Cerebral Hemodynamics in Mild Cognitive Impairment: A Systematic Review

Inclusion criteria

1) Participants with MCI and not dementia; 2) TCD, MR, PET, SPECT, NIRS imaging modality; 3) Marker included a measure of cerebral hemodynamics or oxygenation.

Exclusion criteria

1) Dementia; 2) No measure of cerebral hemodynamics or oxygenation.

Meta-analysis was performed where there were data from two or more studies, with minimal to moderate heterogeneity (CBF, CBFv, CVRi, TOI, PI, and phase). Meta-analysis was performed using RevMan 5 software^© for Windows using a random effects model. All outcomes were continuous variables using the inverse variance method. Studies were excluded from meta-analysis if there were fewer than two studies examining the same vessel, no standard deviations provided, no control data, or control and MCI groups presented as mixed data. Where studies presented data for both the right and left middle cerebral artery (MCA) [47], both data were included in meta-analyses, as were data for proximal and distal MCA measures [48].

Heterogeneity between studies was reported as I² index, where: I 2 = [ Q – df / Q ] × 100

(Q = X² statistic, df = degrees of freedom).

I² index represents the percentage variation due to heterogeneity [49]. I² = 25% was considered low, I² = 50% moderate and I² = 75% high heterogeneity, and was significant if p < 0.01 [49]. Finally, funnel plots were produced to detect publication bias. Meta-regression analysis was not possible to explore heterogeneity as a sub-group analysis as all meta-analyses contained fewer than 10 studies [49].

Summary of characteristics of the studies

Twenty-six studies were suitable for inclusion, of which eight were TCD [18, 50–55], threecombining TCD and NIRS [47, 57], eight NIRS only [58–64], five fMRI [15, 65–67] and two using PET or SPECT [20, 40] (Table 1). Median quality score was 13/15. All studies lacked sample size calculations, and few demonstrated testing for validity and reliability (Supplementary Figures 1 and 2).

Twenty-two of the 26 studies were cross sectional [15, 53–69], and the remaining four studies were longitudinal [50–52, 60] (two mixed design) [52, 60]. Follow-up time in the longitudinal studies ranged from five weeks to two years [50–52, 60]. Sample sizes were small for most studies (total sample size range: 15–127).

The diagnostic classification system used for MCI varied between studies, five had no formal criteria [18, 68], the majority used the Petersen criteria (n = 14) [40, 67], and fewer used the Mayo [50, 55], Winbald [65], Jak [15, 69], and Reisberg [66] criteria. Several studies had extensive exclusion of other neurodegenerative, psychiatric conditions [15, 67] and psychotropic medications [59, 67]. Some studies restricted their population to solely amnestic MCI [47, 67]. Fifteen of the 24 studies collected data on vascular risk factors in addition to age, sex, and education years [15, 67]; four reported antihypertensive use[15, 60].

The mean age of participants varied between 57–80 years, and few had an equal gender split between participants. All studies reported MMSE except four [15, 69]. All participants had an MMSE >24, or equivalent, except one study (mean MMSE 23.7) [54]. One study stratified participants by ApoE4 genotype [55]. The vessel insonated varied; seven measured solely the MCA [47, 57], two insonated the vertebral artery (VA) andinternal carotid artery (ICA) [50, 56], and four examined multiple vessels [18, 68].

Markers and meta-analysis

Cerebral blood flow

Six studies measured baseline or change in CBF [40, 66–68]. Three studies [56, 67] did not find any difference in baseline CBF between controls and MCI. Liu et al. found total CBF was significantly lower in the MCI group, when normalized for total brain volume [56]. Akkawi et al. found that a threshold CBF of 558 ml/min, had a sensitivity of 68.4%, and specificity of 72.2% for detection of MCI [50]. Rivera-Rivera et al. examined multiple intra-cranial vessels using arterial spin labelling and 4D-MRI but only demonstrated significantly lower CBF in the proximal posterior cerebral artery (PCA) compared to age-matched controls [68].

In a meta-analysis of two studies, CBF corrected for brain volume was significantly lower in MCI (SMD: –0.51 (–0.94, –0.07), p = 0.02) (Fig. 2).

