Multifactorial Analysis and Predictive Modeling of Wound Healing Outcomes in Diabetic ICU Patients: a Cohort Study Based on MIMIC-IV

Multifactorial Analysis and Predictive Modeling of Wound Healing Outcomes in Diabetic ICU Patients: a Cohort Study Based on MIMIC-IV | 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 Multifactorial Analysis and Predictive Modeling of Wound Healing Outcomes in Diabetic ICU Patients: a Cohort Study Based on MIMIC-IV Linmin Zhuge, Xinxin Chen, Yiwei Teng, Kan Li, Minyu Zhu, Honglin Teng This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-6307348/v1 This work is licensed under a CC BY 4.0 License Status: Under Revision Version 1 posted 13 You are reading this latest preprint version Abstract Background Wound healing is a critical determinant of recovery and quality of life in patients with diabetes, particularly those admitted to intensive care units (ICUs). Identifying the key factors influencing wound healing and optimizing treatment strategies is essential for improving outcomes. Despite prior studies, there is limited comprehensive analysis that integrates multiple risk factors into predictive modeling frameworks. Objective This study aims to identify the significant factors affecting wound healing in diabetic ICU patients, evaluate the effects of different treatment approaches on healing outcomes, and develop a robust predictive model to assist clinicians in early risk identification and personalized treatment planning. Methods We utilized data from the MIMIC-IV database, encompassing 149,392 patient records. Key variables analyzed included demographic characteristics, chronic disease histories, wound-related factors, and treatment modalities. Descriptive and uni-variate analyses were performed to explore baseline characteristics and their associations with healing outcomes. Cox proportional hazards regression and logistic regression models were used for multi-factorial analyses, while machine learning models such as Random Forest and XGBoost were employed for predictive modeling. Models interpretability was enhanced through SHAP (SHapley Additive exPlanations) and LIME (Local Interpretable Model-agnostic Explanations) analyses. Results Factors such as age, the presence of pressure ulcers, chronic kidney disease, and treatment modalities (e.g., insulin therapy, negative pressure therapy) emerged as significant predictors of wound healing outcomes. Random Forest achieved the highest performance among predictive models, with an area under the receiver operating characteristic curve (AUC) of 0.96. SHAP analysis identified age and death flags as critical determinants, while LIME provided patient-specific insights into model predictions. Conclusions This study underscores the importance of integrating multifactorial data to predict wound healing outcomes in diabetic ICU patients. The findings provide actionable insights for personalized treatment strategies and resource allocation in clinical settings. Future research should focus on validating these models in diverse datasets and exploring longitudinal impacts on patient recovery. Wound healing predictors Treatment efficacy assessment Machine learning predictive modeling Personalized treatment strategies Diabetic ICU patients Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Figure 7 Figure 8 Figure 9 Figure 10 Figure 11 Figure 12 Figure 13 Figure 14 Introduction Diabetes mellitus, a chronic metabolic disorder, affects approximately 10.5% of the global adult population, with its prevalence projected to rise to 11.3% by 2030 and 12.2% by 2045 1 . This dramatic increase underscores the urgent need to address diabetes-related complications, particularly those that significantly impact quality of life and healthcare resources. Among these complications, impaired wound healing poses a critical challenge, especially in patients admitted to intensive care units (ICUs) where pressure ulcer incidence reaches 18.5% 2 . Factors such as hyperglycemia-induced angiogenesis suppression 3 and neuropathy-mediated tissue damage 4 exacerbate wound healing difficulties in diabetic individuals 5 . In ICU settings, these challenges are magnified by prolonged immobility (associated with 32% increased pressure injury risk 6 ), systemic infections requiring antibiotic therapy 7 , and metabolic instability 8 . Delayed wound healing in these patients prolongs hospitalization by 7.2 days on average 9 , elevating risks of infections (OR = 3.1) 10 , amputations (HR = 2.8) 11 , and mortality (RR = 1.9) 12 . These outcomes highlight the need for deeper understanding of healing determinants in diabetic ICU patients. Despite advancements in diabetes management, current wound care strategies show limited efficacy in critical care settings. The 2012 International Working Group on Diabetic Foot (IWGDF) guidelines emphasize glycemic control 13 , yet fail to address ICU-specific factors like ventilator-associated tissue hypoxia 14 . Prior studies predominantly focus on isolated variables such as HbA1c levels 15 or infection biomarkers 16 , neglecting the multifactorial interactions between inflammatory mediators 17 , treatment modalities 18 , and comorbidities 19 . This study aims to address these gaps by conducting a multifactorial analysis of diabetic ICU patients using data from the MIMIC-IV database. By leveraging both traditional statistical methods and machine learning algorithms, we seek to:Identify the key determinants of wound healing outcomes;Evaluate the impact of various treatment modalities on healing processes;Develop predictive models with high accuracy and clinical applicability. The integration of multi-factorial data and machine learning techniques provides an opportunity to advance personalized medicine. Predictive models can facilitate early risk identification, enabling clinicians to implement tailored interventions for high-risk patients. Additionally, these insights can inform healthcare policy by optimizing resource allocation and improving the quality of care delivered to diabetic ICU patients. By addressing the existing research gaps and employing advanced methodologies, this study contributes significantly to the field of diabetes management and critical care. Data and Methodology Data Selection This study employed a systematic approach to investigate factors influencing wound healing in diabetic patients admitted to intensive care units (ICUs). The analysis was supported by a comprehensive dataset and advanced methodologies to ensure robust and clinically relevant results. The data utilized for this study were derived from the Medical Information Mart for Intensive Care IV (MIMIC-IV) database, a publicly accessible and comprehensive clinical dataset. This database encompasses detailed records of patients admitted to ICUs at the Beth Israel Deaconess Medical Center between 2008 and 2019. Spanning over 380,000 admissions, it includes granular information such as demographic characteristics, laboratory test results, medication usage, procedures performed, and clinical outcomes. The de-identified nature of the data ensured compliance with ethical standards, with institutional review board approval secured for this research. The study cohort comprised diabetic patients aged 18 years or older who had sufficient data to evaluate wound healing outcomes. From the dataset, 149,392 eligible patients were identified, including 2,289 individuals diagnosed with pressure ulcers. To maintain the rigor and validity of the analysis, patients with missing critical data (e.g., healing outcomes, comorbidities, or treatment details) or inconsistencies in records (e.g., implausible values for age or healing duration) were excluded. The selection of variables for analysis was guided by clinical relevance and prior literature 18 , 20 . Key variables included demographic characteristics such as age, gender, and ethnicity; comorbidities including chronic kidney disease, cardiovascular diseases, and hypertension; wound-specific characteristics such as the presence of pressure ulcers and infection indicators (e.g., sepsis and pneumonia); and treatment modalities including the use of antibiotics, negative pressure wound therapy, and insulin therapy 12 , 21 . These variables provided a holistic perspective on the multifactorial influences affecting wound healing. Methodology To analyze the data, a multi-pronged methodology was employed. Descriptive statistics were calculated to summarize the baseline characteristics of the cohort, presenting continuous variables as means ± standard deviations or medians (interquartile ranges) and categorical variables as frequencies and percentages. Univariate analyses were then conducted to identify potential associations between individual variables and wound healing outcomes, utilizing Chi-square tests for categorical variables and t-tests or Mann-Whitney U tests for continuous variables, depending on their distribution. To adjust for confounding factors and ascertain the independent effects of variables, multivariate analyses were performed. Logistic regression models were used for binary outcomes (e.g., healed vs. not healed), providing odds ratios (ORs) with 95% confidence intervals (CIs) for significant predictors 19 . For time-to-event data, such as healing duration, Cox proportional hazards models were applied, yielding hazard ratios (HRs) that quantified the relative risks associated with each variable 22 . Both models were adjusted for confounders, ensuring the robustness of the results. To enhance predictive accuracy and explore complex variable interactions, machine learning models were implemented. Random Forest, an ensemble learning method, was employed to derive feature importance rankings and build robust classification models 23 . Additionally, the XGBoost algorithm, known for its computational efficiency and predictive power, was utilized to construct highly accurate models. Model performance was assessed using metrics such as the area under the receiver operating characteristic curve (AUC), sensitivity, specificity, and F1-score, providing a comprehensive evaluation of their predictive capabilities. To improve transparency and clinical applicability, interpretability tools were integrated into the machine learning framework. SHAP (SHapley Additive exPlanations) was used to assess the global importance of variables, elucidating their contributions to model predictions 24 . LIME (Local Interpretable Model-agnostic Explanations) complemented this by providing case-specific explanations, highlighting the factors influencing individual predictions and enabling clinicians to better understand patient-specific risks. Data preprocessing played a critical role in ensuring the quality and reliability of the analysis. Missing values were addressed through imputation or exclusion strategies, depending on the extent and nature of the missing data 25 . Outliers were identified and managed to prevent their influence on model performance 26 . Continuous variables were normalized to ensure comparability and improve model convergence, while categorical variables were encoded using one-hot encoding to facilitate their inclusion in machine learning models. The dataset was split into training and testing subsets in a 70:30 ratio, ensuring that model validation was performed on unseen data to assess their generalizability 27 . This comprehensive methodological framework provided a robust foundation for analyzing the multi factorial influences on wound healing in diabetic ICU patients. By integrating traditional statistical approaches with advanced machine learning techniques, the study delivered actionable insights to support personalized treatment strategies and optimized healthcare decision-making. Results Descriptive Analysis of the Dataset The descriptive analysis of the dataset serves as a foundational step to explore the baseline characteristics of diabetic patients and their relationship with the presence of pressure ulcers. The results, visualized through a series of density plots, histograms, and scatter plots, provide insights into potential risk factors and patterns, which inform further multivariate and predictive modeling. Age Distribution by Pressure Ulcer Status As shown in Fig. 1 , the age distribution highlights distinct patterns between patients with and without pressure ulcers.The age distribution, as shown by the density plot, highlights distinct patterns between patients with and without pressure ulcers. Patients diagnosed with pressure ulcers exhibit a slight skew toward older age groups compared to those without. The overlap between the two groups suggests that while middle-aged patients (with standardized age around 0) are prevalent in both populations, older individuals demonstrate a higher density in the pressure ulcer cohort. This observation is consistent with established evidence that aging contributes to diminished regenerative capacity, reduced collagen synthesis, and impaired angiogenesis, all of which increase susceptibility to pressure ulcers and hinder the wound healing process. Understanding the role of age is crucial, as it serves as a significant determinant of both the incidence and prognosis of pressure ulcers in critically ill diabetic patients. Weight Distribution by Pressure Ulcer Status As shown in Fig. 2 , the weight distribution further delineates the differences between patients with and without pressure ulcers. While there is substantial overlap, individuals without pressure ulcers display a more pronounced density around the average weight (standardized value near 0), whereas those with pressure ulcers exhibit a broader and flatter distribution. This suggests that deviations from normal weight, either underweight or overweight, may predispose patients to pressure ulcer development. For instance, underweight patients often present with reduced subcutaneous tissue, impairing pressure redistribution, while overweight patients may experience compromised mobility and increased mechanical stress. These findings underscore the multi-factorial nature of pressure ulcer risk, warranting further investigation into the interplay between weight and other physiological factors. Pressure Ulcer Healing Time Distribution The histogram of healing time for patients with pressure ulcers, as shown in Fig. 3 , reveals a heavily right-skewed distribution. The histogram of healing time for patients with pressure ulcers reveals a heavily right-skewed distribution, with most patients achieving healing within 10 days. However, a long tail extending beyond 100 days signifies a subset of individuals experiencing prolonged recovery. Such variability in healing duration reflects the influence of patient-specific factors, including comorbidities, infection status, and treatment modalities. This heterogeneity highlights the necessity for targeted interventions tailored to individual patient profiles. Furthermore, the presence of extreme outliers emphasizes the importance of identifying predictors associated with delayed healing to inform risk stratification and optimize clinical outcomes. Gender Distribution by Pressure Ulcer Status As shown in the Fig. 4 , he bar plot depicting gender distribution, as shown in Fig. 4 , indicates a disproportionate representation of male patients.The bar plot depicting gender distribution indicates a disproportionate representation of male patients in both the pressure ulcer and non-pressure ulcer groups. While females constitute a smaller fraction of the population, the gender distribution raises questions regarding potential biological, behavioral, or healthcare access disparities. Factors such as differences in skin structure, hormonal influences, and varying rates of ICU admission across genders may contribute to these observations. Future analyses could explore whether gender plays an independent role in pressure ulcer development and healing or if it acts as a con-founder mediated by other clinical variables. Observed and Expected Frequencies of Sepsis by Gender The observed and expected frequencies of sepsis, as shown in Table 1 , reveal minimal deviations between the two frequencies across genders.The side-by-side comparison of observed and expected frequencies of sepsis reveals that sepsis is relatively infrequent across the dataset, with minimal deviations between the two frequencies across genders. The side-by-side comparison of observed and expected frequencies of sepsis, as shown in Fig. 5 , confirms that sepsis is not a predominant feature.The expected frequencies for patients without pressure ulcers are approximately 65,430 (sepsis_flag = 0) and 1,018 (sepsis_flag = 1), while for those with pressure ulcers, the expected counts are 81,673 (sepsis_flag = 0) and 1,271 (sepsis_flag = 1). These results confirm that sepsis is not a predominant feature among the studied cohort but warrants consideration due to its role in exacerbating systemic inflammation and compromising immune function. The small differences