Abstract TP074: Covert cerebrovascular disease detected by artificial intelligence (C2D2AI): preliminary baseline data for patient enrollment feasibility

Abstract

Introduction: Covert cerebrovascular disease (CCD), including covert brain infarction (CBI) and white matter disease (WMD), is frequently found incidentally on routine neuroimaging. Though without overt symptoms, CCD shares pathophysiology with symptomatic stroke and is a strong predictor of stroke and dementia. We developed a natural language processing (NLP) tool to identify incidentally discovered CCD from radiology reports for head CTs and brain MRIs. This study evaluates the feasibility of recruiting patients with CCD, characterizing their baseline risk, and assessing behavioral changes to support the design of future interventional trials. Methods: Participants are recruited at Tufts Medical Center, a diverse urban, tertiary care, academic hospital. 50 participants will be enrolled, both prospectively (within 30 days of index scan) and retrospectively (scanned within the last 5 years) based on NLP-identified CCD findings. Each neuroimage is reviewed to verify the presence of CCD. Primary care providers are contacted and given the option to exclude patients. Patients are screened by phone, then assessed in person. Baseline assessments include a review of the patient’s medical history and medications, the NIH Stroke Scale, the Montreal Cognitive Assessment (MoCA), and validated questionnaires. Follow-up at 90 days is performed to evaluate changes in medications, attitudes, and health-seeking behavior. Results: Of 382 eligible participants, letters were sent to 328. Phone calls have been completed for 218 participants. 63 participants were scheduled for Visit 1, of whom 18 were no shows. 40 participants completed Visit 1, of whom 37 were eligible for Visit 2. All patients had WMD and 10% had CBI. 2/3 had mild WMD, and 1/3 had moderate or severe WMD. The majority of patients were identified by CT. Baseline characteristics (Table 1) showed a mean age of 71 with a slight female predominance. Measurements of blood pressure and hemoglobin A1c suggest the presence of modifiable stroke risk factors. Median MoCA scores (24.5, IQR 21-26) suggest prevalent mild cognitive impairment. The median Brain Care Score (15, IQR 13-17) suggests potential for improvement in brain health promoting behaviors. Conclusions: Recruitment of CCD patients using NLP is feasible, well-received, and proceeding faster than planned, with strong engagement from patients and providers. These findings support future research on preventive strategies in this neglected, high risk population.

Ähnliche Arbeiten