Methods
Setting
The setting for this study was 3 adult ICUs in the OSUWMC. OSUWMC serves as a major referral center for patients from Ohio and throughout the Midwest. OSUWMC has 2 medical ICUs and 1 surgical ICU. The medical ICUs each have 39 beds, and the surgical ICU has 44 beds. In the ICUs, Braden scores are recorded for each patient on admission and periodically thereafter, with patients scoring at risk for pressure ulcers reassessed more often than patients with lower risk scores.
Data Extraction
For data extraction from the Information Warehouse, the following eligibility criteria were documented and applied: adult patients (age ≥18 years) admitted to ICUs between January 1, 2007, and December 31, 2010, comprised the sample, with 2 exceptions. Patients whose ICU stay was shorter than 3 days were excluded, because it is reported that pressure ulcers usually develop more than 72 hours after admission. Additionally, patients who had a pressure ulcer at the time of admission were excluded, thereby enabling inclusion only of patients who acquired a pressure ulcer during the hospital stay.
Patients who had a pressure ulcer develop were identified by reviewing discharge diagnoses represented by using codes from the International Classification of Diseases,Ninth Revision (ICD-9). For instance, if a patient had an ICD-9 code of 707.05 (Pressure ulcer, Buttock), the patient was considered as a case and included in the pressure ulcer group. On the other hand, if a patient did not have any of the ICD-9 codes representing pressure ulcers, the patient was included in the non–pressure ulcer group (comparison).
Approval for the data extraction was obtained from the institutional review board. Data were de - identified by the staff in the Information Warehouse as the honest broker. Data elements included demographics (age, sex, and race/ethnicity), length of ICU stay, admission and discharge diagnoses, and Braden score at admission. All data included in the present analysis were coded (structured) data; free text data (unstructured) were not included.
Data Analysis
Python scripts (Python Software Foundation) and MySQL database were used for data cleaning and preparation for analysis. For instance, if a patient had more than 1 ICU admission record during the study period, only the first admission record was included in the analysis. Patients' demographics and the incidence of pressure ulcers were summarized by using descriptive statistics. Pressure ulcer and non–pressure ulcer groups were compared by using the χ test for categorical variables and an independent 2-sample t test for continuous variables. Predictive validity was measured by using sensitivity, specificity, positive predictive value, and negative predictive value. The receiver operating characteristic curve was generated, and the area under the curve (AUC) was calculated.
Sensitivity is the probability that the scale classifies a patient as at risk of pressure ulcer development when given the patient has a pressure ulcer, that is, the true-positive rate. A false-positive result occurs when a patient does not have a pressure ulcer, but the scale classifies the patient as at risk of pressure ulcer development. Specificity is the probability that the scale does not classify a patient as at risk of pressure ulcer development given that the patient does not have a pressure ulcer, that is, the truenegative rate. A false-negative result occurs when a patient has a pressure ulcer but the scale does not classify the patient as at risk of pressure ulcer development; thus, true instances are missed by the scale. Positive predictive value is the proportion of patients classified as at risk who actually have pressure ulcers develop, whereas the negative predictive value is the proportion of patients classified as not at risk who do not have pressure ulcers develop. It is ideal that all 4 indicators have high values, but in reality, when the sensitivity goes up, the specificity goes down.
A cutoff point for classification of a patient as at risk is generally determined by considering costs of a false-positive result, the significance of missing a case, and the prevalence of the disease. When it is important to identify patients who are likely to have a pressure ulcer develop and should receive intensive preventive interventions, weight should be given to sensitivity and negative predictive value. Such an emphasis will ensure that most patients at high risk for pressure ulcers will not be missed; however, it may result in overuse of preventive resources on patients who may not have a pressure ulcer develop because of the potential for a high false-negative rate.
The receiver operating characteristic curve displays the trade-off between sensitivity and specificity for a range of test scores. True-positive rate (sensitivity) is plotted on the vertical axis against the false-positive rate (specificity) on the horizontal axis over a range of potential cutoff scores. The AUC is a measure of how well a scale can discriminate between 2 groups, for example, pressure ulcer group vs non–pressure ulcer group. A higher AUC value means a higher discriminating power. An AUC of 1 indicates a perfect accuracy whereas an AUC of 0.5 means no better than random chance. All statistical analyses were performed with SPSS version 19.0 for Windows (SPSS Inc).