docking suggested that small molecules such as Dasatinib, Mifepristone, and Retinoic acid may modulate these pathways.Conclusion This study reveals the critical roles of CXCL1 and IGFBP5 in AD-CD comorbidity, providing a theoretical basis for exploring gut-brain axis mechanisms and potential targeted interventions.
- 25 October 2026
- Published by Korean Neuropsychiatric Association in Psychiatry Investigation
- Vol. 23 (10) , 1278-1297
Open AccessIdentification of Gene Signatures and Molecular Mechanisms Underlying the Comorbidity of Alzheimer’s Disease and Crohn’s Disease Using Machine Learning
Objective Alzheimer’s disease (AD) and Crohn’s disease (CD) both involve inflammation and immune dysregulation, yet the potential molecular mechanisms underlying their comorbidity remain unclear.Methods We integrated transcriptomic data from AD and CD patients and applied differential expression analysis, weighted gene coexpression network analysis, protein-protein interaction networks, and multiple machine learning approaches to identify key comorbidity genes. Functional enrichment, single-cell sequencing validation, and virtual knockout analyses were used to explore their biological roles. Molecular docking was performed to evaluate the binding affinity of candidate small-molecule drugs to the identified core genes.Results CXCL1 and IGFBP5 were identified as core comorbidity genes. CXCL1 was associated with inflammatory signaling, including cytokine receptor binding, neutrophil migration, and NOD-like receptor signaling. IGFBP5 was linked to growth factor binding, smooth muscle cell proliferation, and extracellular matrix-receptor interactions. Single-cell and virtual knockout analyses indicated that these genes play pivotal roles in inflammation, immune regulation, cell migration, and tissue remodeling, potentially bridging central and peripheral inflammation via the gut-brain axis and IgSF CAM signaling. Candidate drug prediction and molecular docking suggested that small molecules such as Dasatinib, Mifepristone, and Retinoic acid may modulate these pathways.Conclusion This study reveals the critical roles of CXCL1 and IGFBP5 in AD-CD comorbidity, providing a theoretical basis for exploring gut-brain axis mechanisms and potential targeted interventions.
- Open Access
- 28 May 2026
- Published by JMIR Publications Inc. in JMIR Aging
- Vol. 9, e78914
Subsequent Hypertension’s Mediation of the Association Between Sleep Duration Trajectories and New-Onset Cardiovascular Disease: Population-Based Cohort Study
Background: Prior research has rarely explored the link between sleep duration trajectories and cardiovascular disease (CVD) either in cross-sectional or longitudinal data. Objective: We aimed to analyze the relationship between sleep duration trajectories and the incidence of CVD. Methods: Data from 5603 participants in the China Health and Retirement Longitudinal Study (CHARLS) were used in the analysis. Sleep duration was self-reported at multiple time points, and group-based trajectory modeling identified distinct patterns of total and nocturnal sleep duration over time. To investigate the link among sleep duration trajectories and CVD, Cox proportional hazards models, restricted cubic splines, and mediation analyses were used, with sensitivity analyses ensuring the results’ robustness. Results: Four sleep duration trajectories were identified, including steady high, inverted U-shaped, steady low, and U-shaped. In fully adjusted models, an inverted U-shaped sleep trajectory was associated with a 34% higher risk of hypertension compared with the steady high group (total sleep: hazard ratio [HR]=1.34, 95% CI 1.07‐1.69; P =.01). Notably, the inverted U-shaped sleep trajectory showed a 110% increased stroke risk during nocturnal periods compared to the steady high group (nocturnal sleep: HR=2.10, 95% CI 1.08‐4.05; P =.03). The findings remained consistent across a range of sensitivity analyses. Hypertension was identified as a partial mediator of the association between sleep instability and stroke risk. Conclusions: The findings indicate that an inverted U-shaped sleep pattern is associated with increased cardiovascular risk, partially mediated by hypertension, highlighting sleep stability as a potential preventive target.
