(PI-022) Initial Evaluation of the Use of AI for Pressure Injury Classification and Staging Compared to Clinicians’ Assessments: Applications in Wound Care Management

RENNER BURKLE, BSc – DIRECTOR OF PROFESSIONAL SERVICES, VANTIQ; DIANA EASTON, RN WCC – NURSE TRAINER, TELEMEDICINE SOLUTIONS LLC; KASHAELYN HOLLINS-HENDERSON, DNP APRN FNP-C – FAMILY NURSE PRACTITIONER, LTH FAMILY MEDICINE; JENN KLEINERT, LPN WCC – NURSE TRAINER, TELEMEDICINE SOLUTIONS LLC; JAMISON LUTEN, MSN APRN FNP-C CWOCN CSWD-C EMT-P – NURSE PRACTITIONER, THE WOUND PROS; BRETT RUDENSTEIN, SE – VP SALES ENGINEERING AND SERVICES, VANTIQ; SHISHIR SHAH, DO, CWS, ABPM – National Wound Medical Director, TEAM HEALTH

Introduction: Pressure injury classification and staging is a challenging element of pressure injury documentation due to variations in clinician training, proficiency, and knowledge. Errors in classification and staging can lead to inappropriate treatments as well as insufficient care team oversight of wounds with greater severity than the patient’s chart shows. While proficient wound clinicians are the gold standard of diagnosis, one approach to address documentation errors related to the shortage of skilled wound clinicians is to use deep learning to train an artificial intelligence (AI) model to predict pressure injury classifications and stages based on image analysis, which has potential as a decision support tool for clinicians in wound care practice. This study assesses the accuracy of the AI model’s predictions, using 12,000 labeled images and a larger data set of training images, to predict pressure injury classifications and stages, compared with a cohort of wound care clinicians.

Methods: An observational cross-sectional study conducted with 10 wound care clinicians was performed to test the prediction accuracy of an AI model. The clinicians documented their classifications and staging assessments on the tested images of pressure injuries including patients of variable age, skin tone, and wound complexity. The range, mean, and inter- and intra-rater reliability of clinicians' assessments of pressure wound classifications and stages will be evaluated and compared to the predictions generated by the AI model.

Results: Descriptive data, including clinicians' backgrounds, years of experience, specialized wound care experience, and the care settings in which they practice, will be presented. The scores of clinicians' assessments of classifications and stages will be compared with the prediction accuracy generated by the AI model, which exceeded 70% for the lowest measured parameter.

Discussion: The study reveals significant variability in the accuracy of pressure injury documentation among wound care clinicians, with some clinicians’ performance falling below the AI model. The implications for wound care practice include consideration of integrating an AI model into quality assurance processes, which could improve wound management oversight via automated AI-driven wound documentation review. Further analysis is needed to quantify clinical, operational and financial impacts on wound care practice.