[Skip to Content]
Sign In
Individual Sign In
Create an Account
Institutional Sign In
OpenAthens Shibboleth
[Skip to Content Landing]
Download PDF
Figure 1.
Initial Analysis of the Data Available for 928 Patients With Mild Cognitive Impairment (MCI)
Initial Analysis of the Data Available for 928 Patients With Mild Cognitive Impairment (MCI)

A, Overview of the 249 variables (listed in the Table and eTable 3 in the Supplement) is given in 8 categories. B, Receiver operating characteristic curves combine cerebrospinal fluid (CSF) β-amyloid 42 (Aβ42), t-tau, and p-tau in modeling progression from MCI to Alzheimer disease (AD) within 1 to 6 years. The areas under the curve (AUCs) (95% CI) are given for each year. Receiver operating characteristic curves for Aβ42, t-tau, and p-tau separately are available in eFigure 2 in the Supplement; the sample size available is shown in eTable 2 in the Supplement. Diagonal line indicates completely random discrimination; MRI, magnetic resonance imaging.

Figure 2.
Associations Among 249 Variables Shown by a Circular Visualization of Correlation (CVC) Plot
Associations Among 249 Variables Shown by a Circular Visualization of Correlation (CVC) Plot

Includes 928 patients with mild cognitive impairment. Most of the variables were unavailable for the 928 patients, as indicated in Figure 1A. We calculated the correlation coefficient between pairwise complete observations. Lines (edges) represent the Spearman rank correlation coefficient (r) between 2 variables (nodes). Variables were grouped in 8 categories. Category names of the variables are indicated; except for the cerebrospinal fluid (CSF) and plasma communicome categories, the 249 variables are labeled by numbers (more details on categories and variables are given in the Table and eTable 3 in the Supplement). The CVC plot was thresholded at |r| > 0.3 to display only the strongest correlations. AD indicates Alzheimer disease; MRI, magnetic resonance imaging.

Figure 3.
Prediction of Progression From Mild Cognitive Impairment (MCI) to Alzheimer Disease (AD) Within 1 to 6 Years
Prediction of Progression From Mild Cognitive Impairment (MCI) to Alzheimer Disease (AD) Within 1 to 6 Years

A, Strategy for modeling progression from MCI to AD uses an elastic net to perform joint modeling and variable selection. We performed 1000-fold resampling and ranked variables according to their number of appearances in the elastic net model across the 1000 permutations. The mean classification accuracy rate, sensitivity, and specificity were calculated across the 1000 resampled test data sets. K indicates the number of variables; n, sample size. B, Sensitivity as a function of the specificity for models includes the 2 to 20 top selected variables for each time point of progression. The number of variables are indicated by the numbers in the lines in the plot. Gray quadrant indicates the area of the figure in which sensitivity and specificity are greater than 0.7. C, Classification accuracy rate as a function of its SD across permutations for models includes the top 2 to 20 selected variables for each time point of progression. A small SD across permutations (close to 0) demonstrates a high stability of the model. D, Heatmap of variables consistently selected across time points of progression. A total of 80 of 224 variables were selected at least once as 1 of the top 20 variables in the different models. Details of the 80 variables are available in eTable 5 in the Supplement. Only variables selected in the top 20 of at least 3 progression time points are represented in the heatmap. The importance index represents the number of times a variable was selected across the 1000 permutations. Aβ indicates β-amyloid; ApoA-II, apolipoprotein A-II; CA-19-9, cancer antigen 19-9; CDR SOB, Clinical Dementia Rating Sum of Boxes composite test score; CSF, cerebrospinal fluid; FGF-4, fibroblast growth factor 4; PAPP-A, pregnancy-associated plasma protein A; TECK, thymus-expressed chemokine; TF, transferrin; TRAIL-R3, tumor necrosis factor–related apoptosis-inducing ligand receptor 3; and vWF, von Willebrand factor.

Figure 4.
Prediction of Progression From Mild Cognitive Impairment to Alzheimer Disease (AD) Within 3 Years
Prediction of Progression From Mild Cognitive Impairment to Alzheimer Disease (AD) Within 3 Years

Prediction includes 7 models combining different subsets of variables. Correct classification rate of the top 20 variables was estimated on the test data set after 1000-fold resampling of the learning and test data sets. Sex and age were included in all models. APOE4 indicates apolipoprotein ε4; CSF, cerebrospinal fluid; and MRI, magnetic resonance imaging.