- 28 May 2026
- Published by JMIR Publications Inc. in JMIR Aging
- Vol. 9, e78914
Open AccessSubsequent Hypertension’s Mediation of the Association Between Sleep Duration Trajectories and New-Onset Cardiovascular Disease: Population-Based Cohort Study
Background: Prior research has rarely explored the link between sleep duration trajectories and cardiovascular disease (CVD) either in cross-sectional or longitudinal data. Objective: We aimed to analyze the relationship between sleep duration trajectories and the incidence of CVD. Methods: Data from 5603 participants in the China Health and Retirement Longitudinal Study (CHARLS) were used in the analysis. Sleep duration was self-reported at multiple time points, and group-based trajectory modeling identified distinct patterns of total and nocturnal sleep duration over time. To investigate the link among sleep duration trajectories and CVD, Cox proportional hazards models, restricted cubic splines, and mediation analyses were used, with sensitivity analyses ensuring the results’ robustness. Results: Four sleep duration trajectories were identified, including steady high, inverted U-shaped, steady low, and U-shaped. In fully adjusted models, an inverted U-shaped sleep trajectory was associated with a 34% higher risk of hypertension compared with the steady high group (total sleep: hazard ratio [HR]=1.34, 95% CI 1.07‐1.69; P =.01). Notably, the inverted U-shaped sleep trajectory showed a 110% increased stroke risk during nocturnal periods compared to the steady high group (nocturnal sleep: HR=2.10, 95% CI 1.08‐4.05; P =.03). The findings remained consistent across a range of sensitivity analyses. Hypertension was identified as a partial mediator of the association between sleep instability and stroke risk. Conclusions: The findings indicate that an inverted U-shaped sleep pattern is associated with increased cardiovascular risk, partially mediated by hypertension, highlighting sleep stability as a potential preventive target.
- Open Access
- 25 May 2026
- Published by Korean Neuropsychiatric Association in Psychiatry Investigation
- Vol. 23 (5) , 685-695
Exploring Body Roundness Index’s Mediating Role in the Depression-Cognitive Performance Relationship Among US Adults (≥60 Years): Evidence From National Health and Nutrition Examination Survey
Objective Late-life depression is a known risk factor for cognitive decline, yet the role of central adiposity, measured by the body roundness index (BRI), remains unclear. This study examined whether BRI mediates the relationship between depressive symptoms and cognitive performance in older adults.Methods Our analysis included 2,580 adults aged 60 and over, using data from NHANES. Depressive symptoms were assessed via the Patient Health Questionnaire-9 (PHQ-9), cognitive function was evaluated using the Consortium to Establish a Registry for Alzheimer’s Disease (CERAD) Word Learning and Delayed Recall (CERAD-WL/DR), Animal Fluency test (AFT), and Digit Symbol Substitution Test (DSST), and BRI was calculated from waist circumference and height. Analyses included weighted multivariable logistic regression, restricted cubic spline analyses, subgroup and sensitivity analyses, and mediation analyses.Results Across all assessment modalities, there was a significant link between high PHQ-9 scores and cognitive performance. For continuous PHQ-9 scores, each 1-point increase corresponded to: CERAD (odds ratio [OR]=1.034, 95% confidence interval [CI]: 1.000–1.070, p=0.049), AFT (OR=1.046, 95% CI: 1.014–1.079, p=0.007), and DSST (OR=1.080, 95% CI: 1.040–1.123, p<0.001). Categorical analysis revealed particularly strong effects for moderate-to-severe depression (PHQ-9 ≥10) versus minimal symptoms (PHQ-9, 0–4): DSST (OR=2.562, 95% CI: 1.432–4.584, p=0.004), AFT (OR=1.834, 95% CI: 1.239–2.714, p=0.005), with CERAD showing marginal significance (OR=1.454, 95% CI: 0.936–2.259, p=0.090). Sensitivity analyses confirmed the stability of these associations in multiply imputed datasets. BRI exhibited a statistical mediation effect, accounting for 7.23% (β=-0.021, p=0.006) of the depression-DSST association, though mediation effects were nonsignificant for CERAD (6.21%, p=0.184) or AFT (4.41%, p=0.448).Conclusion Depression severity was independently linked to lower cognitive performance, with BRI partially mediating this relationship.
