Kivipelto M, Ngandu T, Laatikainen T, Winblad B, Soininen H, Tuomilehto J.
Risk score for the prediction of dementia risk in 20 years among middle aged people: a longitudinal, population-based study.
Lancet Neurol. 2006 Sep;5(9):735-41.
PubMed.
It is very important to identify those people at risk for developing dementia in later life, but few studies have tried to identify risk scores for predicting dementia. The focus these authors place on the middle aged is of great interest, because it can help with the early identification of people at risk, most likely prior to the development of neuropathological burden. In this regard, the paper is a major contribution.
The focus on cardiovascular risks is very interesting as well. First of all, that is mostly what they measured, which puts certain limits on the model, but it also confirms prior studies identifying mid- and late-life cardiovascular disease risk factors for dementia. These risk factors, including hyperlipidemia, hypertension, and obesity, are similar to those linking metabolic syndrome (that comprises these domains as a composite as well) with dementia.
While the paper is useful in that we can start to identify folks at risk, until we have definitive prevention strategies there is not a whole lot we can do other than advise people to control cardiovascular risks. We cannot yet change much about age, education, and ApoE. But we can advise people to increase physical activity, lose weight, reduce cholesterol, and maintain healthy blood pressure.
One thing this study did not address was psychosocial parameters. They really did not include any psychosocial measures that may influence the risk for dementia, such as social support, depression measures, and intellectual activity. It would be interesting to see how these fit into the risk score model.
From a scientific perspective, it is desirable to be able to predict who develops dementia and when. It’s even better to be able to prevent it. Niels Bohr cautioned, however, that “Prediction is very difficult, especially of the future.”
Kivipelto and colleagues offer up a risk assessment tool, derived from a population-based Finnish study, that is reminiscent of the coronary heart disease risk scales hosted on the American Heart Association’s website. The dementia risk scale is similarly based on simple demographic and clinical characteristics, most of which are individually well-established statistical dementia risk factors: age, low education, male gender, high blood pressure, obesity, high cholesterol, decreased physical activities, and in one version of the scale, ApoE genotype.
In principle, any middle-aged person can modify four of these factors in directions that would presumably decrease risk for dementia. Thus, the main value of this scale is as an educational tool. As the authors discuss, it is more heuristic than clinical and can’t really be applied to individuals. If a person scores the highest on the scale, then he has at best only about a 22-23 percent subsequent risk for dementia, and if she scores the lowest, then no more than a 2 percent risk.
At the current stage of the scale’s development, its main clinical utility may be in providing statistical assurance that, for example, if you are a physically active 46-year-old Finnish woman with at least 10 years of schooling, and not hypertensive, obese, or hypercholesterolemic, and do not have an ApoE 4 allele, then your chance for developing dementia before age 80 is less than 1 percent. However, living for 8 more years while becoming less active may more than triple your risk. The heuristic practicality of the scale is that a 54-year-old can be shown that if her systolic blood pressure rises to 140 mm Hg, her body mass index goes to 30 kg/m2, and her cholesterol increases, then her risk can jump to as high as 10 percent.
As usual the devil is in the details, and while the authors propose a scale that is easy to use and makes sense, it is still a work in progress. Some examples: 1) once you reach age 54, if your four modifiable factors stay low, then according to the scale, your risk doesn’t change no matter how old you get; 2) contrary to the authors’ assertions, the scale is not actually validated to predict “late-life” dementia in that they only assessed subjects up to age 80, although half the incidence of dementia is after this age; 3) per their research methods, subjects were assessed for dementia only if they had an MMSE score less than 24, so many of their dementia cases were not recognized. The scale might have better metrical characteristics if the milder dementia cases were identified and if the follow-up age extended beyond age 80. No doubt, the authors are engaged in further development. Maybe the scale will soon appear on the Alzheimer's Association’s website the way the coronary heart disease risks scale appears on the American Heart Association’s. Despite the limitations, the instrument can be a useful teaching tool.
One final caveat: just because obesity, hypertension, hypercholesterolemia, and inactivity are dementia risk factors, that may not necessarily mean that reversing these conditions will lessen one’s subsequent specific dementia risk. The interventional research is still evolving here. But improving these factors clearly is a good thing to do anyway, if you can, because it will lower risks for coronary heart disease. Avoiding early cardiac death gives you a fighting chance to live long enough to try to avoid dementia.