Table.  
Individual Patient Variables Included in the Study
Individual Patient Variables Included in the Study
1.
Morris  JC, Roe  CM, Grant  EA,  et al.  Pittsburgh compound B imaging and prediction of progression from cognitive normality to symptomatic Alzheimer disease. Arch Neurol. 2009;66(12):1469-1475.PubMedArticle
2.
Blennow  K, Dubois  B, Fagan  AM, Lewczuk  P, de Leon  MJ, Hampel  H.  Clinical utility of cerebrospinal fluid biomarkers in the diagnosis of early Alzheimer’s disease. Alzheimers Dement. 2015;11(1):58-69.PubMedArticle
3.
Jack  CR  Jr, Knopman  DS, Jagust  WJ,  et al.  Hypothetical model of dynamic biomarkers of the Alzheimer’s pathological cascade. Lancet Neurol. 2010;9(1):119-128.PubMedArticle
4.
Sperling  RA, Aisen  PS, Beckett  LA,  et al.  Toward defining the preclinical stages of Alzheimer’s disease: recommendations from the National Institute on Aging-Alzheimer’s Association workgroups on diagnostic guidelines for Alzheimer’s disease. Alzheimers Dement. 2011;7(3):280-292.PubMedArticle
5.
Sperling  R, Mormino  E, Johnson  K.  The evolution of preclinical Alzheimer’s disease: implications for prevention trials. Neuron. 2014;84(3):608-622.PubMedArticle
6.
Albert  MS, DeKosky  ST, Dickson  D,  et al.  The diagnosis of mild cognitive impairment due to Alzheimer’s disease: recommendations from the National Institute on Aging-Alzheimer’s Association workgroups on diagnostic guidelines for Alzheimer’s disease. Alzheimers Dement. 2011;7(3):270-279.PubMedArticle
7.
Vadikolias  K, Tsiakiri-Vatamidis  A, Tripsianis  G,  et al.  Mild cognitive impairment: effect of education on the verbal and nonverbal tasks performance decline. Brain Behav. 2012;2(5):620-627.PubMedArticle
8.
Petersen  RC, Smith  GE, Waring  SC, Ivnik  RJ, Tangalos  EG, Kokmen  E.  Mild cognitive impairment: clinical characterization and outcome. Arch Neurol. 1999;56(3):303-308.PubMedArticle
9.
Mitchell  AJ, Shiri-Feshki  M.  Rate of progression of mild cognitive impairment to dementia: meta-analysis of 41 robust inception cohort studies. Acta Psychiatr Scand. 2009;119(4):252-265.PubMedArticle
10.
Bennett  DA, Wilson  RS, Schneider  JA,  et al.  Natural history of mild cognitive impairment in older persons. Neurology. 2002;59(2):198-205.PubMedArticle
11.
Blennow  K, Zetterberg  H.  Cerebrospinal fluid biomarkers for Alzheimer’s disease. J Alzheimers Dis. 2009;18(2):413-417.PubMed
12.
Frisoni  GB, Bocchetta  M, Chételat  G,  et al; ISTAART’s NeuroImaging Professional Interest Area.  Imaging markers for Alzheimer disease: which vs how. Neurology. 2013;81(5):487-500.PubMedArticle
13.
Klunk  WE.  Amyloid imaging as a biomarker for cerebral β-amyloidosis and risk prediction for Alzheimer dementia. Neurobiol Aging. 2011;32(suppl 1):S20-S36.PubMedArticle
14.
Killiany  RJ, Gomez-Isla  T, Moss  M,  et al.  Use of structural magnetic resonance imaging to predict who will get Alzheimer’s disease. Ann Neurol. 2000;47(4):430-439.PubMedArticle
15.
Jack  CR  Jr, Petersen  RC, Xu  YC,  et al.  Prediction of AD with MRI-based hippocampal volume in mild cognitive impairment. Neurology. 1999;52(7):1397-1403.PubMedArticle
16.
Zhang  S, Han  D, Tan  X, Feng  J, Guo  Y, Ding  Y.  Diagnostic accuracy of 18 F-FDG and 11 C-PIB–PET for prediction of short-term conversion to Alzheimer’s disease in subjects with mild cognitive impairment. Int J Clin Pract. 2012;66(2):185-198.PubMedArticle
17.
Shaw  LM, Vanderstichele  H, Knapik-Czajka  M,  et al; Alzheimer’s Disease Neuroimaging Initiative.  Cerebrospinal fluid biomarker signature in Alzheimer’s Disease Neuroimaging Initiative subjects. Ann Neurol. 2009;65(4):403-413.PubMedArticle
18.
Weiner  MW, Veitch  DP, Aisen  PS,  et al; Alzheimer’s Disease Neuroimaging Initiative.  The Alzheimer’s Disease Neuroimaging Initiative: a review of papers published since its inception. Alzheimers Dement. 2013;9(5):e111-e194.PubMedArticle
19.
Folstein  MF, Folstein  SE, McHugh  PR.  “Mini-mental state”: a practical method for grading the cognitive state of patients for the clinician. J Psychiatr Res. 1975;12(3):189-198.PubMedArticle
20.
Rosen  WG, Mohs  RC, Davis  KL.  A new rating scale for Alzheimer’s disease. Am J Psychiatry. 1984;141(11):1356-1364.PubMedArticle
21.
Morris  JC.  The Clinical Dementia Rating (CDR): current version and scoring rules. Neurology. 1993;43(11):2412-2414.PubMedArticle
22.
Pfeffer  RI, Kurosaki  TT, Harrah  CH  Jr, Chance  JM, Filos  S.  Measurement of functional activities in older adults in the community. J Gerontol. 1982;37(3):323-329.PubMedArticle
23.
Sheikh  JI, Yesavage  JA. Recent evidence and development of a shorter version. In: Brink  TL, ed. Clinical Gerontology: A Guide to Assessment and Intervention. New York, NY: The Haworth Press; 1986:165-173.
24.
Rosen  WG, Terry  RD, Fuld  PA, Katzman  R, Peck  A.  Pathological verification of ischemic score in differentiation of dementias. Ann Neurol. 1980;7(5):486-488.PubMedArticle
25.
Kaufer  DI, Cummings  JL, Ketchel  P,  et al.  Validation of the NPI-Q, a brief clinical form of the Neuropsychiatric Inventory. J Neuropsychiatry Clin Neurosci. 2000;12(2):233-239.PubMedArticle
26.
Robin  X, Turck  N, Hainard  A,  et al.  pROC: an open-source package for R and S+ to analyze and compare ROC curves. BMC Bioinformatics. 2011;12:77.PubMedArticle
27.
DeLong  ER, DeLong  DM, Clarke-Pearson  DL.  Comparing the areas under two or more correlated receiver operating characteristic curves: a nonparametric approach. Biometrics. 1988;44(3):837-845.PubMedArticle
28.
Epskamp  S, Cramer AeOJ, Waldorp LJ, Schmittmann VD, Borsboom D.  qgraph: Network visualizations of relationships in psychometric data. J Stat Softw. 2012;48(4):1-18.Article
29.
Friedman  J, Hastie  T, Tibshirani  R.  Regularization paths for generalized linear models via coordinate descent. J Stat Softw. 2010;33(1):1-22.PubMedArticle
30.
Zou  H, Hastie  T.  Regularization and variable selection via the elastic net. J R Stat Soc B. 2005;67:301-320.Article
31.
Ray  S, Britschgi  M, Herbert  C,  et al.  Classification and prediction of clinical Alzheimer’s diagnosis based on plasma signaling proteins. Nat Med. 2007;13(11):1359-1362.PubMedArticle
32.
Coupé  P, Eskildsen  SF, Manjón  JV,  et al; Alzheimer’s Disease Neuroimaging Initiative.  Scoring by nonlocal image patch estimator for early detection of Alzheimer’s disease. Neuroimage Clin. 2012;1(1):141-152.PubMedArticle
33.
Boulesteix  AL, Slawski  M.  Stability and aggregation of ranked gene lists. Brief Bioinform. 2009;10(5):556-568.PubMedArticle
34.
He  Z, Yu  W.  Stable feature selection for biomarker discovery. Comput Biol Chem. 2010;34(4):215-225.PubMedArticle
35.
Michiels  S, Koscielny  S, Hill  C.  Prediction of cancer outcome with microarrays: a multiple random validation strategy. Lancet. 2005;365(9458):488-492.PubMedArticle
