Pressure Ulcers Risk Assessment According to Nursing Criteria

Pressure ulcers (PU) represent a health problem with a significant impact on the morbidity and mortality of immobilized patients, and on the quality of life of affected people and their families. Risk assessment of pressure ulcers incidence must be carried out in a structured and comprehensive manner. The Braden Scale is the result of an analysis of risk factors that includes subscales that define exactly what should be interpreted in each one. The healthcare work with evidence-based practice with an objective criterion by the nursing professional is an essential addition for the application of preventive measures. Explanatory models based on the different subscales of Braden Scale purvey an estimation to level changes in the risk of suffering PU. A binary-response logistic regression model, supported by a study with an analytical, observational, longitudinal, and prospective design in the Granada-Metropolitan Primary Healthcare District (DSGM) in Andalusia (Southern Spain), with a sample of 16,215 immobilized status patients, using a Braden Scale log, is performed. A model that includes the mobility and activity scales achieves a correct classification rate of 86% (sensitivity (S) = 87.57%, specificity (SP) = 81.69%, positive predictive value (PPV) = 91.78%, and negative preventive value (NPV) = 73.78%), while if we add the skin moisture subscale to this model, the correct classification rate is 96% (S = 90.74%, SP = 88.83%, PPV = 95.00%, and NPV = 80.42%). The six subscales provide a model with a 99.5% correct classification rate (S = 99.93%, SP = 98.50%, PPV = 99.36%, and NPV = 99.83%). This analysis provides useful information to help predict this risk in this group of patients through objective nursing criteria.


Introduction
Pressure ulcers (PU) are injuries caused to the skin and/or underlying tissues as a result of continuous pressure on these tissues, or due to the combination with shearing. They are usually located on bony prominences [1]. They have a high average prevalence, both in Europe (10.8%) [2] and in Spain, which stands at 7% in the hospital environment [3] and 4.79% among patients in home care [4]. PU represent a health problem with a significant impact on the morbidity and mortality of immobilized patients and in the quality of life of affected people and their families [5]. Risk assessment of pressure ulcers incidences must be carried out in a structured and comprehensive manner. The Braden Scale for Predicting

Statistical Analysis
Firstly, a graphical exploratory analysis was carried out to identify which variables individually, by pairs, and triplets were able to identify patients at higher risk of PU. Second, various binary-response logistic regression models [20] were used to identify the subscales combinations with higher performance for PU risk prediction. A stepwise forward-backward selection model, without any interactions, was deemed to best fit the records. The goodness-of-fit was compared using the probability ratio test and Stukel's chi-squared test. Wald's test was used to evaluate the significance at the population level of the factors that entered the models. The validation of the model was done by calculating the rate of correct classifications. The ROC curve was used to analyze the performance of the models. The strength ratios for every level regarding the adjacent level were achieved, according to the potential variations in the risk subscales studied. The R Statistical Computing Software (version 4.1.1) (https://www.r-project.org/) (Accessed on 27 July 2022) was used for the statistical analysis.

Results
This section is structured as follows. First, a descriptive analysis of the response and explanatory variables is shown. Then, an exploratory analysis based on graphical outputs allows to probe the power for classification, of being in a risk level (or not) for developing pressure ulcers, by means of individual or certain groups of subscales. Section 3.3 estimates a binary logistic regression model for the risk of PU, based on the six subscales. A detailed analysis of prognostic ability of each subscale is performed. In addition, different measures and graphical outputs of the quality of the model from an inferential, accuracy, and validity point of view are provided. The two following sections estimate, analyze, and validate binary logistic regression models based on the activity and mobility subscales (Section 3.3), and based on activity, mobility, and skin moisture subscales (Section 3.4). These last two models are based on the graphical exploratory analysis performed in Section 3.2.

Sample Description
According to the Braden Scale scores, the individuals were classified into: no risk or risk of developing pressure ulcers. The descriptive analysis of the subscales deemed is shown in Table 1. The overlap histograms for the classification of a patient at risk or not of suffering from pressure ulcer, based on sex and each one of the subscales independently, are dis-played in Figure 1. Activity (BAct) and mobility (BMov) subscales seem to be adequate as independent classifiers. This is reflected in their histograms, because the colors that correspond to each level of PU risk are well separated for the different values of these subscales. The overlap histograms for the classification of a patient at risk or not of sufferi from pressure ulcer, based on sex and each one of the subscales independently, are d played in Figure 1. Activity (BAct) and mobility (BMov) subscales seem to be adequate independent classifiers. This is reflected in their histograms, because the colors that co respond to each level of PU risk are well separated for the different values of these su scales.    Figure 2 shows the potential applicability as classifiers of each pair of variables jointly. In this graphical output it is reflected that activity (BAct) and mobility (BMov) subscales are jointly adequate classifiers because the green and red dots are well separated in the corresponding biplot located (according to a matrix notation) in the 4th row and 5th column. There are more pairs of variables that could be considered jointly as adequate for classification, but keeping in mind that the objective nursing criteria are based on such quick identification and objective information provided by activity and mobility subscales, these are not relevant in this work.