- 25 May 2026
- Published by Korean Neuropsychiatric Association in Psychiatry Investigation
- Vol. 23 (5) , 685-695
Open AccessExploring Body Roundness Index’s Mediating Role in the Depression-Cognitive Performance Relationship Among US Adults (≥60 Years): Evidence From National Health and Nutrition Examination Survey
Objective Late-life depression is a known risk factor for cognitive decline, yet the role of central adiposity, measured by the body roundness index (BRI), remains unclear. This study examined whether BRI mediates the relationship between depressive symptoms and cognitive performance in older adults.Methods Our analysis included 2,580 adults aged 60 and over, using data from NHANES. Depressive symptoms were assessed via the Patient Health Questionnaire-9 (PHQ-9), cognitive function was evaluated using the Consortium to Establish a Registry for Alzheimer’s Disease (CERAD) Word Learning and Delayed Recall (CERAD-WL/DR), Animal Fluency test (AFT), and Digit Symbol Substitution Test (DSST), and BRI was calculated from waist circumference and height. Analyses included weighted multivariable logistic regression, restricted cubic spline analyses, subgroup and sensitivity analyses, and mediation analyses.Results Across all assessment modalities, there was a significant link between high PHQ-9 scores and cognitive performance. For continuous PHQ-9 scores, each 1-point increase corresponded to: CERAD (odds ratio [OR]=1.034, 95% confidence interval [CI]: 1.000–1.070, p=0.049), AFT (OR=1.046, 95% CI: 1.014–1.079, p=0.007), and DSST (OR=1.080, 95% CI: 1.040–1.123, p<0.001). Categorical analysis revealed particularly strong effects for moderate-to-severe depression (PHQ-9 ≥10) versus minimal symptoms (PHQ-9, 0–4): DSST (OR=2.562, 95% CI: 1.432–4.584, p=0.004), AFT (OR=1.834, 95% CI: 1.239–2.714, p=0.005), with CERAD showing marginal significance (OR=1.454, 95% CI: 0.936–2.259, p=0.090). Sensitivity analyses confirmed the stability of these associations in multiply imputed datasets. BRI exhibited a statistical mediation effect, accounting for 7.23% (β=-0.021, p=0.006) of the depression-DSST association, though mediation effects were nonsignificant for CERAD (6.21%, p=0.184) or AFT (4.41%, p=0.448).Conclusion Depression severity was independently linked to lower cognitive performance, with BRI partially mediating this relationship.
- Open Access
- 22 May 2026
- Published by Wiley in European Journal of Neurology
- Vol. 33 (5) , e70618
Diagnostic and Prognostic Utility of Plasma p‐tau217 for Alzheimer's Disease in Chinese Elderly: Insights From the SILCODE Study With a Derived Threshold
Background: Plasma phosphorylated tau 217 (p‐tau217) has emerged as a promising Alzheimer's disease (AD) biomarker, yet its longitudinal associations with neurodegeneration and cognitive decline remain inadequately characterized in Chinese populations, and ethnicity‐specific diagnostic thresholds are lacking for optimal clinical application. Methods: A total of 541 participants (402 cognitively unimpaired [CU]; 139 cognitively impaired [CI]) from the Sino Longitudinal Study on Cognitive Decline (SILCODE) cohort were enrolled. Cross‐sectional and longitudinal associations of plasma p‐tau217 with amyloid‐β (Aβ) pathology, neurodegeneration, and cognition were evaluated. Diagnostic thresholds were derived using receiver operating characteristic analysis, and Cox regression assessed prognostic value for clinical progression. Results: Cross‐sectionally, baseline p‐tau217 was associated with greater Aβ burden, neurodegeneration, and poorer cognition in the whole cohort and CI group; in the CU group, associations were confined to amyloid measures. Longitudinally, accelerated p‐tau217 accumulation was associated with faster neurodegeneration and cognitive decline in the whole cohort, with stage‐dependent patterns: nominal associations with neurodegenerative markers in CU and prominent cognitive associations in CI. Plasma p‐tau217 demonstrated high diagnostic accuracy for amyloid positivity (AUC = 0.891; cutoff: 0.529 pg/mL). Threshold‐based stratification effectively differentiated individuals by Aβ burden, neurodegeneration, and cognitive trajectories. Elevated baseline p‐tau217 predicted higher progression risk (Whole cohort: HR = 2.66 [1.28–5.53], p = 0.009; CU: HR = 2.44 [1.07–5.59], p = 0.034). Conclusion: Plasma p‐tau217 serves as a valuable diagnostic and prognostic biomarker for AD, even among CU individuals, and the ethnicity‐specific threshold of 0.529 pg/mL enhances its clinical applicability for early detection and risk stratification in Chinese populations.
