Hemostasis, Hematoma Expansion, and Outcomes After Intracerebral Hemorrhage

点击翻译按钮获取中文摘要

PROJECT SUMMARY Intracerebral hemorrhage (ICH), bleeding into brain tissue, is often disabling or deadly. Complications of ICH compound the impact on patient outcome. Hematoma expansion is growth of the hematoma from the first (diagnostic) computed tomography (CT) scan to a follow-up CT scan, and occurs in 10-25% of patients. Reducing hematoma expansion reduces death and disability by keeping small hematomas small. Otherwise, hematoma expansion will compress brain tissue in the limited space in the skull. Hemostatic medication (e.g., prothrombin complex concentrate) is effective in reducing hematoma expansion in patients at high risk for hematoma expansion. However, our ability to predict hematoma expansion and select patients for hemostatic medication is limited. Unfortunately, hemostatic medication has potential side effects, such as myocardial infarction. Separately, seizures occur in about 10% of patients and may progress to status epilepticus (continual seizures). Unfortunately, a practice of widespread seizure medications also leads to complications, worse patient outcomes, and reduced health-related quality of life at follow-up. Patient care is harmed because we cannot select patients for potentially life-saving treatments (hemostatic treatment, seizure medications). ML/AI provide a solution to these roadblocks. We have developed ML/AI algorithms to solve these roadblocks of targeting hemostatic treatment and seizure medications in the context of an ongoing award to predict hematoma expansion using ML of multiple measures of blood clotting (biomarkers of hemostasis). However, significant ethical issues remain unresolved. ML/AI algorithms may be subject to unforeseen bias in the patient populations from which they are developed. Bias may lead to different performance in patients considered vulnerable by the NIH (e.g., minority race, rural location, access to healthcare services). Previous qualitative interviews with clinicians (nurses, pharmacists, physicians) have revealed deep ambivalence about the use of ML/AI. There is excitement about consistent treatment recommendations and availability at all hours, but also concerns about bias, accountability, and oversight. Such qualitative research has not been performed in the setting of ICH, which requires emergent decision making because hematoma expansion and seizures occur within hours of ICH symptom onset. In the context of our ongoing award, we will determine if preliminary ML/AI algorithms we have developed have different performance in vulnerable patients; if needed, we will develop methods to correct it, assisted by diverse data sources in hand and established predictors of hematoma expansion and seizures. We will perform qualitative interviews with clinicians, patients, and family members affected by ICH to identify and address concerns that must be addressed if ML/AI is to be trusted in ICH. This proposal will serve as a model for other acute diseases that require emergent decision making.