between observed and expected values further suggest that gender is not a major determinant of sepsis occurrence. However, its potential impact on wound healing outcomes, particularly in conjunction with pressure ulcers, should be explored further. Table 1 Sepsis Frequencies by Pressure Ulcer Status Expected Frequencies: sepsis_flag 0 1 pressure_ulcer_flag 0 65429.876727 1018.123273 1 81673.123273 1270.876727 Weight vs. Age by Pressure Ulcer Flag The scatter plot of weight versus age, as shown in Fig. 6 , provides a granular perspective on the relationship between these variables.The scatter plot of weight versus age, stratified by pressure ulcer status, provides a granular perspective on the relationship between these variables. The distribution reveals no clear linear correlation between weight and age. However, the scatter suggests that pressure ulcers are more prevalent among patients with extreme values in either dimension. Such clustering patterns highlight the multi-factorial nature of pressure ulcer risk, where deviations in demographic and clinical characteristics collectively influence outcomes. This reinforces the need for integrative modeling approaches to capture the complex interactions between variables. Implications of Findings The descriptive analysis yielded several key insights that serve as a foundation for subsequent analytical stages: Age and Weight as Risk Factors: Both age and weight distributions indicate that deviations from normative ranges are associated with increased pressure ulcer prevalence, underscoring their role as critical demographic determinants.Heterogeneity in Healing Times: The wide variability in healing duration highlights the complexity of wound healing processes, emphasizing the need for individualized treatment strategies. Gender Disparities: The disproportionate representation of males in the dataset suggests potential biases or systemic differences that require further investigation. Sepsis and Compounding Risks: While sepsis frequencies are low, its potential role in exacerbating delayed healing and adverse outcomes underscores the importance of incorporating infection status in predictive models. Complex Variable Interactions: The scatter plot of weight and age emphasizes the need to consider multidimensional interactions when assessing pressure ulcer risk. These findings not only inform the selection of variables for multivariate modeling but also provide a basis for developing predictive frameworks aimed at early identification and intervention for high-risk patients. By integrating these descriptive insights into advanced statistical and machine learning models, the study aims to generate actionable knowledge to improve patient outcomes and optimize resource allocation in critical care settings. Univariate Analysis The results of the Cox proportional hazards regression analysis, as shown in Table 2 , highlight the significant interactions between clinical variables.The univariate analysis revealed significant associations between several comorbidities and wound healing outcomes in diabetic ICU patients. Among the variables analyzed, hypertension, chronic kidney disease (CKD), and the presence of pressure ulcers emerged as key determinants that significantly delayed wound healing. Hypertension, present in over 52% of the diabetic cohort, likely exacerbates vascular dysfunction by impairing endothelial function and reducing arterial elasticity. These effects hinder the delivery of oxygen and nutrients to wound sites, a critical requirement for fibroblast activity and collagen deposition. Furthermore, hypertension-induced microvascular damage disrupts angiogenesis, which is essential for forming new capillaries to support the healing process. This physiological interference prolongs the inflammatory phase and delays progression to wound granulation and re-epithelialization. Chronic kidney disease (CKD), which affected approximately 26% of the cohort, demonstrated a similarly significant impact on wound healing. CKD is known to cause systemic oxidative stress and chronic inflammation, which impair multiple cellular processes necessary for wound repair. For instance, the accumulation of uremic toxins disrupts fibroblast proliferation, delays keratinocyte migration, and weakens the mechanical strength of healed tissue. Additionally, CKD-associated metabolic disturbances, such as anemia and hypoalbuminemia, further compromise the wound healing microenvironment by reducing oxygen transport and depriving cells of essential nutrients. The presence of pressure ulcers, affecting 1.5% of patients, was another critical factor. These localized injuries result from prolonged pressure on the skin and underlying tissue, leading to ischemia and tissue necrosis. Pressure ulcers impair the normal healing cascade by creating hypoxic and necrotic environments, which inhibit angiogenesis and collagen remodeling. Additionally, the risk of secondary infections is heightened in pressure ulcer sites, further complicating the healing process. Taken together, these findings emphasize the multifactorial nature of delayed wound healing and the importance of addressing comorbidities as integral components of patient management strategies. Multivariate Analysis The multivariate analysis using Cox proportional hazards regression models provided deeper insights into the independent and interactive effects of clinical variables on wound healing outcomes. Among the variables examined, the presence of pressure ulcers emerged as the most significant predictor of delayed healing, with a hazard ratio (HR) of 2.36 (95% confidence interval: 2.08–2.67). This finding underscores the multifaceted challenges posed by pressure ulcers, including chronic inflammation, impaired angiogenesis, and heightened risk of secondary infections. The combination of localized tissue ischemia and systemic complications creates an environment that is highly resistant to conventional wound healing processes. The log(HR) and 95% CI plot for predictors, as shown in Fig. 7 , illustrates the impact of pressure ulcer flag and associated interactions. Interaction analyses revealed compelling relationships between pressure ulcers, comorbid conditions, and treatment modalities 12 . Notably, the interaction between negative pressure wound therapy (NPWT) and pressure ulcers demonstrated substantial efficacy in mitigating delayed healing outcomes, with an HR of 1.64 (95% CI: 1.38–1.94). This suggests that NPWT can alleviate some of the adverse effects of ischemia by promoting granulation tissue formation and removing exudate from the wound bed. These findings align with prior studies demonstrating the effectiveness of NPWT in managing complex wounds, particularly in diabetic populations. Conversely, the interaction between chronic kidney disease (CKD) and pressure ulcers significantly exacerbated delayed healing outcomes, with an HR of 2.14 (95% CI: 1.81–2.54). This interaction likely reflects the combined effects of CKD-related systemic inflammation and the localized ischemic environment created by pressure ulcers. CKD disrupts the production of growth factors such as vascular endothelial growth factor (VEGF), which are critical for angiogenesis and tissue repair. Additionally, CKD patients often experience anemia and hypoalbuminemia, further compounding the challenges of wound healing by impairing oxygen delivery and reducing the availability of essential proteins. The log(HR) with 95% CI for predictors incorporating pressure ulcer and interaction terms, as shown in Fig. 8 , provides further insights. Another notable interaction was observed between insulin therapy and pressure ulcers. The analysis revealed that insulin therapy’s benefits are partially attenuated in patients with pressure ulcers, with an HR of 1.64 (95% CI: 1.39–1.94). This may be attributed to the systemic inflammatory responses often associated with severe pressure ulcers, which can counteract the metabolic benefits of insulin by inducing insulin resistance at the cellular level. These findings suggest that while insulin therapy remains critical for glycemic control, additional interventions may be required to address the unique challenges posed by pressure ulcers in diabetic patients. The interaction between antibiotic treatment and pressure ulcers, while not statistically significant (HR = 0.98; 95% CI: 0.95–1.01), highlights the complexity of integrating therapeutic modalities into wound management strategies. While antibiotics are essential for infection control, their role in directly influencing wound healing may be limited without addressing other contributing factors such as ischemia and systemic inflammation. Overall, these findings highlight the importance of adopting a personalized and multifactorial approach to wound management in diabetic ICU patients. By accounting for the interplay between treatment modalities and comorbid conditions, clinicians can develop targeted interventions that address the specific challenges faced by high-risk patients. For example, combining NPWT with adjunct therapies such as growth factor supplementation or anti-inflammatory agents may offer synergistic benefits for patients with severe pressure ulcers. Similarly, optimizing nutritional support in CKD patients could help mitigate the systemic factors that impede wound healing. This evidence underscores the need for integrative care models that leverage both clinical and technological innovations to improve outcomes for diabetic ICU patients with complex wound profiles. Table 2 Clinical Variable Interactions and Their Impact on Wound Healing Outcomes in Diabetic ICU Patients: A Cox Proportional Hazards Regression Analysis coef exp(coef) se(coef) coef lower 95% coef upper 95% exp(coef) lower 95% exp(coef) upper 95% cmp to z p -log2(p) age 0.00 1.00 0.01 -0.01 0.02 0.99 1.02 0.00 0.25 0.81 0.31 gender 0.00 1.00 0.02 -0.03 0.03 0.97 1.03 0.00 0.09 0.93 0.11 weight 0.00 1.00 0.01 -0.01 0.02 0.99 1.02 0.00 0.21 0.83 0.26 hypertension_flag 0.01 1.01 0.02 -0.03 0.04 0.98 1.04 0.00 0.41 0.68 0.55 cardiovascular_flag 0.02 1.02 0.03 -0.04 0.07 0.96 1.07 0.00 0.53 0.59 0.75 chronic_kidney_disease_flag 0.01 1.01 0.02 -0.02 0.05 0.98 1.05 0.00 0.67 0.50 1.00 type1_diabetes_flag 0.01 1.01 0.05 -0.09 0.11 0.92 1.11 0.00 0.20 0.84 0.25 type2_diabetes_flag 0.03 1.03 0.03 -0.04 0.09 0.96 1.10 0.00 0.75 0.45 1.15 pressure_ulcer_flag 0.86 2.36 0.06 0.73 0.98 2.08 2.67 0.00 13.63 < 0.005 138.16 antibiotic_treatment_flag -0.02 0.98 0.02 -0.05 0.01 0.95 1.01 0.00 -1.23 0.22 2.19 negative_pressure_flag -0.02 0.98 0.02 -0.05 0.01 0.95 1.01 0.00 -1.23 0.22 2.19 insulin_flag -0.02 0.98 0.02 -0.05 0.01 0.95 1.01 0.00 -1.22 0.22 2.18 pressure_ulcer_treatment_interaction 0.49 1.63 0.09 0.32 0.66 1.38 1.93 0.00 5.75 < 0.005 26.75 pressure_ulcer_negative_pressure_interaction 0.49 1.64 0.09 0.33 0.66 1.38 1.94 0.00 5.76 < 0.005 26.84 pressure_ulcer_insulin_interaction 0.50 1.64 0.09 0.33 0.66 1.39 1.94 0.00 5.79 < 0.005 27.12 pressure_ulcer_ckd_interaction 0.76 2.14 0.09 0.59 0.93 1.81 2.54 0.00 8.82 < 0.005 59.62 Logistic Regression Analysis The ROC curve depicting the performance of a binary classification model, as shown in Fig. 9 , indicates an AUC of 0.76. The logistic regression model served as a foundational method for evaluating the relationships between clinical variables and wound healing outcomes. Despite its simplicity and interpretability, the model demonstrated moderate discriminatory power with an Area Under the Receiver Operating Characteristic Curve (AUC) of 0.7153. The model's classification performance, as reflected in the confusion matrix and classification report, revealed significant challenges in handling the imbalanced nature of the dataset. While the model achieved near-perfect precision and recall for the majority class (healed wounds), it failed to identify any cases of the minority class (non-healed wounds). Specifically, the classification report showed an F1-score of 0.98 for the healed class but 0.00 for the non-healed class. The confusion matrix indicated 458 false negatives and no true positives for the minority class, underscoring the model’s inability to effectively address imbalanced data distributions. These results highlight the limitations of logistic regression in capturing complex, non-linear relationships and its sensitivity to skewed class distributions. Feature importance analysis, derived from logistic regression coefficients, identified infection status, sepsis, pneumonia, and chronic kidney disease (CKD) as the most influential predictors of wound healing outcomes. The strong association between infection-related variables and delayed healing aligns with established clinical findings, emphasizing the need to prioritize infection management in wound care. However, the linear nature of logistic regression limited its ability to account for interaction effects among these variables, further constraining its predictive accuracy. Random Forest Results The ROC curve demonstrating the classification performance of the Random Forest algorithm, as shown in Fig. 10 , indicates an AUC of 0.99.The Random Forest model achieved outstanding performance, far surpassing the logistic regression model. Its ability to handle non-linear relationships and interactions among variables contributed to this exceptional result. The classification report indicated balanced precision, recall, and F1-scores of 0.96 for both healed and non-healed classes, demonstrating the model’s robustness across the dataset. The confusion matrix revealed 1,097 false positives and 1,172 false negatives, representing a substantial reduction in misclassification errors compared to logistic regression. The plot illustrating the association between antibiotic treatment status and infection status, as shown in Fig. 11 , provides insights into the impact of therapeutic interventions. Random Forest’s feature importance analysis corroborated the findings from logistic regression, highlighting infection status, sepsis, CKD, and pressure ulcers as critical predictors. Moreover, the model identified additional important variables, such as antibiotic treatment and negative pressure wound therapy (NPWT), underscoring the multifactorial nature of wound healing. SHAP (SHapley Additive exPlanations) analysis further enhanced the interpretability of the Random Forest model. SHAP values demonstrated that infection-related variables had the most substantial negative impact on healing outcomes, while therapeutic interventions like NPWT exerted consistently positive effects. For example, in one case, the absence of NPWT combined with advanced CKD and infection was identified as the primary contributor to delayed healing, providing actionable insights for clinical decision-making. XGBoost and MLP Results The model performance comparison in terms of AUC, as shown in Fig. 12 , highlights the superior performance of Random Forest and XGBoost models.The XGBoost model also performed exceptionally well, achieving an AUC of 0.96. Its gradient-boosting approach optimized predictive performance by iteratively minimizing errors, making it particularly effective for imbalanced datasets. The model’s classification metrics were comparable to Random Forest, with high precision and recall across both classes. Feature importance analysis from XGBoost aligned closely with Random Forest, further validating the critical role of infection-related variables and therapeutic interventions. The MLP model, a neural network-based approach, achieved an AUC of 0.92. While its performance was slightly lower than Random Forest and XGBoost, it still outperformed logistic regression by a considerable margin. However, the lack of straightforward interpretability in MLP models posed challenges for clinical application, as insights into variable importance and decision-making processes were less accessible. Comparative Analysis The comparative analysis of logistic regression and machine learning models underscores the superior performance of ensemble methods like Random Forest and XGBoost in predicting wound healing outcomes. While logistic regression offered interpretability and simplicity, its reliance on linear assumptions and sensitivity to class imbalance limited its utility in this context. In contrast, Random Forest and XGBoost excelled in capturing non-linear relationships and addressing class imbalances, achieving near-perfect AUC values. The interpretability provided by SHAP analysis in Random Forest and XGBoost further enhanced their clinical applicability, enabling data-driven decision-making and personalized treatment strategies. The machine learning models, particularly Random Forest, demonstrated the potential to integrate into clinical workflows for early risk identification and optimized resource allocation. Future research should focus on incorporating additional data sources, such as genomic and proteomic profiles, to further refine these models and expand their applicability across diverse