Development and Validation of a Machine Learning-Based Nomogram for Predicting Pressure Ulcer Risk in Respiratory Patients.

Development and Validation of a Machine Learning‐Based Nomogram for Predicting Pressure Ulcer Risk in Respiratory Patients - PMC Skip to main content An official website of the United States government Here's how you know Here's how you know Official websites use .gov A .gov website belongs to an official government organization in the United States. Secure .gov websites use HTTPS A lock ( Lock Locked padlock icon ) or https:// means you've safely connected to the .gov website. Share sensitive information only on official, secure websites. Search Log in Dashboard Publications Account settings Log out Search… Search NCBI Primary site navigation Search Logged in as: Dashboard Publications Account settings Log in Search PMC Full-Text Archive Search in PMC Journal List User Guide PERMALINK Copy As a library, NLM provides access to scientific literature. Inclusion in an NLM database does not imply endorsement of, or agreement with, the contents by NLM or the National Institutes of Health. Learn more: PMC Disclaimer | PMC Copyright Notice Int Wound J . 2026 Apr 20;23(4):e70924. doi: 10.1111/iwj.70924 Search in PMC Search in PubMed View in NLM Catalog Add to search Development and Validation of a Machine Learning‐Based Nomogram for Predicting Pressure Ulcer Risk in Respiratory Patients Qian Qian Qian Qian 1 Department of Respiratory and Digestive Department, The Affiliated Hospital of Xuzhou Medical University, Xuzhou, Jiangsu Province, China Find articles by Qian Qian 1 , Lu Han Lu Han 2 Department of Respiratory Department, The Affiliated Hospital of Xuzhou Medical University, Xuzhou, Jiangsu Province, China Find articles by Lu Han 2 , Hao Chen Hao Chen 2 Department of Respiratory Department, The Affiliated Hospital of Xuzhou Medical University, Xuzhou, Jiangsu Province, China Find articles by Hao Chen 2 , Mingyue Gao Mingyue Gao 1 Department of Respiratory and Digestive Department, The Affiliated Hospital of Xuzhou Medical University, Xuzhou, Jiangsu Province, China Find articles by Mingyue Gao 1, ✉ Author information Article notes Copyright and License information 1 Department of Respiratory and Digestive Department, The Affiliated Hospital of Xuzhou Medical University, Xuzhou, Jiangsu Province, China 2 Department of Respiratory Department, The Affiliated Hospital of Xuzhou Medical University, Xuzhou, Jiangsu Province, China * Correspondence: Mingyue Gao ( [email protected] ) ✉ Corresponding author. Revised 2026 Mar 26; Received 2025 Nov 23; Accepted 2026 Apr 10; Collection date 2026 Apr. © 2026 The Author(s). International Wound Journal published by Medicalhelplines.com Inc and John Wiley & Sons Ltd. This is an open access article under the terms of the http://creativecommons.org/licenses/by-nc-nd/4.0/ License, which permits use and distribution in any medium, provided the original work is properly cited, the use is non‐commercial and no modifications or adaptations are made. PMC Copyright notice PMCID: PMC13095859  PMID: 42010668 ABSTRACT Pressure ulcers represent a significant healthcare challenge among respiratory patients. This study aimed to develop and validate a predictive nomogram based on machine learning algorithms to identify patients at high risk for pressure ulcer development. We conducted a retrospective analysis of 263 respiratory patients (166 with pressure ulcers). Patients were randomly divided into training and testing cohorts at a 7:3 ratio. Potential risk factors were identified through univariate logistic regression. Least absolute shrinkage and selection operator (LASSO) regression selected 17 significant predictors, from which 10 variables with optimal predictive values were incorporated into a nomogram model. Model performance was assessed using receiver operating characteristic (ROC) curves, calibration plots and decision curve analysis (DCA). The final nomogram incorporated 10 predictors: age, albumin, C‐reactive protein, serum sodium, history of diabetes, chronic obstructive pulmonary disease, peripheral vascular disease, urinary incontinence, length of hospital stay and Braden sensory perception score. The model demonstrated excellent discriminative ability with AUCs of 0.865 (95% CI: 0.816–0.914) in the training cohort and 0.837 (95% CI: 0.783–0.891) in the testing cohort. Calibration curves showed good agreement between predicted and observed probabilities (Hosmer–Lemeshow test: training cohort χ 2 = 4.257, P = 0.833; testing cohort χ 2 = 12.350, P = 0.142). DCA confirmed the nomogram's superior clinical utility compared to individual predictors across a wide range of threshold probabilities. The machine learning–derived nomogram provides a practical, noninvasive tool for early identification of respiratory patients at risk for pressure ulcers. Implementation of this model could facilitate timely intervention strategies, potentially reducing the incidence of pressure ulcers and improving patient outcomes. Keywords: LASSO regression, machine learning, nomogram, pressure ulcer, preventive care, respiratory patients, risk prediction, risk stratification 1. Introduction Pressure ulcers (PUs), or pressure injuries, are localized injuries to the skin and underlying tissue, typically over a bony prominence, resulting from pressure or pressure combined with shear forces [ 1 , 2 ]. They remain a significant global healthcare challenge, affecting 0.4% to 38% of hospitalized patients despite advances in preventive care, with high prevalence in specialized settings [ 3 ]. Respiratory patients are a uniquely vulnerable population for PU development [ 4 ]. Their conditions often lead to mobility