36.
Zucknick  M, Richardson  S, Stronach  EA.  Comparing the characteristics of gene expression profiles derived by univariate and multivariate classification methods. Stat Appl Genet Mol Biol. 2008;7(1):e7. doi:10.2202/1544-6115.1307.PubMed
37.
Corder  EH, Saunders  AM, Strittmatter  WJ,  et al.  Gene dose of apolipoprotein E type 4 allele and the risk of Alzheimer’s disease in late onset families. Science. 1993;261(5123):921-923.PubMedArticle
38.
Trzepacz  PT, Yu  P, Sun  J,  et al; Alzheimer’s Disease Neuroimaging Initiative.  Comparison of neuroimaging modalities for the prediction of conversion from mild cognitive impairment to Alzheimer’s dementia. Neurobiol Aging. 2014;35(1):143-151.PubMedArticle
39.
Fei  M, Jianhua  W.  Apolipoprotein ε4-allele as a significant risk factor for conversion from mild cognitive impairment to Alzheimer’s disease: a meta-analysis of prospective studies. J Mol Neurosci. 2013;50(2):257-263.PubMedArticle
40.
Winblad  B, Palmer  K, Kivipelto  M,  et al.  Mild cognitive impairment—beyond controversies, towards a consensus: report of the International Working Group on Mild Cognitive Impairment. J Intern Med. 2004;256(3):240-246.PubMedArticle
41.
Young  J, Modat  M, Cardoso  MJ, Mendelson  A, Cash  D, Ourselin  S; Alzheimer’s Disease Neuroimaging Initiative.  Accurate multimodal probabilistic prediction of conversion to Alzheimer’s disease in patients with mild cognitive impairment. Neuroimage Clin. 2013;2:735-745.PubMedArticle
42.
Querfurth  HW, LaFerla  FM.  Alzheimer’s disease. N Engl J Med. 2010;362(4):329-344.PubMedArticle
43.
Henneman  WJ, Sluimer  JD, Barnes  J,  et al.  Hippocampal atrophy rates in Alzheimer disease: added value over whole brain volume measures. Neurology. 2009;72(11):999-1007.PubMedArticle
44.
Vemuri  P, Wiste  HJ, Weigand  SD,  et al; Alzheimer’s Disease Neuroimaging Initiative.  MRI and CSF biomarkers in normal, MCI, and AD subjects: predicting future clinical change. Neurology. 2009;73(4):294-301.PubMedArticle
45.
Davatzikos  C, Bhatt  P, Shaw  LM, Batmanghelich  KN, Trojanowski  JQ.  Prediction of MCI to AD conversion, via MRI, CSF biomarkers, and pattern classification. Neurobiol Aging. 2011;32(12):2322.e19-2322.e27.PubMedArticle
46.
Eskildsen  SF, Coupé  P, García-Lorenzo  D, Fonov  V, Pruessner  JC, Collins  DL; Alzheimer’s Disease Neuroimaging Initiative.  Prediction of Alzheimer’s disease in subjects with mild cognitive impairment from the ADNI cohort using patterns of cortical thinning. Neuroimage. 2013;65:511-521.PubMedArticle
47.
Wee  CY, Yap  PT, Shen  D; Alzheimer’s Disease Neuroimaging Initiative.  Prediction of Alzheimer’s disease and mild cognitive impairment using cortical morphological patterns. Hum Brain Mapp. 2013;34(12):3411-3425.PubMedArticle
48.
Wolz  R, Julkunen  V, Koikkalainen  J,  et al; Alzheimer’s Disease Neuroimaging Initiative.  Multi-method analysis of MRI images in early diagnostics of Alzheimer’s disease. PLoS One. 2011;6(10):e25446.PubMedArticle
49.
Llano  DA, Devanarayan  V, Simon  AJ; Alzheimer’s Disease Neuroimaging Initiative (ADNI).  Evaluation of plasma proteomic data for Alzheimer disease state classification and for the prediction of progression from mild cognitive impairment to Alzheimer disease. Alzheimer Dis Assoc Disord. 2013;27(3):233-243.PubMedArticle
50.
Zhao  X, Lejnine  S, Spond  J,  et al.  A candidate plasma protein classifier to identify Alzheimer’s disease. J Alzheimers Dis. 2015;43(2):549-563.PubMed
51.
Hye  A, Riddoch-Contreras  J, Baird  AL,  et al.  Plasma proteins predict conversion to dementia from prodromal disease. Alzheimers Dement. 2014;10(6):799-807.e2.PubMedArticle
52.
Ferreira  D, Perestelo-Pérez  L, Westman  E, Wahlund  LO, Sarría  A, Serrano-Aguilar  P.  Meta-review of CSF core biomarkers in Alzheimer’s disease: the state-of-the-art after the new revised diagnostic criteria. Front Aging Neurosci. 2014;6:47.PubMed
53.
Jaeger  PA, Villeda  SA, Berdnik  D, Britschgi  M, Wyss-Coray  T. Focused plasma proteomics for the study of brain aging and neurodegeneration. In: Coppola  G, ed. The OMICs: Applications in Neuroscience. Oxford, England: Oxford University Press; 2014:183-191.
54.
Olsson  B, Hertze  J, Ohlsson  M,  et al.  Cerebrospinal fluid levels of heart fatty acid binding protein are elevated prodromally in Alzheimer’s disease and vascular dementia. J Alzheimers Dis. 2013;34(3):673-679.PubMed
55.
Britschgi  M, Rufibach  K, Huang  SL,  et al.  Modeling of pathological traits in Alzheimer’s disease based on systemic extracellular signaling proteome. Mol Cell Proteomics. 2011;10(10):008862. doi:10.1074/mcp.M111.008862.PubMedArticle
56.
Csernansky  JG, Dong  H, Fagan  AM,  et al.  Plasma cortisol and progression of dementia in subjects with Alzheimer-type dementia. Am J Psychiatry. 2006;163(12):2164-2169.PubMedArticle
57.
Soares  HD, Potter  WZ, Pickering  E,  et al; Biomarkers Consortium Alzheimer’s Disease Plasma Proteomics Project.  Plasma biomarkers associated with the apolipoprotein E genotype and Alzheimer disease. Arch Neurol. 2012;69(10):1310-1317.PubMedArticle
58.
Toledo  JB, Toledo  E, Weiner  MW,  et al; Alzheimer’s Disease Neuroimaging Initiative.  Cardiovascular risk factors, cortisol, and amyloid-β deposition in Alzheimer’s Disease Neuroimaging Initiative. Alzheimers Dement. 2012;8(6):483-489.PubMedArticle
59.
Martignoni  E, Petraglia  F, Costa  A, Bono  G, Genazzani  AR, Nappi  G.  Dementia of the Alzheimer type and hypothalamus-pituitary-adrenocortical axis: changes in cerebrospinal fluid corticotropin releasing factor and plasma cortisol levels. Acta Neurol Scand. 1990;81(5):452-456.PubMedArticle
60.
Lei  JK.  Change of serum ACTH and cortisol levels in Alzheimer disease and mild cognition impairment [in Chinese]. Zhonghua Yi Xue Za Zhi. 2010;90(41):2894-2896.PubMed
61.
Calero  M, Rostagno  A, Matsubara  E, Zlokovic  B, Frangione  B, Ghiso  J.  Apolipoprotein J (clusterin) and Alzheimer’s disease. Microsc Res Tech. 2000;50(4):305-315.PubMedArticle
62.
Song  F, Poljak  A, Crawford  J,  et al.  Plasma apolipoprotein levels are associated with cognitive status and decline in a community cohort of older individuals. PLoS One. 2012;7(6):e34078.PubMedArticle
63.
Huang  Y, Mahley  RW.  Apolipoprotein E: structure and function in lipid metabolism, neurobiology, and Alzheimer’s diseases. Neurobiol Dis. 2014;72(pt A):3-12.PubMedArticle
64.
Craig-Schapiro  R, Kuhn  M, Xiong  C,  et al.  Multiplexed immunoassay panel identifies novel CSF biomarkers for Alzheimer’s disease diagnosis and prognosis. PLoS One. 2011;6(4):e18850.PubMedArticle
65.
Heppner  FL, Ransohoff  RM, Becher  B.  Immune attack: the role of inflammation in Alzheimer disease. Nat Rev Neurosci. 2015;16(6):358-372.PubMedArticle
Original Investigation
February 2016

