ABSTRACT
Aims
To systematically evaluate existing prediction models for medical adhesive‐related skin injury and assess their potential contribution to nurse‐led risk assessment and preventive care planning.
Methods
A systematic review and meta‐analysis was conducted. Seven databases were searched from inception to 25 October 2025. Studies developing prediction models for medical adhesive‐related skin injury in adults were assessed using the Prediction Model Risk of Bias Assessment Tool. Random‐effects meta‐analyses were used to synthesize prevalence estimates, model discrimination and common predictors.
Results
Twelve studies comprising 14 prediction models were included. Reported area under the curve or C‐index values ranged from 0.789 to 0.960. In the nine studies included in the discrimination meta‐analysis, the pooled area under the curve was 0.87 (95% confidence interval: 0.84–0.90). All studies had a high overall risk of bias, and none performed external validation. The pooled prevalence of medical adhesive‐related skin injury was 22.0% (95% confidence interval: 18.5%–25.4%). Age over 60 years, a history of medical adhesive‐related skin injury, allergy history, edema, study‐defined hypoproteinemia, and moist skin were associated with higher odds, whereas higher Braden Scale scores were associated with lower odds.
Conclusions
Existing models are not ready for stand‐alone clinical use because of their high risk of bias and lack of external validation. Further well‐designed studies are needed to develop and externally validate clinically useful prediction models for medical adhesive‐related skin injury.
Implications for Nursing Practice
The identified factors may help nurses prioritise closer skin assessment, surveillance, and individualised preventive care. They should be considered within a comprehensive nursing assessment rather than used as a validated risk score or intervention threshold.
No Patient or Public Contribution
None because this review used previously published evidence.
Registration
PROSPERO registration number CRD420251179067.
Keywords: medical adhesive‐related skin injury, meta‐analysis, nursing assessment, prediction model, preventive nursing care
1. Introduction
Medical adhesives are widely used in contemporary healthcare for dressing fixation, catheter maintenance, electrode monitoring, and ostomy care (Shi et al. 2025; Wang et al. 2024). However, improper selection, application, or removal of these adhesives can compromise the skin's surface, resulting in medical adhesive‐related skin injury (MARSI) (McNichol et al. 2013; Shi et al. 2025; Wang et al. 2024). McNichol et al. (2013) defined MARSI as skin damage, such as erythema, erosion, blistering, or tearing, that persists for at least 30 min after adhesive removal.
MARSI compromises the skin barrier and may lead to pain, infection, delayed wound healing, reduced quality of life, and increased nursing workload (Wang et al. 2024). Reported MARSI incidence estimates range from 25% to 42% across adult inpatient and critical care settings (Frota et al. 2023; Wang et al. 2024).
Older age, poor nutritional status, chronic illness, dehydration, and pre‐existing dermatologic conditions may weaken skin integrity (Frota et al. 2023; McNichol et al. 2013; Wang et al. 2024). Prolonged hospitalisation and repeated removal or replacement of adhesives used to secure medical devices may increase MARSI risk (Frota et al. 2023; Y. Zhang et al. 2020). Moisture imbalance, radiation therapy, and medications such as corticosteroids, anticoagulants, immunosuppressants, and some antineoplastic agents may further increase susceptibility to mechanical or irritant injury (Du and Liu 2024; McNichol et al. 2013; Zaballa Canive et al. 2026).
Because nurses routinely apply and remove medical adhesives and monitor related skin changes, they play a central role in MARSI prevention and early identification. Key nurse‐led strategies include systematic skin assessment, appropriate adhesive selection and application, protection of vulnerable skin, regular monitoring of adhesive sites, and safe removal techniques (Du and Liu 2024; McNichol et al. 2013).
Risk prediction models integrate multiple patient‐related and treatment‐related factors to estimate an individual's likelihood of developing MARSI (Grant et al. 2018). In nursing practice, adequately validated models may help identify patients at increased risk before or during adhesive use. Their estimates could be considered alongside clinical assessment when planning closer monitoring and individualised preventive care, including appropriate adhesive selection and protection of vulnerable skin (de Faria et al. 2022; McNichol et al. 2013). Published MARSI models differ in their selected predictors, reported performance, and validation methods (Li et al. 2023; Wu et al. 2024; S. Wu et al. 2025). Evidence directly comparing logistic regression with machine‐learning approaches for MARSI prediction remains limited (S. Wu et al. 2025). This systematic review and meta‐analysis therefore aimed to evaluate the characteristics, methodological quality, predictive performance, and common predictors of existing MARSI risk prediction models. The findings were also considered in relation to their potential relevance for nurse‐led risk assessment, clinical decision‐making, and preventive care planning.
2. Methods
This systematic review was conducted according to a prospectively registered protocol and reported in accordance with the PRISMA 2020 statement (Page et al. 2021). The protocol was registered with PROSPERO under registration number CRD420251179067.
2.1. Search Strategy
A comprehensive search was conducted in seven databases: CNKI, Wanfang, SinoMed, PubMed, Web of Science, the Cochrane Library, and CINAHL. Each database was searched from inception to 25 October 2025. The search strategy combined subject headings and free‐text terms, including terms such as “medical adhesive,” ‘adhesive‐related skin injury,’ ‘tape,’ ‘dressing fixation,’ ‘catheter securement,’ ‘patch,’ ‘ostomy appliance,’ ‘skin injury,’ ‘skin tear,’ ‘epidermal stripping,’ ‘epidermal damage,’ ‘contact dermatitis,’ ‘blister,’ ‘MARSI,’ ‘skin reaction,’ ‘prediction,’ ‘prediction model,’ ‘prediction tool,’ ‘screening tool,’ ‘risk prediction,’ ‘risk assessment,’ ‘risk score,’ ‘nomogram,’ and ‘machine learning.’ Detailed database‐specific search strategies and search dates are provided in Table S1.
2.2. Inclusion and Exclusion Criteria
Studies were eligible if they used a prospective or retrospective cohort, case–control, or cross‐sectional design. Participants had to be aged 18 years or older. Studies were also required to develop a MARSI risk prediction model and report at least one measure of model performance: area under the receiver operating characteristic curve (AUC), sensitivity, specificity, calibration, or discrimination. Full‐text academic dissertations were eligible if they otherwise met these criteria and provided sufficient information on model development and performance. Studies were excluded if they were reviews, protocols, diagnostic accuracy studies, or reports for which the full text was unavailable. Grey literature other than eligible full‐text dissertations, including conference abstracts, preprints, institutional reports, and unpublished studies, was excluded. We also excluded expert‐consensus‐based risk assessment scales not derived from patient‐level multivariable modelling and studies published in languages other than Chinese or English.
2.3. Study Selection and Data Extraction
Two researchers (Yujing Gu, Zhouzhou Lu) independently managed and deduplicated the retrieved records using EndNote. Titles, abstracts, and full texts were screened in stages according to the predefined inclusion and exclusion criteria. Key information was then extracted using the standardised CHARMS checklist for critical appraisal and data extraction in predictive modelling studies (Moons et al. 2014). Disagreements were resolved through discussion with a third researcher (Jun Xie). Extracted data included first author, publication year, study design, sample size, study population, modelling sample size, data source, MARSI prevalence, modelling methods, number of candidate variables, variable selection methods, and outcome events. The complete study‐level data extraction results are presented in Tables 1, 2, 3 of the main manuscript.
TABLE 1.