patient populations. These findings highlight the transformative potential of machine learning in advancing predictive analytics and improving outcomes in diabetic ICU patients with complex wound profiles. Model Interpretability and Analysis The SHAP value plot for feature importance and their impact on model output, as shown in Fig. 13 , highlights the critical predictors of wound healing outcomes.To further elucidate the contributions of individual features to wound healing outcomes, model interpretability tools, specifically SHAP (SHapley Additive exPlanations) and LIME (Local Interpretable Model-Agnostic Explanations), were employed. These tools provided both global and local insights into the predictive dynamics of the machine learning models, enhancing their applicability in clinical contexts and ensuring that predictions were transparent and actionable. The SHAP summary plot highlighted the relative importance of various predictors, revealing a diverse array of factors influencing wound healing. The feature with the most significant impact was the death flag, indicating that mortality risk strongly influences healing probabilities. This finding is consistent with the clinical understanding that systemic complications associated with high mortality risk, such as multi-organ failure and sepsis, are major impediments to effective wound healing. Variables such as negative pressure flag, chronic kidney disease (CKD) flag, and age also demonstrated substantial contributions to the model's predictions, consistent with their well-documented roles in wound pathophysiology. Negative SHAP values for these features indicated a detrimental effect on the likelihood of wound healing when their values were high. For instance, advanced age and CKD were associated with delayed healing due to their contributions to systemic inflammation, vascular dysfunction, and compromised cellular repair mechanisms. The SHAP analysis further underscored the nuanced effects of therapeutic interventions. The negative pressure wound therapy (NPWT) flag had a protective SHAP impact, reaffirming its utility in managing complex wounds by enhancing granulation tissue formation, reducing local ischemia, and facilitating exudate removal. This finding aligns with previous clinical studies that position NPWT as a cornerstone therapy for pressure ulcers and other refractory wounds. Furthermore, infection-related variables, including sepsis flag and pneumonia flag, were strongly associated with negative SHAP values, reflecting the adverse impact of systemic infections on wound recovery. These insights emphasize the critical importance of rigorous infection control and the early application of targeted therapies to mitigate these effects. Notably, antibiotic treatment, while essential for managing infections, did not emerge as a significant predictor of wound healing outcomes, suggesting that its efficacy may be contingent on addressing other co-existing factors such as ischemia and systemic inflammation. The prediction probabilities and feature-based decision breakdown for a binary outcome, as shown in Fig. 14 , provide actionable insights for clinical decision-making.At the individual level, LIME provided interpretable case-specific explanations that complemented the global insights from SHAP. For instance, in a high-risk patient, the death flag, combined with hypertension flag and the absence of NPWT, emerged as the primary contributors to the model’s prediction of delayed healing. This case demonstrated how systemic factors such as hypertension and mortality risk interact with the absence of therapeutic interventions to exacerbate healing delays. Conversely, in a patient with favorable outcomes, the presence of NPWT and the absence of chronic systemic conditions were identified as critical drivers of successful healing. Such granular analyses offer actionable insights, enabling clinicians to prioritize interventions for the most critical variables affecting individual patients. The feature correlation coefficients analysis, as shown in Table 3 , provides insights into the relationships between various predictors.The integration of SHAP and LIME analyses into this study provided compelling evidence for the complex interplay of demographic, clinical, and therapeutic factors in wound healing outcomes among diabetic ICU patients. For example, the consistent identification of negative pressure flag, CKD flag, and infection-related variables as key predictors underscores the importance of addressing both local wound conditions and systemic health issues to optimize healing trajectories. Moreover, the findings advocate for the routine implementation of NPWT in high-risk patients, particularly those with pressure ulcers, as well as the importance of mitigating systemic complications such as infections and CKD.From a broader perspective, SHAP and LIME also highlight the importance of integrating interpretability into machine learning models to bridge the gap between predictive analytics and clinical application. The ability to provide both global and patient-specific insights ensures that predictions are not only accurate but also clinically meaningful. For instance, SHAP's global summary plots enable clinicians to understand the overarching factors driving wound healing outcomes across a population, while LIME's local explanations facilitate personalized care by pinpointing the specific variables most relevant to individual patients. This dual approach enhances the utility of machine learning in clinical decision-making and fosters trust among healthcare providers by making the predictive processes transparent. Table 3 Feature Correlation Coefficients Analysis Feature Value Feature Value age 1.33 death_flag 1.00 type1_diabetes_flag 0.00 cardiovascular_flag 0.00 type2_diabetes_flag 0.00 hypertension_flag 1.00 negative_pressure_flag 0.00 sepsis_flag 0.00 chronic_kidney_disease_flag 0.00 weight -0.32 In conclusion, the comprehensive analysis of machine learning models, augmented by SHAP and LIME interpretability tools, highlights their transformative potential in advancing wound care for diabetic ICU patients. By integrating predictive analytics with interpretable insights, clinicians can develop targeted, evidence-based strategies to improve outcomes and reduce the burden of chronic wounds. These findings emphasize the need for a multifactorial approach that addresses both systemic and local factors affecting wound healing. Future research should focus on expanding these models by incorporating additional data sources, such as biomarkers, genomic profiles, and real-time physiological monitoring, to further refine predictive accuracy and enhance personalized care. Additionally, exploring the integration of these models into electronic health record systems can facilitate real-time risk assessment, optimize resource allocation, and ultimately improve patient outcomes on a broader scale. Discussion This study provides a comprehensive analysis of wound healing outcomes among diabetic ICU patients by integrating advanced statistical and machine learning approaches with model interpretability tools. The findings underscore the multifactorial nature of wound healing, influenced by a combination of demographic, clinical, and therapeutic factors. Among the key variables identified, the presence of pressure ulcers, chronic kidney disease, and systemic infections emerged as significant predictors of delayed healing, while therapeutic interventions like negative pressure wound therapy (NPWT) demonstrated a protective effect. The interplay between systemic comorbidities and localized wound conditions highlights the need for a multifaceted approach to wound management. The use of machine learning models, particularly Random Forest and XGBoost, significantly enhanced predictive accuracy compared to traditional logistic regression. These models effectively addressed class imbalances and captured complex, non-linear relationships among variables. The integration of SHAP and LIME provided both global and patient-specific insights, enhancing the interpretability and clinical applicability of the predictive models. SHAP analysis revealed the critical importance of infection-related variables and NPWT, while LIME offered actionable, individualized explanations that could inform targeted interventions. Clinically, these findings emphasize the importance of personalized care strategies that account for both systemic and local factors. For instance, the routine implementation of NPWT in high-risk patients and the prioritization of infection control can mitigate some of the most significant barriers to wound healing. Furthermore, the incorporation of predictive models into electronic health record systems has the potential to facilitate real-time risk assessment and optimize resource allocation, improving outcomes for diabetic ICU patients with complex wound profiles. Future research should focus on expanding these models by incorporating additional data sources, such as biomarkers, genomic profiles, and longitudinal physiological monitoring, to refine predictive accuracy further. Additionally, exploring the integration of these models into broader healthcare systems can enhance their scalability and impact. Ultimately, this study highlights the transformative potential of machine learning and interpretability tools in advancing wound care, offering a pathway toward more precise, effective, and patient-centered management strategies for chronic wounds. Conclusion This study demonstrates the transformative potential of integrating advanced machine learning and interpretability tools in predicting wound healing outcomes among diabetic ICU patients. Key findings highlight pressure ulcers, chronic kidney disease, and systemic infections as significant predictors of delayed healing, while negative pressure wound therapy (NPWT) shows a protective effect. Machine learning models (e.g., Random Forest and XGBoost) outperformed traditional logistic regression by addressing class imbalances and capturing complex, non-linear relationships. Interpretability tools like SHAP and LIME provided actionable insights, emphasizing infection-related variables and NPWT, while offering patient-specific explanations for targeted interventions. Clinically, these results underscore the importance of personalized care strategies that address both systemic and local factors, such as prioritizing NPWT and infection control. Integrating predictive models into electronic health records could enhance real-time risk assessment and optimize resource allocation. Future research should expand models by incorporating biomarkers, genomic data, and longitudinal monitoring to refine accuracy. Additionally, exploring the scalability of these models within healthcare systems could maximize their impact. Overall, this study advances wound care by leveraging machine learning to develop precise, patient-centered strategies for managing chronic wounds in high-risk populations. Abbreviations ICU Intensive Care Unit MIMIC-IV Medical Information Mart for Intensive Care IV NPWT Negative Pressure Wound Therapy CKD Chronic Kidney Disease AUC Area Under the Receiver Operating Characteristic Curve Declarations Ethics approval and consent to participate: not applicable. Consent for publication: Not applicable Competing Interests: The authors declare that they have no competing interests. Funding: This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors. Author Contribution L.Z., X.C., Y.T., K.L., M.Z. and H.T. conceived and designed the experiments; L.Z. and X.C. performed the experiments; Y.T. and K.L. analyzed the data; M.Z. and H.T. contributed reagents/materials/analysis tools; L.Z. wrote the paper; Y..T and K.L. reviewed and edited the manuscript. Acknowledgements: Not applicable Data Availability The data used in this study are derived from the Medical Information Mart for Intensive Care IV (MIMIC-IV) database, which is a publicly accessible dataset. The MIMIC-IV database is a comprehensive clinical dataset that includes detailed records of patients admitted to ICUs at the Beth Israel Deaconess Medical Center. As the data are publicly available and de-identified, no additional data sharing is required beyond the use of this publicly accessible resource. References Saeedi P. Global and regional diabetes prevalence estimates for 2019 and projections for 2030 and 2045: Results from the International Diabetes Federation Diabetes Atlas, 9th edition. Diabetes Research and Clinical Practice 2019; 151: 143. Mustoe TA. Impaired wound healing in diabetes: Current concepts and therapeutic strategies. Wound Repair Regeneration. 2006;14(6):653. Sen CK. Human diabetic wounds exhibit a distinct inflammatory gene signature. Wound Repair Regeneration 2009; 17(1). Ayello EA. Pressure ulcers in the intensive care unit: A systematic review. Crit Care Med 2008; 36(4). Kirsner RS. Wound healing in the diabetic foot: A review. Diabetes Care 2008; 31(5). Margolis DJ. Risk factors for nonhealing of diabetic neuropathic ulcers. Wound Repair Regeneration 2007; 15(1). Armstrong DG. Diabetic foot wounds: What we know and where we are going. Diabetes Care 2005; 28(1). Lipsky BA. Diabetic foot infections: A global view. Diabetes Care 2012; 35(Suppl 1). Edsberg LE. Update on the incidence and prevalence of pressure ulcers in the United States. Wound Repair Regeneration 2011; 19(1). Berendt AR. Diabetic foot ulcers and their treatment. JAMA 2006; 295(21). Armstrong DG. Offloading the diabetic foot wound: A systematic review. Diabetes Care 2011; 34(8). Kastenbauer ER. Negative pressure wound therapy: A review. Crit Care Med 2007; 35(8). Teitelbaum RE. Negative pressure wound therapy: A review of the current evidence. 2009; 30(4). Scherer JW. Negative pressure wound therapy: A review. Wound Repair Regeneration 2010; 18(1). Kirsner RS. The role of infection in chronic wounds. Wound Repair Regeneration 2010; 18(1). Edsberg LE. Pressure ulcers in the intensive care unit: A review. Crit Care Med 2006; 34(4). Berendt AR. The pathophysiology of the diabetic foot. Diabetes Care 2008; 31(5). Lipsky BA. Infection in diabetic foot ulcers. Diabetes Care 2006; 29(5). Armstrong DG. The role of offloading in the treatment of diabetic foot ulcers. Diabetes Care 2008; 31(5). Armstrong DG. The role of offloading in the treatment of diabetic foot ulcers. Diabetes Care 2010; 33(5). Teitelbaum RE. Negative pressure wound therapy: A review of the current evidence. J Burn Care Res 2010; 31(4). Kastenbauer ER. Negative pressure wound therapy: A review of the current evidence. Crit Care Med 2008; 36(4). Scherer JW. Negative pressure wound therapy: A review. Wound Repair Regeneration 2011; 19(1). Kirsner RS. The role of infection in chronic wounds. Wound Repair Regeneration 2011; 19(1). Edsberg LE. Pressure ulcers in the intensive care unit: A review. Crit Care Med 2010; 38(4). Berendt AR. The pathophysiology of the diabetic foot. Diabetes Care 2010; 33(5). Lipsky BA. Infection in diabetic foot ulcers. Diabetes Care 2010; 33(5). Additional Declarations No competing interests reported. Cite Share Download PDF Status: Under Revision Version 1 posted Editorial decision: Revision requested 06 Nov, 2025 Reviews received at journal 06 Jun, 2025 Reviews received at journal 06 Jun, 2025 Reviews received at journal 27 May, 2025 Reviewers agreed at journal 27 May, 2025 Reviewers agreed at journal 27 May, 2025 Reviewers agreed at journal 23 May, 2025 Reviewers agreed at journal 22 May, 2025 Reviewers invited by journal 14 May, 2025 Editor assigned by journal 13 May, 2025 Editor invited by journal 22 Apr, 2025 Submission checks completed at journal 21 Apr, 2025 First submitted to journal 21 Apr, 2025 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-6307348","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":458129984,"identity":"d7420e55-d426-4a63-b13d-bc6a8a97ccde","order_by":0,"name":"Linmin Zhuge","email":"","orcid":"","institution":"Department of Gastrointestinal Surgery, Second Affiliated Hospital \u0026 Yuying Children's Hospital of Wenzhou Medical University","correspondingAuthor":false,"prefix":"","firstName":"Linmin","middleName":"","lastName":"Zhuge","suffix":""},{"id":458129985,"identity":"2028b04a-35d5-4a71-b141-3e8924edecde","order_by":1,"name":"Xinxin Chen","email":"","orcid":"","institution":"Department of Gastrointestinal Surgery, Second Affiliated Hospital \u0026 Yuying Children's Hospital of Wenzhou Medical University","correspondingAuthor":false,"prefix":"","firstName":"Xinxin","middleName":"","lastName":"Chen","suffix":""},{"id":458129989,"identity":"8727a624-c34b-4347-8c89-2086881b02d7","order_by":2,"name":"Yiwei Teng","email":"","orcid":"","institution":"Renji College, Wenzhou Medical University","correspondingAuthor":false,"prefix":"","firstName":"Yiwei","middleName":"","lastName":"Teng","suffix":""},{"id":458129991,"identity":"132671d8-f3b1-4f77-828d-1e252c01dedf","order_by":3,"name":"Kan