Combined Plasma and Cerebrospinal Fluid Signature for the Prediction of Midterm Progression From Mild Cognitive Impairment to Alzheimer Disease

Author Affiliations
  • 1Department of Neurology and Neurological Sciences, Stanford University School of Medicine, Stanford, California
  • 2Translational Technologies and Bioinformatics, Roche Pharma Research and Early Development, Roche Innovation Center Basel, F. Hoffmann–La Roche, Ltd, Basel, Switzerland
  • 3Roche Pharma Development, F. Hoffmann–La Roche, Ltd, Basel, Switzerland
  • 4Center for Tissue Regeneration, Repair and Restoration, Veterans Affairs Palo Alto Health Care System, Palo Alto, California
  • 5Neuroscience, Ophthalmology, and Rare Diseases Discovery and Translational Areas, Roche Pharma Research and Early Development, Roche Innovation Center Basel, F. Hoffmann–La Roche, Ltd, Basel, Switzerland
JAMA Neurol. 2016;73(2):203-212. doi:10.1001/jamaneurol.2015.3135
Abstract

Importance  A reliable method of detecting Alzheimer disease (AD) in its prodromal state is needed for patient stratification in clinical trials or for personalizing existing or potential upcoming therapies. Current cerebrospinal fluid (CSF)– or imaging-based single biomarkers for AD offer reliable identification of patients with underlying AD but insufficient prediction of the rate of AD progression.

Objective  To optimize prediction of progression from mild cognitive impairment (MCI) to AD dementia by combining information from diverse patient variables.

Design, Setting, and Participants  This cohort study from the Alzheimer Disease Neuroimaging Initiative (ADNI) enrolled 928 patients with MCI at baseline and 249 selected variables available in the ADNI data set. Variables included clinical and demographic data, cognitive scores, magnetic resonance imaging–based brain volumetric data, the apolipoprotein E (APOE) and translocase of outer mitochondrial membrane 40 homolog (TOMM40) genotypes, and analyte levels measured in the CSF and plasma. Data were collected in July 2012 and analyzed from July 1, 2012, to June 1, 2015.

Main Outcomes and Measures  Progression from MCI to AD within 1 to 6 years. To determine whether combinations of markers could predict progression from MCI to AD within 1 to 6 years, the elastic net algorithm was used in an iterative resampling of a training- and test-based variable selection and modeling approach.

Results  Among the 928 patients with MCI in the ADNI database, 94 had 224 of the required variables available for the modeling. The results showed the contributions of age, Clinical Dementia Rating Sum of Boxes composite test score, hippocampal volume, and multiple plasma and CSF factors in modeling progression to AD. A combination of apolipoprotein A-II and cortisol levels in plasma and fibroblast growth factor 4, heart-type fatty acid binding protein, calcitonin, and tumor necrosis factor–related apoptosis-inducing ligand receptor 3 (TRAIL-R3) in CSF allowed for reliable prediction of disease status 3 years from the time of sample collection (80% classification accuracy, 88% sensitivity, and 70% specificity).