- 22 May 2026
- Published by Wiley in European Journal of Neurology
- Vol. 33 (5) , e70618
Open AccessDiagnostic and Prognostic Utility of Plasma p‐tau217 for Alzheimer's Disease in Chinese Elderly: Insights From the SILCODE Study With a Derived Threshold
Background: Plasma phosphorylated tau 217 (p‐tau217) has emerged as a promising Alzheimer's disease (AD) biomarker, yet its longitudinal associations with neurodegeneration and cognitive decline remain inadequately characterized in Chinese populations, and ethnicity‐specific diagnostic thresholds are lacking for optimal clinical application. Methods: A total of 541 participants (402 cognitively unimpaired [CU]; 139 cognitively impaired [CI]) from the Sino Longitudinal Study on Cognitive Decline (SILCODE) cohort were enrolled. Cross‐sectional and longitudinal associations of plasma p‐tau217 with amyloid‐β (Aβ) pathology, neurodegeneration, and cognition were evaluated. Diagnostic thresholds were derived using receiver operating characteristic analysis, and Cox regression assessed prognostic value for clinical progression. Results: Cross‐sectionally, baseline p‐tau217 was associated with greater Aβ burden, neurodegeneration, and poorer cognition in the whole cohort and CI group; in the CU group, associations were confined to amyloid measures. Longitudinally, accelerated p‐tau217 accumulation was associated with faster neurodegeneration and cognitive decline in the whole cohort, with stage‐dependent patterns: nominal associations with neurodegenerative markers in CU and prominent cognitive associations in CI. Plasma p‐tau217 demonstrated high diagnostic accuracy for amyloid positivity (AUC = 0.891; cutoff: 0.529 pg/mL). Threshold‐based stratification effectively differentiated individuals by Aβ burden, neurodegeneration, and cognitive trajectories. Elevated baseline p‐tau217 predicted higher progression risk (Whole cohort: HR = 2.66 [1.28–5.53], p = 0.009; CU: HR = 2.44 [1.07–5.59], p = 0.034). Conclusion: Plasma p‐tau217 serves as a valuable diagnostic and prognostic biomarker for AD, even among CU individuals, and the ethnicity‐specific threshold of 0.529 pg/mL enhances its clinical applicability for early detection and risk stratification in Chinese populations.