Bivariate Graphical Exploratory Analysis
Healthcare 2022, 10, 1438 5 3.2.2. Bivariate Graphical Exploratory Analysis Figure 2 shows the potential applicability as classifiers of each pair of varia jointly. In this graphical output it is reflected that activity (BAct) and mobility (BM subscales are jointly adequate classifiers because the green and red dots are well separ in the corresponding biplot located (according to a matrix notation) in the 4th row and column. There are more pairs of variables that could be considered jointly as adequat classification, but keeping in mind that the objective nursing criteria are based on quick identification and objective information provided by activity and mobility scales, these are not relevant in this work.

Three-Subscale Graphical Exploratory Analysis
The activity (BAct) and mobility (BMov) subscales combined with any one of th mainder of the subscales or the variable sex could provide adequate classifiers. Figu shows six 3D scatterplots where this fact can be analyzed. Considering Figure 3 as a m of graphical outputs with three rows and two columns, it is immediately clear tha scatterplot in position (row = 1, column = 1) is the best concerning the adequate classi tion in one PU risk level or another, because the green and red dots are completely s rated. This plot corresponds to the joint classification of the activity, mobility, and moisture (BHum) subscales.

Three-Subscale Graphical Exploratory Analysis
The activity (BAct) and mobility (BMov) subscales combined with any one of the remainder of the subscales or the variable sex could provide adequate classifiers. Figure 3 shows six 3D scatterplots where this fact can be analyzed. Considering Figure 3 as a matrix of graphical outputs with three rows and two columns, it is immediately clear that the scatterplot in position (row = 1, column = 1) is the best concerning the adequate classification in one PU risk level or another, because the green and red dots are completely separated. This plot corresponds to the joint classification of the activity, mobility, and skin moisture (BHum) subscales.

Explanatory Model for Pressure Ulcer Risk Prognosis Based on the Six Braden Subscales
In this section a binary-response logistic regression model for PU risk classification is performed based on the variable sex and the six Braden subscales. The stepwise forward-backward selection model included the six Braden subscales in the binary logistic regression model as relevant for the prognosis of patients' pressure ulcer risk. The estimated model for risk prognosis has the following form:

Explanatory Model for Pressure Ulcer Risk Prognosis Based on the Six Braden Subscales
In this section a binary-response logistic regression model for PU risk classification is performed based on the variable sex and the six Braden subscales. The stepwise forward-backward selection model included the six Braden subscales in the binary logistic regression model as relevant for the prognosis of patients' pressure ulcer risk. The estimated model for risk prognosis has the following form: The parameters estimated for each subscale in the binary logistic regression model can be found in Table 2 below. Note: BAct = activity, BMov = mobility, BRoc = shearing, BSens = sensibility, BHum = skin moisture, BNut = nutritional status, B = estimated parameter, DT = standard deviation, Z = Z statistic, p = p-value, OR = odds ratio, CI = confidence interval, Lower = lower limit of the CI, Upper = upper limit of the CI.
The chi-square log-likelihood test for this model was X 2 (8, N = 16,215) = 2538.20, p < 0.001. Therefore, when these variables were included in the model, the fit improved significantly compared to a model than only takes the constant into account. The Stukel goodness-of-fit test for this model was X 2 (2, N = 16,215) = 15,063, p < 0.001. These results did not conclude, therefore, that the model produce a good fit at population level for the risk of developing pressure ulcer.
In light of the results of the z-test (see Table 2), all the levels of the subscales are significant at a population-based level (p < 0.001). The prognosis change ratio for the levels considered (no risk vs. risk of development pressure ulcer) was analysed for all the explanatory variables with respect to the baseline category. For instance, with regard to the activity scale, it should be noted that the odds of PU development for bedridden patients is 69-fold the odds of those who frequently wander (OR = 68,871.66; 95% CI: 25,591. 852 Finally, it is worth mentioning that this model has a rate of correct classifications of 99.5%. Figure 4 shows the confusion matrix of the model supporting this fact. Finally, it is worth mentioning that this model has a rate of correct classifications of 99.5%. Figure 4 shows the confusion matrix of the model supporting this fact. In addition, the area under the ROC is of 99.31% which confirms the good discrimination ability of the model for PU risk identification. Figure 5 shows the ROC curve supporting this fact. Therefore, this model produces the same classification as the Braden Scale with high accuracy.