limitations, altered oxygen saturation and impaired tissue perfusion, which are major contributors to PU formation [ 5 , 6 ]. PU prevalence in this group can reach 50%, particularly in those with chronic obstructive pulmonary disease (COPD), pneumonia or respiratory failure [ 7 ]. The consequences are severe, including prolonged hospitalization, infection and elevated mortality rates [ 8 , 9 ]. The economic burden of PUs is substantial, with treatment costs reaching billions annually and exceeding $70 000 for a single severe case [ 10 , 11 ]. This highlights the critical importance of prevention, which is more cost‐effective than treatment. Therefore, early identification of high‐risk patients is essential for implementing targeted interventions and resource allocation [ 12 , 13 ]. While conventional tools like the Braden Scale are widely utilized for risk assessment [ 14 , 15 ] they have limitations in predictive accuracy for specific, complex patient populations [ 16 ]. The Braden Scale's accuracy may be reduced in respiratory patients due to their unique risk factor profiles [ 17 , 18 ]. Machine learning (ML) offers new opportunities to develop more precise, personalized risk models [ 19 ]. ML algorithms can analyse complex clinical data to identify subtle patterns and risk factor interactions missed by traditional statistical methods [ 20 ], integrating diverse variables into tailored prediction models [ 21 , 22 ]. Nomograms are graphical tools that translate complex models into practical clinical use [ 23 , 24 ]. Their visual format allows clinicians to quickly estimate individual risk based on patient characteristics and implement appropriate preventive measures [ 25 , 26 ], potentially overcoming the limitations of existing scales by providing more accurate, personalized risk stratification [ 27 ]. Despite these benefits, research on advanced prediction models for PU risk in respiratory patients is limited. Few studies have applied ML to develop specialized tools for this vulnerable population, and existing models often lack rigorous validation or a clear path to clinical implementation. This study aims to address this gap by developing and validating an ML‐based nomogram for predicting PU risk in respiratory patients. By using LASSO regression to identify key predictors and construct a clinically applicable tool, we seek to enhance early risk identification and facilitate targeted preventive interventions, ultimately aiming to reduce PU incidence, improve patient outcomes and optimize healthcare resource use. 2. Methods 2.1. Study Design and Population This retrospective cohort study was conducted at the Department of Respiratory Medicine, Xuzhou Medical University Affiliated Hospital. We collected and analysed data from respiratory inpatients admitted between January 2023 and January 2025. 2.2. Patient Selection We initially identified 375 adult patients (≥ 18 years) admitted to the respiratory department during the study period. Inclusion criteria were: (1) hospitalization in the respiratory department for at least 48 h; (2) complete medical records and nursing documentation and (3) regular skin assessments documented during the hospital stay. Exclusion criteria included: (1) patients with pressure ulcers present at admission; (2) patients transferred from other hospitals with incomplete medical records; (3) pregnant women; (4) patients who underwent surgical procedures during hospitalization that might affect skin integrity assessment and (5) patients with incomplete data for the variables of interest. After applying these criteria, 263 patients were eligible for final analysis, including 166 patients who developed pressure ulcers during hospitalization and 97 who did not. 2.3. Data Collection Data were extracted from the hospital's electronic medical records system by two trained researchers independently. Any discrepancies were resolved through discussion with a third senior researcher. The collected information included: Demographic characteristics : Age, sex and body mass index (BMI). Biochemical indicators : Albumin, haemoglobin, total protein, blood glucose, C‐reactive protein (CRP), lymphocyte count, electrolytes (sodium, potassium, chlorine, calcium), creatinine and blood urea nitrogen. All laboratory tests were performed at the hospital's central laboratory according to standardized procedures. Disease characteristics : Primary respiratory diagnosis (categorized as COPD‐related, pneumonia‐related, respiratory failure, bronchiectasis or other respiratory diseases) and comorbidities (diabetes, COPD, peripheral vascular disease). Nursing‐related characteristics : Presence of urinary or faecal incontinence, long‐term bedridden status (defined as bed rest for > 20 h per day for at least 7 consecutive days), length of hospital stay, Braden Scale scores (total and subscale scores for sensory perception, moisture, activity, mobility, nutrition and friction/shear), turning frequency, use of pressure‐reducing devices and use of physical restraints. Pressure ulcer characteristics : For patients who developed pressure ulcers, we recorded the stage according to the National Pressure Ulcer Advisory Panel (NPUAP) classification system (Stages I–IV). 2.4. Sample Size Calculation Based on previous studies and considering the number of potential predictors to be included in the model, we determined that a minimum of 200 patients would be required to develop a robust prediction model, with at least 10 events per predictor variable. Our sample of 263 patients with 166 pressure ulcer cases exceeded this requirement, ensuring adequate statistical power. 