OPEN IN VIEWER

Fig.2

Forest plot for a meta- analysis of normalized cerebral blood flow (CBF). Data is pooled from two studies [56, 66]. Results are reported as standardized mean differences (95% confidence interval) as individual results are reported on different scales (ml/min/100 grams, and ml/min/100 ml). Statistical significance set at p < 0.05. Results show significantly reduced CBF in patients with mild cognitive impairment (MCI) (p = 0.02). Heterogeneity was low (I² = 0), and non-significant (p = 0.78). There were no differences in heterogeneity or effect size between random or fixed effects models.

Cerebral blood flow velocity

Eight studies reported CBFv, but only one showed a significant difference between MCI and controls [55]. Six studies [18, 57] were suitable for meta-analysis (n MCI = 219, n controls = 327); CBFv was significantly lower in MCI than controls (MD: –3.92 (–5.29, –2.55), p < 0.01) (Fig. 3).

OPEN IN VIEWER

Fig.3

Forest plot for a meta-analysis of cerebral blood flow velocity (CBFv). Data is pooled from six studies [18, 57]. Roher et al. included multiple measurements [48] and Viola et al. measured both middle cerebral arteries [47], all results for middle cerebral artery are included. Results are mean difference (95% confidence interval). Statistical significance set at p < 0.05. CBFv is significantly lower in mild cognitive impairment (MCI) (p < 0.01). Heterogeneity was low (I² = 0%), and non-significant (p = 0.47). There were no differences in heterogeneity or effect size between random or fixed effects models.

Sun et al. demonstrated significantly lower CBFv across all MCI patients compared with controls (p < 0.005–0.001) [55]. ApoE4 positive carriers had significantly reduced CBFv compared to non-carriers [55].

Roher et al. could not discriminate between MCI and control groups with sufficient sensitivity based on CBF [48], likely due to small sample size and heterogeneity [48]. Results were significant when the MCI group were restricted to amnestic sub typeonly [48].

In a study by Gommer et al. five patients with MCI developed AD during the study and had a significantly lower mean CBFv (35 (5) versus 51 (11) cm/s), than those with stable MCI [52].

Pulsatility index

Five studies reported PI, only two studies demonstrated a significantly higher PI in MCI [47, 68]. Rivera-Rivera et al. found increased PI was only significant in the inferior ICA and proximal PCA in MCI [68]. The remaining three studies did not show any significant differences [48, 54]. In a meta-analysis of three studies, PI was significantly higher in MCI (n = 159), (MD: 0.15 [0.10, 0.20] p < 0.01) (Fig. 4).

OPEN IN VIEWER

Fig.4

Forest plot for a meta-analysis of pulsatility index. Data is pooled from three studies [47, 54]. Roher et al. took multiple measurements, all middle cerebral artery measurements are included [48], and Viola et al. measured both the right and left middle cerebral arteries [47], both measures are included. Results are mean difference (95% confidence interval). Statistical significance set at p < 0.05. Results show significantly higher pulsatility index in mild cognitive impairment (MCI) compared to controls (p < 0.01). Heterogeneity was low and non-significant (I² = 0%, p = 0.78). There were no differences in effect size or heterogeneity between random and fixed effects models.

Viticchi et al. did not find any significant differences in PI between stable (n = 96) and progressive (n = 21) MCI patients over 1 year [51].

Breath holding index

Two studies measured BHI, one cross sectional [18] and the other longitudinal [51]. Zavoreo et al. demonstrated significantly reduced BHI in MCI (n = 20) compared to control (n = 60) (0.69 versus 1.35 respectively, p < 0.05) [18].

Viticchi et al. demonstrated both BHI and carotid wall thickness (IMT) were predictive of those who progressed to dementia (BHI odds ratio: 5.80 (1.83–18.37) p < 0.05, IMT OR: 3.08 (1.02–9.33 p < 0.05) [51]. BHI was a more sensitive marker of progression than carotid wall thickness [51].