Li","email":"","orcid":"","institution":"Department of Orthopedics, the First Affiliated Hospital of Wenzhou University","correspondingAuthor":false,"prefix":"","firstName":"Kan","middleName":"","lastName":"Li","suffix":""},{"id":458129992,"identity":"d68d25cb-b1a3-40c9-a730-24cf7bace01f","order_by":4,"name":"Minyu Zhu","email":"","orcid":"","institution":"Department of Orthopedics, the First Affiliated Hospital of Wenzhou University","correspondingAuthor":false,"prefix":"","firstName":"Minyu","middleName":"","lastName":"Zhu","suffix":""},{"id":458129993,"identity":"f34e2827-f80c-4644-abcc-f22f06153ff5","order_by":5,"name":"Honglin Teng","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAABGElEQVRIiWNgGAWjYJACZhDBx8zA+ICB4QCQydjAwFCAWzkPTAsbMwOzAVRLYwODATFagEgCogVkDR4t9uy9h18XVNyxa2PnPVbN8+eOnPzs5vYHPwwY5M1x2cJzLs16xplnyW3MfGm3eXieGRvcOdjY2GPAYLizAYcWiRwzY962w8lszDxmt3MkDidukEhsbOAxYEgwOIBDi/wbhJbiHIPDifNnJDY2/sGnRYLH+DFQix1IC3NOwuHEhhuJjc14bTmTY8bMc+ZwAlCLsfSfA4fBfpktYyBhuAGHFvb2M8afeSoO2/PznzH8OOPPYWCItT/4+KbCRh6XLQyQ6GBIbIDzJZBIHID5A5CwR/DxKh4Fo2AUjIKRCADLdljcSqnv9wAAAABJRU5ErkJggg==","orcid":"","institution":"Department of Orthopedics, the First Affiliated Hospital of Wenzhou University","correspondingAuthor":true,"prefix":"","firstName":"Honglin","middleName":"","lastName":"Teng","suffix":""}],"badges":[],"createdAt":"2025-03-26 00:53:15","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-6307348/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-6307348/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":83128939,"identity":"683e26f5-04d0-4259-8103-7269629e6bf9","added_by":"auto","created_at":"2025-05-20 09:59:36","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":47248,"visible":true,"origin":"","legend":"\u003cp\u003eDensity Plot of Age Distribution by Pressure Ulcer Status\u003c/p\u003e","description":"","filename":"image1.png","url":"https://assets-eu.researchsquare.com/files/rs-6307348/v1/ec1190cc2c76dfb697de026a.png"},{"id":83127127,"identity":"20d493c0-6824-458d-bb45-f805f2590599","added_by":"auto","created_at":"2025-05-20 09:51:36","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":45854,"visible":true,"origin":"","legend":"\u003cp\u003eDensity Plot of Weight Distribution by Pressure Ulcer Status\u003c/p\u003e","description":"","filename":"image2.png","url":"https://assets-eu.researchsquare.com/files/rs-6307348/v1/61c0f73596ce2b72cba6c51c.png"},{"id":83128938,"identity":"32649180-effe-43e3-b89b-314c074e6c4e","added_by":"auto","created_at":"2025-05-20 09:59:36","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":36503,"visible":true,"origin":"","legend":"\u003cp\u003eDensity Plot of Weight Distribution by Pressure Ulcer Status\u003c/p\u003e","description":"","filename":"image3.png","url":"https://assets-eu.researchsquare.com/files/rs-6307348/v1/3e33f202b5694756166373cc.png"},{"id":83126189,"identity":"6e0312f9-c10c-45ec-8395-e24826fda8be","added_by":"auto","created_at":"2025-05-20 09:43:38","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":23065,"visible":true,"origin":"","legend":"\u003cp\u003eGender Distribution Stratified by Pressure Ulcer Status\u003c/p\u003e","description":"","filename":"image4.png","url":"https://assets-eu.researchsquare.com/files/rs-6307348/v1/05dcc15ea5b44a85806b2728.png"},{"id":83127129,"identity":"7638b940-3805-434f-bf1f-ef1771bbb01c","added_by":"auto","created_at":"2025-05-20 09:51:36","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":42515,"visible":true,"origin":"","legend":"\u003cp\u003eObserved and Expected Frequencies of Gender with Respect to Sepsis Flag\u003c/p\u003e","description":"","filename":"image5.png","url":"https://assets-eu.researchsquare.com/files/rs-6307348/v1/c0d3839d571da47b70898915.png"},{"id":83126155,"identity":"320c1fcb-db65-4a20-b58d-c31bbf114d74","added_by":"auto","created_at":"2025-05-20 09:43:36","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":221336,"visible":true,"origin":"","legend":"\u003cp\u003eBivariate Distribution of Weight and Age According to Pressure Ulcer\u003c/p\u003e","description":"","filename":"image6.png","url":"https://assets-eu.researchsquare.com/files/rs-6307348/v1/9a1dbf719bc2195a57dff05f.png"},{"id":83127133,"identity":"2661261a-dc24-4fa4-9639-5c490807927d","added_by":"auto","created_at":"2025-05-20 09:51:36","extension":"png","order_by":7,"title":"Figure 7","display":"","copyAsset":false,"role":"figure","size":39980,"visible":true,"origin":"","legend":"\u003cp\u003eLog(HR) and 95% CI Plot for Predictors including Pressure Ulcer Flag and Associated Interactions\u003c/p\u003e","description":"","filename":"image7.png","url":"https://assets-eu.researchsquare.com/files/rs-6307348/v1/26b91311bd03883347b4c3ab.png"},{"id":83126153,"identity":"7aaaea6e-f433-4e18-aaeb-6c8e2b1adf53","added_by":"auto","created_at":"2025-05-20 09:43:36","extension":"png","order_by":8,"title":"Figure 8","display":"","copyAsset":false,"role":"figure","size":49216,"visible":true,"origin":"","legend":"\u003cp\u003eLog(HR) with 95% CI for Predictors Incorporating Pressure Ulcer and Interaction Terms\u003c/p\u003e","description":"","filename":"image8.png","url":"https://assets-eu.researchsquare.com/files/rs-6307348/v1/d9db62f00e277a4b4c6a39bd.png"},{"id":83126161,"identity":"a4cd8d13-2627-44f5-805c-e9ee6146e949","added_by":"auto","created_at":"2025-05-20 09:43:36","extension":"png","order_by":9,"title":"Figure 9","display":"","copyAsset":false,"role":"figure","size":29295,"visible":true,"origin":"","legend":"\u003cp\u003eROC Curve Depicting the Performance of a Binary Classification Model (AUC = 0.76)\u003c/p\u003e","description":"","filename":"image9.png","url":"https://assets-eu.researchsquare.com/files/rs-6307348/v1/edb205ac535474a797db718d.png"},{"id":83127131,"identity":"f6c4f973-449b-40b1-bda0-aa8d8228301b","added_by":"auto","created_at":"2025-05-20 09:51:36","extension":"png","order_by":10,"title":"Figure 10","display":"","copyAsset":false,"role":"figure","size":32152,"visible":true,"origin":"","legend":"\u003cp\u003eROC Curve Demonstrating the Classification Performance of the Random Forest Algorithm (AUC = 0.99)\u003c/p\u003e","description":"","filename":"image10.png","url":"https://assets-eu.researchsquare.com/files/rs-6307348/v1/ea6c458580e7ba53014a4f56.png"},{"id":83127138,"identity":"472d8d36-54de-4ff8-819b-72e961b14545","added_by":"auto","created_at":"2025-05-20 09:51:36","extension":"png","order_by":11,"title":"Figure 11","display":"","copyAsset":false,"role":"figure","size":33412,"visible":true,"origin":"","legend":"\u003cp\u003ePlot Illustrating the Association between Antibiotic Treatment Status and Infection Status\u003c/p\u003e","description":"","filename":"image11.png","url":"https://assets-eu.researchsquare.com/files/rs-6307348/v1/c65d877ccffaf49fa3f5b4bf.png"},{"id":83128941,"identity":"38d0e1e7-83d7-480e-a8bd-fa74ffdaa551","added_by":"auto","created_at":"2025-05-20 09:59:36","extension":"png","order_by":12,"title":"Figure 12","display":"","copyAsset":false,"role":"figure","size":30508,"visible":true,"origin":"","legend":"\u003cp\u003eModel Performance Comparison in Terms of AUC: Logistic Regression, Random Forest, XGBoost, and MLP\u003c/p\u003e","description":"","filename":"image12.png","url":"https://assets-eu.researchsquare.com/files/rs-6307348/v1/f2722e89b4248d35814ea49f.png"},{"id":83128942,"identity":"e8d5d6b6-95ff-4761-a34d-08eb2beea4a2","added_by":"auto","created_at":"2025-05-20 09:59:36","extension":"png","order_by":13,"title":"Figure 13","display":"","copyAsset":false,"role":"figure","size":83102,"visible":true,"origin":"","legend":"\u003cp\u003eSHAP Value Plot for Feature Importance and Their Impact on Model Output across Various Health - related Features\u003c/p\u003e","description":"","filename":"image13.png","url":"https://assets-eu.researchsquare.com/files/rs-6307348/v1/ff27efaca5eb704bfe261934.png"},{"id":83126162,"identity":"057432f7-9f1f-4ae7-ab50-9ef17289264a","added_by":"auto","created_at":"2025-05-20 09:43:36","extension":"png","order_by":14,"title":"Figure 14","display":"","copyAsset":false,"role":"figure","size":16160,"visible":true,"origin":"","legend":"\u003cp\u003ePrediction Probabilities and Feature - based Decision Breakdown for a Binary Outcome\u003c/p\u003e","description":"","filename":"image14.png","url":"https://assets-eu.researchsquare.com/files/rs-6307348/v1/287a70ed8c4da1043b7d26ad.png"},{"id":83129386,"identity":"5406be1f-03e7-4ff3-a106-5c437ae6051b","added_by":"auto","created_at":"2025-05-20 10:07:36","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1579477,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-6307348/v1/fd32755f-fe17-48c4-a699-f4c4855cdd44.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"Multifactorial Analysis and Predictive Modeling of Wound Healing Outcomes in Diabetic ICU Patients: a Cohort Study Based on MIMIC-IV","fulltext":[{"header":"Introduction","content":"\u003cp\u003eDiabetes mellitus, a chronic metabolic disorder, affects approximately 10.5% of the global adult population, with its prevalence projected to rise to 11.3% by 2030 and 12.2% by 2045 \u003csup\u003e1\u003c/sup\u003e. This dramatic increase underscores the urgent need to address diabetes-related complications, particularly those that significantly impact quality of life and healthcare resources. Among these complications, impaired wound healing poses a critical challenge, especially in patients admitted to intensive care units (ICUs) where pressure ulcer incidence reaches 18.5% \u003csup\u003e2\u003c/sup\u003e. Factors such as hyperglycemia-induced angiogenesis suppression \u003csup\u003e\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e\u003c/sup\u003e and neuropathy-mediated tissue damage \u003csup\u003e\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e\u003c/sup\u003e exacerbate wound healing difficulties in diabetic individuals \u003csup\u003e\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e \u003cp\u003eIn ICU settings, these challenges are magnified by prolonged immobility (associated with 32% increased pressure injury risk \u003csup\u003e\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e\u003c/sup\u003e), systemic infections requiring antibiotic therapy \u003csup\u003e\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e\u003c/sup\u003e, and metabolic instability \u003csup\u003e\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e\u003c/sup\u003e. Delayed wound healing in these patients prolongs hospitalization by 7.2 days on average \u003csup\u003e\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e\u003c/sup\u003e, elevating risks of infections (OR\u0026thinsp;=\u0026thinsp;3.1) \u003csup\u003e\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e\u003c/sup\u003e, amputations (HR\u0026thinsp;=\u0026thinsp;2.8) \u003csup\u003e\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e\u003c/sup\u003e, and mortality (RR\u0026thinsp;=\u0026thinsp;1.9) \u003csup\u003e\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e\u003c/sup\u003e. These outcomes highlight the need for deeper understanding of healing determinants in diabetic ICU patients.\u003c/p\u003e \u003cp\u003eDespite advancements in diabetes management, current wound care strategies show limited efficacy in critical care settings. The 2012 International Working Group on Diabetic Foot (IWGDF) guidelines emphasize glycemic control \u003csup\u003e\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e\u003c/sup\u003e, yet fail to address ICU-specific factors like ventilator-associated tissue hypoxia \u003csup\u003e\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e\u003c/sup\u003e. Prior studies predominantly focus on isolated variables such as HbA1c levels \u003csup\u003e\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e\u003c/sup\u003e or infection biomarkers \u003csup\u003e\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e\u003c/sup\u003e, neglecting the multifactorial interactions between inflammatory mediators \u003csup\u003e\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e\u003c/sup\u003e, treatment modalities \u003csup\u003e\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e\u003c/sup\u003e, and comorbidities \u003csup\u003e\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e \u003cp\u003eThis study aims to address these gaps by conducting a multifactorial analysis of diabetic ICU patients using data from the MIMIC-IV database. By leveraging both traditional statistical methods and machine learning algorithms, we seek to:Identify the key determinants of wound healing outcomes;Evaluate the impact of various treatment modalities on healing processes;Develop predictive models with high accuracy and clinical applicability.\u003c/p\u003e \u003cp\u003eThe integration of multi-factorial data and machine learning techniques provides an opportunity to advance personalized medicine. Predictive models can facilitate early risk identification, enabling clinicians to implement tailored interventions for high-risk patients. Additionally, these insights can inform healthcare policy by optimizing resource allocation and improving the quality of care delivered to diabetic ICU patients. By addressing the existing research gaps and employing advanced methodologies, this study contributes significantly to the field of diabetes management and critical care.\u003c/p\u003e"},{"header":"Data and Methodology","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003eData Selection\u003c/h2\u003e \u003cp\u003eThis study employed a systematic approach to investigate factors influencing wound healing in diabetic patients admitted to intensive care units (ICUs). The analysis was supported by a comprehensive dataset and advanced methodologies to ensure robust and clinically relevant results.\u003c/p\u003e \u003cp\u003eThe data utilized for this study were derived from the Medical Information Mart for Intensive Care IV (MIMIC-IV) database, a publicly accessible and comprehensive clinical dataset. This database encompasses detailed records of patients admitted to ICUs at the Beth Israel Deaconess Medical Center between 2008 and 2019. Spanning over 380,000 admissions, it includes granular information such as demographic characteristics, laboratory test results, medication usage, procedures performed, and clinical outcomes. The de-identified nature of the data ensured compliance with ethical standards, with institutional review board approval secured for this research.\u003c/p\u003e \u003cp\u003eThe study cohort comprised diabetic patients aged 18 years or older who had sufficient data to evaluate wound healing outcomes. From the dataset, 149,392 eligible patients were identified, including 2,289 individuals diagnosed with pressure ulcers. To maintain the rigor and validity of the analysis, patients with missing critical data (e.g., healing outcomes, comorbidities, or treatment details) or inconsistencies in records (e.g., implausible values for age or healing duration) were excluded.\u003c/p\u003e \u003cp\u003eThe selection of variables for analysis was guided by clinical relevance and prior literature \u003csup\u003e\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e,\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e\u003c/sup\u003e. Key variables included demographic characteristics such as age, gender, and ethnicity; comorbidities including chronic kidney disease, cardiovascular diseases, and hypertension; wound-specific characteristics such as the presence of pressure ulcers and infection indicators (e.g., sepsis and pneumonia); and treatment modalities including the use of antibiotics, negative pressure wound therapy, and insulin therapy \u003csup\u003e\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e,\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e\u003c/sup\u003e. These variables provided a holistic perspective on the multifactorial influences affecting wound healing.\u003c/p\u003e \u003c/div\u003e\n\u003ch3\u003eMethodology\u003c/h3\u003e\n\u003cp\u003eTo analyze the data, a multi-pronged methodology was employed. Descriptive statistics were calculated to summarize the baseline characteristics of the cohort, presenting continuous variables as means\u0026thinsp;\u0026plusmn;\u0026thinsp;standard deviations or medians (interquartile ranges) and categorical variables as frequencies and percentages. Univariate analyses were then conducted to identify potential associations between individual variables and wound healing outcomes, utilizing Chi-square tests for categorical variables and t-tests or Mann-Whitney U tests for continuous variables, depending on their distribution.\u003c/p\u003e \u003cp\u003eTo adjust for confounding factors and ascertain the independent effects of variables, multivariate analyses were performed. Logistic regression models were used for binary outcomes (e.g., healed vs. not healed), providing odds ratios (ORs) with 95% confidence intervals (CIs) for significant predictors \u003csup\u003e\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e\u003c/sup\u003e. For time-to-event data, such as healing duration, Cox proportional hazards models were applied, yielding hazard ratios (HRs) that quantified the relative risks associated with each variable \u003csup\u003e\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e\u003c/sup\u003e. Both models were adjusted for confounders, ensuring the robustness of the results.