2.5. Statistical Analysis 2.5.1. Data Preprocessing Continuous variables were assessed for normality using the Shapiro–Wilk test. Normally distributed variables were presented as mean ± standard deviation, while non‐normally distributed variables were presented as median with interquartile range. Categorical variables were expressed as frequencies with percentages. Missing data, which accounted for less than 5% for any single variable, were handled using multiple imputation by chained equations (MICE). This technique was primarily applied to biochemical indicators where occasional missing values were present, ensuring the integrity of the dataset for subsequent analysis. 2.5.2. Model Development The dataset was randomly divided into a training cohort (70%, n = 184) and a testing cohort (30%, n = 79) using stratified random sampling to maintain the same proportion of pressure ulcer cases in both cohorts. The model was developed using the following steps: Univariate analysis : We performed univariate logistic regression analysis to identify potential risk factors associated with pressure ulcer development. Variables with p < 0.05 were considered for further analysis. Feature selection : Least absolute shrinkage and selection operator (LASSO) regression was employed to select the most significant predictors from variables identified in the univariate analysis. LASSO regression was performed with 10‐fold cross‐validation to determine the optimal lambda value (lambda.1se) that provided the most regularized model with a cross‐validated error within one standard error of the minimum. This process identified 17 potential predictors. Multivariate logistic regression : The 17 predictors selected by LASSO were entered into a multivariate logistic regression model. The variables were further refined based on their receiver operating characteristic (ROC) curves' predictive value, ultimately selecting the 10 most significant predictors for inclusion in the final nomogram. Nomogram construction : A nomogram was constructed based on the final multivariate logistic regression model to provide a graphical representation of the pressure ulcer risk prediction tool. 2.5.3. Model Validation The performance of the nomogram was evaluated using both internal validation (in the training cohort) and external validation (in the testing cohort) through the following methods: Discrimination : The model's ability to distinguish between patients with and without pressure ulcers was assessed using the area under the ROC curve (AUC). An AUC value of 0.5 indicates no discrimination, while a value of 1.0 indicates perfect discrimination. Calibration : The agreement between predicted probabilities and observed outcomes was evaluated using calibration curves and the Hosmer–Lemeshow goodness‐of‐fit test. A non‐significant test result ( p > 0.05) indicates good calibration. Clinical utility : Decision curve analysis (DCA) was performed to assess the net benefit of the prediction model across a range of threshold probabilities, providing information on the clinical usefulness of the nomogram compared to alternative strategies. Comparative analysis : The performance of the nomogram was compared with individual predictors incorporated in the model to demonstrate the added value of the integrated approach. 2.6. Software Statistical analyses were performed using R software (version 4.2.0; R Foundation for Statistical Computing, Vienna, Austria) with the following packages: ‘glmnet’ for LASSO regression, ‘rms’ for nomogram construction and model validation, ‘pROC’ for ROC curve analysis and ‘rmda’ for decision curve analysis. A two‐sided p ‐value < 0.05 was considered statistically significant for all analyses. 3. Results 3.1. Study Population A flowchart depicting the patient selection process is presented in Figure 1 . From an initial pool of 375 respiratory patients screened for eligibility, 112 were excluded based on predetermined criteria: 43 patients had pressure ulcers present at admission, 32 were transferred from other hospitals with incomplete records, 16 underwent surgical procedures affecting skin integrity assessment and 21 had incomplete data for variables of interest. Consequently, 263 patients met all inclusion criteria and were included in the final analysis, comprising 166 patients who developed pressure ulcers during hospitalization (pressure ulcer group) and 97 patients who did not (non‐pressure ulcer group). These patients were randomly divided into training ( n = 184, 70%) and testing ( n = 79, 30%) cohorts for model development and validation. FIGURE 1. Open in a new tab Flowchart with inclusion and exclusion criteria for the study. 3.2. Patient Characteristics A total of 263 respiratory patients were included in the final analysis, comprising 166 patients who developed pressure ulcers during hospitalization (pressure ulcer group) and 97 patients who did not (non‐pressure ulcer group). These patients were randomly divided into training ( n = 184, 70%) and testing ( n = 79, 30%) cohorts. Baseline characteristics of the study population are summarized in Table 1 . TABLE 1. Comparison of baseline characteristics between the pressure ulcer group and the non‐pressure ulcer group. Variables Pressure ulcer group Non‐pressure ulcer group p Demographic characteristics Age (years) 77.0 ± 5.8 62.7 ± 3.2 < 0.01 Female, n (%) 92 (50.0%) 40 (50.6%) 0.926 BMI (kg/m 2 ) 19.3 ± 1.6 22.8 ± 1.0 < 0.01 Biochemical indicators Albumin (g/dL) 3.0 ± 0.3 3.9 ± 0.2 < 0.01 Haemoglobin (g/L) 98.7 ± 8.2 123.3 ± 5.1 < 0.01 Total protein (g/L) 56.6 ± 2.6 66.2 ± 1.4 < 0.01 Blood glucose (mmol/L) 7.2 ± 0.6 5.8 ± 0.2 < 0.01 CRP (mg/L) 22.8 ± 6.3 9.3 ± 0.8 < 0.01 Lymphocyte count (10 9 /L) 1.1 ± 0.3 1.9 ± 0.1 < 0.01 Sodium (mmol/L) 139.1 ± 1.7 139.4 ± 1.1 0.157 Potassium (mmol/L) 3.6 ± 0.3 4.1 ± 0.1 < 0.01 Chlorine (mmol/L) 99.5 ± 1.3 99.6 ± 0.5 0.523 Calcium (mmol/L) 2.06 ± 0.11 2.25 ± 0.04 < 0.01 Creatinine (μmol/L) 68.3 ± 7.2 53.1 ± 2.5 < 0.01 Blood urea nitrogen (mmol/L) 6.9 ± 0.8 5.1 ± 0.2 < 0.01 Disease characteristics Main diagnosis type, n (%) < 0.01 COPD‐related 102 (55.4%) 3 (3.8%) Pneumonia‐related 28 (15.2%) 1 (1.3%) Respiratory failure 16 (8.7%) 1 (1.3%) Bronchiectasis 21 (11.4%) 2 (2.5%) Other respiratory diseases 17 (9.2%) 72 (91.1%) History of diabetes, n (%) 91 (49.5%) 2 (2.5%) < 0.01 Cerebrovascular disease, n (%) 100 (54.3%) 5 (6.3%) < 0.01 Peripheral vascular disease, n (%) 23 (12.5%) 0 (0.0%) 0.01 Nursing‐related characteristics Urinary incontinence, n (%) 128 (69.6%) 0 (0.0%) < 0.01 Faecal incontinence, n (%) 88 (47.8%) 0 (0.0%) < 0.01 Long‐term bedridden, n (%) 184 (100.0%) 0 (0.0%) < 0.01 Length of hospital stay (days) 18.9 ± 4.1 6.8 ± 1.3 < 0.01 Braden total score 11.6 ± 1.6 18.2 ± 0.7 < 0.01 Braden subscale scores Sensory perception 2.1 ± 0.6 4.0 ± 0.4 < 0.01 Moisture 2.5 ± 0.5 4.0 ± 0.5 < 0.01 Activity 1.6 ± 0.5 3.5 ± 0.5 < 0.01 Mobility 1.9 ± 0.4 3.0 ± 0.6 < 0.01 Nutrition 1.4 ± 0.5 1.9 ± 0.5 < 0.01 Friction and shear 2.2 ± 0.4 2.0 ± 0.4 < 0.01 Turning frequency (hours/time) 3.2 ± 0.6 2.0 ± 0.3 < 0.01 Use of pressure‐reducing devices, n (%) 148 (80.4%) 0 (0.0%) < 0.01 Use of restraints, n (%) 68 (37.0%) 0 (0.0%) < 0.01 Pressure ulcer characteristics Pressure ulcer occurrence, n (%) 184 (100.0%) 0 (0.0%) < 0.01 Pressure ulcer stage, n (%) Stage I 39 (21.2%) Stage II 97 (52.7%) Stage III 40 (21.7%) Stage IV 8 (4.3%) Open in a new tab Note: Continuous variables are presented as mean ± standard deviation (normal distribution) or median (interquartile range) (non‐normal distribution); categorical variables are presented as frequency (percentage). p ‐values were calculated using independent samples t ‐test (normally distributed continuous variables), Mann–Whitney U test (non‐normally distributed continuous variables) or chi‐square test/Fisher's exact test (categorical variables). p < 0.05 indicates statistical significance. Patients in the pressure ulcer group were significantly older (77.0 ± 5.8 vs. 62.7 ± 3.2 years, p < 0.01) and had lower BMI (19.3 ± 1.6 vs. 22.8 ± 1.0 kg/m 2 , p < 0.01) compared to those in the non‐pressure ulcer group. Regarding biochemical indicators, the pressure ulcer group exhibited significantly lower levels of albumin (3.0 ± 0.3 vs. 3.9 ± 0.2 g/dL, p < 0.01), haemoglobin (98.7 ± 8.2 vs. 123.3 ± 5.1 g/L, p < 0.01), total protein (56.6 ± 2.6 vs. 66.2 ± 1.4 g/L, p < 0.01), lymphocyte count (1.1 ± 0.3 vs. 1.9 ± 0.1 × 10^9/L, p < 0.01), potassium (3.6 ± 0.3 vs. 4.1 ± 0.1 mmol/L, p < 0.01) and calcium (2.06 ± 0.11 vs. 2.25 ± 0.04 mmol/L, p < 0.01). Conversely, they demonstrated higher levels of blood glucose (7.2 ± 0.6 vs. 5.8 ± 0.2 mmol/L, p < 0.01), CRP (22.8 ± 6.3 vs. 9.3 ± 0.8 mg/L, p < 0.01), creatinine (68.3 ± 7.2 vs. 53.1 ± 2.5 μmol/L, p < 0.01) and blood urea nitrogen (6.9 ± 0.8 vs. 5.1 ± 0.2 mmol/L, p < 0.01). The distribution of primary respiratory diagnoses differed significantly between groups ( p < 0.01), with COPD‐related conditions being most prevalent in the pressure ulcer group (55.4% vs. 3.8%). Patients who developed pressure ulcers had a higher prevalence of comorbidities, including diabetes (49.5% vs. 2.5%, p < 0.01), COPD (54.3% vs. 6.3%, p < 0.01) and peripheral vascular disease (12.5% vs. 0.0%, p = 0.01). Nursing‐related characteristics showed striking differences between groups. All patients in the pressure ulcer group were long‐term bedridden (100% vs. 0%, p < 0.01) and had significantly longer hospital stays (18.9 ± 4.1 vs. 6.8 ± 1.3 days, p < 0.01). Urinary incontinence (69.6% vs. 0%, p < 0.01) and faecal incontinence (47.8% vs. 0%, p < 0.01) were exclusively observed in the pressure ulcer group. The total Braden score was significantly lower in the pressure ulcer group (11.6 ± 1.6 vs. 18.2 ± 0.7, p < 0.01), as were all subscale scores except for friction and shear, which were higher in the pressure ulcer group (2.2 ± 0.4 vs. 2.0 ± 0.0, p < 0.01). Turning frequency was less frequent in the pressure ulcer group (3.2 ± 0.6 vs. 2.0 ± 0.0 h/time, p < 0.01), and the use of pressure‐reducing devices (80.4% vs. 0%, p < 0.01) and physical restraints (37.0% vs. 0%, p < 0.01) was more common in this group. 3.3. Identification of Risk Factors 3.3.1. Univariate Logistic Regression Analysis Univariate logistic regression analysis was performed to identify potential risk factors for pressure ulcer development (Table 2 ). Among demographic and clinical variables, age (OR = 1.627, 95% CI: 1.462–1.810, p < 0.01), BMI (OR = 0.189, 95% CI: 0.132–0.269, p < 0.01), albumin (OR = 0.001, 95% CI: 0.000–0.003, p = 0.032), blood glucose (OR = 18.432, 95% CI: 10.174–33.393, p < 0.01), CRP (OR = 2.371, 95% CI: 1.887–2.979, p < 0.01), lymphocyte count (OR = 0.000, 95% CI: 0.000–0.000, p < 0.01), potassium (OR = 0.001, 95% CI: 0.000–0.005, p < 0.01), calcium (OR = 0.000, 95% CI: 0.000–0.000, p < 0.01), creatinine (OR = 1.615, 95% CI: 1.411–1.849, p = 0.031) and blood urea nitrogen (OR = 23.674, 95% CI: 12.793–43.813, p < 0.01) were significantly associated with pressure ulcer risk. TABLE 2. Univariate logistic regression analysis for pressure ulcer risk factors. Variables OR (95% CI) p Demographic characteristics Female sex 0.977 (0.572–1.667) 0.926 Age (years) 1.627 (1.462–1.810) < 0.01 BMI (kg/m 2 ) 0.189 (0.132–0.269) < 0.01 Biochemical indicators Albumin (g/dL) 0.001 (0.000–0.003) 0.032 Haemoglobin (g/L) 0.782 (0.739–0.827) 0.061 Total protein (g/L) 0.166 (0.111–0.249) 0.052 Blood glucose (mmol/L) 18.432 (10.174–33.393) < 0.01 CRP (mg/L) 2.371 (1.887–2.979) < 0.01 Lymphocyte count (10 