In this systematic review, we searched nearly 20,000 titles and abstracts in order to identify 38 studies that investigated the diagnostic accuracy of MRI or CT for detecting a vascular component to dementia. Only four of these studies assessed both imaging methods. Included studies were generally small and many were at high risk of bias due to the potential for biased selection of patients and possibility that test results were incorporated into the reference standard. However there was little evidence that these sources of bias impacted on estimates of accuracy. Only seven studies used autopsy as the reference standard, and their results were heterogeneous. Among the 31 studies that used a non-autopsy reference standard, no individual imaging finding was assessed in a majority of studies, and results were heterogeneous. White matter hyperintensities were the most frequently assessed imaging finding, but based on summary estimates of sensitivity and specificity this finding had limited utility for ruling in or ruling out a diagnosis of VaD or mixed dementia. The presence of general infarcts showed the greatest potential for ruling in a diagnosis of VaD or mixed dementia, but none of the findings appeared sufficiently sensitive to rule out a diagnosis of AD. Comparative analyses suggested that MRI may be more accurate than CT for distinguishing vascular or mixed dementia from Alzheimer’s disease and other conditions, but confidence intervals on estimated ratios of diagnostic odds ratios were wide.
We performed a comprehensive search, without language restrictions, to identify both published and unpublished literature: thus it is unlikely that relevant studies have been missed. We employed systematic review methods to minimise bias and errors during study selection, data extraction and quality assessment and used the most rigorous methods of meta-analysis for diagnostic accuracy data. We made both direct and indirect comparisons of the accuracy of CT and MRI, but were limited by the substantial between-study heterogeneity and small number of studies that directly compared the imaging methods. We assessed study quality using accepted criteria for diagnostic accuracy studies and investigated the effects of potential sources of bias in the analysis. Most of the included studies did not enrol an appropriate patient spectrum, which we defined as patients with symptoms of dementia in whom the diagnosis had not been confirmed. In practice, MRI and CT will have most clinical value if used at a relatively early stage in the diagnostic work-up of patients with symptoms of dementia, in order to help reach a definitive diagnosis and begin appropriate treatment early in the course of disease. Studies that did not assess MRI and/or CT in this context may produce less applicable, or biased, estimates of diagnostic accuracy: for example if they used a case–control design where cases already have a confirmed diagnosis of dementia subtype, or if they were conducted in patients with a longer duration of illness.
We stratified the analysis based on whether studies used an autopsy or non-autopsy reference standard. Because only a small number of autopsy studies were available, the impact of the type of reference standard on estimated diagnostic accuracy could not be evaluated with precision: no more than three autopsy studies assessed any individual imaging finding and most findings were assessed in only one or two studies. Although we considered autopsy to be the least biased reference standard there is a potential risk of disease progression bias as there will be a time lapse between the imaging and the autopsy examination. This means that some patients may not have had VaD orAD when they were assessed by MRI/CT but have developed one of these conditions before they died. This has the potential to impact on estimates of sensitivity and specificity, depending on whether the original reference standard is more likely to wrongly classify patients as VaD or AD. There was a risk of bias due to incorporation of test results in the reference standard, in many studies that used a non-autopsy reference standard. There was a suggestion that incorporation bias resulted in greater diagnostic accuracy for studies of CT but this was not found for studies of MRI. We would expect incorporation bias to increase agreement between the index test and reference standard leading to inflated estimates of sensitivity and specificity [21
In the United States, the use of either CT or MRI as part of the diagnostic work-up of a dementia patient is recommended [22
]. The UK National Institute for Health and Clinical Excellence (NICE) guidelines on dementia diagnosis state that structural imaging should be used in the assessment of suspected dementia to exclude other cerebral pathologies and to help establish the subtype diagnosis [6
]. MRI is referred to as the preferred method to detect subcortical vascular changes, although it is acknowledged that CT could also be used. A 1988 narrative review by Joyce and Lishman [23
], which discussed 9 studies, concluded that neither CT nor MRI are reliable in the differential diagnosis of AD and VaD.
Both CT and MRI technology have developed considerably in the time since the majority of the included studies were conducted. For example, helical CT with multiplanar reconstruction is now routinely used and has higher image resolution than the CT scans evaluated in the included studies. Modern CT may be considered to be preferential to MRI because it is quicker and much cheaper to buy and run, it is more comfortable for the patient and there are fewer contraindications to its use. It can be reconstructed in the coronal plane for direct visual assessment of hippocampal volume. These factors should be weighed against increased exposure to ionising radiation exposure with CT. In the future, fluorodeoxyglucose (FDG) - positron emission tomography (PET) may be useful in predicting decline in normal subjects and individuals with mild cognitive impairment [24
]. Abeta-PET appears most useful in distinguishing AD from other dementias, although it has recently been suggested that a combination of Abeta- and FDG-PET may be more accurate. However, neither of these techniques is widely available in many hospital settings [25
New diagnostic accuracy studies are needed to compare the utility of the latest generation of MRI and CT techniques in detecting a vascular component to dementia. The design of studies should aim to avoid the weaknesses of the studies located for this review. They should assess both MRI and CT in the same group of patients with symptoms of early dementia. Study size should be large enough to allow precise estimates of relative diagnostic accuracy. The reference standard should consist of accepted diagnostic criteria, without incorporating imaging findings, ideally supplemented by autopsy confirmation. Global assessment criteria for MRI and CT, based on the most useful individual imaging findings that are indicative of a vascular component to dementias, should be established, and their diagnostic accuracy quantified.