Overview of basic data of the included studies (n = 12).
| Authors (year) | Study design | Research subject | Modelling sample size | Data source | MARSI prevalence rate (%) |
|---|---|---|---|---|---|
| Lai (2022) | Cross‐sectional study | PICC catheterization patients | 806 | Multicentre | 21.46 |
| Y. Wu et al. (2024) | Retrospective study | Patients with infusion ports for malignancy | 387 | Single‐centre | 24.55 |
| S. Wu et al. (2025) | Retrospective study | Lung cancer patients with infusion ports | 500 | Single‐centre | 31.00 |
| Huang et al. (2020) | Retrospective study | Colon cancer patients with PICC lines | 158 | Single‐centre | 31.01 |
| Li et al. (2023) | Retrospective study | Patients with CVADs | 1172 | Single‐centre | 24.06 |
| Zhu et al. (2025) | Retrospective study | Elderly patients undergoing surgery | 249 | Single‐centre | 16.87 |
| Feng et al. (2024) | Retrospective study | ICU patients | 108 | Single‐centre | 21.30 |
| Z. Li (2024) | Cross‐sectional study | ICU patients | 321 | Single‐centre | 21.50 |
| Xie et al. (2024) | Retrospective study | ICU patients | 261 | Single‐centre | 14.94 |
| Y. Zhang (2025) | Retrospective study | ICU patients | 128 | Single‐centre | 18.75 |
| Long (2022) | Retrospective study | Patients with PICC lines | 197 | Single‐centre | 28.43 |
| Liu et al. (2023) | Retrospective study | Patients with PICC lines | 350 | Single‐centre | 11.71 |
Abbreviations: CVAD, central venous access device; ICU, intensive care unit; PICC, peripherally inserted central catheter.
TABLE 2.
Characteristics of MARSI risk prediction model development (n = 12).
| Authors | Modelling methods | Number of candidate variables | Predictor selection method | Number of outcome events | EPV | Model presentation format | Missing data handling method |
|---|---|---|---|---|---|---|---|
| Lai (2022) | Logistic regression | 38 | Univariate analysis | 173 | 4.55 | Regression equation | Deletion |
| Y. Wu et al. (2024) | Logistic regression | 21 | Univariate analysis | 95 | 4.52 | Nomogram | Deletion |
| S. Wu et al. (2025) |
(1) Logistic regression (2) CART (3) Random forest model |
35 | Univariate analysis | 155 | 4.43 | Tree diagram | — |
| Huang et al. (2020) | Logistic regression | 32 | Univariate analysis | 49 | 1.53 | Regression equation | Deletion |
| Li et al. (2023) | — | 8 | Univariate analysis | 282 | 35.25 | Nomogram | Deletion |
| Zhu et al. (2025) | Logistic regression | 16 | Univariate analysis | 42 | 2.63 | Nomogram | — |
| Feng et al. (2024) | Logistic regression | 19 | Univariate analysis | 23 | 1.21 | Nomogram | Deletion |
| Z. Li (2024) | Logistic regression | 19 | Univariate analysis | 69 | 3.63 | Nomogram | — |
| Xie et al. (2024) | Logistic regression | 23 | Univariate analysis | 39 | 1.70 | Nomogram | — |
| Y. Zhang (2025) | Logistic regression | 14 | Univariate analysis | 24 | 1.71 | Nomogram | — |
| Long (2022) | Logistic regression | 18 | Univariate analysis | 56 | 3.11 | Nomogram | Deletion |
| Liu et al. (2023) | Logistic regression | 25 | Univariate analysis | 41 | 1.64 | Nomogram | Deletion |
Abbreviations: CART, classification and regression tree; EPV, the ratio of events to predictor variables —, not reported.
TABLE 3.
MARSI risk factors, model performance, and validation (n = 12).
| Authors | Predictors | Calibration assessment | AUC/C‐index (95% CI) | Sensitivity/Specificity | Validation method |
|---|---|---|---|---|---|
| Lai (2022) | Nine predictors: age, history of alcohol use, prior MARSI, allergy history, surgical history, corticosteroid use, edema, puncture site infection, and catheter dwell time. | H‐L + Calibration plot | 0.909 (0.883–0.936) | 80.9%/86.4% | — |
| Y. Wu et al. (2024) | Nine predictors: BMI, allergy history, history of radiotherapy, use of targeted therapy, use of immunotherapy, white blood cell count, haemoglobin, albumin, and total protein. | H‐L + Calibration curve | 0.816 (0.766–0.866) | — | Internal validation |
| S. Wu et al. (2025) | (1) Logistic regression (4 predictors): skin toxicity, Braden score, prior MARSI, and adhesive film moulding pressure. (2) CART and random forest model (7 predictors): skin toxicity, Braden score, prior MARSI, adhesive film moulding pressure, hypertension, dressing type, and history of medical adhesive allergy. | — | (1) Logistic regression: 0.83 (2) CART: 0.87 (3) Random forest: 0.86 | (1) Logistic regression: 54%/97% (2) CART: 59%/98% (3) Random forest: 61%/97% | Internal validation |
| Huang et al. (2020) | Five predictors: female sex, age, allergy history, history of radiotherapy, and albumin. | — | 0.830 (0.700–0.900) | (1) Modelling cohort: 85.7%/68.8%. (2) Validation cohort: 83.7%/66.7% | Internal validation |
| Li et al. (2023) | Five predictors: prior MARSI, requirement for total parenteral nutrition, other catheter‐related complications, allergy history, and PICC placement. | Calibration curve | C‐index: 0.96 | — | — |
| Zhu et al. (2025) | Five predictors: age, low preoperative serum albumin, use of adhesive tension method, presence of more than three skin issues, and use of dressings or other categories. | H‐L + Calibration curve | 0.931 (0.883–0.980) | — | Internal validation |
| Feng et al. (2024) | Six predictors: age, BMI, Braden score, skin moisture, NRS2002 score, and APACHE II score. | H‐L + Calibration curve | — | 78.94%/97.18% | Internal validation |
| Z. Li (2024) | Five predictors: age, edema, diabetes, skin moisture, and Braden score. | H‐L + Calibration curve | 0.855 (0.807–0.902) | (1) Modelling cohort: 84.1%/71.8% (2) Validation cohort: 69.0%/79.4% | Internal validation |
| Xie et al. (2024) | Eight predictors: age, mechanical ventilation, APACHE II score, allergy history, fever, hypoproteinemia, edema, and use of sedatives. | Calibration curve | 0.85 (0.824–0.886) | 76.9%/76.1% | Internal validation |
| Y. Zhang (2025) | Four predictors: age, diabetes, edema, and hypoproteinemia. | Calibration curve | C‐index: 0.919 (0.850–0.988) | — | Internal validation |
| Long (2022) | Six predictors: edema, prior MARSI, allergy history, season, skin moisture, and patient self‐care ability. | Calibration curve | 0.789 (0.721–0.857) | 75.0%/73.8% | Internal validation |
| Liu et al. (2023) | Five predictors: age, maintenance method, moist and sweaty skin, dry skin, and type of medical adhesive. | Calibration curve | 0.901 (0.840–0.962) | 95.0%/98.3% | Internal validation |
Abbreviations: APACHE II score, acute physiology and chronic health evaluation; BMI, body mass index; CART, classification and regression tree; H‐L, Hosmer‐Lemeshow test; NRS2002, nutrition risk screening 2002; PICC, peripherally inserted central catheter; — not reported.