Conclusions and Relevance  These study findings suggest that a combination of markers measured in plasma and CSF, distinct from β-amyloid and tau, could prove useful in predicting midterm progression from MCI to AD dementia. Such a large-scale, multivariable-based analytical approach could be applied to other similar large data sets involving AD and beyond.

Introduction

The use of biomarkers for early diagnosis of Alzheimer disease (AD) has been investigated widely, and several studies showed that cognitively normal individuals who later develop AD dementia can be identified earlier in the disease course by use of imaging and cerebrospinal fluid (CSF) biomarkers.1,2 Levels of β-amyloid (Aβ) in the brain demonstrate the earliest AD-type changes3 and can be detected by measuring CSF levels of Aβ42 or using positron emission tomography (PET). Unfortunately, the temporal resolution of Aβ-based biomarkers is too weak to accurately predict progression of the disease course from mild cognitive impairment (MCI) to AD dementia.

Alzheimer disease can be represented by a continuum from cognitively normal-appearing individuals with evidence of accumulation of Aβ in the brain to those with severe dementia.4,5 Mild cognitive impairment is an intermediate stage between normal cognitive decline with aging and dementia; during this stage, patients have a greater cognitive decline than expected for their age and educational level.68 Current data indicate that rates of MCI progression to AD are estimated at approximately 10% per year9 and represent the population with the highest risks for progression to AD.10 Other causes of MCI include dementia with Lewy bodies, Parkinson disease, frontotemporal dementia, and stroke.6 Owing to the heterogeneity of the causes of MCI, research criteria and biomarkers were defined; according to the most recently published diagnostic criteria,6 neurodegeneration and cerebral amyloidosis are necessary to determine which individual with MCI has underlying AD. In the search for diagnostic and prognostic biomarkers of AD, the field has focused heavily on CSF11 and neuroimaging markers.12 For example, CSF levels of Aβ4211 or Aβ levels on PET imaging13 are both indicative of cerebral amyloidosis, and magnetic resonance imaging (MRI), fludeoxyglucose F 18–labeled (FDG)–PET,12 or CSF levels of t-tau11 are usually used to assess neurodegeneration. Although several studies showed that progression of individuals with MCI to dementia within a few years could be predicted by MRI findings,14,15 FDG-PET,16 or CSF examination,17 no unified or combination of biomarkers has emerged to predict the time to progression.

Herein, we used data from the Alzheimer’s Disease Neuroimaging Initiative (ADNI), which is an ongoing, longitudinal, multicenter study.18 In this data set, we identified combinations of specific individual patient variables that predict progression from MCI to AD 1 to 6 years before the clinical diagnosis of AD.

Methods
Patient Data

Data used in preparing this article were produced by ADNI (eMethods in the Supplement) and were obtained in July 2012. A complete list of the 249 individual patient variables used for the analysis is provided in the Table.1925 A summary of patient variables among the 928 study participants with MCI are available in eTable 1 in the Supplement. Data preprocessing is detailed in the eMethods in the Supplement. The individual baseline patient variables used in this study were only available for participants in ADNI acquired until 2010. For the multivariable analysis to model progression to AD with baseline data, 94 patients with MCI had all 224 required variables (Figure 1A). The number of patients with MCI included in the follow-up declined during the 6 years (eFigure 1 in the Supplement). In the most recent version of the ADNI database, 24 of the 94 patients with MCI had a longer follow-up. The ADNI study was approved by institutional review boards of all participating institutions. Informed written consent was obtained from all participants at each site.

Utility of Established Biomarkers of AD

To test the utility of the established AD CSF biomarkers Aβ42, t-tau, and p-tau to predict progression from MCI to AD in the ADNI cohort, we used these variables to model the binary end point of stable MCI vs progression to AD within 1 to 6 years (progression was defined by an on-site physician). We calculated a receiver operating characteristic curve using the R pROC package.26 We computed 95% CIs for the area under the curve using the approach of DeLong et al.27 Not all 928 participants underwent lumbar puncture, and the exact sample size available for each CSF biomarker and progression time point is shown in eTable 2 in the Supplement.

Visualization of the Associations Among Large Panels of Variables

To depict associations within and between the 8 categories of variables, a circular visualization of the correlation plot was generated using the qgraph package for R.28 This plot is based on calculating pairwise rank correlations between complete observations. The plot displays a network with nodes representing the variables and edges linking any pairs of variables based on their correlation coefficient with each other. A threshold of |r| > 0.3 was used to display only the strongest correlations. The circular visualization of the correlation plot includes all 249 variables and data available from 928 patients.

Selection and Combination of Specific Individual Patient Variables

Prediction of progression from MCI to AD used 224 variables in only 94 of the 928 patients because data were missing for most of the patients (Figure 1A, Table, and eMethods and eTable 3 in the Supplement). To identify variables associated with progression from MCI to AD within 1 to 6 years, we used an elastic net algorithm29,30 in an iterative resampling of a training- and test-based variable selection and modeling approach. Briefly, elastic net was applied on the training subset to select variables that best discriminate between patients with stable and progressive MCI. The quality of each model was estimated on the test data set using the classification accuracy rate, sensitivity, specificity, and stability. One thousand resamplings of the learning and test data sets were performed, and the variables were ranked according to their number of appearances across permutations in the elastic net models to select the top variables. To refine these predictive models to a simpler final model, we then used a forward classification strategy and compared our results with those obtained by chance. Data were analyzed from July 1, 2012, to June 1, 2015. A complete description of the approach used is detailed in the eMethods in the Supplement.

Results
Utility of Established Biomarkers of AD to Predict Progression to AD

First, we evaluated the utility of CSF levels of Aβ42, t-tau, and p-tau to predict progression to AD within 1 to 6 years and performed a receiver operating characteristic analysis. These established diagnostic AD biomarkers combined (Figure 1B) or separately (eFigure 2 in the Supplement) cannot be used for reliable prediction of progression from MCI to AD. Hence, exploring other fluid markers and clinical variables is indicated for this purpose.

Associations Among Large Panels of Variables

To get an overview of the associations among the 249 variables available across 928 patients with MCI from the ADNI database (Table), we generated a circular visualization of correlation plot (Figure 2). This plot revealed complex relationships within and between groups of variables. For instance, concentrations of secreted proteins involved in intercellular communication (previously termed the communicome31) measured in plasma or CSF samples were strongly correlated within and between these 2 body fluids. Age and sex were correlated with the communicome but not with the established CSF biomarkers Aβ42 and t-tau. In contrast, the apolipoprotein E (APOE) genotype was linked to established CSF biomarkers but not to the communicome. Proteins of the communicome are linked differentially with established AD CSF biomarkers and MRI-based brain volumes, which suggests that variables from different categories could carry complementary information about the disease. This information may help to predict progression from MCI to AD.

Prediction of Progression From MCI to AD Within 1 to 6 Years

To determine whether combinations of markers could predict progression from MCI to AD within 1 to 6 years, we had to restrict our analysis to 94 patients with MCI and 224 variables with sufficient available baseline data (Table and eTable 3 in the Supplement). Of these 94 patients, a growing number had progression to AD within 1 to 6 years, whereas the follow-up of individuals dropped below 50% after 4 years (eFigure 1 in the Supplement; cohort characteristics are given in eTable 4 in the Supplement).

We built predictive models for each progression time point in 6- or 12-month increments (Figure 3A and eMethods in the Supplement); for each model we calculated sensitivity and specificity using the top 2 to 20 selected variables (Figure 3B). Among all models that used 20 variables, those predicting 2 or 3 years were most accurate (sensitivity/specificity, 76%/70% for 2 years; 87%/73% for 3 years), whereas the other models had low sensitivity (for 1, 1.5, and 6 years) or low specificity (for 4 and 5 years). As indicated by the SD across permutations, prediction of progression within 4 years or later was less stable than earlier prediction of progression to AD (Figure 3C).