9/L ) 0.000 (0.000–0.000) < 0.01 Sodium (mmol/L) 0.889 (0.757–1.044) 0.152 Potassium (mmol/L) 0.001 (0.000–0.005) < 0.01 Chlorine (mmol/L) 0.915 (0.702–1.192) 0.508 Calcium (mmol/L) 0.000 (0.000–0.000) < 0.01 Creatinine (μmol/L) 1.615 (1.411–1.849) 0.031 Blood urea nitrogen (mmol/L) 23.674 (12.793–43.813) < 0.01 Disease characteristics Main diagnosis COPD‐related 31.000 (9.483–101.339) < 0.01 Pneumonia‐related 27.067 (3.654–200.558) 0.01 Respiratory failure 1.429 (1.019–17.896) 0.08 Bronchiectasis 10.125 (2.339–43.823) 0.002 Other respiratory diseases Reference History of diabetes 38.099 (9.148–158.678) < 0.01 Cerebrovascular disease 17.391 (6.730–44.935) < 0.01 Peripheral vascular disease 0.01 a Nursing‐related characteristics Urinary incontinence < 0.01 a Faecal incontinence < 0.01 a Long‐term bedridden < 0.01 a Length of hospital stay (days) 2.394 (1.957–2.928) 0.048 Braden total score 0.004 (0.001–0.012) 0.061 Braden subscale scores Sensory perception 0.000 (0.000–0.003) 0.051 Moisture 0.000 (0.000–0.000) 0.042 Activity 0.005 (0.002–0.015) 0.049 Mobility 0.000 (0.000–0.000) 0.032 Nutrition 0.044 (0.020–0.098) 0.068 Friction and shear 9.429 (2.321–28.576) < 0.01 Turning frequency (hours/time) < 0.01 a Use of pressure‐reducing devices < 0.01 a Use of restraints < 0.01 a Open in a new tab a Odds ratios for some variables could not be calculated due to complete separation (0 cell counts in non‐pressure ulcer group), but these factors are significantly associated with pressure ulcer development ( p < 0.001 by Fisher's exact test). Regarding disease characteristics, COPD‐related diagnoses (OR = 31.000, 95% CI: 9.483–101.339, p < 0.01), pneumonia‐related diagnoses (OR = 27.067, 95% CI: 3.654–200.558, p = 0.01) and bronchiectasis (OR = 10.125, 95% CI: 2.339–43.823, p = 0.002) were associated with increased pressure ulcer risk compared to other respiratory diseases. History of diabetes (OR = 38.099, 95% CI: 9.148–158.678, p < 0.01), COPD (OR = 17.391, 95% CI: 6.730–44.935, p < 0.01) and peripheral vascular disease ( p = 0.01) also showed significant associations. All nursing‐related characteristics, including urinary incontinence, faecal incontinence, long‐term bedridden status, length of hospital stay (OR = 2.394, 95% CI: 1.957–2.928, p = 0.048) and most Braden subscale scores, demonstrated significant associations with pressure ulcer development. 3.3.2. LASSO Regression Analysis To address multicollinearity and select the most influential predictors, 29 variables with p < 0.05 in univariate analysis were entered into LASSO regression with 10‐fold cross‐validation. The optimal lambda value was determined, and 17 variables were selected: age, albumin, CRP, sodium, potassium, history of diabetes, COPD, peripheral vascular disease, urinary incontinence, long‐term bedridden status, length of hospital stay, Braden sensory perception score, Braden moisture score, Braden friction/shear score, turning frequency, use of pressure‐reducing devices and use of restraints. Figure 2 illustrates the LASSO coefficient profiles of the 29 variables (Figure 2A ) and the selection of the tuning parameter (lambda) in the LASSO regression (Figure 2B ). The optimal model with a lambda value equal to lambda.1se included 17 non‐zero coefficients. FIGURE 2. Open in a new tab LASSO regression analysis. 3.3.3. Multivariate Logistic Regression Analysis The 17 variables identified by LASSO regression were further analysed using multivariate logistic regression. All variables remained statistically significant independent predictors of pressure ulcer development in the training cohort. Figure 3 presents a forest plot of the multivariate logistic regression analysis, displaying the odds ratios and 95% confidence intervals for each of the 17 independent predictors. FIGURE 3. Open in a new tab Forest plot of multifactor logistic regression analysis. 3.4. Development of the Nomogram Based on the results of multivariate logistic regression and further selection according to ROC curve analysis of individual predictive value, we incorporated the 10 most significant predictors into the final nomogram. These included age, albumin, CRP, sodium, history of diabetes, COPD, peripheral vascular disease, urinary incontinence, length of hospital stay and Braden sensory perception score. Figure 4 displays the nomogram for predicting pressure ulcer risk in Respiratory patients. To use the nomogram, a vertical line is drawn from each variable to the ‘Points’ scale at the top. The sum of these points is then located on the ‘Total Points’ scale, and a vertical line is drawn down to the ‘Risk of Pressure Ulcer’ scale to determine the probability of pressure ulcer development. FIGURE 4. Open in a new tab Nomogram for estimating the risk of pressure ulcers. 3.5. Validation of the Nomogram 3.5.1. Discrimination The nomogram demonstrated excellent discriminative ability, with an AUC of 0.865 (95% CI: 0.816–0.914) in the training cohort and 0.837 (95% CI: 0.783–0.891) in the testing cohort (Figure 5 ). FIGURE 5. Open in a new tab ROC curves of the nomogram prediction model for the training group (A) and testing group (B). 3.5.2. Calibration Calibration curves were constructed to assess the agreement between predicted and observed probabilities of pressure ulcer development. As shown in Figure 6 , the calibration curves for both the training cohort (Figure 6A ) and testing cohort (Figure 6B ) closely aligned with the ideal diagonal line, indicating good calibration of the nomogram. The Hosmer–Lemeshow test yielded non‐significant P ‐values for both the training cohort ( χ 2 = 4.257, p = 0.833) and the testing cohort ( χ 2 = 12.350, p = 0.142), further confirming the good calibration of the model. FIGURE 6. Open in a new tab Calibration curves of the nomogram prediction model for the training group (A) and testing group (B). 