2.4. Quality Assessment
The methodological quality and applicability of the included prediction models were evaluated using PROBAST (Moons et al. 2019), which assesses four risk‐of‐bias domains (participants, predictors, outcome, and analysis) and three applicability domains (participants, predictors, and outcome). Each domain was rated using the PROBAST signalling questions. A domain was considered low risk when all questions were answered ‘yes’ or ‘probably yes’. It was rated as unclear when at least one question was answered ‘no information’ and all remaining questions were answered ‘yes’ or ‘probably yes’. Any answer of ‘no’ or ‘probably no’ resulted in a high‐risk rating. Overall risk of bias was classified as low when all domains were rated as low risk and as high when any domain was rated as high risk. Studies with at least one unclear domain and no high‐risk domains were classified as having an unclear overall risk of bias. Applicability was assessed per domain and rated as low concern, high concern, or unclear, with the overall rating taking the most unfavourable judgement. Two researchers (Yujing Gu, Zhouzhou Lu) independently assessed each study and compared their ratings. Disagreements were resolved through discussion with a third researcher (Jun Xie).
2.5. Statistical Analysis
Meta‐analyses of MARSI prevalence, model discrimination, and common predictive factors were performed using Stata 17.0. Prevalence estimates were summarized as proportions with 95% confidence intervals (CIs). AUC values with sufficient variance information were pooled to summarize discrimination; the pooled AUC was not interpreted as evidence of calibration or clinical utility. Predictive factors reported in at least three studies with extractable data were summarized as odds ratios (ORs) with 95% CIs. Because clinical and methodological heterogeneity was anticipated across populations, adhesive exposures, predictor definitions, and modelling approaches, random‐effects models were used. Statistical heterogeneity was evaluated using Cochran's Q test and the I2 statistic. Egger's test was used to explore small‐study effects for the prevalence analysis, which included at least 10 studies. All tests were two‐sided, and p < 0.05 was considered statistically significant. Owing to the small number of studies available for individual predictors, subgroup analysis, meta‐regression, and formal assessment of small‐study effects were not undertaken for the predictor syntheses.
3. Results
3.1. Study Selection
Figure 1 presents an overview of the study selection process and results. The initial search identified 1222 records, of which 212 duplicates were removed. After title and abstract screening, 967 records were excluded, leaving 43 reports for full‐text assessment. Of these, 31 reports were excluded: 22 did not develop a prediction model, two did not meet the population criteria, six did not evaluate MARSI as the outcome, and one developed a consensus‐based risk assessment scale without statistical prediction‐model development. Ultimately, 12 studies (Feng et al. 2024; Huang et al. 2020; Lai 2022; Li et al. 2023; Z. Li 2024; Liu et al. 2023; Long 2022; S. Wu et al. 2025; Y. Wu et al. 2024; Xie et al. 2024; Y. Zhang 2025; Zhu et al. 2025) encompassing 14 prediction models were included.
FIGURE 1.
Flowchart of literature search and selection.
3.2. Study Characteristics
Tables 1, 2, 3 summarize the characteristics of the included studies and MARSI prediction models. Twelve studies published between 2020 and 2025 were included. Eleven (Feng et al. 2024; Huang et al. 2020; Lai 2022; Z. Li 2024; Liu et al. 2023; Long 2022; S. Wu et al. 2025; Y. Wu et al. 2024; Xie et al. 2024; Y. Zhang 2025; Zhu et al. 2025) were published in Chinese, and one (Li et al. 2023) was published in English. Ten studies were retrospective (Feng et al. 2024; Huang et al. 2020; Li et al. 2023; Liu et al. 2023; Long 2022; S. Wu et al. 2025; Y. Wu et al. 2024; Xie et al. 2024; Y. Zhang 2025; Zhu et al. 2025), and two were cross‐sectional (Lai 2022; Z. Li 2024). Eleven studies (Y. Wu et al. 2024; S. Wu et al. 2025; Huang et al. 2020; Li et al. 2023; Zhu et al. 2025; Feng et al. 2024; Z. Li 2024; Xie et al. 2024; Y. Zhang 2025; Long 2022; Liu et al. 2023) were conducted at a single center. Sample sizes ranged from 108 to 1172 participants, and MARSI prevalence ranged from 11.71% to 31.01%.
The same eligibility criteria were applied to Chinese‐ and English‐language records, without geographic restrictions. Among the English‐language MARSI records retrieved, most addressed prevalence, risk factors, prevention, or consensus recommendations rather than the development or validation of multivariable prediction models.
The 12 studies reported 14 models, most of which were developed using logistic regression (Feng et al. 2024; Huang et al. 2020; Lai 2022; Z. Li 2024; Liu et al. 2023; Long 2022; S. Wu et al. 2025; Y. Wu et al. 2024; Xie et al. 2024; Y. Zhang 2025; Zhu et al. 2025). Reported AUC or C‐index values ranged from 0.789 to 0.960. Ten studies conducted internal validation (Feng et al. 2024; Huang et al. 2020; Z. Li 2024; Liu et al. 2023; Long 2022; S. Wu et al. 2025; Y. Wu et al. 2024; Xie et al. 2024; Y. Zhang 2025; Zhu et al. 2025), whereas none performed external validation. Reporting of calibration, missing‐data handling, sensitivity, and specificity varied across studies.
3.3. Results of Quality Assessment
All included studies (Feng et al. 2024; Huang et al. 2020; Lai 2022; Li et al. 2023; Z. Li 2024; Liu et al. 2023; Long 2022; S. Wu et al. 2025; Y. Wu et al. 2024; Xie et al. 2024; Y. Zhang 2025; Zhu et al. 2025) were judged to have a high overall risk of bias (Table 4). In the participant domain, 11 studies were rated as high risk (Feng et al. 2024; Huang et al. 2020; Li et al. 2023; Z. Li 2024; Liu et al. 2023; Long 2022; S. Wu et al. 2025; Y. Wu et al. 2024; Xie et al. 2024; Y. Zhang 2025; Zhu et al. 2025). These ratings were mainly attributable to retrospective or single‐center designs and limited sample representativeness.
TABLE 4.
Bias risk and applicability assessment.
| Authors | Risk of bias | Applicability | Overall | ||||||
|---|---|---|---|---|---|---|---|---|---|
| Participants | Predictors | Outcome | Analysis | Participants | Predictors | Outcome | Risk of bias | Applicability | |
| Lai (2022) | + | ? | ? | − | + | + | + | − | + |
| Y. Wu et al. (2024) | − | ? | ? | − | + | + | + | − | + |
| S. Wu et al. (2025) | − | ? | ? | − | + | + | + | − | + |
| Huang et al. (2020) | − | ? | ? | − | + | + | + | − | + |
| Li et al. (2023) | − | ? | ? | − | + | + | + | − | + |
| Zhu et al. (2025) | − | ? | ? | − | + | + | + | − | + |
| Feng et al. (2024) | − | ? | ? | − | + | + | + | − | + |
| Z. Li (2024) | − | + | ? | − | + | + | + | − | + |
| Xie et al. (2024) | − | ? | ? | − | + | + | + | − | + |
| Y. Zhang (2025) | − | ? | ? | − | + | + | + | − | + |
| Long (2022) | − | ? | ? | − | + | + | + | − | + |
| Liu et al. (2023) | − | ? | ? | − | + | + | + | − | + |
Note: “+”, low risk of bias/low concern regarding applicability; “−”, high risk of bias/high concern regarding applicability; “?”, unclear.
Risk of bias in the predictor domain was unclear in 11 studies (Feng et al. 2024; Huang et al. 2020; Lai 2022; Li et al. 2023; Liu et al. 2023; Long 2022; S. Wu et al. 2025; Wu et al. 2024; Xie et al. 2024; Zhang 2025; Zhu et al. 2025). Risk of bias in the outcome domain was unclear in all 12 studies (Feng et al. 2024; Huang et al. 2020; Lai 2022; Li et al. 2023; Z. Li 2024; Liu et al. 2023; Long 2022; S. Wu et al. 2025; Y. Wu et al. 2024; Xie et al. 2024; Y. Zhang 2025; Zhu et al. 2025). The main concerns were insufficient reporting of operational definitions, assessor blinding, and the relative timing of predictor and outcome assessments.