Cerebrovascular resistance/index and resistance index

Six studies examined CVR, but there was significant heterogeneity between outcome measures [15, 68]. Four of the six studies did not find any significant difference in CVR/CVRi between groups [15, 57]. Two studies measured CVRi (mean arterial pressure/mean blood flow velocity), and are described in a meta-analysis below [52, 57]. Liu et al. measured CVR (mean arterial pressure/CBF), which was increased by 13% in MCI relative to controls [56]. Although Gommer et al. did not demonstrate any difference in baseline CVRi, it was predictive in a small group of MCI patients (n = 5) who progressed to dementia [52]. In a 4D-MRI study, resistance index was found to be significantly increased only in the inferior ICA, but there were no differences in vessel cross-sectional area in MCI [68].

In a meta-analysis of two studies [52, 57], CVRi was significantly higher in MCI (n = 46) patients compared to controls (n = 35), (0.17 [0.06, 0.29], p = 0.03) (Fig. 5).

OPEN IN VIEWER

Fig.5

Forest plot for a meta-analysis of the cerebrovascular resistance index (CVRi). Data is pooled from two studies [52, 57]. Results are mean difference (95% confidence interval). Statistical significance set at p < 0.05. Results show CVRi is significantly higher in mild cognitive impairment (MCI) versus controls (p = 0.003). Heterogeneity is low, and non-significant (I² = 24%, p = 0.25). In a fixed effects model, heterogeneity remained the same, effect size: 0.19 (0.13, 0.25), p < 0.001.

Autoregulatory index

One study reported the autoregulatory index, with no difference between MCI and control groups [52].

Vasomotor reactivity

Eight studies examined cerebrovascular reactivity, but varied in imaging type; MR [52, 65–67], PET/SPECT [20, 40], TCD [53, 54], NIRS [59], and method of inducing changes in hemodynamics, which precluded meta-analysis. Five studies used CO₂ inhalation, two studies used intravenous acetazolamide [20, 40], and one utilized breath holding to induce hypercapnia [54]. Five studies demonstrated reduced reactivity in MCI compared to controls [54, 65–67], and one reported no difference [53].

Babiloni et al. combined NIRS and electroencephalogram (EEG) to investigate VMR (in response to CO₂) and coherence [59]. Subjects with amnestic MCI showed lower EEG coherence both before and after CO₂ inhalation and demonstrated poorer reactivity of coherence during CO₂ inhalation [59], suggesting an early loss of neurovascular coupling in amnestic MCI [59].

Ponto et al. used PET to measure vasoactivation using intravenous acetazolamide and neurovascular coupling by neuroactivation with cognitive tasks in MCI (n = 15) [40]. Participants were sub-classified into better or worse cognitive function but CBF, VMR, and acetazolamide challenge did not differ between the two groups [40]. Change in CBF was a better predictor of cognition than VMR [40]; in contrast to Glodzik et al. who found reactivity rather than CBF was a better discriminator from normal aging [66].

Transfer function analysis

Two studies used transfer function analysis to report phase, gain, and coherence [52, 57]. Both studies found intact dynamic autoregulation in MCI compared to healthy controls [52, 57]. Tarumi et al. reported higher gain between TOI, and CBFv was related to poorer memory performance in MCI [57]. Meta-analysis of gain between two studies [52, 57] was non-significant, but phase was significantly higher in MCI (MD: 0.1 (0.07, 0.14), p < 0.01) (Fig. 6).

OPEN IN VIEWER

Fig.6

Forest plot for meta-analysis for phase. Data is pooled from two studies [52, 57]. Results are mean difference (95% confidence interval). Statistical significance set at p < 0.05. Results show significantly increased phase in mild cognitive impairment (MCI) compared to controls (p < 0.01). Heterogeneity was low (I² = 0%, p = 0.44). There were no differences in effect size or heterogeneity between random and fixed effects models.

Tissue oxygenation index

Three studies examined TOI using NIRS [47, 57]; which was significantly lower in MCI patients compared to controls (–7.51 [–13.80, –1.23] p = 0.02), but with significant heterogeneity at meta-analysis (I² = 85%, p < 0.01) (Fig. 7).

