Health technology assessment of artificial intelligence in radiology within Kenya; multicriteria decision analysis

Abstract

Enhanced technology in healthcare improves patient outcomes, stream lines operations and enhances diagnostic capacity. Enhanced diagnostic capacity is fueled by advanced development and adoption of artificial intelligence (AI) solutions. Europe and North America have made bigger strides in the regulation and implementation of AI in radiology compared to Africa and Asia. AI technology can be developed locally or imported for implementation. Assessment of software as a medical device locally remains unclear which affects comprehensive standardization and articulation of its value proposition. This informs the need for a health technology assessment (HTA) tool for systemic appraisal of economic, social, ethical and clinical health care priorities in Kenya. Radiologists in America and Europe demonstrate that embedding user centered artificial intelligence in radiology improves efficiency and effectiveness in the radiology workflow. Few radiologists use AI in radiology within Kenya and more are willing to train artificial intelligence models despite multi factorial barriers. A normative assessment framework remains unclear in the design, development and clinical implementation of AI in radiology within Kenya. This study developed a health technology assessment (HTA) tool for AI in radiology using a hypothetical AI lesion detection software for pulmonary embolism (PE) on computed tomography pulmonary embolism (CTPA) images. The study evaluated AI in radiology using radiology multi-society practical considerations (MSC), the radiology AI deployment and assessment rubric (RADAR), and developed a local HTA tool for AI in radiology. Relevant institutional ethical permission was granted, data privacy and confidentiality was observed. The tool was piloted among 3 domain experts for construct and face validity, data was collected from 54 decision makers through online surveys and extracted into spreadsheets for analysis. This was an action research methodology on deductive themes analyzed using a multicriteria decision analysis (MCDA). Descriptive proportions were presented using categorical data where majority of the participants favored the third iteration of the tool. Inferential statistics analyzed using SPSSv25 confirmed intra class correlation co efficient using Cronbach α>0.7 with significant reliability (p=0.001) for each deductive theme. Each item on the final HTA tool demonstrated statistical significance (p>0.05). Despite hypothetical application of the AI software, HTA tool provides a foundation for comprehensive evaluation of AI in radiology with good reliability. The HTA tool iteratively supports multicriteria decision making among strategic decision makers on AI in radiology against global standards.

Key words: Health technology assessment, Artificial Intelligence, Radiology, Multicriteria Decision Analysis