Investigação do nexo de causalidade na imputação da responsabilidade civil do médico por falha de software como dispositivo médico (Software as a Medical Device - SaMD)

Abstract

The dissertation examines the impact of artificial intelligence (AI) in medicine, specifically investigating the causal nexus in the attribution of medical civil liability. The study focuses on Software as a Medical Device (SaMD) used for diagnostic support and is structured into three main areas, reflecting the interdisciplinary nature of the topic. The first chapter presents the medical perspective, discussing the historical and current use of AI in healthcare, with an emphasis on applications, benefits, and limitations of SaMD. The second chapter adopts a technological approach, analyzing key technical aspects such as AI hallucinations, explainability, biases, emergent behavior, and autonomy. The third chapter explores the regulatory landscape of AI at both national and international levels, while the fourth chapter introduces the legal dimension, beginning with classical civil liability doctrine, with a focus on the causal nexus. The study is guided by key research questions, including whether every AI application in healthcare requires medical supervision, whether SaMD can be considered the author of the recommendation or if this responsibility remains with the physician, whether the use of SaMD transforms medical practice into an obligation of result, and how the introduction of this technology impacts the assessment of causal nexus in cases of diagnostic errors resulting from the interaction between technology and medical practice. The research employs an exploratory legal approach, combining literature review and normative analysis. The dissertation concludes that not all AI applications in healthcare require human supervision (Human-in-the-Loop - HITL), as the need for oversight must be assessed on a case-by-case basis. In low-risk contexts, such as simple laboratory tests, supervision may be waived, whereas in high-risk scenarios, such as oncological diagnoses, human oversight is essential to mitigate risks and ensure patient safety. Regarding the possibility of SaMD being considered the author of the recommendation in the physician-patient relationship, the study determines that the physician remains the ultimate decision-maker in clinical decisions, in accordance with guidelines from the Federal Council of Medicine (CFM), using SaMD strictly as a decision-support tool. Concerning the potential transformation of medical practice into an obligation of result, the dissertation reaffirms that medical practice remains an obligation of means, considering the unpredictable factors inherent in therapeutic outcomes. Finally, regarding causal nexus analysis in cases involving SaMD, the dissertation concludes that the interaction between technology and medical practice necessitates a detailed assessment of professional diligence, taking into account factors such as clinical protocols, software validation, physician training, justification of decisions, and informed patient consent. Additionally, the research recommends that professional councils establish guidelines for the use of AI-based SaMD in diagnostic activities, while also encouraging the training of healthcare professionals in the use of these tools. It further suggests active participation from regulatory agencies such as ANVISA. As a contribution, the dissertation offers an integrated approach combining law, medicine, and technology, proposing practical solutions to the challenges posed by AI in the healthcare sector.