Explanatory Model for Risk Pressure Ulcer Prognosis Based on Activity and Mobility Subscales
The estimated model for risk prognosis based on the activity and mobility subscales has the following form (Table 3 bellow includes the estimated parameters for this model): In addition, the area under the ROC is of 99.31% which confirms the good discrimination ability of the model for PU risk identification. Figure 5 shows the ROC curve supporting this fact. Finally, it is worth mentioning that this model has a rate of correct classifications of 99.5%. Figure 4 shows the confusion matrix of the model supporting this fact. In addition, the area under the ROC is of 99.31% which confirms the good discrimination ability of the model for PU risk identification. Figure 5 shows the ROC curve supporting this fact. Therefore, this model produces the same classification as the Braden Scale with high accuracy.

Explanatory Model for Risk Pressure Ulcer Prognosis Based on Activity and Mobility Subscales
The estimated model for risk prognosis based on the activity and mobility subscales has the following form (Table 3 bellow includes the estimated parameters for this model): Therefore, this model produces the same classification as the Braden Scale with high accuracy.

Explanatory Model for Risk Pressure Ulcer Prognosis Based on Activity and Mobility Subscales
The estimated model for risk prognosis based on the activity and mobility subscales has the following form (Table 3 bellow includes the estimated parameters for this model): Note: BAct = Activity, BMov = Mobility, B = estimated parameter, DT = standard deviation, Z = Z statistic, p = p-value, OR = odds ratio, CI = confidence interval, Lower = lower limit of the CI, Upper = upper limit of the CI.
The Chi-square log-likelihood test for this model was X 2 (8, N = 16,215) = 10,259.70, p < 0.001. Therefore, when these variables were included in the model, the fit improved significantly compared to a model than only takes the constant into account. The Stukel goodness-of-fit test for this model was X 2 (2, N = 16,215) = 2.90, p = 0.234 > 0.05. These results point out that the model produces a good fit, at the population level, for the prediction of PU risk.
Once again, in light of the results of the z-test (see Table 3), all the levels of the subscales are significant at a population-based level (p < 0.001). For the activity scale, it should be noted that the prognosis ratio of PU development for b patients is 177 times that in those who frequently wander (OR = 177.68; 95% CI: 117.92-275.89), 23 times higher for patients in the chair-wandering category (OR = 23.34; 95% CI: 16.44-34.47), and 4 times higher for patients who occasionally wander (OR = 4.14; 95% CI: 2.89-6.05). As regards the scale of mobility, the advantage of developing a pressure ulcer was 148-fold in completely limited patients when compared to those with no limitation (OR = 148.41; 95% CI: 70.11-368.71), 30 times higher in very limited patients (OR 29.96; 95% CI: 14.73-72.24), and 3 times higher in slightly limited patients (OR = 2.97; 95% CI: 1.46-7.24).
As for the previous model, it is relevant to mention that this model has a rate of correct classifications of 85.8%. Figure 6 shows the confusion matrix supporting this fact.  Note: BAct = Activity, BMov = Mobility, B = estimated parameter, DT = standard deviation, Z = Z statistic, p = p-value, OR = odds ratio, CI = confidence interval, Lower = lower limit of the CI, Upper = upper limit of the CI.
The Chi-square log-likelihood test for this model was X 2 (8, N = 16,215) = 10,259.70, p < 0.001. Therefore, when these variables were included in the model, the fit improved significantly compared to a model than only takes the constant into account. The Stukel goodness-of-fit test for this model was X 2 (2, N = 16,215) = 2.90, p = 0.234 > 0.05. These results point out that the model produces a good fit, at the population level, for the prediction of PU risk.
Once again, in light of the results of the z-test (see Table 3), all the levels of the subscales are significant at a population-based level (p < 0.001). For the activity scale, it should be noted that the prognosis ratio of PU development for b patients is 177 times that in those who frequently wander (OR = 177.68; 95% CI: 117.92-275.89), 23 times higher for patients in the chair-wandering category (OR = 23.34; 95% CI: 16.44-34.47), and 4 times higher for patients who occasionally wander (OR = 4.14; 95% CI: 2.89-6.05). As regards the scale of mobility, the advantage of developing a pressure ulcer was 148-fold in completely limited patients when compared to those with no limitation (OR = 148.41; 95% CI: 70.11-368.71), 30 times higher in very limited patients (OR 29.96; 95% CI: 14.73-72.24), and 3 times higher in slightly limited patients (OR = 2.97; 95% CI: 1.46-7.24).
As for the previous model, it is relevant to mention that this model has a rate of correct classifications of 85.8%. Figure 6 shows the confusion matrix supporting this fact.  In addition, the area under the ROC curve is of 91.68% which confirms the good discrimination of the model to identify PU risk using only these two subscales. Figure 7 shows the ROC curve supporting this fact.
In addition, the area under the ROC curve is of 91.68% which confirms the good discrimination of the model to identify PU risk using only these two subscales. Figure 7 shows the ROC curve supporting this fact.

