TL;DR
This paper outlines a pilot study protocol for an AI-based application designed to assist nurses in assessing and documenting two common but visually similar skin conditions: Incontinence-Associated Dermatitis (IAD) and Pressure Injuries (PI). The goal is to evaluate if this AI tool can reduce assessment time, improve guideline adherence, and decrease nurse workload, ultimately enhancing patient care.
What Problem Does This Paper Address?
Nurses often face challenges differentiating between Incontinence-Associated Dermatitis (IAD) and Pressure Injuries (PI) due to their similar appearance and location, particularly in the pelvic area. This can lead to misclassification, delayed or inappropriate treatment, and increased patient burden. Current wound assessment and documentation methods can also be time-consuming and prone to inaccuracies, especially when relying on manual measurements and subjective assessments. The KIADEKU project aims to address these issues by developing an AI-powered application to support nurses in this critical aspect of care.
Methods
The study employs a controlled, non-randomized, pilot intervention design with a pre-post approach. Nurses in the control group will continue with standard wound care and documentation practices, while the intervention group will utilize the AI-based KIADEKU application. This application, developed through a technical sub-study, uses AI models trained on wound images to assist in differentiating, assessing, and documenting PI and IAD. Key measurements include:
- Duration of wound assessment, dressing change, and documentation: Measured via direct observation with standardized start and endpoints.
- Nurse task load: Assessed using the validated NASA TLX questionnaire.
- Guideline adherence: Evaluated using a set of five criteria derived from expert workshops and established guidelines.
- Accuracy of AI predictions: Validated against an on-site expert assessment.
- Usability of the AI application and standard documentation systems: Assessed using the UMUX-Lite questionnaire.
The study will also account for covariates such as nurse's professional experience, wound severity, and overall shift workload. Statistical analysis will primarily involve multiple linear regression models.
Datasets
The paper mentions that the AI model was trained using a "literature-based minimal data set for wound documentation of PI and IAD," which included 1,555 existing wound images. These images, along with associated characteristics, were extracted from the hospital information systems of two German university hospitals. The study team then validated, segmented, and annotated these images. The input data for AI prediction in the clinical setting includes wound images captured by the application and additional patient data entered by the nurse (e.g., immobility, incontinence, sensibility).
Key Insights and Ideas
- Bridging the IAD/PI diagnostic gap: The core innovation is the AI's potential to objectively differentiate between IAD and PI, a common point of confusion for nurses, thereby improving diagnostic accuracy.
- Streamlining documentation: By automating wound measurement and pre-populating wound characteristics, the AI aims to significantly reduce the time nurses spend on documentation.
- Enhancing evidence-based care: The AI application can provide guideline-based nursing interventions, promoting adherence to best practices.
- User-centered AI development: The project emphasizes involving nurses and wound experts throughout the development and evaluation process to ensure the AI tool meets practical needs and promotes user acceptance.
- Pilot study for real-world validation: This protocol outlines a crucial step in moving AI from research to clinical practice by testing its efficacy and usability in a real-world hospital setting.
Key Results
As this is a study protocol, no results are presented yet. The paper details the planned methodology for data collection and analysis. The sample size calculation suggests a need for 85 nurses, with a recruitment target of 95 to account for potential dropouts.
Limitations and Future Work
- Monocentric study: The study is conducted at a single university hospital, limiting the generalizability of findings to other healthcare settings.
- Non-randomized design and absence of blinding: Due to the nature of the intervention, blinding participants and observers is not possible, which could introduce bias.
- Consecutive observation bias: The consecutive observation of control and intervention groups might lead to awareness of pressure injury and IAD care, potentially influencing outcomes.
- Organizational delays: Inevitable clinical processes may cause delays in data collection.
- Limited scope of AI: The AI application is a prototype and its predictions require validation by nurses. It does not replace expert nursing judgment.
- Future work: The authors emphasize the need for further research to draw general conclusions about AI's potential in nursing wound management, particularly regarding patient outcomes and the impact on nurses' critical thinking. They also aim to actively involve nurses in future AI development and implementation.
Summary updated