OPEN IN VIEWER

Fig.7

Forest plot for meta-analysis of tissue oxygenation index (TOI). Data is pooled from three studies [47, 57]. Viola et al. measured both right and left frontal cortex, and both measures were included. Results are mean difference (95% confidence interval). Statistical significance set at p < 0.05. Results show a significantly lower TOI in mild cognitive impairment (MCI) compared to controls (p = 0.02). Heterogeneity was high, and statistically significant (I² = 85%, p < 0.01). In a fixed effects model, heterogeneity remained the same, and the effect size was smaller: –4.16 (–6.24, –2.08), p < 0.001.

Tarumi et al. measured functional coupling between CBFv (TCD) and TOI (NIRS) [57]. TOI was lower at rest in MCI and correlated with memory and executive function [57], suggesting a pathological de-coupling between oxygenation and CBFv in MCI [57]. Sit-to-stand maneuvers made differences in tissue oxygenation between MCI and controls more apparent (p = 0.047 versus p = 0.03) [57].

Viola et al. also found tissue oxygenation was significantly reduced in MCI, most obviously in the temporo-parietal cortex [47]. Using receiver-operator characteristic (ROC) curves, TOI at a cut off of 59%, had a sensitivity of 81% and specificity of 95% in distinguishing MCI from controls, and was more sensitive than PI (cut off of 0.96, Sensitivity: 69%, Specificity: 98%) [47].

In contrast, Liu et al. did not find differences in TOI, or oxygen extraction fraction [56]. However, cerebral metabolic rate of oxygen was significantly reduced by 11% in MCI [56]. Furthermore, there was a significant correlation between CBF and cerebral metabolic rate of oxygen in controls but not MCI, suggesting early loss of neurovascularcoupling [56].

Activation and neurovascular coupling

Nine studies measured activation to cognitive tasks; seven NIRS [58, 60–64], one PET [40], and one SPECT [20]. Five studies demonstrated reduced activation during retrieval task [58], verbal fluency task [61], working memory task [62, 69], and verbal fluency task during walking [63]. Uemura et al. demonstrated reduced activation bilaterally in the dorso-lateral prefrontal cortex during a retrieval task, with no differences in memory encoding [58]. Yeung et al. demonstrated significant loss of lateralization during the category fluency task in MCI, which was more significant in the non-amnestic sub-group [64]. In a later study by Yeung et al. MCI patients had similar performance on working memory tasks, and comparable frontal activation pattern to matched-controls at low task load [69]. On increasing the working memory task from low to high load, the increase in frontal activation in healthy controls was not apparent in the MCI group [69].

Vermeij et al. found no differences in activation in patients with MCI pre and post a 5-week computer based memory training program [60]. The MCI group did have improved behavioral performance at lower load working memory task, indicating a potential clinical benefit from cognitive training [60].

Arai et al. found oxygenation was significantly lower in the right parietal region during the verbal fluency task compared to controls (p = 0.0012) [25, 61]. Using the right parietal area, sensitivity for distinguishing MCI from controls was only 40%, but specificity was high at 94% [61].

Doi et al. compared cortical oxygenation during a normal walking and walking whilst performing the verbal fluency task in MCI [63], where the latter resulted in increased prefrontal activation. Executive function was significantly correlated with walking during the verbal fluency task, but not with normal walking [63].

In a neuroactivation PET study, more challenging cognitive tasks (memory versus counting task) demonstrated a more significant relationship between change in CBF and the ability to learn from experiences [40]. Performance on the Trails B test was predicted by change in CBF during a more challenging task, which was lost in the non-learning group [40]. Change in CBF was found to be a better predictor of cognition than VMR [40].

Knapp et al. demonstrated 30% of MCI patients had perfusional defects in the parieto-temporal regions with SPECT [20]. Vasoactivation with acetazolamide, decreased the number of defects [20], whereas neuroactivation with cognitive tasks increased the number of defects, and may be more useful clinically [20].

Other vessel measures

A number of studies measured CBFv and other parameters in the anterior cerebral artery (ACA), PCA, vertebral and basilar arteries (VA, BA), and the ICA. The number of studies examining these vessels was fewer, and therefore meta-analyses included small numbers of studies and could only be performed for ACA, ICA, and VA measurements of CBFv.