Hypertension, obesity, and high cholesterol levels, signs also reported in hypercortisolism, all seem to point to high cortisol levels as a predictor for the development of AD. Of interest is the study by Peskind et al. (1) which reports higher CSF cortisol concentrations associated with increased frequency of the ApoE-ε4 allele. Cohen and colleagues (2) also report increased cortisol levels in those with lower income and education. The Cushingoid features of Down syndrome may help explain the association with DS and AD.
Perhaps chronic hypercortisolism negatively regulates ACTH secretion and may explain the downregulation of the ACTH responsive gene seladin-1 (selective Alzheimer’s disease indicator-1) in AD (3). It is encoded by the DHCR24 gene and converts desmosterol to cholesterol. The Crameri group (4) finds that overexpression of seladin-1 resulted in both reduced BACE processing of APP and Aβ formation. It would be interesting to see whether ACTH treatment normalized BACE processing of APP and Aβ formation in the seladin-1 deficient mouse brains.
References:
Peskind ER, Wilkinson CW, Petrie EC, Schellenberg GD, Raskind MA.
Increased CSF cortisol in AD is a function of APOE genotype.
Neurology. 2001 Apr 24;56(8):1094-8.
PubMed.
Cohen S, Doyle WJ, Baum A.
Socioeconomic status is associated with stress hormones.
Psychosom Med. 2006 May-Jun;68(3):414-20.
PubMed.
Greeve I, Hermans-Borgmeyer I, Brellinger C, Kasper D, Gomez-Isla T, Behl C, Levkau B, Nitsch RM.
The human DIMINUTO/DWARF1 homolog seladin-1 confers resistance to Alzheimer's disease-associated neurodegeneration and oxidative stress.
J Neurosci. 2000 Oct 1;20(19):7345-52.
PubMed.
Crameri A, Biondi E, Kuehnle K, Lütjohann D, Thelen KM, Perga S, Dotti CG, Nitsch RM, Ledesma MD, Mohajeri MH.
The role of seladin-1/DHCR24 in cholesterol biosynthesis, APP processing and Abeta generation in vivo.
EMBO J. 2006 Jan 25;25(2):432-43.
PubMed.
Several recent studies, including the large, NIH-sponsored Alzheimer’s Disease Neuroimaging Initiative, have been launched to clarify the utility of using different imaging modalities as potential surrogate markers for disease processes. Considerable excitement was generated when initial reports appeared showing the 1H MRS measurements of N-acetylaspartate/creatine (NAA/Cr) ratio differed in patients with Alzheimer disease compared to normal elderly. The question of whether these techniques could discern chemical changes associated with earlier stages in the illness, such as those associated with mild cognitive impairment (MCI), would indeed be of great interest.
The authors of the current study have taken a well-characterized, large population (total n = 197) of control, MCI, and Alzheimer patients and studied them longitudinally in order to address these questions. Recent evidence has shown that in their target region, the posterior cingulate, changes that are observable with voxel-based morphometry and fMRI occur in MCI patients who convert to Alzheimer disease—these changes are not present in those who do not convert. With all of these design strengths, it is a little disappointing that the effects they observe are as modest as they are. The annual percent change of NAA/Cr ratio in MCI patients who convert to AD is only 2.5 percent different from the annual change in the normal comparison group, with other results being similar in magnitude. It is possible that the bilateral posterior cingulate gyrus may not be an ideal location to see bigger differences at this early stage of the disease represented by MCI. The posterior cingulate does present excellent features for spectroscopic examination, while the mesotemporal region, which is damaged earlier in the disease process, is clearly a much harder target to work with, due to disease-related shrinkage and CSF artifact. However, the chemical changes may be more robust in this location. Nevertheless, this very well-designed, longitudinal study provides significant and important information to the spectroscopic literature of MCI and Alzheimer disease.