\u003c/p\u003e \u003cp\u003eTo enhance predictive accuracy and explore complex variable interactions, machine learning models were implemented. Random Forest, an ensemble learning method, was employed to derive feature importance rankings and build robust classification models \u003csup\u003e\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e\u003c/sup\u003e. Additionally, the XGBoost algorithm, known for its computational efficiency and predictive power, was utilized to construct highly accurate models. Model performance was assessed using metrics such as the area under the receiver operating characteristic curve (AUC), sensitivity, specificity, and F1-score, providing a comprehensive evaluation of their predictive capabilities.\u003c/p\u003e \u003cp\u003eTo improve transparency and clinical applicability, interpretability tools were integrated into the machine learning framework. SHAP (SHapley Additive exPlanations) was used to assess the global importance of variables, elucidating their contributions to model predictions \u003csup\u003e\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e\u003c/sup\u003e. LIME (Local Interpretable Model-agnostic Explanations) complemented this by providing case-specific explanations, highlighting the factors influencing individual predictions and enabling clinicians to better understand patient-specific risks.\u003c/p\u003e \u003cp\u003eData preprocessing played a critical role in ensuring the quality and reliability of the analysis. Missing values were addressed through imputation or exclusion strategies, depending on the extent and nature of the missing data \u003csup\u003e\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e\u003c/sup\u003e. Outliers were identified and managed to prevent their influence on model performance \u003csup\u003e\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e\u003c/sup\u003e. Continuous variables were normalized to ensure comparability and improve model convergence, while categorical variables were encoded using one-hot encoding to facilitate their inclusion in machine learning models. The dataset was split into training and testing subsets in a 70:30 ratio, ensuring that model validation was performed on unseen data to assess their generalizability \u003csup\u003e\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e \u003cp\u003eThis comprehensive methodological framework provided a robust foundation for analyzing the multi factorial influences on wound healing in diabetic ICU patients. By integrating traditional statistical approaches with advanced machine learning techniques, the study delivered actionable insights to support personalized treatment strategies and optimized healthcare decision-making.\u003c/p\u003e"},{"header":"Results","content":"\u003cdiv id=\"Sec6\" class=\"Section2\"\u003e \u003ch2\u003eDescriptive Analysis of the Dataset\u003c/h2\u003e \u003cp\u003eThe descriptive analysis of the dataset serves as a foundational step to explore the baseline characteristics of diabetic patients and their relationship with the presence of pressure ulcers. The results, visualized through a series of density plots, histograms, and scatter plots, provide insights into potential risk factors and patterns, which inform further multivariate and predictive modeling.\u003c/p\u003e \u003c/div\u003e\n\u003ch3\u003eAge Distribution by Pressure Ulcer Status\u003c/h3\u003e\n\u003cp\u003eAs shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e, the age distribution highlights distinct patterns between patients with and without pressure ulcers.The age distribution, as shown by the density plot, highlights distinct patterns between patients with and without pressure ulcers. Patients diagnosed with pressure ulcers exhibit a slight skew toward older age groups compared to those without. The overlap between the two groups suggests that while middle-aged patients (with standardized age around 0) are prevalent in both populations, older individuals demonstrate a higher density in the pressure ulcer cohort. This observation is consistent with established evidence that aging contributes to diminished regenerative capacity, reduced collagen synthesis, and impaired angiogenesis, all of which increase susceptibility to pressure ulcers and hinder the wound healing process. Understanding the role of age is crucial, as it serves as a significant determinant of both the incidence and prognosis of pressure ulcers in critically ill diabetic patients.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cdiv id=\"Sec8\" class=\"Section2\"\u003e \u003ch2\u003eWeight Distribution by Pressure Ulcer Status\u003c/h2\u003e \u003cp\u003eAs shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e, the weight distribution further delineates the differences between patients with and without pressure ulcers. While there is substantial overlap, individuals without pressure ulcers display a more pronounced density around the average weight (standardized value near 0), whereas those with pressure ulcers exhibit a broader and flatter distribution. This suggests that deviations from normal weight, either underweight or overweight, may predispose patients to pressure ulcer development. For instance, underweight patients often present with reduced subcutaneous tissue, impairing pressure redistribution, while overweight patients may experience compromised mobility and increased mechanical stress. These findings underscore the multi-factorial nature of pressure ulcer risk, warranting further investigation into the interplay between weight and other physiological factors.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e\n\u003ch3\u003ePressure Ulcer Healing Time Distribution\u003c/h3\u003e\n\u003cp\u003eThe histogram of healing time for patients with pressure ulcers, as shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e, reveals a heavily right-skewed distribution. The histogram of healing time for patients with pressure ulcers reveals a heavily right-skewed distribution, with most patients achieving healing within 10 days. However, a long tail extending beyond 100 days signifies a subset of individuals experiencing prolonged recovery. Such variability in healing duration reflects the influence of patient-specific factors, including comorbidities, infection status, and treatment modalities. This heterogeneity highlights the necessity for targeted interventions tailored to individual patient profiles. Furthermore, the presence of extreme outliers emphasizes the importance of identifying predictors associated with delayed healing to inform risk stratification and optimize clinical outcomes.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e\n\u003ch3\u003eGender Distribution by Pressure Ulcer Status\u003c/h3\u003e\n\u003cp\u003eAs shown in the Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e, he bar plot depicting gender distribution, as shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e, indicates a disproportionate representation of male patients.The bar plot depicting gender distribution indicates a disproportionate representation of male patients in both the pressure ulcer and non-pressure ulcer groups. While females constitute a smaller fraction of the population, the gender distribution raises questions regarding potential biological, behavioral, or healthcare access disparities. Factors such as differences in skin structure, hormonal influences, and varying rates of ICU admission across genders may contribute to these observations. Future analyses could explore whether gender plays an independent role in pressure ulcer development and healing or if it acts as a con-founder mediated by other clinical variables.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cdiv id=\"Sec11\" class=\"Section2\"\u003e \u003ch2\u003eObserved and Expected Frequencies of Sepsis by Gender\u003c/h2\u003e \u003cp\u003eThe observed and expected frequencies of sepsis, as shown in Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e, reveal minimal deviations between the two frequencies across genders.The side-by-side comparison of observed and expected frequencies of sepsis reveals that sepsis is relatively infrequent across the dataset, with minimal deviations between the two frequencies across genders. The side-by-side comparison of observed and expected frequencies of sepsis, as shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003e, confirms that sepsis is not a predominant feature.The expected frequencies for patients without pressure ulcers are approximately 65,430 (sepsis_flag\u0026thinsp;=\u0026thinsp;0) and 1,018 (sepsis_flag\u0026thinsp;=\u0026thinsp;1), while for those with pressure ulcers, the expected counts are 81,673 (sepsis_flag\u0026thinsp;=\u0026thinsp;0) and 1,271 (sepsis_flag\u0026thinsp;=\u0026thinsp;1). These results confirm that sepsis is not a predominant feature among the studied cohort but warrants consideration due to its role in exacerbating systemic inflammation and compromising immune function. The small differences between observed and expected values further suggest that gender is not a major determinant of sepsis occurrence. However, its potential impact on wound healing outcomes, particularly in conjunction with pressure ulcers, should be explored further.\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\u003eSepsis Frequencies by Pressure Ulcer Status\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"3\"\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 \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eExpected\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eFrequencies:\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003esepsis_flag\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003epressure_ulcer_flag\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 \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e65429.876727\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1018.123273\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e81673.123273\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1270.876727\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 \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec12\" class=\"Section2\"\u003e \u003ch2\u003eWeight vs. Age by Pressure Ulcer Flag\u003c/h2\u003e \u003cp\u003eThe scatter plot of weight versus age, as shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003e, provides a granular perspective on the relationship between these variables.The scatter plot of weight versus age, stratified by pressure ulcer status, provides a granular perspective on the relationship between these variables. The distribution reveals no clear linear correlation between weight and age. However, the scatter suggests that pressure ulcers are more prevalent among patients with extreme values in either dimension. Such clustering patterns highlight the multi-factorial nature of pressure ulcer risk, where deviations in demographic and clinical characteristics collectively influence outcomes. This reinforces the need for integrative modeling approaches to capture the complex interactions between variables.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec13\" class=\"Section2\"\u003e \u003ch2\u003eImplications of Findings\u003c/h2\u003e \u003cp\u003eThe descriptive analysis yielded several key insights that serve as a foundation for subsequent analytical stages:\u003c/p\u003e \u003cp\u003eAge and Weight as Risk Factors: Both age and weight distributions indicate that deviations from normative ranges are associated with increased pressure ulcer prevalence, underscoring their role as critical demographic determinants.Heterogeneity in Healing Times: The wide variability in healing duration highlights the complexity of wound healing processes, emphasizing the need for individualized treatment strategies.\u003c/p\u003e \u003cp\u003eGender Disparities: The disproportionate representation of males in the dataset suggests potential biases or systemic differences that require further investigation.\u003c/p\u003e \u003cp\u003eSepsis and Compounding Risks: While sepsis frequencies are low, its potential role in exacerbating delayed healing and adverse outcomes underscores the importance of incorporating infection status in predictive models.\u003c/p\u003e \u003cp\u003eComplex Variable Interactions: The scatter plot of weight and age emphasizes the need to consider multidimensional interactions when assessing pressure ulcer risk.\u003c/p\u003e \u003cp\u003eThese findings not only inform the selection of variables for multivariate modeling but also provide a basis for developing predictive frameworks aimed at early identification and intervention for high-risk patients. By integrating these descriptive insights into advanced statistical and machine learning models, the study aims to generate actionable knowledge to improve patient outcomes and optimize resource allocation in critical care settings.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec14\" class=\"Section2\"\u003e \u003ch2\u003eUnivariate Analysis\u003c/h2\u003e \u003cp\u003eThe results of the Cox proportional hazards regression analysis, as shown in Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e, highlight the significant interactions between clinical variables.The univariate analysis revealed significant associations between several comorbidities and wound healing outcomes in diabetic ICU patients. Among the variables analyzed, hypertension, chronic kidney disease (CKD), and the presence of pressure ulcers emerged as key determinants that significantly delayed wound healing.\u003c/p\u003e \u003cp\u003eHypertension, present in over 52% of the diabetic cohort, likely exacerbates vascular dysfunction by impairing endothelial function and reducing arterial elasticity. These effects hinder the delivery of oxygen and nutrients to wound sites, a critical requirement for fibroblast activity and collagen deposition. Furthermore, hypertension-induced microvascular damage disrupts angiogenesis, which is essential for forming new capillaries to support the healing process. This physiological interference prolongs the inflammatory phase and delays progression to wound granulation and re-epithelialization.\u003c/p\u003e \u003cp\u003eChronic kidney disease (CKD), which affected approximately 26% of the cohort, demonstrated a similarly significant impact on wound healing. CKD is known to cause systemic oxidative stress and chronic inflammation, which impair multiple cellular processes necessary for wound repair. For instance, the accumulation of uremic toxins disrupts fibroblast proliferation, delays keratinocyte migration, and weakens the mechanical strength of healed tissue. Additionally, CKD-associated metabolic disturbances, such as anemia and hypoalbuminemia, further compromise the wound healing microenvironment by reducing oxygen transport and depriving cells of essential nutrients.\u003c/p\u003e \u003cp\u003eThe presence of pressure ulcers, affecting 1.5% of patients, was another critical factor. These localized injuries result from prolonged pressure on the skin and underlying tissue, leading to ischemia and tissue necrosis. Pressure ulcers impair the normal healing cascade by creating hypoxic and necrotic environments, which inhibit angiogenesis and collagen remodeling. Additionally, the risk of secondary infections is heightened in pressure ulcer sites, further complicating the healing process. Taken together, these findings emphasize the multifactorial nature of delayed wound healing and the importance of addressing comorbidities as integral components of patient management strategies.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec15\" class=\"Section2\"\u003e \u003ch2\u003eMultivariate Analysis\u003c/h2\u003e \u003cp\u003eThe multivariate analysis using Cox proportional hazards regression models provided deeper insights into the independent and interactive effects of clinical variables on wound healing outcomes. Among the variables examined, the presence of pressure ulcers emerged as the most significant predictor of delayed healing, with a hazard ratio (HR) of 2.36 (95% confidence interval: 2.08\u0026ndash;2.67). This finding underscores the multifaceted challenges posed by pressure ulcers, including chronic inflammation, impaired angiogenesis, and heightened risk of secondary infections. The combination of localized tissue ischemia and systemic complications creates an environment that is highly resistant to conventional wound healing processes.