- Open Access
- 1 January 2026
- Published by Wiley in Behavioural Neurology
- Vol. 2026 (1) , e1282354
Circadian Syndrome Mediates the Association Between Body Roundness Index and Incident Stroke Risk in Middle‐Aged and Older Chinese Adults: A Nationwide Prospective Cohort Study
Background: Despite BRI being a novel measure for chronic disease risk, the relationship between BRI and stroke risk under varying circadian syndrome (CircS) conditions is still unclear. Methods: The analysis utilized data from 7535 participants involved in the China Health and Retirement Longitudinal Study (CHARLS). CircS was defined by at least four concurrent metabolic disturbances due to circadian rhythm disruption. To ensure robustness, the BRI‐stroke association was assessed using Cox models, restricted cubic splines, Kaplan–Meier curves, and mediation analyses, along with subgroup and sensitivity analyses. The receiver operating characteristic (ROC) analysis was used to compare the stroke risk prediction performance of BRI with conventional metabolic indices. Results: Following full adjustment, higher log (BRI) values were significantly associated with an increased risk of stroke (HR = 2.50, 95% CI: 1.81–3.44, p < 0.001), with Q4 showing a greater risk than Q1 (HR = 2.23, 95% CI: 1.63–3.04, p < 0.001). In CircS patients, no significant association was observed (log (BRI) HR = 1.23, 95% CI: 0.55–2.76; Q4 vs. Q1 HR = 0.61, 95% CI: 0.21–1.77; all p > 0.05). However, the association remained significant in non‐CircS individuals (log (BRI) HR = 2.36, 95% CI: 1.61–3.47; Q4 vs. Q1 HR = 2.08, 95% CI: 1.46–2.98; all p < 0.001). Subgroup and sensitivity analyses bolstered the robustness of our findings. CircS mediated 21.0% of the association between BRI and stroke risk (95% CI: 18%–91%, p < 0.001). With an area under the curve (AUC) of 0.656, BRI demonstrated moderate predictive capability for stroke, outperforming other anthropometric indices. Conclusions: An increased BRI is associated with a higher stroke risk, partially mediated by CircS, highlighting the potential role of circadian health in stroke prevention.
- 1 January 2026
- Published by Wiley in Behavioural Neurology
- Vol. 2026 (1) , e1282354
Open AccessCircadian Syndrome Mediates the Association Between Body Roundness Index and Incident Stroke Risk in Middle‐Aged and Older Chinese Adults: A Nationwide Prospective Cohort Study
Background: Despite BRI being a novel measure for chronic disease risk, the relationship between BRI and stroke risk under varying circadian syndrome (CircS) conditions is still unclear. Methods: The analysis utilized data from 7535 participants involved in the China Health and Retirement Longitudinal Study (CHARLS). CircS was defined by at least four concurrent metabolic disturbances due to circadian rhythm disruption. To ensure robustness, the BRI‐stroke association was assessed using Cox models, restricted cubic splines, Kaplan–Meier curves, and mediation analyses, along with subgroup and sensitivity analyses. The receiver operating characteristic (ROC) analysis was used to compare the stroke risk prediction performance of BRI with conventional metabolic indices. Results: Following full adjustment, higher log (BRI) values were significantly associated with an increased risk of stroke (HR = 2.50, 95% CI: 1.81–3.44, p < 0.001), with Q4 showing a greater risk than Q1 (HR = 2.23, 95% CI: 1.63–3.04, p < 0.001). In CircS patients, no significant association was observed (log (BRI) HR = 1.23, 95% CI: 0.55–2.76; Q4 vs. Q1 HR = 0.61, 95% CI: 0.21–1.77; all p > 0.05). However, the association remained significant in non‐CircS individuals (log (BRI) HR = 2.36, 95% CI: 1.61–3.47; Q4 vs. Q1 HR = 2.08, 95% CI: 1.46–2.98; all p < 0.001). Subgroup and sensitivity analyses bolstered the robustness of our findings. CircS mediated 21.0% of the association between BRI and stroke risk (95% CI: 18%–91%, p < 0.001). With an area under the curve (AUC) of 0.656, BRI demonstrated moderate predictive capability for stroke, outperforming other anthropometric indices. Conclusions: An increased BRI is associated with a higher stroke risk, partially mediated by CircS, highlighting the potential role of circadian health in stroke prevention.