All 12 studies were rated as high risk in the analysis domain (Feng et al. 2024; Huang et al. 2020; Lai 2022; Li et al. 2023; Z. Li 2024; Liu et al. 2023; Long 2022; S. Wu et al. 2025; Y. Wu et al. 2024; Xie et al. 2024; Y. Zhang 2025; Zhu et al. 2025). The main concerns were limited sample sizes, categorization of continuous predictors, inadequate handling of missing data, univariate predictor screening, incomplete performance assessment, and insufficient validation. Applicability concerns were low across all studies (Feng et al. 2024; Huang et al. 2020; Lai 2022; Li et al. 2023; Z. Li 2024; Liu et al. 2023; Long 2022; S. Wu et al. 2025; Y. Wu et al. 2024; Xie et al. 2024; Y. Zhang 2025; Zhu et al. 2025).
3.4. Meta‐Analysis Results
A pooled analysis of the 12 included studies revealed substantial heterogeneity in MARSI prevalence across studies (I 2 = 86.9%, p < 0.001), and a random‐effects model was applied for synthesis. Meta‐analysis indicated a pooled MARSI prevalence of 22.0% (95% CI: 18.5%–25.4%) (Figure 2). Egger's test yielded t = 0.31 (p = 0.766), indicating no evidence of small‐study effects.
FIGURE 2.
Forest plot of MARSI prevalence.
Three studies lacked AUC or C‐index confidence intervals or other information required for pooling (Feng et al. 2024; Li et al. 2023; S. Wu et al. 2025). The remaining nine studies were included in the discrimination meta‐analysis (Huang et al. 2020; Lai 2022; Z. Li 2024; Liu et al. 2023; Long 2022; Y. Wu et al. 2024; Xie et al. 2024; Y. Zhang 2025; Zhu et al. 2025). Heterogeneity was observed (I 2 = 71.7%, p < 0.001), and a random‐effects model was used for pooling. The pooled AUC was 0.87 (95% CI: 0.84–0.90) (Figure 3), providing an exploratory summary of the reported discrimination.
FIGURE 3.
Forest plot of AUC/C‐index values.
For the predictor meta‐analysis, only variables reported in at least three studies with extractable data were included. The study by (Li et al. 2023) did not report extractable multivariable odds ratios and was therefore not included in the predictor meta‐analysis. In the study by Wu et al. (2025), predictive variables from the decision tree and random forest models lacked quantifiable data, and thus only results from the logistic regression model were included in the meta‐analysis.
Liu et al. (2023) reported moist and sweaty skin as a significant predictor in the multivariable logistic regression analysis (OR = 0.006, 95% CI: 0.000–0.012, p = 0.001). The variable was originally coded as ‘yes = 1’ and ‘no = 2.’ The comparison direction needed to be reversed to harmonize this variable with the other studies. However, a reliable reciprocal estimate could not be calculated because the original lower confidence limit was reported as 0.000. Consequently, this estimate was not included in the skin‐moisture meta‐analysis.
Age > 60 years (OR = 3.59, 95% CI: 2.31–5.57), history of MARSI (OR = 5.95, 95% CI: 2.10–16.86), and allergy history (OR = 6.35, 95% CI: 3.70–10.89) were associated with higher odds of MARSI. Edema (OR = 3.95, 95% CI: 2.64–5.91), study‐defined hypoproteinemia (OR = 5.50, 95% CI: 3.10–9.75), and moist skin (OR = 3.74, 95% CI: 1.33–10.47) were also associated with higher odds of MARSI. Higher Braden Scale scores, analysed as continuous raw values, were associated with lower odds of MARSI (OR = 0.46, 95% CI: 0.24–0.90). The included studies did not consistently specify the type of allergy, the grade or location of edema, laboratory thresholds for hypoproteinemia, or standardised definitions of skin moisture. Accordingly, no post hoc definitions or cutoff values were introduced. Because relatively few studies contributed to each predictor analysis, subgroup analyses, meta‐regression, and formal assessments of small‐study effects were not undertaken. Detailed findings are presented in Table 5, and the corresponding forest plots are provided in Figure S1–S7.
TABLE 5.
Meta‐analysis results of common predictors associated with MARSI.
| Predictors | Number of studies | Heterogeneity test | Meta‐analysis | |||
|---|---|---|---|---|---|---|
| I2 (%) | p | OR (95% CI) | Z | p | ||
| Age over 60 years (Feng et al. 2024; Huang et al. 2020; Z. Li 2024; Xie et al. 2024) | 4 | 0.0 | 0.995 | 3.59 (2.31–5.57) | 5.690 | < 0.001 |
| History of MARSI (Lai 2022; Long 2022; S. Wu et al. 2025) | 3 | 81.3 | 0.005 | 5.95 (2.10–16.86) | 3.355 | 0.001 |
| Allergy history (Huang et al. 2020; Lai 2022; Long 2022; Y. Wu et al. 2024; Xie et al. 2024) | 5 | 45.1 | 0.121 | 6.35 (3.70–10.89) | 6.702 | < 0.001 |
| Edema (Lai 2022; Z. Li 2024; Long 2022; Xie et al. 2024; Y. Zhang 2025) | 5 | 0.0 | 0.648 | 3.95 (2.64–5.91) | 6.665 | < 0.001 |
| Hypoproteinemia (Huang et al. 2020; Y. Wu et al. 2024; Xie et al. 2024; Y. Zhang 2025) | 4 | 0.0 | 0.843 | 5.50 (3.10–9.75) | 5.840 | < 0.001 |
| Braden score (Feng et al. 2024; Z. Li 2024; S. Wu et al. 2025) | 3 | 90.7 | < 0.001 | 0.46 (0.24–0.90) | −2.283 | 0.022 |
| Skin moisture (Feng et al. 2024; Z. Li 2024; Long 2022) | 3 | 82.8 | 0.003 | 3.74 (1.33–10.47) | 2.509 | 0.012 |
Note: Braden scale scores were analysed as continuous raw values. Definitions of allergy history, edema, hypoproteinemia, and skin moisture varied across the included studies and were retained as reported in the original articles.
4. Discussion
To our knowledge, this is the first systematic review and meta‐analysis to evaluate existing MARSI risk prediction models. The review is relevant to nursing practice because nurses are involved in skin assessment, identification of patients at risk, adhesive selection, and the correct application and removal of adhesive products for MARSI prevention (de Faria et al. 2022; McNichol et al. 2013; Yang et al. 2025). The synthesis identifies commonly reported predictors that may warrant closer nursing assessment. The methodological appraisal also highlights considerations for the future development of MARSI risk assessment tools.
4.1. Predictive Performance and Risk of Bias Analysis
The 12 included studies reported 14 MARSI risk prediction models, all of which were developed using data from Chinese populations. Reported AUC or C‐index values ranged from 0.789 to 0.960. The summary AUC across nine studies was 0.87 (95% CI: 0.84–0.90). Because the included studies evaluated different independently developed models and substantial heterogeneity was present, the pooled AUC represents only an exploratory summary of reported discrimination. It should not be interpreted as the expected performance of any specific MARSI prediction model. Furthermore, most estimates were derived from model‐development datasets or internal validation, and none of the included models had undergone external validation. Consequently, the reported performance may be optimistic (Moons et al. 2015). The available evidence does not establish model calibration, clinical utility, or generalizability to other populations or care settings.