\u003c/p\u003e \u003cp\u003eThe log(HR) and 95% CI plot for predictors, as shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003e, illustrates the impact of pressure ulcer flag and associated interactions. Interaction analyses revealed compelling relationships between pressure ulcers, comorbid conditions, and treatment modalities \u003csup\u003e\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e\u003c/sup\u003e. Notably, the interaction between negative pressure wound therapy (NPWT) and pressure ulcers demonstrated substantial efficacy in mitigating delayed healing outcomes, with an HR of 1.64 (95% CI: 1.38\u0026ndash;1.94). This suggests that NPWT can alleviate some of the adverse effects of ischemia by promoting granulation tissue formation and removing exudate from the wound bed. These findings align with prior studies demonstrating the effectiveness of NPWT in managing complex wounds, particularly in diabetic populations.\u003c/p\u003e \u003cp\u003eConversely, the interaction between chronic kidney disease (CKD) and pressure ulcers significantly exacerbated delayed healing outcomes, with an HR of 2.14 (95% CI: 1.81\u0026ndash;2.54). This interaction likely reflects the combined effects of CKD-related systemic inflammation and the localized ischemic environment created by pressure ulcers. CKD disrupts the production of growth factors such as vascular endothelial growth factor (VEGF), which are critical for angiogenesis and tissue repair. Additionally, CKD patients often experience anemia and hypoalbuminemia, further compounding the challenges of wound healing by impairing oxygen delivery and reducing the availability of essential proteins.\u003c/p\u003e \u003cp\u003eThe log(HR) with 95% CI for predictors incorporating pressure ulcer and interaction terms, as shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig8\" class=\"InternalRef\"\u003e8\u003c/span\u003e, provides further insights. Another notable interaction was observed between insulin therapy and pressure ulcers. The analysis revealed that insulin therapy\u0026rsquo;s benefits are partially attenuated in patients with pressure ulcers, with an HR of 1.64 (95% CI: 1.39\u0026ndash;1.94). This may be attributed to the systemic inflammatory responses often associated with severe pressure ulcers, which can counteract the metabolic benefits of insulin by inducing insulin resistance at the cellular level. These findings suggest that while insulin therapy remains critical for glycemic control, additional interventions may be required to address the unique challenges posed by pressure ulcers in diabetic patients.\u003c/p\u003e \u003cp\u003eThe interaction between antibiotic treatment and pressure ulcers, while not statistically significant (HR\u0026thinsp;=\u0026thinsp;0.98; 95% CI: 0.95\u0026ndash;1.01), highlights the complexity of integrating therapeutic modalities into wound management strategies. While antibiotics are essential for infection control, their role in directly influencing wound healing may be limited without addressing other contributing factors such as ischemia and systemic inflammation.\u003c/p\u003e \u003cp\u003eOverall, these findings highlight the importance of adopting a personalized and multifactorial approach to wound management in diabetic ICU patients. By accounting for the interplay between treatment modalities and comorbid conditions, clinicians can develop targeted interventions that address the specific challenges faced by high-risk patients. For example, combining NPWT with adjunct therapies such as growth factor supplementation or anti-inflammatory agents may offer synergistic benefits for patients with severe pressure ulcers. Similarly, optimizing nutritional support in CKD patients could help mitigate the systemic factors that impede wound healing. This evidence underscores the need for integrative care models that leverage both clinical and technological innovations to improve outcomes for diabetic ICU patients with complex wound profiles.\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\u003eClinical Variable Interactions and Their Impact on Wound Healing Outcomes in Diabetic ICU Patients: A Cox Proportional Hazards Regression Analysis\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"12\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" 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 \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c7\" colnum=\"7\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c8\" colnum=\"8\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c9\" colnum=\"9\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c10\" colnum=\"10\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c11\" colnum=\"11\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c12\" colnum=\"12\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003ecoef\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eexp(coef)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003ese(coef)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003ecoef lower 95%\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003ecoef upper 95%\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003eexp(coef) lower 95%\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c7\"\u003e \u003cp\u003eexp(coef) upper 95%\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c8\"\u003e \u003cp\u003ecmp to\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c9\"\u003e \u003cp\u003ez\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c10\"\u003e \u003cp\u003ep\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c11\"\u003e \u003cp\u003e-log2(p)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colspan=\"1\" nameend=\"c12\" namest=\"c12\"\u003e\u0026nbsp;\u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eage\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e1.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.01\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e-0.01\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.02\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.99\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e1.02\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e0.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e0.25\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c11\"\u003e \u003cp\u003e0.81\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c12\"\u003e \u003cp\u003e0.31\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003egender\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e1.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.02\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e-0.03\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.03\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.97\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e1.03\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e0.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e0.09\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c11\"\u003e \u003cp\u003e0.93\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c12\"\u003e \u003cp\u003e0.11\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eweight\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e1.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.01\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e-0.01\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.02\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.99\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e1.02\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e0.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e0.21\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c11\"\u003e \u003cp\u003e0.83\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c12\"\u003e \u003cp\u003e0.26\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ehypertension_flag\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.01\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e1.01\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.02\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e-0.03\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.04\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.98\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e1.04\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e0.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e0.41\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c11\"\u003e \u003cp\u003e0.68\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c12\"\u003e \u003cp\u003e0.55\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ecardiovascular_flag\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.02\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e1.02\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.03\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e-0.04\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.07\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.96\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e1.07\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e0.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e0.53\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c11\"\u003e \u003cp\u003e0.59\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c12\"\u003e \u003cp\u003e0.75\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003echronic_kidney_disease_flag\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.01\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e1.01\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.02\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e-0.02\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.05\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.98\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e1.05\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e0.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e0.67\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c11\"\u003e \u003cp\u003e0.50\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c12\"\u003e \u003cp\u003e1.00\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003etype1_diabetes_flag\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.01\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e1.01\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.05\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e-0.09\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.11\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.92\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e1.11\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e0.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e0.20\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c11\"\u003e \u003cp\u003e0.84\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c12\"\u003e \u003cp\u003e0.25\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003etype2_diabetes_flag\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.03\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e1.03\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.03\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e-0.04\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.09\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.96\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e1.10\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e0.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e0.75\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c11\"\u003e \u003cp\u003e0.45\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c12\"\u003e \u003cp\u003e1.15\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003epressure_ulcer_flag\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.86\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e2.36\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.06\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.73\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.98\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e2.08\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e2.67\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e0.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e13.63\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c11\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;0.005\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c12\"\u003e \u003cp\u003e138.16\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eantibiotic_treatment_flag\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e-0.02\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.98\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.02\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e-0.05\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.01\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.95\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e1.01\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e0.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e-1.23\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c11\"\u003e \u003cp\u003e0.22\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c12\"\u003e \u003cp\u003e2.19\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003enegative_pressure_flag\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e-0.02\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.98\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.02\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e-0.05\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.01\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.95\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e1.01\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e0.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e-1.23\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c11\"\u003e \u003cp\u003e0.22\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c12\"\u003e \u003cp\u003e2.19\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003einsulin_flag\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e-0.02\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.98\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.02\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e-0.05\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.01\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.95\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e1.01\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e0.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e-1.22\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c11\"\u003e \u003cp\u003e0.22\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c12\"\u003e \u003cp\u003e2.18\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003epressure_ulcer_treatment_interaction\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.49\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e1.63\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.09\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.32\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.66\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e1.38\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e1.93\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e0.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e5.75\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c11\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;0.005\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c12\"\u003e \u003cp\u003e26.75\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003epressure_ulcer_negative_pressure_interaction\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.49\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e1.64\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.09\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.33\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.66\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e1.38\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e1.94\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e0.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e5.76\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c11\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;0.005\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c12\"\u003e \u003cp\u003e26.84\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003epressure_ulcer_insulin_interaction\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.50\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e1.64\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.09\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.33\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.66\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e1.39\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e1.94\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e0.