3.5.3. Clinical Utility Decision curve analysis was performed to evaluate the clinical utility of the nomogram. Figure 7 presents the decision curves for the training cohort (Figure 7A ) and testing cohort (Figure 7B ). The nomogram showed higher net benefit across a wide range of threshold probabilities compared to the ‘treat all’ or ‘treat none’ strategies, demonstrating its clinical utility in guiding decision‐making for pressure ulcer prevention. FIGURE 7. Open in a new tab DCA of the nomogram prediction model for the training group (A) and testing group (B). 3.5.4. Comparison With Individual Predictors To further validate the superiority of the integrated nomogram, we compared its performance with that of each individual predictor included in the model (Figure 8 ). FIGURE 8. Open in a new tab The nomogram prediction model achieved a higher AUC value compared to each individual predictor incorporated in the model. Figure 9 presents the decision curve analysis comparing the nomogram with each individual predictor. The nomogram consistently demonstrated higher net benefit across a broad range of threshold probabilities, further supporting its clinical superiority over single‐predictor approaches. FIGURE 9. Open in a new tab DCA of the nomogram prediction model and each individual predictor incorporated in the model. In summary, our results demonstrate that the developed nomogram, incorporating 10 clinically relevant predictors, provides excellent discriminative ability, good calibration and superior clinical utility for predicting pressure ulcer risk in respiratory patients compared to individual risk factors alone. 4. Discussion In this study, we developed and validated a machine learning‐based nomogram that uses 10 key variables—age, albumin, CRP, sodium, history of diabetes, COPD, peripheral vascular disease, urinary incontinence, length of hospital stay and Braden sensory perception score—to predict PU risk in respiratory patients. The resulting nomogram demonstrated excellent discrimination, calibration and clinical utility, highlighting its value as a clinical decision support tool. Our findings align with the multifactorial aetiology of PUs, which involves both intrinsic and extrinsic factors [ 28 ], while offering specific insights for respiratory patients. Advanced age was a significant predictor, consistent with previous research showing that age‐related changes in skin integrity, such as decreased elasticity and perfusion, increase susceptibility to pressure damage [ 29 ]. This underscores the need for intensive prevention in elderly respiratory patients. Nutritional and inflammatory markers were also prominent. Lower albumin, reflecting poor nutrition or chronic inflammation, was strongly associated with PU development, as protein deficiency impairs tissue repair. Similarly, elevated CRP, an indicator of systemic inflammation, was a key predictor, likely because chronic inflammation can compromise microcirculation and tissue perfusion. These results highlight the potential of nutritional and anti‐inflammatory interventions in PU prevention. Hyponatremia also emerged as a significant predictor. Although less discussed in PU literature, electrolyte abnormalities can affect cellular function and tissue integrity [ 30 ]. Hyponatremia may lead to cellular edema and altered tissue pressure [ 31 ], suggesting that monitoring and correcting electrolytes is an important, perhaps underappreciated, aspect of PU prevention. Comorbidities like diabetes, COPD and peripheral vascular disease were strong predictors, as they share pathophysiological mechanisms that compromise tissue perfusion and oxygen delivery. The inclusion of COPD is noteworthy, as its associated hypoxemia, systemic inflammation and impaired mobility create a unique vulnerability to pressure‐related tissue damage. Among care‐related factors, urinary incontinence and length of hospital stay were significant. Urinary incontinence compromises skin integrity by causing maceration and irritation [ 32 ], while prolonged hospitalization increases risk through extended pressure exposure and the cumulative effects of immobility and acute illness. This emphasizes the importance of continence management and minimizing hospital stay duration. Notably, our analysis identified the Braden sensory perception subscore as the only component of the Braden Scale included in our final model. This suggests that for respiratory patients—who may have altered consciousness from hypoxemia, hypercapnia or sedation—impaired sensory perception and the inability to respond to pressure‐related discomfort may be the most critical risk factor. This highlights the need for interventions that specifically target this deficit, such as more frequent repositioning schedules. The superior performance of our integrated nomogram over individual predictors underscores the multifactorial nature of PU development and the limitations of single‐factor assessments. By capturing the interplay between multiple factors, our model provides a more comprehensive and accurate risk profile, advancing beyond traditional methods. From a clinical perspective, our nomogram has several advantages. It uses readily available clinical data, facilitating implementation without specialized equipment. Its visual format enables rapid risk estimation at the bedside, and by providing individualized probabilities rather than broad categories, it allows for more nuanced intervention planning. The model's strong performance in the testing cohort also suggests good generalizability. Several limitations of our study should be