ESJ Featured Article of the Month – April 2025

Prediction of dementia using CT imaging  in Stroke

Stroke contributes to a substantial proportion of dementia.  Patients who experience more severe stroke are at higher risk for developing dementia over the following years.¹ MRI imaging shows promise to identify imaging biomarkers which signify an increased risk of dementia. In a systematic review of 27 studies, including 13,144 participants with MRI performed at stroke diagnosis, cerebral atrophy, microbleeds and increasing severity of white matter hyperintensities were all associated with post-stroke dementia.²

While no disease modifying treatments exist to date for use in post-stroke dementia, improved identification of those at highest risk is important to allow time for future care planning, communication of risk, and resource allocation.

This month in ESJ, Hafdi and colleagues aimed to identify CT prognostic markers of dementia.³ While the images obtained from CT are a lot less detailed than MRI, CT has some practical advantages: CT is widely accessible, especially in resource limited contexts, is quicker and cheaper, carries less contraindications, and may be more accessible and require less training for non-radiologist readers. Furthermore, as the burden of stroke grows in low and middle income countries, development and validation of a dementia-prediction tool using CT is necessary.

Hafdi and colleagues recruited patients with stroke or TIA from 6 Scottish hospitals from 2016 to 2019. All participants had a CT scan performed after their admission for stroke. Scans were read for evidence of old infarction, white matter lesions, medial temporal atrophy, and overall atrophy. Presence of each markers was assigned a point to construct a ‘brain frailty’ score, which ranged from 0-4 with higher scores indicating a more frail brain. Participants were followed up 6-monthly with a comprehensive battery of neurocognitive assessments. A diagnosis of dementia (according to DSM-5 criteria) was established by consensus of the investigators based on cognitive tests, informant questionnaires and clinical records. The analysis was done using a cox proportional hazards model with adjustment for age, sex, cardiovascular disease, pre-stroke modified Rankin Score (mRS), history of prior stroke and NIHSS score.

Participants were  69-years-old on average, just over half were men, and vascular risk factors were common. Most participants had ischaemic strokes, but 13 participants with haemorrhagic strokes were included. The median NIHSS was 2 (indicating mildly disabling stroke) and the average prestroke mRS was 1, indicating mild baseline disability. Of 195 participants, 64 were diagnosed with dementia over 3 years follow-up. The mean brain frailty score was higher in participants with dementia (3.24±0.89 SD) compared to those without (2.37±1.22 SD). Increasing brain frailty score was associated with an increased risk of dementia (aHR 1.38; CI 1.00-1.89, per 1-point increase). Results of a sensitivity analysis which included death as a competing risk were consistent with the main results. When each component of the brain frailty score was assessed, only the association between severe medial temporal atrophy and dementia remained statistically significant after full adjustment (aHR 2.09, CI 1.07-4.08).

The investigators then constructed a predictive model comprising clinical parameters (age, cardiovascular disease, pre-stroke mRS, NIHSS, and AMT-10) in combination with brain frailty. The combined model had a c-statistic of 0.77 (0.71-0.83) to predict future dementia. (The c-statistic measures the discrimination of a predictive model, that is how well it separates individuals who develop the outcome from individuals who don’t. ⁴ A c-statistic of 0.5 indicates discrimination which is no better than chance or tossing a coin. C-statistics between 0.6 and 0.8 are accepted to have moderate discrimination, while c-statistics >0.8 have good discrimination.) The combined model demonstrated a marginal improvement in discrimination compared with clinical factors alone (c-statistic 0.76 [0.71-0.82]) and brain frailty markers alone (c-statistic 0.69 [0.62-0.76]).

This work highlights that widely accessible CT imaging features of prior infarcts, white matter disease and atrophy are markers of increased risk of dementia. However, the added value of these markers in addition to clinically assessed risk-factors is, according to these results,  marginal. Advantages of this work are its generalisability, with its broad inclusion criteria, and the robust and systematic follow-up with cognitive testing, which backs-up the clinical relevance of the dementia outcome. Additionally, the brain frailty score is straight-forward to calculate especially for non-radiologists and CT imaging is widely available. Disadvantages are that this model has not been validated in independent data, which is a necessary step to confirm its performance prior to clinical use. Predictive tools which are tested only in the data in which they were derived, are at risk of over-estimating the performance of the model. External validation of this approach may provide clinicians with a practical and useful tool to better identify patients at highest risk of dementia post-stroke.

Prediction of dementia using CT imaging in stroke

Hafdi et al.

Online first March 2025

—

References:

1. Pendlebury ST, Rothwell PM. Incidence and prevalence of dementia associated with transient ischaemic attack and stroke: analysis of the population-based Oxford Vascular Study. Lancet Neurol 2019; 18(3): 248-58.
2. Ball EL, Shah M, Ross E, et al. Predictors of post-stroke cognitive impairment using acute structural MRI neuroimaging: A systematic review and meta-analysis. Int J Stroke 2023; 18(5): 543-54.
3. Hafdi M, Taylor-Rowan M, Drozdowska B, et al. Prediction of dementia using CT imaging in stroke (PRODUCTS). Eur Stroke J 2025: 23969873251325076.
4. Pencina MJ, D’Agostino RB, Sr. Evaluating Discrimination of Risk Prediction Models: The C Statistic. Jama 2015; 314(10): 1063-4.