- Open Access
- 1 January 2026
- Published by Wiley in Mediators of Inflammation
- Vol. 2026 (1)
Gut Microbiota‐Derived Metabolites Regulate CASP3 and Neuroimmune Pathways in Multiple Sclerosis: An Integrative Multiomics Study
Background: Multiple sclerosis (MS) is a chronic autoimmune disease of the central nervous system, closely associated with neuroinflammation, immune dysregulation, and gut microbiota imbalance. Gut microbiota‐derived metabolites may modulate key targets involved in MS pathogenesis. Methods: This study integrated network pharmacology, machine learning (ML), and single‐cell transcriptome analysis to identify MS‐related differentially expressed genes (DEGs) and potential targets of gut microbial metabolites. Feature contributions were evaluated using the SHapley Additive exPlanations (SHAP) method, and causal relationships were validated via Mendelian randomization (MR). Single‐cell analysis, molecular docking, and assessments of drug‐likeness and toxicity were also performed. Results: Caspase‐3 (CASP3) was identified as a core target interacting with multiple gut microbial metabolites, including L‐isoleucine, aromatic lactic acid derivatives, 3‐hydroxyphenethyl alcohol, and D‐xylose, potentially regulating neuroimmune responses via TNF, MAPK, IL‐17, and galectin pathways. Specific microbial taxa, such as Akkermansia , Bacteroides , and Bifidobacterium , were closely associated with these metabolites. The metabolites exhibited favorable drug‐likeness and low predicted toxicity, indicating potential therapeutic value. Conclusion: Gut microbial dysbiosis and its metabolites play a significant role in MS onset and progression, providing a theoretical basis for identifying therapeutic targets and gut–CNS axis interventions. Experimental validation is needed to confirm mechanisms and translational potential.
- 1 January 2026
- Published by Wiley in Mediators of Inflammation
- Vol. 2026 (1)
Open AccessGut Microbiota‐Derived Metabolites Regulate CASP3 and Neuroimmune Pathways in Multiple Sclerosis: An Integrative Multiomics Study
Background: Multiple sclerosis (MS) is a chronic autoimmune disease of the central nervous system, closely associated with neuroinflammation, immune dysregulation, and gut microbiota imbalance. Gut microbiota‐derived metabolites may modulate key targets involved in MS pathogenesis. Methods: This study integrated network pharmacology, machine learning (ML), and single‐cell transcriptome analysis to identify MS‐related differentially expressed genes (DEGs) and potential targets of gut microbial metabolites. Feature contributions were evaluated using the SHapley Additive exPlanations (SHAP) method, and causal relationships were validated via Mendelian randomization (MR). Single‐cell analysis, molecular docking, and assessments of drug‐likeness and toxicity were also performed. Results: Caspase‐3 (CASP3) was identified as a core target interacting with multiple gut microbial metabolites, including L‐isoleucine, aromatic lactic acid derivatives, 3‐hydroxyphenethyl alcohol, and D‐xylose, potentially regulating neuroimmune responses via TNF, MAPK, IL‐17, and galectin pathways. Specific microbial taxa, such as Akkermansia , Bacteroides , and Bifidobacterium , were closely associated with these metabolites. The metabolites exhibited favorable drug‐likeness and low predicted toxicity, indicating potential therapeutic value. Conclusion: Gut microbial dysbiosis and its metabolites play a significant role in MS onset and progression, providing a theoretical basis for identifying therapeutic targets and gut–CNS axis interventions. Experimental validation is needed to confirm mechanisms and translational potential.