This report by Kantarci et al. is a welcome addition to the literature on neuroimaging measures that are of potential interest in diagnosis and charting of the longitudinal course of mild cognitive impairment (MCI). The authors used single voxel proton magnetic resonance spectroscopy (MRS) of the posterior cingulate region, and combined it with structural volumetric MRI, to study MCI and Alzheimer disease (AD) patients and healthy control subjects. At baseline, MRS measures were significantly different between MCI and controls as well as AD and controls, with reduced N-acetylaspartate/creatine (NAA/Cr) ratios and increased choline/creatine (Cho/Cr) and myoinositol/creatine (mI/Cr) ratios in both MCI and AD groups. These findings are consistent with the published literature (1). Interestingly, baseline MRS data did not distinguish the MCI subjects who went on to develop dementia from those who did not. The authors did not examine the NAA/mI and NAA/Cho ratios as discriminators, although the former has been reported to be the most accurate discriminator between AD and controls (2). The fact that NAA/Cr values decrease and mI/Cr levels increase with progression of MCI would arguably make NAA/mI ratio useful to distinguish between MCI converters and non-converters. This potential was thwarted in the study by the lack of stability of the mI/Cr values in the control subjects, the reasons for which are not clear. An increase in mI values with age has been reported previously (3).
One limitation of the analysis is the use of relative values of the metabolites. Absolute quantitation methods are more accurate in this respect, but results from different centers have not been consistent on this (4). An alternative approach, which might have been instructive in this study, would have been to examine concordant findings based on referencing of metabolite peaks to both the cerebrospinal fluid (CSF)-corrected water signal and the Cr signal, which might reflect the real biochemical status. Another limitation is the lack of control for the partial volume effect of CSF in the voxel. While the authors used ratios rather than absolute quantities, these ratios can be influenced by the percentage of the voxel that is occupied by CSF, and this is likely to change as atrophy progresses (5). Furthermore, as has been argued previously (1), using at least two voxels of interest, with the second voxel being in a white matter region, could overcome some of the problems in interpreting the data in relation to relative change in gray and white matter.
The comparable association of MRS and volumetric estimation of the ventricles with cognitive change in AD and MCI subjects in this study argues for the usefulness of MRS in the longitudinal examination of subjects. However, it is uncertain whether the two measures are together superior to each alone in predicting decline. If their contributions are not additive or complementary, the volumetric measure, which is technically easier to obtain, will of course be clinically more acceptable. MRS, while readily available on most clinical scanners, continues to pose technical challenges that make it less suitable for routine clinical use. Accurate shimming, the placement of the voxel, acquisition parameters, and analytical software are likely to influence the data such that findings differ from scanner to scanner, and from one center to another center with the same type of scanner. For this reason, it continues to remain in the research domain, and its clear clinical utility remains to be established. The findings of this study show that with technical advances, and better standardization, it has the potential to add functional information to the excellent structural data obtained with MRI.
MRS does have the potential to inform hypotheses about the pathogenesis of disease. The finding of a decline in Cho/Cr in MCI non-converters, if replicated, warrants an examination of the underlying pathophysiology in the early stages of MCI. A study of longer duration may well show that this is but one phase in the longitudinal course of MCI. In the same vein, the mI and Cho changes in MCI and AD have implications for the understanding of the pathophysiology of AD, and may well suggest possible interventions (1).
In conclusion, the paper by Kantarci et al. highlights the potential of MRS, and points to a future in which a combination of structural and functional imaging techniques will be used to inform us about the progression of AD and permit more accurate early diagnosis and prognosis. Study of MCI using these techniques is therefore to be commended. The clinical applicability of MRS must, however, await further standardization of techniques and more longitudinal studies.
References:
Valenzuela MJ, Sachdev P.
Magnetic resonance spectroscopy in AD.
Neurology. 2001 Mar 13;56(5):592-8.
PubMed.
Miller BL, Moats RA, Shonk T, Ernst T, Woolley S, Ross BD.
Alzheimer disease: depiction of increased cerebral myo-inositol with proton MR spectroscopy.
Radiology. 1993 May;187(2):433-7.
PubMed.
Ross AJ, Sachdev PS, Wen W, Brodaty H.
Longitudinal changes during aging using proton magnetic resonance spectroscopy.
J Gerontol A Biol Sci Med Sci. 2006 Mar;61(3):291-8.
PubMed.
Henriksen O.
In vivo quantitation of metabolite concentrations in the brain by means of proton MRS.
NMR Biomed. 1995 Jun;8(4):139-48.
PubMed.
Ross AJ, Sachdev PS, Wen W, Brodaty H, Joscelyne A, Lorentz LM.
Prediction of cognitive decline after stroke using proton magnetic resonance spectroscopy.
J Neurol Sci. 2006 Dec 21;251(1-2):62-9. Epub 2006 Nov 7
PubMed.