Explanatory Model for Pressure Ulcer Risk Prognosis Based on Activity, Mobility, and Skin Moisture Subscales
The estimated model for risk prognosis bases on the activity, mobility, and skin moisture subscales has the following form (Table 4 bellow   Note: BAct = activity, BMov = mobility, BHum = skin moisture, B = estimated parameter, DT = standard deviation, Z = Z statistic, p = p-value, OR = odds ratio, CI = confidence interval, Lower = lower limit of the CI, Upper = upper limit of the CI.
The chi-square log-likelihood test for this model was X 2 (8, N = 16,215) = 7207.69, p < 0.001. Therefore, when these variables were included in the model, the fit improved significantly compared to a model than only takes the constant into account. The Stukel goodness-of-fit test for this model was X 2 (2, N = 16,215) = 33.54, p < 0.001. These results concluded, therefore, that the model did not produce a good fit at population level with the risk of developing pressure ulcer.
Once again, in light of the results of the z-test (see Table 3), all the levels of the subscales are significant at a population-based level (p < 0.001 or p < 0.05). For the activity scale, it should be note that the prognosis ratio of PU development for bedridden patients is 190-fold that of patients who frequently wander (OR = 190.57; 95% CI: 79.04-533.79), 30 times higher for patients in the chair-wandering category (OR = 29.96; 95% CI: 13.20-80.64), and 3 times higher for those who occasionally wander (OR = 2.72; 95% CI: 1.20-7.32). As regards the scale of mobility, the advantage of developing a pressure ulcer was 190-fold in completely limited patients when compared to those with no limitation (OR = 190.57; 95% CI: 117.92-314.19), 18 times higher in very limited patients (OR 17.64; 95% CI: 11.70-27.39), and 3 times higher in slightly limited patients (OR = 3.16; 95% CI: 2.12-4.90). For the skin moisture scale, the prognosis ratio for risk of pressure ulcer is multiplied by 450 if the patient is constantly wet, when compared to those with no problems of skin moisture (OR = 450.34; 95% CI: 304.90-678.58), the risk is multiplied by 79 if the patient is often web (OR = 79.04; 95% CI: 63.43-99.48), and by 10 times if the patient is occasionally wet (OR = 10.28; 95% CI: 8.50-12.68).
As for the previous model, is relevant to mention that this model has a rate of correct classifications of 90.2%. Figure 8 shows the confusion matrix of this model supporting this fact.
concluded, therefore, that the model did not produce a good fit at population level with the risk of developing pressure ulcer.
Once again, in light of the results of the z-test (see Table 3), all the levels of the subscales are significant at a population-based level (p < 0.001 or p < 0.05). For the activity scale, it should be note that the prognosis ratio of PU development for bedridden patients is 190-fold that of patients who frequently wander (OR = 190.57; 95% CI: 79.04-533.79), 30 times higher for patients in the chair-wandering category (OR = 29.96; 95% CI: 13.20-80.64), and 3 times higher for those who occasionally wander (OR = 2.72; 95% CI: 1.20-7.32). As regards the scale of mobility, the advantage of developing a pressure ulcer was 190-fold in completely limited patients when compared to those with no limitation (OR = 190.57; 95% CI: 117.92-314.19), 18 times higher in very limited patients (OR 17.64; 95% CI: 11.70-27.39), and 3 times higher in slightly limited patients (OR = 3.16; 95% CI: 2.12-4.90). For the skin moisture scale, the prognosis ratio for risk of pressure ulcer is multiplied by 450 if the patient is constantly wet, when compared to those with no problems of skin moisture (OR = 450.34; 95% CI: 304.90-678.58), the risk is multiplied by 79 if the patient is often web (OR = 79.04; 95% CI: 63.43-99.48), and by 10 times if the patient is occasionally wet (OR = 10.28; 95% CI: 8.50-12.68).
As for the previous model, is relevant to mention that this model has a rate of correct classifications of 90.2%. Figure 8 shows the confusion matrix of this model supporting this fact. In addition, the area under the ROC is of 96.06%, which confirms the good discrimination of the model to identify risk or absence of risk regarding pressure ulcers with this subscale only. Figure 9 shows the ROC curve supporting this fact. In addition, the area under the ROC is of 96.06%, which confirms the good discrimination of the model to identify risk or absence of risk regarding pressure ulcers with this subscale only. Figure 9 shows the ROC curve supporting this fact.

