Study design: Psychometric study based on retrospectively collected data.
Objective: Development of a pressure injury (PI) risk screening instrument for use during spinal cord injury (SCI) rehabilitation.
Setting: Tertiary rehabilitation center.
Methods: Medical charts of 807 inpatients participating in SCI rehabilitation were reviewed. Two models (recursive partitioning and logistic regression) were developed with demographic and Functional Independence Measure (FIM) variables and compared with the SCI Pressure Ulcer Scale (SCIPUS, n = 603) and Braden scale (n = 100) using modeling (n = 615) and validation (n = 192) datasets. Sensitivity and specificity analyses were completed for each model. Models yielding high sensitivity and area under the curve (AUC), while minimizing false negatives (FN < 0.5%) were preferred.
Results: In the modeling dataset, a single dichotomized FIM variable, Bed/Chair Transfers <4, was predictive of PI incidence (sensitivity = 97%, AUC = 74%, FN = 0.49%) and had similar metrics as the logistic regression model (sensitivity = 97%, AUC = 76%, FN = 0.49%). The recursive partitioning model had fewer FN (sensitivity = 98%, AUC = 75%, FN = 0.33%). When applied to the validation dataset, both models performed similarly. The SCIPUS performed poorly (AUC < 70%). When analyses were limited to cases with available Braden data and no admission PI, recursive partitioning outperformed the other methods for PI risk screening.
Conclusion: A recursive partitioning model, named the SCI-PreSORS (SCI Pressure Sore Onset Risk Screening), demonstrated promise for PI risk screening during inpatient SCI rehabilitation. Prospective validation of the new model is warranted.