A total of 80 of 224 variables were selected at least once as one of the top 20 variables in the different models (eTable 5 in the Supplement). The most frequently used top variable was the neuropsychometric Clinical Dementia Rating Sum of Boxes composite test score (CDR SOB), which was included in 6 of 7 models. The molecular marker CSF tumor necrosis factor–related apoptosis-inducing ligand receptor 3 (TRAIL-R3) was included in 5 models, plasma apolipoprotein A-II (ApoA-II) and CSF fibroblast growth factor 4 (FGF-4) were included in 4 models, and 13 additional variables were selected at least 3 times, including CSF Aβ42, hippocampal volume, and age (the importance index for each variable is given in Figure 3D).

Importance of Specific Categories of Patient Variables to Prediction of Progression Within 3 Years

To investigate whether patient variables from specific categories are sufficient by themselves to predict progression to AD, we tested 7 models using only certain types of variables instead of variables from all groups as used above (Figure 4). We focused on progression to AD within 3 years because the best trade-off between sensitivity and specificity was obtained for this point; for clinical trial planning, this point is also the most relevant. Clinical and demographic characteristics and the number of APOE4 alleles between patients with stable MCI and those with progression to AD within 3 years were similar (eTable 4 in the Supplement). Those with progression to AD had lower levels of CSF Aβ42 (P = .02, unpaired 2-tailed t test).

Classification accuracy and sensitivity (Figure 4 and eFigure 3A in the Supplement) were similarly high for all models except for those using AD CSF biomarkers and the APOE4 genotype, MRI-based brain regional volumes, or cognitive scores. Models using the plasma and CSF communicomes or all variables were superior in accuracy and sensitivity to the other models. Specificity was relatively low for all models but again better for models using the plasma and CSF communicomes or all variables (eFigure 3B in the Supplement), making these 2 models the best overall.

Remarkably, only 6 plasma and CSF analytes were necessary to reach a maximal sensitivity of 88% and a specificity of 70% (eFigure 3A and B in the Supplement). The 6 analytes outperformed a model that included AD CSF biomarkers Aβ42, t-tau, and p-tau together with APOE4 allele carrier status (Figure 4). Furthermore, this top model clearly performed better than randomly generated variables and is not improved by including the plasma Aβ42:Aβ40 ratio (eTable 6 in the Supplement). This signature was composed of 2 analytes measured in plasma (ApoA-II and cortisol) and 4 proteins measured in the CSF (FGF-4, heart-type fatty acid binding protein [FABP-heart], calcitonin, and TRAIL-R3). The mean level of TRAIL-R3 was significantly decreased for patients with progression to AD (mean, 0.63 [95% CI, 0.56-0.70] vs 0.81 [0.71-0.92]; P < .01, unpaired 2-tailed t test), whereas mean levels of ApoA-II (531 [478-583] vs 445 [402-487]), cortisol (165 [149-180] vs 141 [125-158]), and FGF-4 (49 [44-54] vs 39 [33-46]) were significantly increased (eFigure 4 in the Supplement; P < .05, unpaired 2-tailed t tests). Levels of calcitonin and FABP-heart were not significantly different between patients with stable MCI and those with progression to AD (t tests, P = .07 and P = .14, respectively). The levels of 4 of these 6 analytes in addition to CSF Aβ42 were significantly different between patients with stable MCI and progressive MCI, whereas CSF levels of t-tau and p-tau were not (eFigure 5 in the Supplement; P = .28 and P = .70, respectively). In summary, the levels of only 6 plasma and CSF analytes were sufficient for the prediction of progression from MCI to AD within 3 years.

Discussion

In this study, we used data from the ADNI database and combined clinical and demographic data, cognitive measurements, brain volumetric data, APOE and translocase of outer mitochondrial membrane 40 homolog (TOMM40) genotypes, and a large number of analyte measurements in the CSF and plasma to predict progression from MCI to AD during 6 years of patient follow-up. The ADNI database is a valuable resource to better understand AD clinically, particularly with respect to the integration of imaging data.18 In contrast, comprehensive analyses with a broader range of clinical or biological variables, such as the one reported herein, are challenging owing to the rather patchwork collection of the data over time (Figure 1A). As exemplified herein, this approach results in a relatively small sample size for applicable patients with MCI (94 herein owing to the limited availability of the CSF multiplex data, for example), and identification of an appropriate independent validation data set is difficult or even impossible. These 2 points are probably the major limitations of our study. Despite these limitations, a network visualization approach allowed us to produce an overview of the complex relationships between and within categories of variables in ADNI (Figure 2). To overcome the shortcoming of missing data for modeling progression of MCI to AD, we performed thorough cross validation (eg, 1000 times resampling of the patients included in the learning and test data sets for an unbiased and robust estimate of the accuracy of the models32), evaluation of stability of the models to assess the potential generalization of findings from this study to other data sets,33,34 and a final forward classification step to avoid overfitting of the predictive models. Altogether, the validation strategies applied by us are based on current standards in the field,3336 and we believe their combination resulted in the most rigorous validation that can be performed in the absence of an additional cohort.

A key question in modeling progression from MCI to AD relates to the temporal utility of measured variables. Of the features most consistently selected across progression time points (Figure 2D), 7 were analytes measured in CSF (including Aβ42) and 7 were analytes measured in plasma. Only 3 non–body fluid variables (CDR SOB, hippocampal volume, and age) were selected, and the APOE4 genotype was not included despite the APOE4 allele being the major genetic risk factor for AD.37 The utility of APOE4 allele status in predicting time to progression to AD is not clear because results have been inconsistent.3841 In contrast, age, which is the strongest environmental risk factor for developing sporadic AD,42 was included in the model of long-term progression. A high rate of decline in hippocampal size is known to be one of the best MRI-based biomarkers of AD,43,44 and hippocampal atrophy was selected as a predictor of short-term progression. The composite CDR SOB score was in the top features for almost all of the progressive time points studied (6 of 7). Our data indicate that the baseline CDR SOB composite test score combined with other variables could be useful for predicting short-term, midterm, and long-term progression from MCI to AD and support its use for planning and analyzing clinical trials.

Being able to select patients with midterm progression from MCI to AD is of major interest for assessing the efficacy of new AD therapies or for stratifying clinical trial cohorts. Herein, we demonstrated that signatures relying on prediction of progression within 2 and 3 years were more robust than those relying on other progressive time points in terms of sensitivity, specificity, and stability. At least for the time points after 4 years, this outcome may be influenced in large part by the lack of a sufficient sample size. Several other teams32,38,41,4548 analyzed the ADNI data and identified methods to predict progression from MCI to AD within 3 years, primarily focusing on imaging data. The best model so far had a classification accuracy rate close to what we report herein but a sensitivity of only 53%.38 Although imaging is one of the best methods for monitoring AD, a blood test that predicts progression from MCI to AD within a defined period of time would be immensely useful because blood samples are easy to collect. However, other investigators49 reported that using the ADNI data set plasma analytes alone could not adequately predict midterm progression to AD. Recently, 2 candidate signatures of progression to AD were proposed.50,51 These studies were, however, limited to the prediction of progression to AD 1 year before its clinical diagnosis. In addition to plasma biomarkers, CSF-based biomarkers may be particularly representative of the disease progression because CSF is in close contact with the central nervous system. For instance, low CSF concentrations of Aβ42 in combination with high levels of t-tau and p-tau are sensitive and specific diagnostic biomarkers of AD.17,52 In the entire population with MCI in the ADNI cohort, however, our study shows that CSF concentrations of these markers cannot be used to reliably predict time to progression to AD (Figure 1B). Consequently, inclusion of additional markers needs to be investigated, and we found that markers in plasma or CSF used separately for modeling provided a relatively high sensitivity in detecting progression to AD (Figure 4). Once further validated, the marker sets in each of these fluids may be useful for patient enrichment in clinical trials, albeit perhaps with lower accuracy. Our data indicate that combining as few as 6 specific communicome markers measured in CSF and plasma may be more powerful in predicting the progression from MCI to AD and in identifying patients with stable MCI.