Comments
University California, San Francisco
It is very important to identify those people at risk for developing dementia in later life, but few studies have tried to identify risk scores for predicting dementia. The focus these authors place on the middle aged is of great interest, because it can help with the early identification of people at risk, most likely prior to the development of neuropathological burden. In this regard, the paper is a major contribution.
The focus on cardiovascular risks is very interesting as well. First of all, that is mostly what they measured, which puts certain limits on the model, but it also confirms prior studies identifying mid- and late-life cardiovascular disease risk factors for dementia. These risk factors, including hyperlipidemia, hypertension, and obesity, are similar to those linking metabolic syndrome (that comprises these domains as a composite as well) with dementia.
While the paper is useful in that we can start to identify folks at risk, until we have definitive prevention strategies there is not a whole lot we can do other than advise people to control cardiovascular risks. We cannot yet change much about age, education, and ApoE. But we can advise people to increase physical activity, lose weight, reduce cholesterol, and maintain healthy blood pressure.
One thing this study did not address was psychosocial parameters. They really did not include any psychosocial measures that may influence the risk for dementia, such as social support, depression measures, and intellectual activity. It would be interesting to see how these fit into the risk score model.
View all comments by Kristine YaffeUniversity of Southern California Keck School of Medicine
From a scientific perspective, it is desirable to be able to predict who develops dementia and when. It’s even better to be able to prevent it. Niels Bohr cautioned, however, that “Prediction is very difficult, especially of the future.”
Kivipelto and colleagues offer up a risk assessment tool, derived from a population-based Finnish study, that is reminiscent of the coronary heart disease risk scales hosted on the American Heart Association’s website. The dementia risk scale is similarly based on simple demographic and clinical characteristics, most of which are individually well-established statistical dementia risk factors: age, low education, male gender, high blood pressure, obesity, high cholesterol, decreased physical activities, and in one version of the scale, ApoE genotype.
In principle, any middle-aged person can modify four of these factors in directions that would presumably decrease risk for dementia. Thus, the main value of this scale is as an educational tool. As the authors discuss, it is more heuristic than clinical and can’t really be applied to individuals. If a person scores the highest on the scale, then he has at best only about a 22-23 percent subsequent risk for dementia, and if she scores the lowest, then no more than a 2 percent risk.
At the current stage of the scale’s development, its main clinical utility may be in providing statistical assurance that, for example, if you are a physically active 46-year-old Finnish woman with at least 10 years of schooling, and not hypertensive, obese, or hypercholesterolemic, and do not have an ApoE 4 allele, then your chance for developing dementia before age 80 is less than 1 percent. However, living for 8 more years while becoming less active may more than triple your risk. The heuristic practicality of the scale is that a 54-year-old can be shown that if her systolic blood pressure rises to 140 mm Hg, her body mass index goes to 30 kg/m2, and her cholesterol increases, then her risk can jump to as high as 10 percent.
As usual the devil is in the details, and while the authors propose a scale that is easy to use and makes sense, it is still a work in progress. Some examples: 1) once you reach age 54, if your four modifiable factors stay low, then according to the scale, your risk doesn’t change no matter how old you get; 2) contrary to the authors’ assertions, the scale is not actually validated to predict “late-life” dementia in that they only assessed subjects up to age 80, although half the incidence of dementia is after this age; 3) per their research methods, subjects were assessed for dementia only if they had an MMSE score less than 24, so many of their dementia cases were not recognized. The scale might have better metrical characteristics if the milder dementia cases were identified and if the follow-up age extended beyond age 80. No doubt, the authors are engaged in further development. Maybe the scale will soon appear on the Alzheimer's Association’s website the way the coronary heart disease risks scale appears on the American Heart Association’s. Despite the limitations, the instrument can be a useful teaching tool.
One final caveat: just because obesity, hypertension, hypercholesterolemia, and inactivity are dementia risk factors, that may not necessarily mean that reversing these conditions will lessen one’s subsequent specific dementia risk. The interventional research is still evolving here. But improving these factors clearly is a good thing to do anyway, if you can, because it will lower risks for coronary heart disease. Avoiding early cardiac death gives you a fighting chance to live long enough to try to avoid dementia.