Logistic regression was the most frequently used modelling method (Feng et al. 2024; Huang et al. 2020; Lai 2022; Z. Li 2024; Liu et al. 2023; Long 2022; S. Wu et al. 2025; Y. Wu et al. 2024; Xie et al. 2024; Y. Zhang 2025; Zhu et al. 2025). S. Wu et al. (2025) compared a logistic regression model with a classification and regression tree model and a random forest model. The two machine‐learning models yielded slightly higher AUC values than the logistic regression model. However, this single within‐study comparison is insufficient to conclude that machine‐learning approaches are superior for MARSI prediction. Model performance is also influenced by sample size, predictor handling, missing‐data methods, overfitting, and validation procedures (Moons et al. 2015).
PROBAST assessment indicated that all included studies had a high overall risk of bias. In the participant domain, 10 studies used retrospective data (Feng et al. 2024; Huang et al. 2020; Li et al. 2023; Liu et al. 2023; Long 2022; S. Wu et al. 2025; Y. Wu et al. 2024; Xie et al. 2024; Y. Zhang 2025; Zhu et al. 2025). Eleven studies were conducted at a single center (Feng et al. 2024; Huang et al. 2020; Li et al. 2023; Z. Li 2024; Liu et al. 2023; Long 2022; S. Wu et al. 2025; Y. Wu et al. 2024; Xie et al. 2024; Y. Zhang 2025; Zhu et al. 2025). No study used a prospective longitudinal design. These characteristics limit sample representativeness and may introduce selection bias and incomplete documentation (Z. Zhang et al. 2025). In the predictor domain, operational definitions and measurement procedures were frequently insufficiently reported. Several predictors, including edema, skin moisture, allergy history, and skin condition, may depend on subjective assessment or local documentation practices. For example, Liu et al. (2023) included ‘moist and sweaty skin’ and ‘dry skin’ without reporting detailed assessment criteria or inter‐rater reliability. Such inconsistencies in predictor definition and assessment may introduce predictor measurement heterogeneity, potentially affecting model calibration, discrimination, and overall predictive accuracy across settings (Luijken et al. 2019). Future studies should assess observation‐based risk factors using standardised and, where feasible, objective measurement tools, apply explicit criteria for MARSI classification, and report inter‐rater agreement to improve measurement consistency and comparability across settings (Li et al. 2023; Yang et al. 2025).
The analysis domain was the principal source of bias. Only one study (Li et al. 2023) achieved an events‐per‐variable ratio greater than 20, indicating a substantial risk of model overfitting in most studies (Moons et al. 2019; Z. Zhang et al. 2025). Continuous variables, including age and body mass index, were frequently categorised without adequate clinical or statistical justification (Feng et al. 2024; Huang et al. 2020; Lai 2022; Z. Li 2024; Liu et al. 2023; Long 2022; S. Wu et al. 2025; Y. Wu et al. 2024; Xie et al. 2024; Zhu et al. 2025). This practice may result in information loss and unstable cutoff points (Chen et al. 2025; Z. Zhang et al. 2025). Missing data were commonly handled by complete‐case analysis (Lai 2022; Li et al. 2023) or were not adequately reported (Feng et al. 2024; Huang et al. 2020; Liu et al. 2023; Long 2022; Y. Wu et al. 2024). All 12 studies used univariate analysis to screen candidate predictors (Feng et al. 2024; Huang et al. 2020; Lai 2022; Li et al. 2023; Z. Li 2024; Liu et al. 2023; Long 2022; S. Wu et al. 2025; Y. Wu et al. 2024; Xie et al. 2024; Y. Zhang 2025; Zhu et al. 2025). This approach may exclude important predictors and increase uncertainty in the resulting model structure (Moons et al. 2015). Future MARSI model studies should follow established methodological guidance for sample‐size determination, continuous‐variable handling, and missing‐data management (Moons et al. 2015). Candidate predictors should be selected using previous evidence and clinical reasoning rather than univariate statistical significance alone (Moons et al. 2015).
Performance assessment and validation were also incomplete. Four studies did not report sensitivity or specificity (Li et al. 2023; Wu et al. 2024; Y. Zhang 2025; Zhu et al. 2025). No study evaluated clinical utility. Two studies developed models without validation (Lai 2022; Li et al. 2023). The remaining 10 studies performed internal validation only (Feng et al. 2024; Huang et al. 2020; Z. Li 2024; Liu et al. 2023; Long 2022; S. Wu et al. 2025; Y. Wu et al. 2024; Xie et al. 2024; Y. Zhang 2025; Zhu et al. 2025). No study performed external validation (Feng et al. 2024; Huang et al. 2020; Lai 2022; Li et al. 2023; Z. Li 2024; Liu et al. 2023; Long 2022; S. Wu et al. 2025; Y. Wu et al. 2024; Xie et al. 2024; Y. Zhang 2025; Zhu et al. 2025). Bootstrap resampling or cross‐validation can be used for internal validation, whereas temporal and geographic validation are needed to evaluate model performance in new patients and clinical settings (Moons et al. 2015). Before implementation, future MARSI studies should report discrimination and calibration and assess clinical utility (Grant et al. 2018).
4.2. Analysis of Common Predictors in MARSI Risk Prediction Models
The pooled MARSI prevalence in this review was 22.0% (95% CI: 18.5%–25.4%), with substantial heterogeneity (I2 = 86.9%). The predictor meta‐analysis examined seven factors: age > 60 years, history of MARSI, allergy history, edema, study‐defined hypoproteinemia, Braden Scale score, and moist skin.
Previous studies have consistently associated older age with greater susceptibility to MARSI (de Faria et al. 2022; Frota et al. 2023; Shi et al. 2025). In the present review, patients aged > 60 years had 3.59 times the odds of MARSI compared with those aged ≤ 60 years (OR = 3.59, 95% CI: 2.31–5.57). Wang et al. (2024) also identified age > 50 years as a risk factor (OR = 1.21, 95% CI: 1.05–1.41). Although the age thresholds and study populations differed, the direction of the association was consistent. Age‐related reductions in collagen, skin elasticity, and moisture retention may increase vulnerability to friction and adhesive‐related mechanical stress (Frota et al. 2023; Shi et al. 2025).
Three studies evaluated a history of MARSI as a predictor (Lai 2022; Long 2022; S. Wu et al. 2025). The pooled analysis showed that patients with a previous MARSI episode had higher odds of subsequent MARSI (OR = 5.95, 95% CI: 2.10–16.86), although heterogeneity was substantial (I2 = 81.3%). Shi et al. (2025) similarly identified a history of MARSI as a risk factor among patients with cancer (OR = 11.22, 95% CI: 5.60–22.47). A previous episode may indicate persistent skin vulnerability, while chemotherapy‐related impairment of keratinocyte proliferation may further compromise the skin barrier in patients with cancer (Shi et al. 2025). These findings support closer skin surveillance in patients with a history of MARSI (Shi et al. 2025).
Allergy history was associated with higher odds of MARSI (OR = 6.35, 95% CI: 3.70–10.89) (Huang et al. 2020; Lai 2022; Long 2022; Y. Wu et al. 2024; Xie et al. 2024). Wang et al. (2024) reported a similar association (OR = 3.82, 95% CI: 1.92–7.57). However, the included studies did not consistently distinguish among allergies to drugs, foods, disinfectants, dressings, and medical adhesives. The pooled association therefore relates to allergy history in general rather than specifically to medical adhesive allergy. In susceptible individuals, exposure to disinfectants or adhesive components may trigger localized contact dermatitis, including itching, erythema, or blistering (Lai 2022; Long 2022).