Discussion
This work aimed at identifying groups of Braden subscales that provide efficient classification models, and at quantifying the effect of each subscale within the model for prognosis at a level of worsening risk of developing pressure ulcers for immobilized patients. With regard to the first objective and the Braden subscales as explanatory variables, three models that provide a first approximation of level changes in the risk of developing pressure ulcers were obtained. These models included all the Braden subscales considered as relevant to fit them. They are able to predict the probability of an individual being at risk of developing a pressure ulcer. The validation of the model including all the scales showed the same behaviour as the Braden Scale for pressure ulcers risk assessment. The mobility and activity subscales are relevant risk factors involving increasing risk of pressure ulcers. Indeed, these two scales provide a reliable model for risk classification due to the high values of the different internal validity parameters and safety indices. The mobility, activity, and skin moisture scales, jointly, even improve the reliability due to higher values of these parameters. With regard to the second objective, the result showed that high levels of all the scales were associated to situations of greater risk of a pressure ulcer developing. It is relevant to highlight the mobility and activity subscales that were associated to these risk situations for all the models. The Braden's mobility subscale is considered as an independent risk factor for PU development [21,22]. The activity subscale is considered with an overall pooled effect in [23]. However, the subscale skin moisture is also considered in [23] but it did not reach significance. Mobility and activity, jointly, are considered in [16] with patients residing in the long-term care. According to the results of this study, immobilized patients characterized by high levels of activity and mobility, jointly, understanding this fact as a higher limitation, suffer greater risk of developing pressure ulcers.
After the results obtained in this work, regarding the validity of the mobility and activity subscales of the Braden Scale as predictors of pressure ulcer risk, it is reasonable to consider how professionals make decisions using a methodology that adds little value, versus other simpler and more efficient ones, that involved the nurse as an expert on the patient [24,25]. In this sense, a change in practice could be proposed using these subscales due to their high predictive value to identify the risk of pressure ulcers quickly and efficiently [1]. Fast decision-making results in the implementation of an adequate care plan from practically the first signs of risk of developing pressure ulcers. Consequently, the complete Braden Scale could ultimately be used for the categorization and origin of risk as an essential tool for organizing the different resources for patients.
Finally, it is very important to encourage nursing professionals to use their experience to assess the risk of developing pressure ulcers in these patients in order to develop strategies for PU prevention and thus reduce the health burden associated with pressure ulcers.

Clinical Implications
Nurses are essential in the early care of patients. This is the reason for a great demand for care that overloads professionals with work. On the other hand, PUs affect the quality of life of patients and their families. This study provides nurses with confidence in their professional criteria based on the mobility and activity subscales, jointly. This evidencebased practice can be improved with the skin moisture subscale, along with the previous ones. In addition, the six subscales can be also considered by means of a quick-evaluation model for risk assessment of developing pressure ulcers.

Conclusions
Immobilized patients are at greater risk of PU incidence. This represents a health problem with a significant impact in the quality of life of affected people and their families. Therefore, quick decision making by health care professionals becomes paramount for the application of preventive measures. In this sense, the objective nursing criteria is an essential addition. The SIRUPP study provides relevant information to ensure the trustworthiness of a diagnosis of PU risk based only in the objective experience of the health professional. Funding: This study has been totally funded by Consejería de Salud, Junta de Andalucía (Fundación Pública Andaluza Progreso y Salud, Proyecto AP-0086-2016). Funding has been received for data collection, analysis, interpretation of data, and manuscript writing.

Institutional Review Board Statement:
The study was conducted in accordance with the Declaration of Helsinki and approved by the Ethics Committee of the Andalusian Public Health System (AP-0086-2016).

Informed Consent Statement:
This study complies with the basic ethical principles governing responsible conduct in research involving human subjects. Written informed consent was sought from all participants. The patients' data were anonymized in compliance with Spanish regulations. Data Availability Statement: Data available upon request to the authors.