Only one study examined the PCA [48], and there were no significant differences in CBFv or PI between MCI and controls [48]. Two studies measured CBFv in the BA [48, 55], but meta-analysis was precluded as one study did not provide data for MCI and control groups [55]. Both studies report no statistically significant differences in CBFv in the BA between groups [48, 55]. In a meta-analysis of two studies for CBFv in the ACA [18, 48], CBFv was significantly lower in MCI compared to controls (MD: –5.00 [–8.64, –1.36], p < 0.05) (Fig. 8). Furthermore, although Sun et al. did not provide the data for these measures, they also report significantly lower flow velocities (p < 0.05) in the ACA in MCI, but not between ApoE4 carriers and non-carriers [55]. Two studies were included in meta-analyses of ICA and VA measures [48, 56], but neither showed any significant differences in flow velocities between MCI and controls in these vessels (Figs. 9 and 10).

OPEN IN VIEWER

Fig.8

Forest plot for meta-analysis of cerebral blood flow velocity (CBFv) (anterior cerebral artery). Data is pooled from two studies [18, 48]. Roher et al. [48] measured both right and left anterior cerebral artery CBFv, and both measures were included. Results are mean difference (95% confidence interval). Statistical significance was set at p < 0.05. Results show significantly reduced CBFv in mild cognitive impairment (MCI) compared to controls (p = 0.007). Heterogeneity was low and non-significant (p = 0.83). The effect size was no different between random or fixed effects models.

OPEN IN VIEWER

Fig.9

Forest plot for meta-analysis of cerebral blood flow velocity (CBFv) (vertebral artery). Data is pooled from two studies [48, 56]. Both studies measured the right and left vertebral arteries, and both measures are included in the analysis. Results are mean difference (95% confidence interval). Statistical significance was set at p < 0.05. Results show no significant differences between mild cognitive impairment (MCI) and controls (p = 0.78). Heterogeneity was low and non-significant (p = 0.91). The effect size was no different between random or fixed effects models.

OPEN IN VIEWER

Fig.10

Forest plot for meta-analysis of cerebral blood flow velocity (CBFv) (internal carotid artery). Data is pooled from two studies [48, 56]. Both studies measured the right and left internal carotid arteries, and both measures are included in the analysis. Results are mean difference (95% confidence interval). Statistical significance was set at p < 0.05. Results show no significant differences between mild cognitive impairment (MCI) and controls (p = 0.39). Heterogeneity was low and non-significant (p = 0.94). The effect size was no different between random or fixed effects models.

Summary of results

On meta-analysis, CBF, CBFv, and tissue oxygenation were all significantly lower in MCI, with increased PI, CVRi, and phase. Reactivity to CO₂ and the BHI were found to be lower in most studies of MCI, and neuroactivation in response to cognitive tasks was generally reduced [58, 62], with only one study demonstrating an increase in activation [63]. Few studies have examined validity, but thresholds have been determined for CBF, TOI, and A-index (amplitude of change in the HbO₂ waveform) to either predict progression or discriminate between MCI and normal aging [47, 61].

Studies of activation in MCI show mixed findings, with both increased [63] and reduced prefrontal activation [58]. These inconsistencies are likely to be explained by methodological differences, particularly activation techniques, imaging modality, pre-training and population heterogeneity (i.e., phenotype, education level, co-morbidities) [69].

Comparable to results from studies of AD and vascular dementia [23, 24], measures using reactivity to CO₂, small vessel resistance, or neurovascular coupling may be more consistent and reliable markers of MCI, than CBF or CBFv, but further studies are required for confirmation.