Edema was associated with higher odds of MARSI in the present review (OR = 3.95, 95% CI: 2.64–5.91). Wang et al. (2024) reported a similar pooled estimate (OR = 3.59, 95% CI: 1.52–8.47). The contributing studies generally recorded edema only as present or absent and provided limited information about its anatomical location or severity. Edema increases interstitial fluid accumulation and tissue tension and may weaken skin cohesion and local perfusion (Lai 2022; Wang et al. 2024). These changes can increase the vulnerability of edematous skin to stripping and other mechanical injuries during adhesive removal (Lai 2022; Long 2022).
Study‐defined hypoproteinemia was associated with higher odds of MARSI (OR = 5.50, 95% CI: 3.10–9.75) (Huang et al. 2020; Y. Wu et al. 2024; Xie et al. 2024; Y. Zhang 2025). Its operational definition varied across the four contributing studies: one used a cutoff of < 30 g/L (Y. Wu et al. 2024), two used or referred to a cutoff of < 35 g/L (Huang et al. 2020; Xie et al. 2024), and one did not report a numerical threshold (Y. Zhang 2025). The pooled estimate therefore does not support a single laboratory cutoff for MARSI risk assessment. Hypoproteinemia may reduce plasma colloid osmotic pressure, promote edema, and increase tissue fragility, thereby reducing the skin's tolerance to adhesive‐related mechanical forces (Frota et al. 2023; Long 2022).
Higher Braden Scale scores were associated with lower odds of MARSI (OR = 0.46, 95% CI: 0.24–0.90), consistent with the findings of Frota et al. (2023). Lower scores may indicate impaired mobility, sensory perception, nutrition, or moisture control. These conditions may increase susceptibility to friction and mechanical forces from medical adhesives (Feng et al. 2024; Frota et al. 2023). However, this analysis included only three studies and showed substantial heterogeneity (I2 = 90.7%). Because the included studies analysed Braden Scale scores as continuous values, established pressure‐injury risk categories should not be directly applied to MARSI risk classification.
Moist skin was associated with higher odds of MARSI (OR = 3.74, 95% CI: 1.33–10.47) (Feng et al. 2024; Z. Li 2024; Long 2022). Skin moisture was assessed clinically in the contributing studies and was generally described as moist or sweaty skin. Prolonged exposure to sweat, exudate, or residual moisture increases stratum corneum hydration and permeability, weakens the skin barrier, and increases vulnerability to mechanical stress and chemical irritation (Long 2022; Wang et al. 2024). Excessive moisture may also alter adhesion between medical products and the skin, increasing friction during removal and contributing to maceration and local inflammation (Z. Li 2024; Long 2022; Wang et al. 2024).
Taken together, these findings suggest that MARSI risk is associated with both underlying skin vulnerability and observable clinical conditions. However, variability in predictor definitions and measurement methods, together with substantial heterogeneity in several analyses, limits the precision of the pooled estimates. The identified factors should therefore be considered candidate predictors requiring further evaluation rather than components of an established MARSI risk classification system.
4.3. Implications for Nursing Practice
4.3.1. Workflow Integration, Feasibility, and Electronic Documentation
The predictors identified in this review may help nurses recognize patients who require closer skin assessment and individualised preventive care, consistent with existing MARSI prevention recommendations (de Faria et al. 2022; McNichol et al. 2013). Nurses may assess these predictors alongside direct skin examination and information about current adhesive use. When MARSI occurs, it should be documented in the patient's medical record, and skin assessment should be repeated when adhesive products are replaced or the patient's skin condition changes (de Faria et al. 2022; McNichol et al. 2013).
To support routine use, future MARSI prediction tools should be brief, interpretable, and based primarily on routinely collected data. They should minimize duplicate data entry and manual calculation and provide concise outputs linked to specific nursing actions (Hants et al. 2023; Sutton et al. 2020). However, because currently available models have a high risk of bias and lack external validation, they should not be used as stand‐alone decision tools or as substitutes for nurses' professional judgement.
Following adequate external validation and usability testing, MARSI prediction tools could be integrated into electronic nursing documentation systems. Routinely recorded variables could be retrieved automatically, while nurses could verify observation‐based findings and document the condition of adhesive sites. The system could then support the selection of preventive actions and the scheduling of reassessment, consistent with broader efforts to incorporate nursing assessment and clinical decision support into digital health systems (Hants et al. 2023; Sutton et al. 2020). Before clinical implementation, the effects of these tools on nursing workload, adherence to preventive care, and patient outcomes should be evaluated.
4.3.2. Nurse‐Led MARSI Prevention Bundle
Drawing on the MARSI prevention literature, a proposed nurse‐led bundle for future evaluation includes measures before, during, and after adhesive use. Before application, nurses should conduct a structured risk and skin assessment. Adhesive selection should consider the patient's skin condition, application site, required fixation strength, expected wear time, and replacement frequency (de Faria et al. 2022; McNichol et al. 2013). Vulnerable skin should be gently cleansed and thoroughly dried, and an appropriate barrier product may be applied when indicated. Adhesives should be applied without tension, while repeated application to the same site and unnecessary use of highly adherent products should be avoided (de Faria et al. 2022; McNichol et al. 2013).
During adhesive use, nurses should regularly inspect the site for erythema, blistering, skin stripping, maceration, pain, itching, and other early signs of MARSI. During removal, the adhesive should be peeled back slowly and close to the skin in the direction of hair growth while the surrounding skin is supported. A medical adhesive remover may be used when appropriate (McNichol et al. 2013). After removal or replacement, the skin should be reassessed and the findings documented (de Faria et al. 2022; McNichol et al. 2013).
4.3.3. Proposed Nurse‐Led Pathway for MARSI Risk Assessment and Prevention
Figure 4 translates the review findings and the proposed prevention bundle into a cyclical, nurse‐led pathway comprising initial assessment, risk identification, risk‐informed decision‐making, individualised intervention, documentation and communication, and reassessment. The pathway is intended as a practice‐oriented framework rather than a validated prediction algorithm. Because no validated MARSI risk thresholds were identified, the presence or number of risk factors should not be interpreted as a risk category or intervention threshold. Preventive decisions should instead consider these factors alongside current skin findings and changes in the patient's condition.
FIGURE 4.
Proposed Nurse‐led Pathway for MARSI Risk Assessment and Prevention. This proposed nurse‐led pathway has not yet been clinically validated. MARSI, medical adhesive‐related skin injury.
Neither the pathway nor the proposed bundle was evaluated in the studies included in this review. They should therefore be adapted to local nursing protocols and prospectively tested before wider implementation. Evaluation should address feasibility, usability, nursing workload, adherence to preventive care, and patient outcomes.
4.4. Implications for Nursing Education and Training
MARSI prevention requires competencies in structured skin assessment, early recognition of MARSI, interpretation of risk information, appropriate adhesive selection and application, safe removal, documentation, and timely escalation (de Faria et al. 2022; McNichol et al. 2013). Survey findings indicate that nurses' MARSI‐related knowledge and practices vary according to educational level, clinical experience, and previous MARSI training (Rudek et al. 2025; L. Wu et al. 2023). These findings support the need for structured MARSI education.
MARSI education could be incorporated into undergraduate nursing curricula, clinical orientation, and continuing professional development. Educational content could cover MARSI definition and classification, risk and skin assessment, skin care, adhesive selection, application and removal, prevention, early management, and documentation (de Faria et al. 2022; McNichol et al. 2013). Where prediction models are introduced, education should also address the interpretation and limitations of model outputs. Future studies should evaluate whether different educational approaches improve nurses' knowledge, practical skills, adherence to evidence‐based adhesive practices, documentation quality, and patient outcomes.
