The AD-CSF-Index Discriminates Alzheimer s Disease Patients from Healthy Controls: A Validation Study

Journal of Alzheimer s Disease 36 (2013) 67 77 DOI 10.3233/JAD-130203 IOS Press 67 The AD-CSF-Index Discriminates Alzheimer s Disease Patients from Healthy Controls: A Validation Study Jose L. Molinuevo a,b,c,1,, Juan Domingo Gispert c,1, Bruno Dubois d, Michael T. Heneka e, Alberto Lleo f, Sebastiaan Engelborghs g, Jesús Pujol h, Leonardo Cruz de Souza d, Daniel Alcolea f, Frank Jessen i, Marie Sarazin d, Foudil Lamari j, Mircea Balasa a, Anna Antonell a,b and Lorena Rami a,b a Alzheimer s Disease and Other Cognitive Disorders Unit, Neurology Service, Hospital Clínic, Barcelona, Spain b Institut d Investigacions Biomèdiques August Pi i Sunyer (IDIBAPS), Spain c BarcelonaBeta Brain Research Centre, Pasqual Maragall Foundation, Barcelona, Spain d Alzheimer Institute, Research and Resource Memory Centre, AP-HP, Pitié-Salpêtrière Hospital, Paris, France e Neurology Department, University of Bonn, Germany f Neurology Department, Hospital Sant Pau, Barcelona, CIBERNED, Spain g Reference Centre for Biological Markers of Dementia (BIODEM), University of Antwerp, Antwerp, Belgium h CRC-Mar, Hospital del Mar, CRC Corporació Sanitària, Barcelona, Spain i Department of Psychiatry, University of Bonn, Bonn, Germany j Department of Metabolic Biochemistry, AP-HP, Pitiá-Salpêtrière Hospital, Paris, France Handling Associate Editor: Piotr Lewczuk Accepted 2 March 2013 Abstract. Background: Cerebrospinal fluid (CSF) biomarkers of Alzheimer s disease (AD) show an acceptable diagnostic sensitivity and specificity; however, their interpretation and ease of use is far from optimal. Objective: To study and validate the diagnostic accuracy of an easy-to-use normalized CSF biomarker index, the AD-CSF-index, in different European populations. Methods: A total of 342 subjects, 103 healthy controls and 239 AD patients, from four European memory clinics were included. The AD-CSF-index was constructed from the addition of normalized values between the minimum and maximum of amyloid and tau protein levels. The diagnostic accuracy, receiver operating characteristic, and regression analysis of the AD-CSF-index and other composite indices were evaluated in this study. Results: AD patients presented a significantly higher AD-CSF-index than healthy subjects (control = 0.5204; AD = 1.2272; p < 0.001). The AD-CSF-index obtained a sensitivity of 88.6% at 85% specificity and also showed a significantly higher diagnostic power (p < 0.05) than the individual CSF biomarkers and other studied indices. The performance of the AD-CSF-index was very similar between ELISA and MesoScale measurements. Cut-off values of approximately 0.75 provided the lowest achievable overall classification errors and a cut-off point of about 0.95 consistently provided specificities above 85%. 1 Both authors contributed equally to this manuscript. Correspondence to: José Luis Molinuevo, M.D., Alzheimer s Disease and Other Cognitive Disorders Unit, Hospital Clínic, Villarroel 170, and Fundació Pasqual Maragall, 08036, Barcelona, Spain. Tel.: +34 932275785; Fax: +34 932275783; E-mail: jlmoli@clinic.ub.es. ISSN 1387-2877/13/$27.50 2013 IOS Press and the authors. All rights reserved