noted. First, its retrospective design may introduce selection bias. Second, being a single‐center study, its generalizability may be limited. Third, larger multicenter studies are needed for more robust validation. Fourth, we did not assess the clinical impact of implementing the nomogram. Fifth, a significant limitation arises from the baseline differences between our cohorts, particularly the complete separation observed in variables such as ‘long‐term bedridden status’. While this contributes to the model's high discriminative ability, it may also indicate that the model is heavily influenced by this single dominant factor, potentially masking the more subtle predictive contributions of other variables. Future research should aim to include a more balanced control group, such as long‐term bedridden patients who did not develop pressure ulcers, to build a more nuanced model. Sixth, our model does not capture dynamic changes in risk factors during hospitalization. Finally, its performance in other patient populations requires further investigation. Future research should include prospective validation, assessment of the model's impact on clinical outcomes and the development of dynamic prediction tools. Exploring novel biomarkers to further enhance predictive accuracy is also a promising area for future work. In conclusion, our machine learning‐based nomogram shows excellent performance in predicting PU risk among respiratory patients. By integrating multiple risk factors into a user‐friendly visual tool, it offers a practical method for identifying high‐risk individuals. Implementation of this model in clinical practice could contribute to reducing PU incidence, improving patient outcomes and optimizing resource allocation in respiratory care settings. Funding This work was supported by the Key Research and Development Project of Xuzhou Program (KC20063). Ethics Statement The study protocol was approved by the Ethics Committee of The Affiliated Hospital of Xuzhou Medical University (Ethics No. XYFY2022‐KL083). The study was conducted in accordance with the Declaration of Helsinki and all relevant institutional guidelines. The requirement for informed consent was waived by the Ethics Committee of The Affiliated Hospital of Xuzhou Medical University due to the retrospective nature of the study, which involved the analysis of anonymized patient data. Conflicts of Interest The authors declare no conflicts of interest. Data Availability Statement The data that support the findings of this study are available on request from the corresponding author. The data are not publicly available due to privacy or ethical restrictions. References 1. Mervis J. S. and Phillips T. J., “Pressure Ulcers: Pathophysiology, Epidemiology, Risk Factors, and Presentation,” Journal of the American Academy of Dermatology 81, no. 4 (2019): 881–890. [ DOI ] [ PubMed ] [ Google Scholar ] 2. Jaul E., “Assessment and Management of Pressure Ulcers in the Elderly: Current Strategies,” Drugs & Aging 27, no. 4 (2010): 311–325. [ DOI ] [ PubMed ] [ Google Scholar ] 3. Anthony D., Alosoumi D., and Safari R., “Prevalence of Pressure Ulcers in Long‐Term Care: A Global Review,” Journal of Wound Care 28, no. 11 (2019): 702–709. [ DOI ] [ PubMed ] [ Google Scholar ] 4. Aloweni F., Ang S. Y., Fook‐Chong S., et al., “A Prediction Tool for Hospital‐Acquired Pressure Ulcers Among Surgical Patients: Surgical Pressure Ulcer Risk Score,” International Wound Journal 16, no. 1 (2019): 164–175. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 5. Manzano F., Navarro M. J., Roldán D., et al., “Pressure Ulcer Incidence and Risk Factors in Ventilated Intensive Care Patients,” Journal of Critical Care 25, no. 3 (2010): 469–476. [ DOI ] [ PubMed ] [ Google Scholar ] 6. Zajac K. K., Schubauer K., and Simman R., “The Unavoidable Pressure Injury/Ulcer: A Review of Skin Failure in Critically Ill Patients,” Journal of Wound Care 33, no. Sup9 (2024): S18–S22. [ Google Scholar ] 7. Oliveira F. V., Coltro P. S., Nunes A. A., Biaziolo C. F. B., Ferreira M. C., and Farina‐Junior J. A., “Comparative Cohort Analysis of Pressure Ulcer/Injury in Intensive Care Unit Patients Before and During the COVID‐19 Pandemic,” Journal of Plastic, Reconstructive & Aesthetic Surgery 85 (2023): 98–103. [ Google Scholar ] 8. Schallom M., Cracchiolo L., Falker A., et al., “Pressure Ulcer Incidence in Patients Wearing Nasal‐Oral Versus Full‐Face Noninvasive Ventilation Masks,” American Journal of Critical Care 24, no. 4 (2015): 349–357. [ DOI ] [ PubMed ] [ Google Scholar ] 9. Magnan M. A. and Maklebust J., “The Nursing Process and Pressure Ulcer Prevention: Making the Connection,” Advances in Skin & Wound Care 22, no. 2 (2009): 83–94. [ DOI ] [ PubMed ] [ Google Scholar ] 10. Hill J. E., Edney S., Hamer O., Williams A., and Harris C., “Interventions for the Treatment and Prevention of Pressure Ulcers,” British Journal of Community Nursing 27, no. Sup6 (2022): S28–S36. [ Google Scholar ] 11. Demarré L., Van Lancker A., Van Hecke A., et al., “The Cost of Prevention and Treatment of Pressure Ulcers: A Systematic Review,” International Journal of Nursing Studies 52, no. 11 (2015): 1754–1774. [ DOI ] [ PubMed ] [ Google Scholar ] 12. Oozageer Gunowa N., Oti K. A., and Jackson D., “Early Identification of Pressure Injuries in People With Dark Skin Tones: Qualitative Perspectives From Community‐Based Patients and Their Carers,” Journal of Clinical Nursing 33, no. 11 (2024): 4434–4444. [ DOI ] [ PubMed ] [ Google Scholar ] 13. McCray S. and Donaldson A., “Early Identification, Intervention, and Prevention of Hospital‐Acquired Pressure Injuries Using a Nurse‐Driven Pressure Injury Prevention Program,” Clinical Nurse Specialist 38, no. 