5. Limitations
This review has several limitations. First, all included studies were judged to have a high overall risk of bias, particularly in the participants and analysis domains. Most studies used retrospective, single‐center designs, had limited sample sizes, and inadequately addressed missing data or model validation. These methodological limitations may have biased the reported estimates of model performance, thereby reducing confidence in the pooled findings (Moons et al. 2019). Second, all included studies were conducted in China, limiting the generalizability of the findings to other populations and healthcare settings. Multicenter external validation in other regions and countries is therefore needed.
Third, although multiple major databases were searched and eligible full‐text dissertations were included, other sources of grey literature were not systematically searched. Relevant unpublished evidence may therefore have been missed, increasing the possibility of publication bias. Fourth, the operational definitions and classifications of MARSI outcomes and predictors varied across studies. This measurement heterogeneity may have reduced comparability and affected the pooled estimates. Random‐effects models cannot eliminate bias arising from inconsistent measurement; therefore, the pooled findings should be interpreted cautiously.
Fifth, several studies were excluded from specific meta‐analyses because key statistical indicators were unavailable or outcome reporting was incomplete. Their exclusion may have affected the pooled estimates and introduced bias. Finally, restricting inclusion to studies published in Chinese or English may have introduced language bias and reduced the comprehensiveness of the evidence.
6. Conclusion
This systematic review and meta‐analysis included 12 studies describing 14 MARSI risk prediction models. Reported discrimination was generally good. However, all included studies had a high risk of bias, and none performed external validation. These models are therefore not ready for routine use in nursing practice. The pooled analyses showed higher odds of MARSI among patients aged over 60 years and those with a history of MARSI, allergy history, edema, study‐defined hypoproteinemia, or moist skin. Higher Braden Scale scores were associated with lower odds of MARSI. These findings may help nurses prioritise skin assessment and preventive care but should not be interpreted as a validated risk score or intervention threshold.
Future studies should use adequately sized, multicenter prospective cohorts and conduct temporal and geographic external validation in diverse populations and adhesive‐use settings. Candidate modelling methods, including logistic regression and machine‐learning approaches, should be compared using the same development and validation datasets. Model evaluation should include discrimination, calibration, clinical utility, and interpretability. Adherence to established model‐development and reporting standards is needed to improve transparency and reproducibility before implementation in nursing practice.
Author Contributions
Yujing Gu: writing – original draft, conceptualization, methodology, software, formal analysis, data curation. Zhouzhou Lu: writing – review and editing, resources, data curation, validation, supervision. Yun Zhao: methodology, software, data curation. Shujun Fan: methodology, software. Jun Xie: writing – review and editing, supervision, funding acquisition, investigation.
Funding
This study was supported by the Top Talent Support Program for young and middle‐aged people of Wuxi Health Committee (No. BJ2023091) and the Wuxi Science and Technology Development Fund (No. K20253007).
Ethics Statement
Ethical approval and informed consent were not required, as this investigation was a systematic review and meta‐analysis of publicly available, previously published data with no direct involvement of human participants.
Conflicts of Interest
The authors declare no conflicts of interest.
Supporting information
Table S1: Search strategy.
Figure S1: Forest plot of age.
Figure S2: Forest plot of MARSI history.
Figure S3: Forest plot of allergy history.
Figure S4: Forest plot of edema.
Figure S5: Forest plot of hypoalbuminemia.
Figure S6: Forest plot of Braden score.
Figure S7: Forest plot of skin moisture.
Acknowledgements
The authors have nothing to report. The study protocol was prospectively registered on the international prospective register of systematic reviews (PROSPERO) under registration number CRD420251179067.
Data Availability Statement
The data that support the findings of this study are available from the corresponding authors upon reasonable request.
References
- Chen, Z. , Zhu L., Ma C., et al. 2025. “Frailty Risk Prediction Models in Maintenance Hemodialysis Patients: A Systematic Review and Meta‐Analysis of Model Performance and Methodological Quality.” Renal Failure 47, no. 1: 2522329. 10.1080/0886022X.2025.2522329. [DOI] [PMC free article] [PubMed] [Google Scholar]
- de Faria, M. F. , Ferreira M. B. G., dos Santos Felix M. M., Bessa R. M. V., and Barbosa M. H.. 2022. “Prevention of Medical Adhesive‐Related Skin Injury During Patient Care: A Scoping Review.” International Journal of Nursing Studies Advances 4: 100078. 10.1016/j.ijnsa.2022.100078. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Du, M. , and Liu M.. 2024. “Prevalence, Risk Factors, Causes, Assessments, and Prevention of Medical Adhesive‐Related Skin Injury: A Scoping Review.” Advances in Skin & Wound Care 37, no. 11–12: 1–10. 10.1097/ASW.0000000000000235. [DOI] [PubMed] [Google Scholar]
- Feng, Y. , Zhou Y., and Shen Z.. 2024. “Construction and Validation of a Risk Prediction Model for Medical Adhesive‐Related Skin Injury in ICU Patients.” Chinese Evidence‐Based Nursing 10, no. 19: 3561–3567. [Google Scholar]
- Frota, O. P. , Pinho J. N., Ferreira‐Júnior M. A., Sarti E., Paula F. M., and Ferreira D. N.. 2023. “Incidence and Risk Factors for Medical Adhesive‐Related Skin Injury in Catheters of Critically Ill Patients: A Prospective Cohort Study.” Australian Critical Care 36, no. 6: 997–1003. 10.1016/j.aucc.2023.02.005. [DOI] [PubMed] [Google Scholar]
- Grant, S. W. , Collins G. S., and Nashef S. A. M.. 2018. “Statistical Primer: Developing and Validating a Risk Prediction Model.” European Journal of Cardio‐Thoracic Surgery 54, no. 2: 203–208. 10.1093/ejcts/ezy180. [DOI] [PubMed] [Google Scholar]
- Hants, L. , Bail K., and Paterson C.. 2023. “Clinical Decision‐Making and the Nursing Process in Digital Health Systems: An Integrated Systematic Review.” Journal of Clinical Nursing 32, no. 19–20: 7010–7035. 10.1111/jocn.16823. [DOI] [PubMed] [Google Scholar]
- Huang, S. , Wang W., and Xin C.. 2020. “Risk of Medical Adhesive‐Related Skin Injury in Patients With Colon Cancer Undergoing Chemotherapy With an Indwelling Peripherally Inserted Central Catheter.” Journal of Clinical and Pathological Research 40, no. 1: 107–115. [Google Scholar]
- Lai, Y. 2022. “Risk Factor Analysis and Construction of a Prediction Model for Medical Adhesive‐Related Skin Injury at PICC Insertion Sites in Chengdu.”
- Li, J. , Hao N., Han J., Zhang M., and Li X.. 2023a. “Incidence and Predictive Model of Medical Adhesive‐Related Skin Injury in Cancer Patients Managed With Central Venous Access Devices: A Retrospective Study.” Journal of Wound, Ostomy, and Continence Nursing 50, no. 3: 209–213. 10.1097/WON.0000000000000971. [DOI] [PubMed] [Google Scholar]
- Li, Y. , Zhang H., Zhang S., Hou X., and Feng L.. 2023b. “Development and Validation of Medical Adhesive‐Related Skin Injury Risk Assessment Scale at Peripherally Inserted Central Catheter Insertion Site in Oncology Patients.” Nursing Open 10, no. 8: 5244–5251. 10.1002/nop2.1762. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Li, Z. 2024. “Construction and Validation of a Risk Prediction Model for Medical Adhesive‐Related Skin Injury in ICU Patients.”
- Liu, X. , Li R., Zhang Y., Hou S., and Liu Y.. 2023. “Construction and Validation of a Risk Prediction Model for Medical Adhesive‐Related Skin Injury at PICC Insertion Sites.” Modern Journal of Integrated Traditional Chinese and Western Medicine 32, no. 23: 3349–3354. [Google Scholar]
- Long, G. 2022. “Analysis of Influencing Factors and Construction of a Prediction Model for Medical Adhesive‐Related Skin Injury in Home‐Based Patients With PICCs.”