Implications of findings

Despite the heterogeneity of studies examining MCI, on pooled analysis of individual studies, clear abnormalities in vascular physiology become apparent. In particular, reduced levels of CBF and tissue oxygenation indicate that hypoperfusion may be an important pathological mechanism influencing the deposition or acceleration of tau and amyloid plaques [6, 11]. The complete picture remains elusive on the exact interplay of vascular pathology and that of tau and amyloid deposition, but it is likely to be a complex relationship [11], where both mechanisms exacerbate one another and thus resulting in progressive,cyclical neurodegeneration [7]. Reduced CBF not only reduces the delivery of essential nutrients and oxygen to neurons, but it also impairs the important clearance of neurotoxic metabolites and proteins that subsequently accumulate [11, 68]. Indeed, other disorders of amyloid, such as cerebral amyloid angiopathy, demonstrate microvascular abnormalities that result in increased amyloid formation [11]. Hypoperfusion can occur with, or independently, to blood-brain barrier degeneration, but neurotoxicity is less marked with the latter [11]. Blood-brain barrier dysfunction leads to the accumulation of circulating proteins and the generation of reactive oxygen species resulting in microvascular dysfunction [9, 11]. Once established, hypoperfusion disrupts key cellular processes within highly metabolically active neurons, causing; altered pH, loss of action potentials, and electrolyte disturbance [11]. As a result, deposition of neurotoxic proteins, such as amyloid is exacerbated [11]. Animal models using artificial induction of hypoperfusion have demonstrated enhanced oligomerization and deposition of amyloid [9, 11]. The sequence in which these pathological mechanisms occur remains unclear, and further studies have demonstrated that amyloid can also exacerbate hypoperfusion through vasoconstriction [9, 11]. According to the two-hit hypothesis, tau hyperphosphorylation can result independently from both amyloid accumulation and hypoperfusion, but it is most likely to occur as a mixed disease process [11]. In addition, cognitively intact, at–risk individuals, can have impaired cerebral perfusion and functional neurovascular de-coupling, prior to amyloid deposition or clinically detectable cognitive deficits [11]. This early de-coupling is supported by the neuroactivation studies in this review demonstrating a reduced ability of the brain to increase blood flow, despite introducing increasing cognitive demand [58, 62]. Furthermore, the reduced ability of cerebral vessels in MCI to rapidly respond to intravascular changes, such as rising CO₂ [59], suggests the mechanisms to protect neurons under these conditions have become compromised. In line with recent studies of AD [15, 68], small vessel resistance was increased on meta-analysis, and is thought to occur as a combination of amyloid mediated vasoconstriction, arterial stiffening and reduced compliance, capillary microstructural changes, small vessel disease and atherosclerosis [11, 68]. The interpretation of the apparent rise in PI is more complex, with numerous hemodynamic parameters contributing to the observed change [19]. A recent study suggests that the PI can reflect the cerebral perfusion pressure [19], and the higher indices demonstrated in this review, may be indicative of a compensatory increase in CPP to counteract the falling CBFv as a result of structural microvascular changes [70]. Lower BHIs are in line with studies demonstrating reduced CO₂ reactivity, but this measure needs to be interpreted with caution given the limitations to this method [17, 71]. The higher phase demonstrated in the meta-analysis reported here, is suggestive of better autoregulation in MCI, but is not consistent with the overall pattern of results suggestive of poorer cerebral hemodynamics in this review. This may be as a result of the small number of studies and sample sizes included in this analysis. Higher phase values can also be due to a lack of attention to phase “wrap around”, with negative values of phase more likely to occur in healthy subjects, being added to averages, thus reducing the values reported in controls [72]. However, the two studies conclude that dynamic autoregulation remains intact in MCI [52, 57]. Therefore, increased phase could be due to compensatory mechanisms in early disease, in keeping with studies demonstrating hyperactivation [63]. The meta-analyses from this review support the notion that hypoperfusion occurs early in cognitive impairment, although compensatory hyperperfusion has also been demonstrated in MCI [63]. In reality this is likely to represent a spectrum of disease, where individuals with MCI present at various stages of compensation through to decompensation (i.e., hyper to hypoperfused states). Given that not all patients with MCI progress to dementia, longitudinal studies examining the predictive power of individual hemodynamic or autoregulatory markers would provide additional information on the specific vascular pathological mechanisms that are implicated in this conversion from mild decline to fulminant dementia. Furthermore, this could provide the opportunity for the development of therapeutics targeting specific stages of vascular dysfunction. Vermeij et al. demonstrated that greater pre-frontal activation at higher working memory load predicted greater training gains from a cognitive training program [60]. Therefore, hemodynamic biomarkers offer the potential to predict the utility of interventions in specific patient groups.

Limitations

The majority of the meta-analyses had low heterogeneity; however, the TOI analysis is limited by a relatively high and statistically significant heterogeneity.