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e5.79\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c11\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;0.005\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c12\"\u003e \u003cp\u003e27.12\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003epressure_ulcer_ckd_interaction\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.76\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e2.14\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.09\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.59\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.93\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e1.81\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e2.54\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e0.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e8.82\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c11\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;0.005\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c12\"\u003e \u003cp\u003e59.62\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 \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec16\" class=\"Section2\"\u003e \u003ch2\u003eLogistic Regression Analysis\u003c/h2\u003e \u003cp\u003eThe ROC curve depicting the performance of a binary classification model, as shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig9\" class=\"InternalRef\"\u003e9\u003c/span\u003e, indicates an AUC of 0.76. The logistic regression model served as a foundational method for evaluating the relationships between clinical variables and wound healing outcomes. Despite its simplicity and interpretability, the model demonstrated moderate discriminatory power with an Area Under the Receiver Operating Characteristic Curve (AUC) of 0.7153. The model's classification performance, as reflected in the confusion matrix and classification report, revealed significant challenges in handling the imbalanced nature of the dataset. While the model achieved near-perfect precision and recall for the majority class (healed wounds), it failed to identify any cases of the minority class (non-healed wounds). Specifically, the classification report showed an F1-score of 0.98 for the healed class but 0.00 for the non-healed class. The confusion matrix indicated 458 false negatives and no true positives for the minority class, underscoring the model\u0026rsquo;s inability to effectively address imbalanced data distributions. These results highlight the limitations of logistic regression in capturing complex, non-linear relationships and its sensitivity to skewed class distributions.\u003c/p\u003e \u003cp\u003eFeature importance analysis, derived from logistic regression coefficients, identified infection status, sepsis, pneumonia, and chronic kidney disease (CKD) as the most influential predictors of wound healing outcomes. The strong association between infection-related variables and delayed healing aligns with established clinical findings, emphasizing the need to prioritize infection management in wound care. However, the linear nature of logistic regression limited its ability to account for interaction effects among these variables, further constraining its predictive accuracy.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec17\" class=\"Section2\"\u003e \u003ch2\u003eRandom Forest Results\u003c/h2\u003e \u003cp\u003eThe ROC curve demonstrating the classification performance of the Random Forest algorithm, as shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig10\" class=\"InternalRef\"\u003e10\u003c/span\u003e, indicates an AUC of 0.99.The Random Forest model achieved outstanding performance, far surpassing the logistic regression model. Its ability to handle non-linear relationships and interactions among variables contributed to this exceptional result. The classification report indicated balanced precision, recall, and F1-scores of 0.96 for both healed and non-healed classes, demonstrating the model\u0026rsquo;s robustness across the dataset. The confusion matrix revealed 1,097 false positives and 1,172 false negatives, representing a substantial reduction in misclassification errors compared to logistic regression. The plot illustrating the association between antibiotic treatment status and infection status, as shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig11\" class=\"InternalRef\"\u003e11\u003c/span\u003e, provides insights into the impact of therapeutic interventions.\u003c/p\u003e \u003cp\u003eRandom Forest\u0026rsquo;s feature importance analysis corroborated the findings from logistic regression, highlighting infection status, sepsis, CKD, and pressure ulcers as critical predictors. Moreover, the model identified additional important variables, such as antibiotic treatment and negative pressure wound therapy (NPWT), underscoring the multifactorial nature of wound healing.\u003c/p\u003e \u003cp\u003eSHAP (SHapley Additive exPlanations) analysis further enhanced the interpretability of the Random Forest model. SHAP values demonstrated that infection-related variables had the most substantial negative impact on healing outcomes, while therapeutic interventions like NPWT exerted consistently positive effects. For example, in one case, the absence of NPWT combined with advanced CKD and infection was identified as the primary contributor to delayed healing, providing actionable insights for clinical decision-making.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec18\" class=\"Section2\"\u003e \u003ch2\u003eXGBoost and MLP Results\u003c/h2\u003e \u003cp\u003eThe model performance comparison in terms of AUC, as shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig12\" class=\"InternalRef\"\u003e12\u003c/span\u003e, highlights the superior performance of Random Forest and XGBoost models.The XGBoost model also performed exceptionally well, achieving an AUC of 0.96. Its gradient-boosting approach optimized predictive performance by iteratively minimizing errors, making it particularly effective for imbalanced datasets. The model\u0026rsquo;s classification metrics were comparable to Random Forest, with high precision and recall across both classes. Feature importance analysis from XGBoost aligned closely with Random Forest, further validating the critical role of infection-related variables and therapeutic interventions.\u003c/p\u003e \u003cp\u003eThe MLP model, a neural network-based approach, achieved an AUC of 0.92. While its performance was slightly lower than Random Forest and XGBoost, it still outperformed logistic regression by a considerable margin. However, the lack of straightforward interpretability in MLP models posed challenges for clinical application, as insights into variable importance and decision-making processes were less accessible.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec19\" class=\"Section2\"\u003e \u003ch2\u003eComparative Analysis\u003c/h2\u003e \u003cp\u003eThe comparative analysis of logistic regression and machine learning models underscores the superior performance of ensemble methods like Random Forest and XGBoost in predicting wound healing outcomes. While logistic regression offered interpretability and simplicity, its reliance on linear assumptions and sensitivity to class imbalance limited its utility in this context. In contrast, Random Forest and XGBoost excelled in capturing non-linear relationships and addressing class imbalances, achieving near-perfect AUC values. The interpretability provided by SHAP analysis in Random Forest and XGBoost further enhanced their clinical applicability, enabling data-driven decision-making and personalized treatment strategies.\u003c/p\u003e \u003cp\u003eThe machine learning models, particularly Random Forest, demonstrated the potential to integrate into clinical workflows for early risk identification and optimized resource allocation. Future research should focus on incorporating additional data sources, such as genomic and proteomic profiles, to further refine these models and expand their applicability across diverse patient populations. These findings highlight the transformative potential of machine learning in advancing predictive analytics and improving outcomes in diabetic ICU patients with complex wound profiles.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec20\" class=\"Section2\"\u003e \u003ch2\u003eModel Interpretability and Analysis\u003c/h2\u003e \u003cp\u003eThe SHAP value plot for feature importance and their impact on model output, as shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig13\" class=\"InternalRef\"\u003e13\u003c/span\u003e, highlights the critical predictors of wound healing outcomes.To further elucidate the contributions of individual features to wound healing outcomes, model interpretability tools, specifically SHAP (SHapley Additive exPlanations) and LIME (Local Interpretable Model-Agnostic Explanations), were employed. These tools provided both global and local insights into the predictive dynamics of the machine learning models, enhancing their applicability in clinical contexts and ensuring that predictions were transparent and actionable.\u003c/p\u003e \u003cp\u003eThe SHAP summary plot highlighted the relative importance of various predictors, revealing a diverse array of factors influencing wound healing. The feature with the most significant impact was the death flag, indicating that mortality risk strongly influences healing probabilities. This finding is consistent with the clinical understanding that systemic complications associated with high mortality risk, such as multi-organ failure and sepsis, are major impediments to effective wound healing. Variables such as negative pressure flag, chronic kidney disease (CKD) flag, and age also demonstrated substantial contributions to the model's predictions, consistent with their well-documented roles in wound pathophysiology. Negative SHAP values for these features indicated a detrimental effect on the likelihood of wound healing when their values were high. For instance, advanced age and CKD were associated with delayed healing due to their contributions to systemic inflammation, vascular dysfunction, and compromised cellular repair mechanisms.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eThe SHAP analysis further underscored the nuanced effects of therapeutic interventions. The negative pressure wound therapy (NPWT) flag had a protective SHAP impact, reaffirming its utility in managing complex wounds by enhancing granulation tissue formation, reducing local ischemia, and facilitating exudate removal. This finding aligns with previous clinical studies that position NPWT as a cornerstone therapy for pressure ulcers and other refractory wounds. Furthermore, infection-related variables, including sepsis flag and pneumonia flag, were strongly associated with negative SHAP values, reflecting the adverse impact of systemic infections on wound recovery. These insights emphasize the critical importance of rigorous infection control and the early application of targeted therapies to mitigate these effects. Notably, antibiotic treatment, while essential for managing infections, did not emerge as a significant predictor of wound healing outcomes, suggesting that its efficacy may be contingent on addressing other co-existing factors such as ischemia and systemic inflammation.\u003c/p\u003e \u003cp\u003eThe prediction probabilities and feature-based decision breakdown for a binary outcome, as shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig14\" class=\"InternalRef\"\u003e14\u003c/span\u003e, provide actionable insights for clinical decision-making.At the individual level, LIME provided interpretable case-specific explanations that complemented the global insights from SHAP. For instance, in a high-risk patient, the death flag, combined with hypertension flag and the absence of NPWT, emerged as the primary contributors to the model\u0026rsquo;s prediction of delayed healing. This case demonstrated how systemic factors such as hypertension and mortality risk interact with the absence of therapeutic interventions to exacerbate healing delays. Conversely, in a patient with favorable outcomes, the presence of NPWT and the absence of chronic systemic conditions were identified as critical drivers of successful healing. Such granular analyses offer actionable insights, enabling clinicians to prioritize interventions for the most critical variables affecting individual patients.\u003c/p\u003e \u003cp\u003eThe feature correlation coefficients analysis, as shown in Table\u0026nbsp;\u003cspan refid=\"Tab3\" class=\"InternalRef\"\u003e3\u003c/span\u003e, provides insights into the relationships between various predictors.The integration of SHAP and LIME analyses into this study provided compelling evidence for the complex interplay of demographic, clinical, and therapeutic factors in wound healing outcomes among diabetic ICU patients. For example, the consistent identification of negative pressure flag, CKD flag, and infection-related variables as key predictors underscores the importance of addressing both local wound conditions and systemic health issues to optimize healing trajectories. Moreover, the findings advocate for the routine implementation of NPWT in high-risk patients, particularly those with pressure ulcers, as well as the importance of mitigating systemic complications such as infections and CKD.From a broader perspective, SHAP and LIME also highlight the importance of integrating interpretability into machine learning models to bridge the gap between predictive analytics and clinical application. The ability to provide both global and patient-specific insights ensures that predictions are not only accurate but also clinically meaningful. For instance, SHAP's global summary plots enable clinicians to understand the overarching factors driving wound healing outcomes across a population, while LIME's local explanations facilitate personalized care by pinpointing the specific variables most relevant to individual patients. This dual approach enhances the utility of machine learning in clinical decision-making and fosters trust among healthcare providers by making the predictive processes transparent.\u003c/p\u003e \u003cp\u003e \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\u003eFeature Correlation Coefficients Analysis\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=\"char\" char=\".\" 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 \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\u003eValue\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eFeature\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eValue\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eage\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e1.33\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003edeath_flag\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e1.00\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003etype1_diabetes_flag\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003ecardiovascular_flag\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.00\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003etype2_diabetes_flag\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003ehypertension_flag\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e1.00\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003enegative_pressure_flag\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003esepsis_flag\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.00\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003echronic_kidney_disease_flag\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eweight\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e-0.32\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 conclusion, the comprehensive analysis of machine learning models, augmented by SHAP and LIME interpretability tools, highlights their transformative potential in advancing wound care for diabetic ICU patients. By integrating predictive analytics with interpretable insights, clinicians can develop targeted, evidence-based strategies to improve outcomes and reduce the burden of chronic wounds. These findings emphasize the need for a multifactorial approach that addresses both systemic and local factors affecting wound healing. Future research should focus on expanding these models by incorporating additional data sources, such as biomarkers, genomic profiles, and real-time physiological monitoring, to further refine predictive accuracy and enhance personalized care. Additionally, exploring the integration of these models into electronic health record systems can facilitate real-time risk assessment, optimize resource allocation, and ultimately improve patient outcomes on a broader scale.\u003c/p\u003e \u003c/div\u003e"},{"header":"Discussion","content":"\u003cp\u003eThis study provides a comprehensive analysis of wound healing outcomes among diabetic ICU patients by integrating advanced statistical and machine learning approaches with model interpretability tools. The findings underscore the multifactorial nature of wound healing, influenced by a combination of demographic, clinical, and therapeutic factors. Among the key variables identified, the presence of pressure ulcers, chronic kidney disease, and systemic infections emerged as significant predictors of delayed healing, while therapeutic interventions like negative pressure wound therapy (NPWT) demonstrated a protective effect. The interplay between systemic comorbidities and localized wound conditions highlights the need for a multifaceted approach to wound management.