View all comments by Lon S. Schneider�
Hypertension, obesity, and high cholesterol levels, signs also reported in hypercortisolism, all seem to point to high cortisol levels as a predictor for the development of AD. Of interest is the study by Peskind et al. (1) which reports higher CSF cortisol concentrations associated with increased frequency of the ApoE-ε4 allele. Cohen and colleagues (2) also report increased cortisol levels in those with lower income and education. The Cushingoid features of Down syndrome may help explain the association with DS and AD.
Perhaps chronic hypercortisolism negatively regulates ACTH secretion and may explain the downregulation of the ACTH responsive gene seladin-1 (selective Alzheimer’s disease indicator-1) in AD (3). It is encoded by the DHCR24 gene and converts desmosterol to cholesterol. The Crameri group (4) finds that overexpression of seladin-1 resulted in both reduced BACE processing of APP and Aβ formation. It would be interesting to see whether ACTH treatment normalized BACE processing of APP and Aβ formation in the seladin-1 deficient mouse brains.
References:
Peskind ER, Wilkinson CW, Petrie EC, Schellenberg GD, Raskind MA. Increased CSF cortisol in AD is a function of APOE genotype. Neurology. 2001 Apr 24;56(8):1094-8. PubMed.
Cohen S, Doyle WJ, Baum A. Socioeconomic status is associated with stress hormones. Psychosom Med. 2006 May-Jun;68(3):414-20. PubMed.
Greeve I, Hermans-Borgmeyer I, Brellinger C, Kasper D, Gomez-Isla T, Behl C, Levkau B, Nitsch RM. The human DIMINUTO/DWARF1 homolog seladin-1 confers resistance to Alzheimer's disease-associated neurodegeneration and oxidative stress. J Neurosci. 2000 Oct 1;20(19):7345-52. PubMed.
Crameri A, Biondi E, Kuehnle K, Lütjohann D, Thelen KM, Perga S, Dotti CG, Nitsch RM, Ledesma MD, Mohajeri MH. The role of seladin-1/DHCR24 in cholesterol biosynthesis, APP processing and Abeta generation in vivo. EMBO J. 2006 Jan 25;25(2):432-43. PubMed.
View all comments by Mary ReidHarvard Medical School
Several recent studies, including the large, NIH-sponsored Alzheimer’s Disease Neuroimaging Initiative, have been launched to clarify the utility of using different imaging modalities as potential surrogate markers for disease processes. Considerable excitement was generated when initial reports appeared showing the 1H MRS measurements of N-acetylaspartate/creatine (NAA/Cr) ratio differed in patients with Alzheimer disease compared to normal elderly. The question of whether these techniques could discern chemical changes associated with earlier stages in the illness, such as those associated with mild cognitive impairment (MCI), would indeed be of great interest.
The authors of the current study have taken a well-characterized, large population (total n = 197) of control, MCI, and Alzheimer patients and studied them longitudinally in order to address these questions. Recent evidence has shown that in their target region, the posterior cingulate, changes that are observable with voxel-based morphometry and fMRI occur in MCI patients who convert to Alzheimer disease—these changes are not present in those who do not convert. With all of these design strengths, it is a little disappointing that the effects they observe are as modest as they are. The annual percent change of NAA/Cr ratio in MCI patients who convert to AD is only 2.5 percent different from the annual change in the normal comparison group, with other results being similar in magnitude. It is possible that the bilateral posterior cingulate gyrus may not be an ideal location to see bigger differences at this early stage of the disease represented by MCI. The posterior cingulate does present excellent features for spectroscopic examination, while the mesotemporal region, which is damaged earlier in the disease process, is clearly a much harder target to work with, due to disease-related shrinkage and CSF artifact. However, the chemical changes may be more robust in this location. Nevertheless, this very well-designed, longitudinal study provides significant and important information to the spectroscopic literature of MCI and Alzheimer disease.
