Abstract Body: Introduction: Covert brain infarcts (CBIs) are associated with risk of stroke and dementia. It is unknown whether surveillance for CBIs and medical management can mitigate this risk. Stratification tools to identify patients who have CBIs may prove useful in clinical and research settings. We devised a vascular risk-factor-based scoring system to detect CBIs on MRI using a retrospective analysis of the Northern Manhattan Stroke Study (NOMAS) population-based cohort and compared it to the Atherosclerotic Cardiovascular Disease Score (ASCVD).

Methods: Inclusion criteria were age > 40 years, no prior stroke or TIA, and available MRI brain scan. Exclusion criteria: missing information on risk-factor variables. We retrieved demographics and vascular risk factor information at the time-point closest to the MRI. Logistic regression modeling was used to identify variables that had strongest association with CBIs on MRI. Three models (CBI score, CBI score 1, CBI score 2) with varying number of variables were devised to identify abbreviated versions of the score. We estimated cut-offs for the three scores, sensitivity, specificity, positive and negative predictive values (PPV, NPV), and area under the receiver operating characteristic curve (AUROC).

Results: Among 1,290 included patients, 237 (18.4%) had CBI. The CBI, CBI score 1, and CBI score 2 scores had 8, 6, and 4 risk-factor variables, respectively (Table 1). The median CBI score was higher in patients with CBIs compared to those without (Table 2). Two cut-offs, ≥ 2 and ≥ 3, were chosen for sensitivity analysis based on the median scores. The sensitivity of the three scores for identifying CBIs was higher at a cut-off ≥ 2 (0.81) compared to ≥ 3 (0.59) (Table 3). The AUROC indicated moderate accuracy (0.66) for predicting CBI. The Atherosclerotic Cardiovascular Disease (ASCVD) risk score (>7.5%) had higher sensitivity (0.96) but similar accuracy (0.66) in comparison to the CBI scores. Compared to the CBI score, the shorter CBI score 2 had lower sensitivity (0.63) and similar accuracy (0.66) at a cut-off of ≥ 2.

Conclusion: The NOMAS-CBI score uses vascular risk factor information and detects CBI on MRI with high sensitivity and moderate accuracy. A point-of-care variant, the CBI score 2, had lower sensitivity but similar accuracy. The score is comparable to the ASCVD and is easier to perform. Further analysis on the validity of the scores across other prospective cohorts is underway.