This finding is in line with what a group of investigators previously introduced as the communicome being “a reductionist approach to study brain aging and disease.”53(p185) Because the plasma and CSF proteome is particularly challenging for unbiased proteomics approaches, such as mass spectrometry, measuring the secreted communication factors of cells is a straightforward way to explore the integrated response of cellular communication between tissues in physiological and pathophysiological states. Although this method is biased and restricted, it focuses on the proteome of key biological communication factors. Still, future studies should examine the diagnostic utility of the 6 markers to discriminate AD from other causes of dementia and to assess this signature and each candidate marker as a potential biomarker of cognitive decline in independent and larger sample sets. Indeed, individual communicome plasma and CSF factors—and thereby the proposed signature—can be influenced by variables such as age, sex, or ethnicity.

So far, independent evidence reported in other data sets supports an association of the top CSF and plasma markers identified in this study with AD. Of the 6 markers predicting progression to AD within 3 years, CSF levels of FABP-heart and TRAIL-R3 and plasma levels of cortisol and ApoA-II have already been reported by others to be involved in AD.5464 Levels of FABP-heart are increased in CSF samples from patients with progression of MCI to AD54 and highly associated with t-tau and p-tau levels and the ratio of Aβ42 to tau.55 Plasma cortisol levels reflect the degree of cognitive impairment in AD,56 are associated with the presence of the APOE4 allele,57 and correlate with Aβ-plaque brain burden measured by Pittsburgh compound B–labeled PET.58 High cortisol levels were also reported previously in plasma, serum, or CSF in patients with MCI and AD compared with controls56,59,60 and are also associated with more rapidly increasing symptoms of dementia.56,60 Apolipoproteins have been implicated in the cause of AD,6163 and low levels of plasma ApoA-II are associated with an increased risk for cognitive decline in cognitively normal individuals.62 The TRAIL-R3 marker is involved in the regulation of apoptosis and upregulated in cognitively impaired individuals compared with controls.64

The apparent discrepancy between our finding for ApoA-II and TRAIL-R3 in patients with MCI and findings by others62,64 in cognitively normal individuals can perhaps be explained by a possiblly different and so far unknown contribution of ApoA-II and TRAIL-R3 in the disease progression from normal cognition to MCI and from MCI to AD. We report herein an increase in ApoA-II levels in patients with progression of MCI to AD compared with stable MCI. This finding suggests that individual factors, such as ApoA-II levels, show a biphasic association with disease progression, an aspect that should be further explored in other data sets and biological experiments. Finally, to the best of our knowledge, no direct link among calcitonin, FGF-4, and AD or MCI has been reported previously. These proteins have multiple functions, including use in immune pathways that could link them to altered immune function in AD.65

Conclusions

We performed an integrative statistical analysis of the MCI data subset in the ADNI database and showed that the combination of selected plasma and CSF markers may be sufficient for the prediction of midterm progression from MCI to AD. We propose that such a large-scale analytical approach using the ADNI database could be applied to other similar large data sets in AD and beyond. Markers or signatures thereby identified could become helpful for early diagnosis and monitoring of patients, patient stratification in clinical trials, or personalizing existing or upcoming therapies.

Back to top
Article Information

Corresponding Author: Markus Britschgi, PhD, Neuroscience, Ophthalmology, and Rare Diseases Discovery and Translational Areas, Roche Pharma Research and Early Development, Roche Innovation Center Basel, F. Hoffmann–La Roche, Ltd, Grenzacherstrasse 124, CH-4070 Basel, Switzerland (markus.britschgi@roche.com).

Accepted for Publication: August 31, 2015.

Published Online: December 14, 2015. doi:10.1001/jamaneurol.2015.3135.

Author Contributions: Dr Lehallier had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Study concept and design: Lehallier, Essioux, Nikolcheva, Wyss-Coray, Britschgi.

Acquisition, analysis, or interpretation of data: All authors.

Drafting of the manuscript: Lehallier, Essioux, Gayan, Nikolcheva, Wyss-Coray, Britschgi.

Critical revision of the manuscript for important intellectual content: All authors.

Statistical analysis: Lehallier, Essioux, Gayan, Alexandridis.

Obtained funding: Wyss-Coray, Britschgi.

Administrative, technical, or material support: Wyss-Coray, Britschgi.

Study supervision: Essioux, Gayan, Wyss-Coray, Britschgi.

Conflict of Interest Disclosures: Drs Essioux, Gayan, Nikolcheva, and Britschgi are full-time employees of F. Hoffmann-La Roche, Ltd. No other disclosures were reported.

Funding/Support: This study was supported by the Roche Postdoctoral Fellowship program, by grant U01 AG024904 from the National Institutes of Health, and award number W81XWH-12-2-0012 from the Department of Defense to the Alzheimer’s Disease Neuroimaging Initiative (ADNI) (data collection and sharing for this project). ADNI is supported by the National Institute on Aging, the National Institute of Biomedical Imaging and Bioengineering, and generous contributions from AbbVie, Alzheimer’s Association; Alzheimer’s Drug Discovery Foundation; Araclon Biotech; BioClinica, Inc; Biogen; Bristol-Myers Squibb Company; CereSpir, Inc; Eisai, Inc; Elan Pharmaceuticals, Inc; Eli Lilly and Company; EuroImmun; F. Hoffmann–La Roche, Ltd, and its affiliated company Genentech, Inc; Fujirebio; GE Healthcare; IXICO, Ltd; Janssen Alzheimer Immunotherapy Research and Development, LLC; Johnson and Johnson Pharmaceutical Research and Development, LLC; Lumosity; Lundbeck; Merck and Co, Inc; Meso Scale Diagnostics, LLC; NeuroRx Research; Neurotrack Technologies; Novartis Pharmaceuticals Corporation; Pfizer, Inc; Piramal Imaging; Servier; Takeda Pharmaceutical Company; and Transition Therapeutics. The Canadian Institutes of Health Research provides funds to support ADNI clinical sites in Canada. Private sector contributions are facilitated by the Foundation for the National Institutes of Health (http://www.fnih.org). The grantee organization is the Northern California Institute for Research and Education, and the study is coordinated by the Alzheimer’s Disease Cooperative Study at the University of California, San Diego. ADNI data are disseminated by the Laboratory for Neuroimaging at the University of Southern California.

Role of the Funder/Sponsor: Roche employees were involved in the design and conduct of the study; collection, management, analysis, or interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.

Group Information: A complete list of ADNI investigators can be found at https://adni.loni.usc.edu/wp-content/uploads/how_to_apply/ADNI_Acknowledgement_List.pdf.

Additional Information: Data used in preparation of this article were obtained from the ADNI database (http://adni.loni.usc.edu). As such, the investigators within the ADNI contributed to the design and implementation of ADNI and/or provided data but did not participate in the analysis or writing of this report.