View all comments by Perry RenshawUniversity of New South Wales
This report by Kantarci et al. is a welcome addition to the literature on neuroimaging measures that are of potential interest in diagnosis and charting of the longitudinal course of mild cognitive impairment (MCI). The authors used single voxel proton magnetic resonance spectroscopy (MRS) of the posterior cingulate region, and combined it with structural volumetric MRI, to study MCI and Alzheimer disease (AD) patients and healthy control subjects. At baseline, MRS measures were significantly different between MCI and controls as well as AD and controls, with reduced N-acetylaspartate/creatine (NAA/Cr) ratios and increased choline/creatine (Cho/Cr) and myoinositol/creatine (mI/Cr) ratios in both MCI and AD groups. These findings are consistent with the published literature (1). Interestingly, baseline MRS data did not distinguish the MCI subjects who went on to develop dementia from those who did not. The authors did not examine the NAA/mI and NAA/Cho ratios as discriminators, although the former has been reported to be the most accurate discriminator between AD and controls (2). The fact that NAA/Cr values decrease and mI/Cr levels increase with progression of MCI would arguably make NAA/mI ratio useful to distinguish between MCI converters and non-converters. This potential was thwarted in the study by the lack of stability of the mI/Cr values in the control subjects, the reasons for which are not clear. An increase in mI values with age has been reported previously (3).
One limitation of the analysis is the use of relative values of the metabolites. Absolute quantitation methods are more accurate in this respect, but results from different centers have not been consistent on this (4). An alternative approach, which might have been instructive in this study, would have been to examine concordant findings based on referencing of metabolite peaks to both the cerebrospinal fluid (CSF)-corrected water signal and the Cr signal, which might reflect the real biochemical status. Another limitation is the lack of control for the partial volume effect of CSF in the voxel. While the authors used ratios rather than absolute quantities, these ratios can be influenced by the percentage of the voxel that is occupied by CSF, and this is likely to change as atrophy progresses (5). Furthermore, as has been argued previously (1), using at least two voxels of interest, with the second voxel being in a white matter region, could overcome some of the problems in interpreting the data in relation to relative change in gray and white matter.
The comparable association of MRS and volumetric estimation of the ventricles with cognitive change in AD and MCI subjects in this study argues for the usefulness of MRS in the longitudinal examination of subjects. However, it is uncertain whether the two measures are together superior to each alone in predicting decline. If their contributions are not additive or complementary, the volumetric measure, which is technically easier to obtain, will of course be clinically more acceptable. MRS, while readily available on most clinical scanners, continues to pose technical challenges that make it less suitable for routine clinical use. Accurate shimming, the placement of the voxel, acquisition parameters, and analytical software are likely to influence the data such that findings differ from scanner to scanner, and from one center to another center with the same type of scanner. For this reason, it continues to remain in the research domain, and its clear clinical utility remains to be established. The findings of this study show that with technical advances, and better standardization, it has the potential to add functional information to the excellent structural data obtained with MRI.
MRS does have the potential to inform hypotheses about the pathogenesis of disease. The finding of a decline in Cho/Cr in MCI non-converters, if replicated, warrants an examination of the underlying pathophysiology in the early stages of MCI. A study of longer duration may well show that this is but one phase in the longitudinal course of MCI. In the same vein, the mI and Cho changes in MCI and AD have implications for the understanding of the pathophysiology of AD, and may well suggest possible interventions (1).
In conclusion, the paper by Kantarci et al. highlights the potential of MRS, and points to a future in which a combination of structural and functional imaging techniques will be used to inform us about the progression of AD and permit more accurate early diagnosis and prognosis. Study of MCI using these techniques is therefore to be commended. The clinical applicability of MRS must, however, await further standardization of techniques and more longitudinal studies.
References:
Valenzuela MJ, Sachdev P. Magnetic resonance spectroscopy in AD. Neurology. 2001 Mar 13;56(5):592-8. PubMed.
Miller BL, Moats RA, Shonk T, Ernst T, Woolley S, Ross BD. Alzheimer disease: depiction of increased cerebral myo-inositol with proton MR spectroscopy. Radiology. 1993 May;187(2):433-7. PubMed.
Ross AJ, Sachdev PS, Wen W, Brodaty H. Longitudinal changes during aging using proton magnetic resonance spectroscopy. J Gerontol A Biol Sci Med Sci. 2006 Mar;61(3):291-8. PubMed.
Henriksen O. In vivo quantitation of metabolite concentrations in the brain by means of proton MRS. NMR Biomed. 1995 Jun;8(4):139-48. PubMed.
Ross AJ, Sachdev PS, Wen W, Brodaty H, Joscelyne A, Lorentz LM. Prediction of cognitive decline after stroke using proton magnetic resonance spectroscopy. J Neurol Sci. 2006 Dec 21;251(1-2):62-9. Epub 2006 Nov 7 PubMed.
View all comments by Perminder SachdevMake a Comment
To make a comment you must login or register.