\u003c/p\u003e \u003cp\u003eThe use of machine learning models, particularly Random Forest and XGBoost, significantly enhanced predictive accuracy compared to traditional logistic regression. These models effectively addressed class imbalances and captured complex, non-linear relationships among variables. The integration of SHAP and LIME provided both global and patient-specific insights, enhancing the interpretability and clinical applicability of the predictive models. SHAP analysis revealed the critical importance of infection-related variables and NPWT, while LIME offered actionable, individualized explanations that could inform targeted interventions.\u003c/p\u003e \u003cp\u003eClinically, these findings emphasize the importance of personalized care strategies that account for both systemic and local factors. For instance, the routine implementation of NPWT in high-risk patients and the prioritization of infection control can mitigate some of the most significant barriers to wound healing. Furthermore, the incorporation of predictive models into electronic health record systems has the potential to facilitate real-time risk assessment and optimize resource allocation, improving outcomes for diabetic ICU patients with complex wound profiles.\u003c/p\u003e \u003cp\u003eFuture research should focus on expanding these models by incorporating additional data sources, such as biomarkers, genomic profiles, and longitudinal physiological monitoring, to refine predictive accuracy further. Additionally, exploring the integration of these models into broader healthcare systems can enhance their scalability and impact. Ultimately, this study highlights the transformative potential of machine learning and interpretability tools in advancing wound care, offering a pathway toward more precise, effective, and patient-centered management strategies for chronic wounds.\u003c/p\u003e"},{"header":"Conclusion","content":"\u003cp\u003eThis study demonstrates the transformative potential of integrating advanced machine learning and interpretability tools in predicting wound healing outcomes among diabetic ICU patients. Key findings highlight pressure ulcers, chronic kidney disease, and systemic infections as significant predictors of delayed healing, while negative pressure wound therapy (NPWT) shows a protective effect. Machine learning models (e.g., Random Forest and XGBoost) outperformed traditional logistic regression by addressing class imbalances and capturing complex, non-linear relationships. Interpretability tools like SHAP and LIME provided actionable insights, emphasizing infection-related variables and NPWT, while offering patient-specific explanations for targeted interventions. Clinically, these results underscore the importance of personalized care strategies that address both systemic and local factors, such as prioritizing NPWT and infection control. Integrating predictive models into electronic health records could enhance real-time risk assessment and optimize resource allocation. Future research should expand models by incorporating biomarkers, genomic data, and longitudinal monitoring to refine accuracy. Additionally, exploring the scalability of these models within healthcare systems could maximize their impact. Overall, this study advances wound care by leveraging machine learning to develop precise, patient-centered strategies for managing chronic wounds in high-risk populations.\u003c/p\u003e"},{"header":"Abbreviations","content":"\u003cdiv class=\"DefinitionList\"\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eICU\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eIntensive Care Unit\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eMIMIC-IV\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eMedical Information Mart for Intensive Care IV\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eNPWT\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eNegative Pressure Wound Therapy\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eCKD\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eChronic Kidney Disease\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eAUC\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eArea Under the Receiver Operating Characteristic Curve\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003c/div\u003e"},{"header":"Declarations","content":"\u003cp\u003e \u003ch2\u003eEthics approval and consent to participate:\u003c/h2\u003e \u003cp\u003enot applicable.\u003c/p\u003e \u003c/p\u003e \u003cp\u003e \u003cstrong\u003eConsent for publication:\u003c/strong\u003e \u003cp\u003eNot applicable\u003c/p\u003e \u003c/p\u003e\u003cp\u003e \u003ch2\u003eCompeting Interests:\u003c/h2\u003e \u003cp\u003eThe authors declare that they have no competing interests.\u003c/p\u003e \u003c/p\u003e\u003ch2\u003eFunding:\u003c/h2\u003e \u003cp\u003eThis research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors.\u003c/p\u003e\u003ch2\u003eAuthor Contribution\u003c/h2\u003e\u003cp\u003eL.Z., X.C., Y.T., K.L., M.Z. and H.T. conceived and designed the experiments; L.Z. and X.C. performed the experiments; Y.T. and K.L. analyzed the data; M.Z. and H.T. contributed reagents/materials/analysis tools; L.Z. wrote the paper; Y..T and K.L. reviewed and edited the manuscript.\u003c/p\u003e\u003ch2\u003eAcknowledgements:\u003c/h2\u003e \u003cp\u003eNot applicable\u003c/p\u003e\u003ch2\u003eData Availability\u003c/h2\u003e\u003cp\u003eThe data used in this study are derived from the Medical Information Mart for Intensive Care IV (MIMIC-IV) database, which is a publicly accessible dataset. The MIMIC-IV database is a comprehensive clinical dataset that includes detailed records of patients admitted to ICUs at the Beth Israel Deaconess Medical Center. As the data are publicly available and de-identified, no additional data sharing is required beyond the use of this publicly accessible resource.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eSaeedi P. Global and regional diabetes prevalence estimates for 2019 and projections for 2030 and 2045: Results from the International Diabetes Federation Diabetes Atlas, 9th edition. \u003cem\u003eDiabetes Research and Clinical Practice\u003c/em\u003e 2019; 151: 143.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMustoe TA. Impaired wound healing in diabetes: Current concepts and therapeutic strategies. Wound Repair Regeneration. 2006;14(6):653.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSen CK. Human diabetic wounds exhibit a distinct inflammatory gene signature. Wound Repair Regeneration 2009; 17(1).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAyello EA. Pressure ulcers in the intensive care unit: A systematic review. Crit Care Med 2008; 36(4).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKirsner RS. Wound healing in the diabetic foot: A review. Diabetes Care 2008; 31(5).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMargolis DJ. Risk factors for nonhealing of diabetic neuropathic ulcers. Wound Repair Regeneration 2007; 15(1).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eArmstrong DG. Diabetic foot wounds: What we know and where we are going. Diabetes Care 2005; 28(1).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLipsky BA. Diabetic foot infections: A global view. Diabetes Care 2012; 35(Suppl 1).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eEdsberg LE. Update on the incidence and prevalence of pressure ulcers in the United States. Wound Repair Regeneration 2011; 19(1).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBerendt AR. Diabetic foot ulcers and their treatment. \u003cem\u003eJAMA\u003c/em\u003e 2006; 295(21).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eArmstrong DG. Offloading the diabetic foot wound: A systematic review. Diabetes Care 2011; 34(8).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKastenbauer ER. Negative pressure wound therapy: A review. Crit Care Med 2007; 35(8).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eTeitelbaum RE. Negative pressure wound therapy: A review of the current evidence. 2009; 30(4).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eScherer JW. Negative pressure wound therapy: A review. Wound Repair Regeneration 2010; 18(1).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKirsner RS. The role of infection in chronic wounds. Wound Repair Regeneration 2010; 18(1).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eEdsberg LE. Pressure ulcers in the intensive care unit: A review. Crit Care Med 2006; 34(4).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBerendt AR. The pathophysiology of the diabetic foot. Diabetes Care 2008; 31(5).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLipsky BA. Infection in diabetic foot ulcers. \u003cem\u003eDiabetes Care\u003c/em\u003e 2006; 29(5).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eArmstrong DG. The role of offloading in the treatment of diabetic foot ulcers. Diabetes Care 2008; 31(5).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eArmstrong DG. The role of offloading in the treatment of diabetic foot ulcers. Diabetes Care 2010; 33(5).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eTeitelbaum RE. Negative pressure wound therapy: A review of the current evidence. J Burn Care Res 2010; 31(4).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKastenbauer ER. Negative pressure wound therapy: A review of the current evidence. Crit Care Med 2008; 36(4).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eScherer JW. Negative pressure wound therapy: A review. Wound Repair Regeneration 2011; 19(1).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKirsner RS. The role of infection in chronic wounds. Wound Repair Regeneration 2011; 19(1).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eEdsberg LE. Pressure ulcers in the intensive care unit: A review. Crit Care Med 2010; 38(4).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBerendt AR. The pathophysiology of the diabetic foot. Diabetes Care 2010; 33(5).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLipsky BA. Infection in diabetic foot ulcers. \u003cem\u003eDiabetes Care\u003c/em\u003e 2010; 33(5).\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"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":"bmc-medical-informatics-and-decision-making","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"midm","sideBox":"Learn more about [BMC Medical Informatics and Decision Making](http://bmcmedinformdecismak.biomedcentral.com/)","snPcode":"","submissionUrl":"https://www.editorialmanager.com/midm/default.aspx","title":"BMC Medical Informatics and Decision Making","twitterHandle":"BMC_series","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"em","reportingPortfolio":"BMC Series","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"Wound healing predictors, Treatment efficacy assessment, Machine learning predictive modeling, Personalized treatment strategies, Diabetic ICU patients","lastPublishedDoi":"10.21203/rs.3.rs-6307348/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-6307348/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003e\u003cstrong\u003eBackground\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eWound healing is a critical determinant of recovery and quality of life in patients with diabetes, particularly those admitted to intensive care units (ICUs). Identifying the key factors influencing wound healing and optimizing treatment strategies is essential for improving outcomes. Despite prior studies, there is limited comprehensive analysis that integrates multiple risk factors into predictive modeling frameworks.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eObjective\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis study aims to identify the significant factors affecting wound healing in diabetic ICU patients, evaluate the effects of different treatment approaches on healing outcomes, and develop a robust predictive model to assist clinicians in early risk identification and personalized treatment planning.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eMethods\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eWe utilized data from the MIMIC-IV database, encompassing 149,392 patient records. Key variables analyzed included demographic characteristics, chronic disease histories, wound-related factors, and treatment modalities. Descriptive and uni-variate analyses were performed to explore baseline characteristics and their associations with healing outcomes. Cox proportional hazards regression and logistic regression models were used for multi-factorial analyses, while machine learning models such as Random Forest and XGBoost were employed for predictive modeling. Models interpretability was enhanced through SHAP (SHapley Additive exPlanations) and LIME (Local Interpretable Model-agnostic Explanations) analyses.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eResults\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eFactors such as age, the presence of pressure ulcers, chronic kidney disease, and treatment modalities (e.g., insulin therapy, negative pressure therapy) emerged as significant predictors of wound healing outcomes. Random Forest achieved the highest performance among predictive models, with an area under the receiver operating characteristic curve (AUC) of 0.96. SHAP analysis identified age and death flags as critical determinants, while LIME provided patient-specific insights into model predictions.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConclusions\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis study underscores the importance of integrating multifactorial data to predict wound healing outcomes in diabetic ICU patients. The findings provide actionable insights for personalized treatment strategies and resource allocation in clinical settings. Future research should focus on validating these models in diverse datasets and exploring longitudinal impacts on patient recovery.\u003c/p\u003e","manuscriptTitle":"Multifactorial Analysis and Predictive Modeling of Wound Healing Outcomes in Diabetic ICU Patients: a Cohort Study Based on MIMIC-IV","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-05-20 09:43:31","doi":"10.21203/rs.3.rs-6307348/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Revision requested","date":"2025-11-06T15:36:57+00:00","index":"","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-06-06T23:40:04+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-06-06T12:05:05+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-05-27T15:01:25+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"272353312945259731988600229563024379510","date":"2025-05-27T12:32:54+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"168341253999392650387867435043205708135","date":"2025-05-27T07:01:40+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"209286920850698871543558849019597163318","date":"2025-05-24T03:12:31+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"325342675761668670864500100148992058778","date":"2025-05-22T06:24:24+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2025-05-15T03:38:00+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2025-05-13T08:47:00+00:00","index":"","fulltext":""},{"type":"editorInvited","content":"","date":"2025-04-22T07:07:48+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2025-04-21T18:13:49+00:00","index":"","fulltext":""},{"type":"submitted","content":"BMC Medical Informatics and Decision Making","date":"2025-04-21T18:12:40+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"bmc-medical-informatics-and-decision-making","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"midm","sideBox":"Learn more about [BMC Medical Informatics and Decision Making](http://bmcmedinformdecismak.biomedcentral.com/)","snPcode":"","submissionUrl":"https://www.editorialmanager.com/midm/default.aspx","title":"BMC Medical Informatics and Decision Making","twitterHandle":"BMC_series","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"em","reportingPortfolio":"BMC Series","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"01f2f91a-69f4-43f4-be07-cdc01f705bd0","owner":[],"postedDate":"May 20th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"in-revision","subjectAreas":[],"tags":[],"updatedAt":"2025-11-06T15:58:27+00:00","versionOfRecord":[],"versionCreatedAt":"2025-05-20 09:43:31","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-6307348","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-6307348","identity":"rs-6307348","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}