The imaging techniques reviewed here all have their limitations. It is important to note that CBFv measures using TCD rely on the assumption that the vessel diameter remains relatively constant, despite significant fluctuations in blood pressure or CO₂ [14, 74]. A number of recent studies have shown that BHI measured by TCD can vary significantly during large fluctuations in blood pressure or CO₂, can be inadequately performed, and is subject to a number of confounding factors, therefore limiting the usability of this marker over other protocols for the assessment of CO₂ reactivity with TCD [17, 74]. PI reflects a complex construct of multiple hemodynamic parameters and this can limit the clinical application [19].

Small sample sizes, heterogeneous MCI populations, and the cross-sectional design of the majority of individual studies are further limitations. Furthermore, some study populations were restricted to solely amnestic MCI [47, 67]. Chao et al. demonstrated important differences in perfusion between subtypes of MCI, and therefore differences in MCI populations could introduce significantheterogeneity [75]. The classification systems for MCI vary between studies; as the concept of MCI has evolved, this is reflected by the variability in inclusion and exclusion criteria. The Petersen criteria [5] were the most widely used, but others included the Mayo criteria [5] and DSM-V classification of neurocognitive disorders [76, 77]. Some studies included participants with subjective memory complaints, but no objective deficit on testing, and so variants of normal aging are likely to be included in analyses [20]. Severity of cognitive impairment also varied; one study even recruited patients with an average MMSE score below the threshold for AD [54]. Changes in cerebral hemodynamics are likely to be more detectable in these patients, which may confound thresholds used to distinguish AD, MCI and normal aging.

Concerns have been raised about the use of MMSE as a diagnostic tool for MCI and for determining inclusion into clinical studies [78]. In a recent study, MMSE had only 18% sensitivity for detecting MCI, compared to Montreal Cognitive Assessment, which had 90% sensitivity [78].

Information on comorbidities was often limited and, given the increased prevalence of cardiovascular risk factors amongst those with MCI and impaired autoregulation [9, 80], is an important confounder. A recent pilot randomized controlled trial demonstrated reduced CO₂ reactivity and CBFv in hypertensive patients with executive dysfunction randomized to lisinopril or hydrochlorothiazide [81], but few studies reported the number of participants taking antihypertensive or vasoactive medications. Furthermore, asymptomatic carotid stenosis, which can occur in up to 10% of older people and significantly affect CBF, was not routinelyassessed [66].

There were limited analyses on drop-out rates, and frequently these will be patients with more severe cognitive deficits [66]. The setting for recruitment is also important; the majority of participants were recruited from memory clinics. These participants are likely to present with a more advanced form of MCI, and be subject to more detailed and specialist evaluation than those presenting in the community, limiting the generalizability of the findings [5].

Only one of three studies reporting validity of imaging methods used receiver-operator characteristic curves to determine sensitivity and specificity [47]. ROC curves are the most robust method of determining diagnostic validity, and is therefore an important limitation to studies using single measures [82].

Further work

Longitudinal studies with larger sample sizes are required to confirm the findings of this review, with a more comprehensive analysis of co-morbidities and concurrent medications. Consensus is required on the classification of MCI, and a move to more accurate neuropsychological testing should provide greater homogeneity in populations being studied. Analysis of drop-outs from studies will help identify bias and patients that are currently under-represented [66].

Few studies have examined changes in CBFv in MCI during cognitive tasks with TCD. This may be a clinically useful method to examine neurovascular coupling. The feasibility of this has been demonstrated in a number of SPECT, PET, and NIRS studies in this review [20, 40], but the accessibility and ease of use of TCD makes this an attractive option.

Conclusions

Studies of cerebral hemodynamics and oxygenation in MCI are small and heterogeneous. Despite this, pathological changes are demonstrated when compared to normal aging and can predict those who progress to dementia. Meta-analyses suggest that small studies have been underpowered to detect these differences, and comprehensive larger studies are now required to confirm these findings. Studies thus far are likely to have underestimated the scope and significance of hemodynamic and oxygenation changes in MCI, and there is significant potential for the use of cerebral hemodynamics or oxygenation as an early and sensitive marker of cognitive decline.