Majjouti Khalid, Priester Vanessa, Tapp-Herrenbrueck Michaela, Brehmer Alexander, Pinnekamp Hannah, Aleithe Michael, Fischer Uli, Kleesiek Jens, Hosters Bernadette
Department of Nursing Development and Nursing Research, University Hospital Essen, Essen, Germany.
Department of Clinical Nursing Research and Quality Management, Nursing Department, Hospital of the Ludwig Maximilian University (LMU) Munich, Munich, Germany.
BMC Nurs. 2025 Jul 1;24(1):808. doi: 10.1186/s12912-025-03448-4.
Differentiating between stage 1 or 2 pressure ulcer/pressure injury (PU/PI) and incontinence-associated dermatitis (IAD) poses a significant challenge for healthcare professionals, due to their visual similarity. Incorrect assessments may trigger inappropriate interventions, potentially resulting in delayed treatment. KIADEKU is a multi-center research project aimed at supporting the assessment and documentation of PU/PI and IAD, as well as the implementation of evidence-based care through an AI-based application in nursing care. This paper investigates how to integrate evidence from nursing science and clinical practice into the development of the proposed AI system.
We conducted a literature review of nursing criteria for wound assessment. Nursing experts iteratively evaluated the findings, leading to the definition of a Minimum Data Set (MDS) that the research team used to annotate wound images for AI training. We collected a data set of wound images from the medical records of two university hospitals. To ensure high data quality, we implemented a validation process involving up to four independent expert assessments of each wound image. We calculated Krippendorff's alpha to assess the internal consistency of the annotation process for reliability analysis. This study adhered to the TRIPOD-AI guidelines.
The differentiation between PU/PI and IAD primarily relies on clinical observation and visual inspection, with key factors including aetiology, anatomical location, and wound morphology. The validated MDS encompasses 18 wound-related and four aetiological categories, including visual and contextual patient data. The AI system consequently integrates wound images with categorical patient information. The reliability analysis of 1,521 annotated wound images indicates substantial agreement for wound type classification (α = 0.64, 95% CI 0.62-0.68) and fair to moderate agreement for PU/PI (α = 0.57, 95% CI 0.55-0.63) and IAD categorization (α = 0.27, 95% CI 0.20-0.36).
The integration of evidence from nursing science and practice into the AI development process using a mixed-methods approach, established a robust, evidence-based foundation. This approach yielded an innovative implementation of routine care data for AI training, advancing the field of AI-driven wound care solutions.
Registered with the German Clinical Trials Register (DRKS) on 2023-09-05.
DRKS-ID: DRKS00029961.
由于1期或2期压疮/压力性损伤（PU/PI）与失禁相关性皮炎（IAD）在外观上相似，因此对医护人员而言，区分二者是一项重大挑战。错误的评估可能会引发不恰当的干预措施，从而可能导致治疗延误。KIADEKU是一个多中心研究项目，旨在支持PU/PI和IAD的评估与记录，并通过在护理中基于人工智能的应用来实施循证护理。本文探讨如何将护理科学和临床实践的证据整合到所提议的人工智能系统的开发中。
我们对伤口评估的护理标准进行了文献综述。护理专家对研究结果进行了反复评估，从而确定了一个最小数据集（MDS），研究团队使用该数据集为人工智能训练标注伤口图像。我们从两家大学医院的病历中收集了伤口图像数据集。为确保数据质量高，我们实施了一个验证过程，对每张伤口图像进行多达四次独立专家评估。我们计算了克里彭多夫阿尔法系数，以评估标注过程的内部一致性用于可靠性分析。本研究遵循TRIPOD-AI指南。
PU/PI与IAD的区分主要依靠临床观察和目视检查，关键因素包括病因、解剖位置和伤口形态。经过验证的MDS涵盖18个与伤口相关的类别和4个病因类别，包括患者的视觉和背景数据。因此，人工智能系统将伤口图像与分类的患者信息整合在一起。对1521张标注的伤口图像进行的可靠性分析表明，伤口类型分类具有高度一致性（α=0.64，95%CI 0.62-0.68），PU/PI分类具有中等一致性（α=0.57，95%CI 0.55-0.63），IAD分类具有一般到中等一致性（α=0.27，95%CI 0.20-0.36）。
采用混合方法将护理科学与实践的证据整合到人工智能开发过程中，建立了一个强大的循证基础。这种方法为人工智能训练创新性地实施了常规护理数据，推动了人工智能驱动伤口护理解决方案领域的发展。
于2023年9月5日在德国临床试验注册中心（DRKS）注册。
DRKS编号：DRKS000299
