ContinuOuS Monitoring Tool for Delayed Cerebral IsChemia (COSMIC)

点击翻译按钮获取中文摘要

Project Summary/Abstract Approximately 30,000 Americans suffer an aneurysmal subarachnoid hemorrhage (SAH) each year, at a mean age in the mid-50s leading to many years of lost productivity. Delayed cerebral ischemia (DCI) occurs in every fifth patient with SAH with onset between 3-7 days after aneurysm rupture, and is the leading cause of morbidity. Identifying the onset of DCI is challenging even though patients are closely monitored in intensive care units, and too often DCI is only recognized in retrospect. There are several reasons for this: (1) cerebral ischemia results in loss of function and is not passively observable in a neurologically injured patient, (2) can be mistaken for mimics such as seizure or delirium and delay diagnosis, (3) confirmatory testing is resource heavy and carries potential risk which necessitates surpassing a high threshold of suspicion. Existing DCI prediction models do not offer the necessary timeliness nor precision. Improving timeliness and precision of DCI prediction would enable interventions to prevent strokes in patients with SAH as well as reduce overly aggressive treatment. Leveraging the impact that the inflammatory pathomechanism of DCI has on systemic physiology, we created an artificial intelligence (AI) risk score for DCI using features derived from universally available vital signs that updates with new information. In a pseudo-prospective experiment on data from external institutions, this risk score uniquely met the criteria for an ideal situational monitor that does not yet exist: continuous, non-invasive, independent of pretest probability, operator-independent, quantitative, and timely (12 hours before clinical diagnosis). The World Health Organization standard of ethics for AI in healthcare decrees that algorithms should be tested rigorously in the setting in which the technology will be used, and ensure that it meets standards of safety and efficacy. The risks of an untested AI based clinical decision support are misinterpretation and over-trusting with harm to patients at worst, and inconsequence at best. This proposal encompasses the necessary steps to translate this promising model into a tool that can be integrated into clinical practice. In Aim 1, we will perform a Silent Validation and Simulation Study to evaluate the accuracy and acceptance of this novel AI technology in a realistic clinical setting. In Aim 2, we will use Contextual Design methodology for user-centered participatory design and rapid agile prototyping to refine the optimal implementation in clinician workflow. In Aim 3, we will produce an open standards-based interoperable architecture that will be plug and play for implementation at external institutions. The translation of a DCI risk model into a continuous monitor fills an important gap in the management of patients with SAH, and an open standards architecture enables affordable and rapidly achievable dissemination of this novel technology, while providing an essential validation for the standards community.