5 (2024): 210–220. [ DOI ] [ PubMed ] [ Google Scholar ] 14. Jansen R. C. S., Silva K. B. A., and Moura M. E. S., “Braden Scale in Pressure Ulcer Risk Assessment,” Revista Brasileira de Enfermagem 73, no. 6 (2020): e20190413. [ DOI ] [ PubMed ] [ Google Scholar ] 15. Chung M. L., Widdel M., Kirchhoff J., et al., “Risk Factors for Pressure Ulcers in Adult Patients: A Meta‐Analysis on Sociodemographic Factors and the Braden Scale,” Journal of Clinical Nursing 32, no. 9–10 (2023): 1979–1992. [ DOI ] [ PubMed ] [ Google Scholar ] 16. Wei M., Wu L., Chen Y., Fu Q., Chen W., and Yang D., “Predictive Validity of the Braden Scale for Pressure Ulcer Risk in Critical Care: A Meta‐Analysis,” Nursing in Critical Care 25, no. 3 (2020): 165–170. [ DOI ] [ PubMed ] [ Google Scholar ] 17. Moore Z. E. and Patton D., “Risk Assessment Tools for the Prevention of Pressure Ulcers,” Cochrane Database of Systematic Reviews 1, no. 1 (2019): CD006471. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 18. Baris N., Karabacak B. G., and Alpar Ş. E., “The Use of the Braden Scale in Assessing Pressure Ulcers in Turkey: A Systematic Review,” Advances in Skin & Wound Care 28, no. 8 (2015): 349–357. [ DOI ] [ PubMed ] [ Google Scholar ] 19. Deo R. C., “Machine Learning in Medicine,” Circulation 132, no. 20 (2015): 1920–1930. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 20. Uddin S., Khan A., Hossain M. E., and Moni M. A., “Comparing Different Supervised Machine Learning Algorithms for Disease Prediction,” BMC Medical Informatics and Decision Making 19, no. 1 (2019): 281. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 21. Wang Y., Ni B., Xiao Y., Lin Y., Jiang Y., and Zhang Y., “Application of Machine Learning Algorithms to Construct and Validate a Prediction Model for Coronary Heart Disease Risk in Patients With Periodontitis: A Population‐Based Study,” Frontiers in Cardiovascular Medicine 10 (2023): 1296405. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 22. Zhu T. and Tao C., “Prediction Models With Multiple Machine Learning Algorithms for POPs: The Calculation of PDMS‐Air Partition Coefficient From Molecular Descriptor,” Journal of Hazardous Materials 423 (2022): 127037. [ DOI ] [ PubMed ] [ Google Scholar ] 23. Wu J., Zhang H., Li L., et al., “A Nomogram for Predicting Overall Survival in Patients With Low‐Grade Endometrial Stromal Sarcoma: A Population‐Based Analysis,” Cancer Commun (Lond) 40, no. 7 (2020): 301–312. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 24. Kailavasan M., Berridge C., Yuan Y., Turner A., Donaldson J., and Biyani C. S., “A Systematic Review of Nomograms Used in Urolithiasis Practice to Predict Clinical Outcomes in Paediatric Patients,” Journal of Pediatric Urology 18, no. 4 (2022): 448–462. [ DOI ] [ PubMed ] [ Google Scholar ] 25. Jalali A., Alvarez‐Iglesias A., Roshan D., and Newell J., “Visualising Statistical Models Using Dynamic Nomograms,” PLoS One 14, no. 11 (2019): e0225253. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 26. Zhang T., Lai M., Wei Y., et al., “Nomograms for Predicting Overall Survival and Cancer‐Specific Survival in Patients With Invasive Micropapillary Carcinoma: Based on the SEER Database,” Asian Journal of Surgery 46, no. 9 (2023): 3734–3740. [ DOI ] [ PubMed ] [ Google Scholar ] 27. Zhou Y., Lin C., Zhu L., Zhang R., Cheng L., and Chang Y., “Nomograms and Scoring System for Forecasting Overall and Cancer‐Specific Survival of Patients With Prostate Cancer,” Cancer Medicine 12, no. 3 (2023): 2600–2613. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 28. Leigh I. H. and Bennett G., “Pressure Ulcers: Prevalence, Etiology, and Treatment Modalities. A Review,” American Journal of Surgery 167, no. 1A (1994): 25S–30S. [ DOI ] [ PubMed ] [ Google Scholar ] 29. Mathus‐Vliegen E. M., “Old Age, Malnutrition, and Pressure Sores: An Ill‐Fated Alliance,” Journals of Gerontology. Series A, Biological Sciences and Medical Sciences 59, no. 4 (2004): 355–360. [ DOI ] [ PubMed ] [ Google Scholar ] 30. Tatsumi H., “Enteral Tolerance in Critically Ill Patients,” Journal of Intensive Care 7 (2019): 30. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 31. Seethapathy H., Rusibamayila N., Chute D. F., et al., “Hyponatremia and Other Electrolyte Abnormalities in Patients Receiving Immune Checkpoint Inhibitors,” Nephrology, Dialysis, Transplantation 36, no. 12 (2021): 2241–2247. [ Google Scholar ] 32. Irwin G. M., “Urinary Incontinence,” Primary Care 46, no. 2 (2019): 233–242. [ DOI ] [ PubMed ] [ Google Scholar ] Associated Data This section collects any data citations, data availability statements, or supplementary materials included in this article. Data Availability Statement The data that support the findings of this study are available on request from the corresponding author. The data are not publicly available due to privacy or ethical restrictions. Articles from International Wound Journal are provided here courtesy of Wiley ACTIONS View on publisher site PDF (1.4 MB) Cite Collections Permalink PERMALINK Copy RESOURCES Similar articles Cited by other articles Links to NCBI Databases Cite Copy Download .nbib .nbib Format: AMA APA MLA NLM Add to Collections Create a new collection Add to an existing collection Name your collection * Choose a collection Unable to load your collection due to an error Please try again Add Cancel Follow NCBI NCBI on X (formerly known as Twitter) NCBI on Facebook NCBI on LinkedIn NCBI on GitHub NCBI RSS feed Connect with NLM NLM on X (formerly known as Twitter) NLM on Facebook NLM on YouTube National Library of Medicine 8600 Rockville Pike Bethesda, MD 20894 Web Policies FOIA HHS Vulnerability Disclosure Help Accessibility Careers NLM NIH HHS USA.gov Back to Top