Internet-connected medical devices introduce complex cybersecurity risks that challenge the established practice of informed consent. It remains unclear how patients weigh these abstract, dynamic threats against concrete clinical benefits. We present findings from a large-scale (N=2,666) vignette-based experiment designed to uncover the factors driving patient decision-making. Participants chose whether to adopt a connected pacemaker, weighing its enhanced clinical outcomes against potential vulnerabilities. We systematically varied communication factors, including the source of risk information (e.g., clinician, FDA), risk framing, and the details of a subsequent vulnerability disclosure. Our results reveal patient choice hinges on pre-existing physician trust and risk framing. We did not observe any effect from the information’s source. We also find initial choices act as powerful anchors, and that detailed disclosures increase security confidence. Our work provides crucial empirical evidence on this trade-off, offering actionable guidance to better support informed consent for life-critical connected technologies.

OBJECTIVES: To compare performances of a classifier that leverages language models when trained on synthetic versus authentic clinical notes. MATERIALS AND METHODS: A classifier using language models was developed to identify acute renal failure. Four types of training data were compared: (1) notes from MIMIC-III; and (2, 3, and 4) synthetic notes generated by ChatGPT of varied text lengths of 15 (GPT-15 sentences), 30 (GPT-30 sentences), and 45 (GPT-45 sentences) sentences, respectively. The area under the receiver operating characteristics curve (AUC) was calculated from a test set from MIMIC-III. RESULTS: With RoBERTa, the AUCs were 0.84, 0.80, 0.84, and 0.76 for the MIMIC-III, GPT-15, GPT-30- and GPT-45 sentences training sets, respectively. DISCUSSION: Training language models to detect acute renal failure from clinical notes resulted in similar performances when using synthetic versus authentic training data. CONCLUSION: The use of training data derived from protected health information may not be needed.