References
1.
Morris  JC, Roe  CM, Grant  EA,  et al.  Pittsburgh compound B imaging and prediction of progression from cognitive normality to symptomatic Alzheimer disease. Arch Neurol. 2009;66(12):1469-1475.PubMedArticle
2.
Blennow  K, Dubois  B, Fagan  AM, Lewczuk  P, de Leon  MJ, Hampel  H.  Clinical utility of cerebrospinal fluid biomarkers in the diagnosis of early Alzheimer’s disease. Alzheimers Dement. 2015;11(1):58-69.PubMedArticle
3.
Jack  CR  Jr, Knopman  DS, Jagust  WJ,  et al.  Hypothetical model of dynamic biomarkers of the Alzheimer’s pathological cascade. Lancet Neurol. 2010;9(1):119-128.PubMedArticle
4.
Sperling  RA, Aisen  PS, Beckett  LA,  et al.  Toward defining the preclinical stages of Alzheimer’s disease: recommendations from the National Institute on Aging-Alzheimer’s Association workgroups on diagnostic guidelines for Alzheimer’s disease. Alzheimers Dement. 2011;7(3):280-292.PubMedArticle
5.
Sperling  R, Mormino  E, Johnson  K.  The evolution of preclinical Alzheimer’s disease: implications for prevention trials. Neuron. 2014;84(3):608-622.PubMedArticle
6.
Albert  MS, DeKosky  ST, Dickson  D,  et al.  The diagnosis of mild cognitive impairment due to Alzheimer’s disease: recommendations from the National Institute on Aging-Alzheimer’s Association workgroups on diagnostic guidelines for Alzheimer’s disease. Alzheimers Dement. 2011;7(3):270-279.PubMedArticle
7.
Vadikolias  K, Tsiakiri-Vatamidis  A, Tripsianis  G,  et al.  Mild cognitive impairment: effect of education on the verbal and nonverbal tasks performance decline. Brain Behav. 2012;2(5):620-627.PubMedArticle
8.
Petersen  RC, Smith  GE, Waring  SC, Ivnik  RJ, Tangalos  EG, Kokmen  E.  Mild cognitive impairment: clinical characterization and outcome. Arch Neurol. 1999;56(3):303-308.PubMedArticle
9.
Mitchell  AJ, Shiri-Feshki  M.  Rate of progression of mild cognitive impairment to dementia: meta-analysis of 41 robust inception cohort studies. Acta Psychiatr Scand. 2009;119(4):252-265.PubMedArticle
10.
Bennett  DA, Wilson  RS, Schneider  JA,  et al.  Natural history of mild cognitive impairment in older persons. Neurology. 2002;59(2):198-205.PubMedArticle
11.
Blennow  K, Zetterberg  H.  Cerebrospinal fluid biomarkers for Alzheimer’s disease. J Alzheimers Dis. 2009;18(2):413-417.PubMed
12.
Frisoni  GB, Bocchetta  M, Chételat  G,  et al; ISTAART’s NeuroImaging Professional Interest Area.  Imaging markers for Alzheimer disease: which vs how. Neurology. 2013;81(5):487-500.PubMedArticle
13.
Klunk  WE.  Amyloid imaging as a biomarker for cerebral β-amyloidosis and risk prediction for Alzheimer dementia. Neurobiol Aging. 2011;32(suppl 1):S20-S36.PubMedArticle
14.
Killiany  RJ, Gomez-Isla  T, Moss  M,  et al.  Use of structural magnetic resonance imaging to predict who will get Alzheimer’s disease. Ann Neurol. 2000;47(4):430-439.PubMedArticle
15.
Jack  CR  Jr, Petersen  RC, Xu  YC,  et al.  Prediction of AD with MRI-based hippocampal volume in mild cognitive impairment. Neurology. 1999;52(7):1397-1403.PubMedArticle
16.
Zhang  S, Han  D, Tan  X, Feng  J, Guo  Y, Ding  Y.  Diagnostic accuracy of 18 F-FDG and 11 C-PIB–PET for prediction of short-term conversion to Alzheimer’s disease in subjects with mild cognitive impairment. Int J Clin Pract. 2012;66(2):185-198.PubMedArticle
17.
Shaw  LM, Vanderstichele  H, Knapik-Czajka  M,  et al; Alzheimer’s Disease Neuroimaging Initiative.  Cerebrospinal fluid biomarker signature in Alzheimer’s Disease Neuroimaging Initiative subjects. Ann Neurol. 2009;65(4):403-413.PubMedArticle
18.
Weiner  MW, Veitch  DP, Aisen  PS,  et al; Alzheimer’s Disease Neuroimaging Initiative.  The Alzheimer’s Disease Neuroimaging Initiative: a review of papers published since its inception. Alzheimers Dement. 2013;9(5):e111-e194.PubMedArticle
19.
Folstein  MF, Folstein  SE, McHugh  PR.  “Mini-mental state”: a practical method for grading the cognitive state of patients for the clinician. J Psychiatr Res. 1975;12(3):189-198.PubMedArticle
20.
Rosen  WG, Mohs  RC, Davis  KL.  A new rating scale for Alzheimer’s disease. Am J Psychiatry. 1984;141(11):1356-1364.PubMedArticle
21.
Morris  JC.  The Clinical Dementia Rating (CDR): current version and scoring rules. Neurology. 1993;43(11):2412-2414.PubMedArticle
22.
Pfeffer  RI, Kurosaki  TT, Harrah  CH  Jr, Chance  JM, Filos  S.  Measurement of functional activities in older adults in the community. J Gerontol. 1982;37(3):323-329.PubMedArticle
23.
Sheikh  JI, Yesavage  JA. Recent evidence and development of a shorter version. In: Brink  TL, ed. Clinical Gerontology: A Guide to Assessment and Intervention. New York, NY: The Haworth Press; 1986:165-173.
24.
Rosen  WG, Terry  RD, Fuld  PA, Katzman  R, Peck  A.  Pathological verification of ischemic score in differentiation of dementias. Ann Neurol. 1980;7(5):486-488.PubMedArticle
25.
Kaufer  DI, Cummings  JL, Ketchel  P,  et al.  Validation of the NPI-Q, a brief clinical form of the Neuropsychiatric Inventory. J Neuropsychiatry Clin Neurosci. 2000;12(2):233-239.PubMedArticle
26.
Robin  X, Turck  N, Hainard  A,  et al.  pROC: an open-source package for R and S+ to analyze and compare ROC curves. BMC Bioinformatics. 2011;12:77.PubMedArticle
27.
DeLong  ER, DeLong  DM, Clarke-Pearson  DL.  Comparing the areas under two or more correlated receiver operating characteristic curves: a nonparametric approach. Biometrics. 1988;44(3):837-845.PubMedArticle
28.
Epskamp  S, Cramer AeOJ, Waldorp LJ, Schmittmann VD, Borsboom D.  qgraph: Network visualizations of relationships in psychometric data. J Stat Softw. 2012;48(4):1-18.Article
29.
Friedman  J, Hastie  T, Tibshirani  R.  Regularization paths for generalized linear models via coordinate descent. J Stat Softw. 2010;33(1):1-22.PubMedArticle
30.
Zou  H, Hastie  T.  Regularization and variable selection via the elastic net. J R Stat Soc B. 2005;67:301-320.Article
31.
Ray  S, Britschgi  M, Herbert  C,  et al.  Classification and prediction of clinical Alzheimer’s diagnosis based on plasma signaling proteins. Nat Med. 2007;13(11):1359-1362.PubMedArticle
32.
Coupé  P, Eskildsen  SF, Manjón  JV,  et al; Alzheimer’s Disease Neuroimaging Initiative.  Scoring by nonlocal image patch estimator for early detection of Alzheimer’s disease. Neuroimage Clin. 2012;1(1):141-152.PubMedArticle
33.
Boulesteix  AL, Slawski  M.  Stability and aggregation of ranked gene lists. Brief Bioinform. 2009;10(5):556-568.PubMedArticle
34.
He  Z, Yu  W.  Stable feature selection for biomarker discovery. Comput Biol Chem. 2010;34(4):215-225.PubMedArticle
35.
Michiels  S, Koscielny  S, Hill  C.  Prediction of cancer outcome with microarrays: a multiple random validation strategy. Lancet. 2005;365(9458):488-492.PubMedArticle
36.
Zucknick  M, Richardson  S, Stronach  EA.  Comparing the characteristics of gene expression profiles derived by univariate and multivariate classification methods. Stat Appl Genet Mol Biol. 2008;7(1):e7. doi:10.2202/1544-6115.1307.PubMed
37.
Corder  EH, Saunders  AM, Strittmatter  WJ,  et al.  Gene dose of apolipoprotein E type 4 allele and the risk of Alzheimer’s disease in late onset families. Science. 1993;261(5123):921-923.PubMedArticle
38.
Trzepacz  PT, Yu  P, Sun  J,  et al; Alzheimer’s Disease Neuroimaging Initiative.  Comparison of neuroimaging modalities for the prediction of conversion from mild cognitive impairment to Alzheimer’s dementia. Neurobiol Aging. 2014;35(1):143-151.PubMedArticle
39.
Fei  M, Jianhua  W.  Apolipoprotein ε4-allele as a significant risk factor for conversion from mild cognitive impairment to Alzheimer’s disease: a meta-analysis of prospective studies. J Mol Neurosci. 2013;50(2):257-263.PubMedArticle
40.
Winblad  B, Palmer  K, Kivipelto  M,  et al.  Mild cognitive impairment—beyond controversies, towards a consensus: report of the International Working Group on Mild Cognitive Impairment. J Intern Med. 2004;256(3):240-246.PubMedArticle
41.
Young  J, Modat  M, Cardoso  MJ, Mendelson  A, Cash  D, Ourselin  S; Alzheimer’s Disease Neuroimaging Initiative.  Accurate multimodal probabilistic prediction of conversion to Alzheimer’s disease in patients with mild cognitive impairment. Neuroimage Clin. 2013;2:735-745.PubMedArticle
42.
Querfurth  HW, LaFerla  FM.  Alzheimer’s disease. N Engl J Med. 2010;362(4):329-344.PubMedArticle
43.
Henneman  WJ, Sluimer  JD, Barnes  J,  et al.  Hippocampal atrophy rates in Alzheimer disease: added value over whole brain volume measures. Neurology. 2009;72(11):999-1007.PubMedArticle
44.
Vemuri  P, Wiste  HJ, Weigand  SD,  et al; Alzheimer’s Disease Neuroimaging Initiative.  MRI and CSF biomarkers in normal, MCI, and AD subjects: predicting future clinical change. Neurology. 2009;73(4):294-301.PubMedArticle
45.
Davatzikos  C, Bhatt  P, Shaw  LM, Batmanghelich  KN, Trojanowski  JQ.  Prediction of MCI to AD conversion, via MRI, CSF biomarkers, and pattern classification. Neurobiol Aging. 2011;32(12):2322.e19-2322.e27.PubMedArticle
46.
Eskildsen  SF, Coupé  P, García-Lorenzo  D, Fonov  V, Pruessner  JC, Collins  DL; Alzheimer’s Disease Neuroimaging Initiative.  Prediction of Alzheimer’s disease in subjects with mild cognitive impairment from the ADNI cohort using patterns of cortical thinning. Neuroimage. 2013;65:511-521.PubMedArticle
47.
Wee  CY, Yap  PT, Shen  D; Alzheimer’s Disease Neuroimaging Initiative.  Prediction of Alzheimer’s disease and mild cognitive impairment using cortical morphological patterns. Hum Brain Mapp. 2013;34(12):3411-3425.PubMedArticle
48.
Wolz  R, Julkunen  V, Koikkalainen  J,  et al; Alzheimer’s Disease Neuroimaging Initiative.  Multi-method analysis of MRI images in early diagnostics of Alzheimer’s disease. PLoS One. 2011;6(10):e25446.PubMedArticle
49.
Llano  DA, Devanarayan  V, Simon  AJ; Alzheimer’s Disease Neuroimaging Initiative (ADNI).  Evaluation of plasma proteomic data for Alzheimer disease state classification and for the prediction of progression from mild cognitive impairment to Alzheimer disease. Alzheimer Dis Assoc Disord. 2013;27(3):233-243.PubMedArticle
50.
Zhao  X, Lejnine  S, Spond  J,  et al.  A candidate plasma protein classifier to identify Alzheimer’s disease. J Alzheimers Dis. 2015;43(2):549-563.PubMed
51.
Hye  A, Riddoch-Contreras  J, Baird  AL,  et al.  Plasma proteins predict conversion to dementia from prodromal disease. Alzheimers Dement. 2014;10(6):799-807.e2.PubMedArticle
52.
Ferreira  D, Perestelo-Pérez  L, Westman  E, Wahlund  LO, Sarría  A, Serrano-Aguilar  P.  Meta-review of CSF core biomarkers in Alzheimer’s disease: the state-of-the-art after the new revised diagnostic criteria. Front Aging Neurosci. 2014;6:47.PubMed
53.
Jaeger  PA, Villeda  SA, Berdnik  D, Britschgi  M, Wyss-Coray  T. Focused plasma proteomics for the study of brain aging and neurodegeneration. In: Coppola  G, ed. The OMICs: Applications in Neuroscience. Oxford, England: Oxford University Press; 2014:183-191.
54.
Olsson  B, Hertze  J, Ohlsson  M,  et al.  Cerebrospinal fluid levels of heart fatty acid binding protein are elevated prodromally in Alzheimer’s disease and vascular dementia. J Alzheimers Dis. 2013;34(3):673-679.PubMed
55.
Britschgi  M, Rufibach  K, Huang  SL,  et al.  Modeling of pathological traits in Alzheimer’s disease based on systemic extracellular signaling proteome. Mol Cell Proteomics. 2011;10(10):008862. doi:10.1074/mcp.M111.008862.PubMedArticle
56.
Csernansky  JG, Dong  H, Fagan  AM,  et al.  Plasma cortisol and progression of dementia in subjects with Alzheimer-type dementia. Am J Psychiatry. 2006;163(12):2164-2169.PubMedArticle
57.
Soares  HD, Potter  WZ, Pickering  E,  et al; Biomarkers Consortium Alzheimer’s Disease Plasma Proteomics Project.  Plasma biomarkers associated with the apolipoprotein E genotype and Alzheimer disease. Arch Neurol. 2012;69(10):1310-1317.PubMedArticle
58.
Toledo  JB, Toledo  E, Weiner  MW,  et al; Alzheimer’s Disease Neuroimaging Initiative.  Cardiovascular risk factors, cortisol, and amyloid-β deposition in Alzheimer’s Disease Neuroimaging Initiative. Alzheimers Dement. 2012;8(6):483-489.PubMedArticle
59.
Martignoni  E, Petraglia  F, Costa  A, Bono  G, Genazzani  AR, Nappi  G.  Dementia of the Alzheimer type and hypothalamus-pituitary-adrenocortical axis: changes in cerebrospinal fluid corticotropin releasing factor and plasma cortisol levels. Acta Neurol Scand. 1990;81(5):452-456.PubMedArticle
60.
Lei  JK.  Change of serum ACTH and cortisol levels in Alzheimer disease and mild cognition impairment [in Chinese]. Zhonghua Yi Xue Za Zhi. 2010;90(41):2894-2896.PubMed
61.
Calero  M, Rostagno  A, Matsubara  E, Zlokovic  B, Frangione  B, Ghiso  J.  Apolipoprotein J (clusterin) and Alzheimer’s disease. Microsc Res Tech. 2000;50(4):305-315.PubMedArticle
62.
Song  F, Poljak  A, Crawford  J,  et al.  Plasma apolipoprotein levels are associated with cognitive status and decline in a community cohort of older individuals. PLoS One. 2012;7(6):e34078.PubMedArticle
63.
Huang  Y, Mahley  RW.  Apolipoprotein E: structure and function in lipid metabolism, neurobiology, and Alzheimer’s diseases. Neurobiol Dis. 2014;72(pt A):3-12.PubMedArticle
64.
Craig-Schapiro  R, Kuhn  M, Xiong  C,  et al.  Multiplexed immunoassay panel identifies novel CSF biomarkers for Alzheimer’s disease diagnosis and prognosis. PLoS One. 2011;6(4):e18850.PubMedArticle
65.
Heppner  FL, Ransohoff  RM, Becher  B.  Immune attack: the role of inflammation in Alzheimer disease. Nat Rev Neurosci. 2015;16(6):358-372.PubMedArticle
×