- Open Access
- 30 December 2025
- Published by Wiley in Food Science & Nutrition
- Vol. 14 (1)
Using SHAP and LIME to Explain Machine Learning Models Predicting Comorbid Depression and Stroke From Daily Dietary Nutrient Intake in a US Population‐Based Study
While comorbid depression and stroke are a major concern for public health, the effect of dietary nutrient patterns on their concurrent occurrence is still largely unexplored. From NHANES, a survey of the U.S. civilian, non‐institutionalized population, we included 814 participants with complete data on diet, depression, and stroke. Of these, 140 were identified with comorbid depression and stroke. Baseline characteristics were compared between groups, and Weighted Quantile Sum (WQS) regression was used to evaluate the collective effects of nutrient mixtures. Machine learning models aimed at predicting comorbid conditions were developed, incorporating Synthetic Minority Oversampling Technique (SMOTE) for oversampling and Boruta for selecting features. The interpretability of these models was analyzed using SHapley Additive exPlanations (SHAP) and Local Interpretable Model‐agnostic Explanations (LIME). Participants with comorbidities were younger and had lower socioeconomic status, along with reduced intake of thiamin, vitamin B6, total folate, added vitamin B12, and vitamin C. Although neither WQS‐negative nor WQS‐positive indices showed statistically significant associations with comorbidity risk, specific nutrients such as alcohol, alpha‐carotene, added vitamin B12, theobromine, and vitamin E emerged as predominant contributors within the mixture models. The Random Forest classifier achieved the highest area under the receiver operating characteristic curve (AUC = 0.945) when adjusted for covariates and maintained consistently high performance in the unadjusted setting. SHAP and LIME analyses consistently identified vitamin B1, vitamin B12, zinc, vitamin C, and caffeine as influential predictors, with SHAP plots revealing mirrored feature contribution patterns depending on comorbidity status. Covariate adjustment improved directional stability and interpretability, particularly in SHAP dependence plots and waterfall visualizations. LIME explanations at the individual level corroborated these findings, showing consistent yet class‐dependent feature effects. Although the overall mixture effect was not significant, machine learning identified nutrient‐specific signals associated with comorbid depression and stroke. These results indicate that integrating dietary indicators with explainable artificial intelligence may improve transparency in risk prediction and guide future longitudinal and interventional research.
- 30 December 2025
- Published by Wiley in Food Science & Nutrition
- Vol. 14 (1)
Open AccessUsing SHAP and LIME to Explain Machine Learning Models Predicting Comorbid Depression and Stroke From Daily Dietary Nutrient Intake in a US Population‐Based Study
While comorbid depression and stroke are a major concern for public health, the effect of dietary nutrient patterns on their concurrent occurrence is still largely unexplored. From NHANES, a survey of the U.S. civilian, non‐institutionalized population, we included 814 participants with complete data on diet, depression, and stroke. Of these, 140 were identified with comorbid depression and stroke. Baseline characteristics were compared between groups, and Weighted Quantile Sum (WQS) regression was used to evaluate the collective effects of nutrient mixtures. Machine learning models aimed at predicting comorbid conditions were developed, incorporating Synthetic Minority Oversampling Technique (SMOTE) for oversampling and Boruta for selecting features. The interpretability of these models was analyzed using SHapley Additive exPlanations (SHAP) and Local Interpretable Model‐agnostic Explanations (LIME). Participants with comorbidities were younger and had lower socioeconomic status, along with reduced intake of thiamin, vitamin B6, total folate, added vitamin B12, and vitamin C. Although neither WQS‐negative nor WQS‐positive indices showed statistically significant associations with comorbidity risk, specific nutrients such as alcohol, alpha‐carotene, added vitamin B12, theobromine, and vitamin E emerged as predominant contributors within the mixture models. The Random Forest classifier achieved the highest area under the receiver operating characteristic curve (AUC = 0.945) when adjusted for covariates and maintained consistently high performance in the unadjusted setting. SHAP and LIME analyses consistently identified vitamin B1, vitamin B12, zinc, vitamin C, and caffeine as influential predictors, with SHAP plots revealing mirrored feature contribution patterns depending on comorbidity status. Covariate adjustment improved directional stability and interpretability, particularly in SHAP dependence plots and waterfall visualizations. LIME explanations at the individual level corroborated these findings, showing consistent yet class‐dependent feature effects. Although the overall mixture effect was not significant, machine learning identified nutrient‐specific signals associated with comorbid depression and stroke. These results indicate that integrating dietary indicators with explainable artificial intelligence may improve transparency in risk prediction and guide future longitudinal and interventional research.