- Luijken, K. , Groenwold R. H. H., Van Calster B., Steyerberg E. W., and van Smeden M.. 2019. “Impact of Predictor Measurement Heterogeneity Across Settings on the Performance of Prediction Models: A Measurement Error Perspective.” Statistics in Medicine 38, no. 18: 3444–3459. 10.1002/sim.8183. [DOI] [PMC free article] [PubMed] [Google Scholar]
- McNichol, L. , Lund C., Rosen T., and Gray M.. 2013. “Medical Adhesives and Patient Safety: State of the Science: Consensus Statements for the Assessment, Prevention, and Treatment of Adhesive‐Related Skin Injuries.” Orthopaedic Nursing 32, no. 5: 267–281. 10.1097/NOR.0b013e3182a39caf. [DOI] [PubMed] [Google Scholar]
- Moons, K. G. M. , Altman D. G., Reitsma J. B., et al. 2015. “Transparent Reporting of a Multivariable Prediction Model for Individual Prognosis or Diagnosis (TRIPOD): Explanation and Elaboration.” Annals of Internal Medicine 162, no. 1: W1–W73. 10.7326/M14-0698. [DOI] [PubMed] [Google Scholar]
- Moons, K. G. M. , de Groot J. A. H., Bouwmeester W., et al. 2014. “Critical Appraisal and Data Extraction for Systematic Reviews of Prediction Modelling Studies: The CHARMS Checklist.” PLoS Medicine 11, no. 10: e1001744. 10.1371/journal.pmed.1001744. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Moons, K. G. M. , Wolff R. F., Riley R. D., et al. 2019. “PROBAST: A Tool to Assess Risk of Bias and Applicability of Prediction Model Studies: Explanation and Elaboration.” Annals of Internal Medicine 170, no. 1: W1–W33. 10.7326/M18-1377. [DOI] [PubMed] [Google Scholar]
- Page, M. J. , McKenzie J. E., Bossuyt P. M., et al. 2021. “The PRISMA 2020 Statement: An Updated Guideline for Reporting Systematic Reviews.” BMJ (Clinical Research Ed.) 372: n71. 10.1136/bmj.n71. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Rudek, J. , Majda A., Naworska B., and Sawicka M.. 2025. “Nurses' Knowledge on Medical Adhesive‐Related Skin Injuries (MARSI).” Nursing Problems / Problemy Pielęgniarstwa 33, no. 2: 66–75. 10.5114/ppiel.2025.152631. [DOI] [Google Scholar]
- Shi, W. , Leng Y., Li T., Li Q., Wang N., and Wang G.. 2025. “The Incidence and Risk Factors for Medical Adhesive‐Related Skin Injury in Cancer Patients in China: A Systematic Review and Meta‐Analysis.” Research in Nursing & Health 48, no. 4: 497–507. 10.1002/nur.22473. [DOI] [PubMed] [Google Scholar]
- Sutton, R. T. , Pincock D., Baumgart D. C., Sadowski D. C., Fedorak R. N., and Kroeker K. I.. 2020. “An Overview of Clinical Decision Support Systems: Benefits, Risks, and Strategies for Success.” npj Digital Medicine 3: 17. 10.1038/s41746-020-0221-y. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Wang, P. , Luo X., Chen H., Feng Q., and Song H.. 2024. “The Prevalence, Incidence and Risk Factors of Medical Adhesive‐Related Skin Injury in Adult Inpatients: A Systematic Review and Meta‐Analysis.” Journal of Tissue Viability 33, no. 4: 960–967. 10.1016/j.jtv.2024.10.007. [DOI] [PubMed] [Google Scholar]
- Wu, L. , Deng S., Yu L., and Rong H.. 2023. “Nurses' Knowledge, Attitude and Behaviour on Medical Adhesive Related Skin Injury in Neonatal Department: A Survey.” Nursing Open 10, no. 7: 4713–4720. 10.1002/nop2.1721. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Wu, S. , Liu K., and Tang T.. 2025. “Construction of a Risk Prediction Model for Medical Adhesive‐Related Skin Injury Associated With Implanted Venous Access Ports During Treatment in Patients With Lung Cancer.” Chinese Nursing Research 39, no. 15: 2525–2534. [Google Scholar]
- Wu, Y. , Lu Y., Xu R., Yuan L., Luo Y., and Qian H.. 2024. “Construction of a Risk Prediction Model for Medical Adhesive‐Related Skin Injury After Implantable Venous Access Port Placement in Patients With Malignant Tumors.” Medical Journal of Communications 38, no. 1: 99–105. 10.19767/j.cnki.32-1412.2024.01.029. [DOI] [Google Scholar]
- Xie, T. , Zhang G., Zhang J., Li Y., and Zou X.. 2024. “Development and Validation of a Risk Prediction Model for Medical Adhesive‐Related Skin Injury in ICU Patients.” Journal of Nurses Training 39, no. 20: 2173–2178. 10.16821/j.cnki.hsjx.2024.20.009. [DOI] [Google Scholar]
- Yang, T. Y. , Park J. E., Kim J. B., and Lee N. H.. 2025. “Incidence and Characteristics of Medical Adhesive‐Related Skin Injuries in Patients Following Spinal Surgery: A Prospective Observational Study.” International Wound Journal 22, no. 4: e70457. 10.1111/iwj.70457. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Zaballa Canive, M. , González Blas L., Lurueña Rodríguez S., et al. 2026. “Medical Adhesive‐Related Skin Injuries in Oncology and Haematology Patients With Peripherally Inserted Central Catheters: A Prospective Descriptive Study.” Journal of Clinical Nursing 35, no. 2: 839–849. 10.1111/jocn.70035. [DOI] [PubMed] [Google Scholar]
- Zhang, Y. 2025. “Construction and Validation of a Risk Prediction Model for Medical Adhesive‐Related Skin Injury in ICU Patients.” Journal of Tibet Medicine 46, no. 1: 44–46. [Google Scholar]
- Zhang, Y. , Wang S., Zhang X., Zhang W., and Wang X.. 2020. “Incidence and Influencing Factors of Medical Adhesive‐Related Skin Injury in Critically Ill Patients.” Advances in Skin & Wound Care 33, no. 5: 260–266. 10.1097/01.ASW.0000658584.09988.fa. [DOI] [PubMed] [Google Scholar]
- Zhang, Z. , Wang S., Xu Z., et al. 2025. “Frailty Risk Prediction Models in Maintenance Hemodialysis Patients: A Systematic Review and Meta‐Analysis of Studies From China.” Renal Failure 47, no. 1: 2500663. 10.1080/0886022X.2025.2500663. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Zhu, Y. , Huang M., Zhang Q., Luo H., and Zhong W.. 2025. “Construction and Validation of a Risk Prediction Model for Medical Adhesive‐Related Skin Injury in Older Patients During Surgery.” Progress in Modern Biomedicine 25, no. 4: 781–787. 10.13241/j.cnki.pmb.2025.04.024. [DOI] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Table S1: Search strategy.
Figure S1: Forest plot of age.
Figure S2: Forest plot of MARSI history.
Figure S3: Forest plot of allergy history.
Figure S4: Forest plot of edema.
Figure S5: Forest plot of hypoalbuminemia.
Figure S6: Forest plot of Braden score.
Figure S7: Forest plot of skin moisture.
Data Availability Statement
The data that support the findings of this study are available from the corresponding authors upon reasonable request.