Predicting post-stroke dementia: mechanisms and markers

It is well recognised that a stroke triggers a significantly increased risk of accelerated dementia onset1,2. Moreover, while many of these patients present with post-stroke cognitive impairment, and all receive clinical post-stroke scans (CT or 1.5T MR), current methods lack the ability to discriminate those who will decline further from those who will retain, or regain, cognitive function. Highly sensitive cognitive assessment will be used in combination with Machine learning (ML) algorithms, which generate pattern classifications from large, complex datasets3, to establish a method to detect those at high risk of cognitive decline and dementia. The middle cerebral artery (MCA) is the most common artery involved in acute stroke, and MCA strokes very frequently impact the inferior parietal lobe (IPL) and result in accelerated dementia. The IPL is recognised for its role in the allocation of attention, and clinically, is associated with visuospatial neglect. However, this region is also fundamental to integrating remembered information into a subjective memory experience4,5. IPL damage can prevent connectivity with the medial temporal lobe (MTL) and impair memory and memory experience, thus contributing substantially to post-stroke cognitive impairment and accelerated dementia onset. We have recently developed a new method for estimating IPL memory function by measuring the region's influence on attention allocation. Performance on this bi-hemispheric test of IPL attention function is closely associated with hippocampally-dependent recognition memory. We therefore expect a patient's attention profile to indicate the integrity of their IPL memory function and IPL-MTL connectivity, and thus identify individuals at risk of further cognitive decline. Alongside traditional measures of cognitive decline and our bi-hemispheric IPL assessments, we will track cognition using highly sensitive measures of pattern separation-dependent, hippocampally-driven memory function in parietal stroke patients. This memory process is particularly vulnerable in very early dementia, and its measurement will therefore allow us to detect the emerging memory-component of cognitive decline with high levels of sensitivity and specificity. We will use this data to develop supervised ML algorithms (e.g., random forest classifier) to predict accelerated dementia from standard clinical scans. As stroke patients rarely receive high-resolution scans (e.g., 3T MR), using standardly available CT and 1.5 MR data in the development of this prognostic pathway maximises translational value.