68 J.L. Molinuevo et al. / Validation of the AD-CSF-Index Conclusion: The AD-CSF-index represents a novel approach, combining normalized CSF values, for the biological diagnosis of AD. The AD-CSF-index presents an optimal AUC with high sensitivity and specificity and seems to be a simple and intuitive way to interpret AD CSF biomarker results even from different analytical platforms. Keywords: Alzheimer s disease, biomarkers, cerebrospinal fluid, diagnostic accuracy, sensitivity, specificity INTRODUCTION Alzheimer s disease (AD) has been traditionally conceptualized as a clinico-pathological disease, requiring for its definite diagnosis the presence of a characteristic pathology together with a dementia clinical picture [1]. The main AD pathological lesions are neurofibrillary tangles and senile plaques formed by neuronal accumulations of abnormal tau filaments and extracellular deposits of amyloid fibrils respectively. Cerebrospinal fluid (CSF) studies measuring A 1-42 (A ) and total tau (t-tau) protein levels in AD patients with confirmed pathology have demonstrated that abnormal levels of both biomarkers constitute a specific signature of the underlying AD-related pathology [2, 3]. In that sense, a promising approach to increase diagnostic accuracy for AD is the use of CSF biomarkers, together with specific clinical symptomatology [4, 5]. Therefore, implementation of new criteria that captures both clinical and biological aspects of the disease has shifted the diagnostic paradigm toward a clinico-biological approach allowing an earlier and more specific detection of the disease [6]. However, although CSF biomarkers show an acceptable sensitivity and specificity to diagnose AD [7, 8], tau, phosphorylated tau (p-tau), and A concentrations considerably overlap between controls and patients [9]. To enhance discrimination power, combinations of these biomarkers into a single value have been proposed [10]. Examples of this approach are the ratios of tau or p-tau to A [11, 12] or the AD CSF profile derived by Hulstaert and colleagues [8]. These indices present higher diagnostic power than the individual CSF biomarkers [6, 13, 14] and prognostic capacity to predict cognitive decline in non-demented elderly subjects [15 17]. In that sense, we have recently developed a novel and easy way to calculate a CSF biomarker index that captures the biochemical changes along the AD continuum [18]. The AD-CSF-index is basically the normalization of A, tau, or p-tau values, which are then added into a single value (range from 0 to 2) to express the possibility of having AD with high sensitivity and specificity [18]. AD patients had a significantly higher AD-CSF-index than healthy subjects, which resulted in a sensitivity of 92% and a specificity of 89% in the discrimination of AD patients from controls [18]. Therefore, the main objective of this work is to validate the diagnostic accuracy of a normalized CSF biomarker index, the AD-CSF-index, in different European populations and with different methods of detection for CSF biomarkers. MATERIAL AND METHODS Subjects Subjects were recruited from four experienced European memory clinics from September 2009 to April 2012. All subjects underwent clinical and neuropsychological assessment, lumbar puncture and CSF analysis at the local laboratory. The study was approved by the local ethics committee and all participants gave written informed consent to participate in the study. A total of 341 subjects, 103 controls and 238 AD patients, were included: 78 participants (44 controls, 34 AD patients) were consecutively recruited at the Alzheimer s Ddisease and other Cognitive Disorders Unit, from the Hospital Clinic i Universitari, Barcelona (Site 1); 59 participants (12 controls, 47 AD patients) were consecutively recruited from the Memory Unit, from the Sant Pau Hospital, Barcelona (Site 2); 60 participants (60 AD patients) were consecutively recruited from the Salpetriere (Site 3); 145 participants (47 controls, 98 AD patients) were consecutively recruited from the memory clinic of the University of at Bonn Medical Centre (Site 4). Normal controls were defined as cognitive healthy elderly that presented no cognitive complaints and no evidence of cognitive impairment (below 1.5 standard deviations) in any of the neuropsychological tests administered. They were recruited as research volunteers and thoughtfully studied at each memory clinic. AD patients were diagnosed according to the NINCDS-ADRDA criteria. All patients included were in the mild and moderate stage of the disease. Clinical diagnoses were made independently of the CSF results. The demographic characteristics of the population studied are shown in Table 1.

J.L. Molinuevo et al. / Validation of the AD-CSF-Index 69 Table 1 Demographic characteristics of the studied population Clinic St. Pau Bonn Paris Control AD Control AD Control AD AD mean (SD) mean (SD) mean (SD) mean (SD) mean (SD) mean (SD) mean (SD) Age 65.6 (9.0) 67.6 (9.2) 63.5 (8.1) 70.3 (8.1) 66.7 (11.3) 71.3 (8.5) 65.0 (9.8) Gender (% female) 50% 64.8% 33.3% 66.7% 26.7% 42.3% 51.7% Education (y) 10.7 (4.5) 8.7 (3.4) 11.6 (4.4) 9.5 (4.6) 14.1 (3.5) 11.7 (3.4) 9.8 (5.1) MMSE 28.5 (1.4) 22.7 (4.1) 28.9 (0.9) 21.2 (7.4) 26.1 (5.4) 21.1 (5.4) 20.0 (5.4) CDR 0.06 (0.0) 3.6 (2.0) 0 (0) 0.9 (0.4) n.a. n.a. n.a. APOE % 4 18.2% 53.7% 16.7% 52.1% n.a. n.a. n.a. p < 0.05. SD, standard deviation; n.a., not available; MMSE, Mini-Mental Status Exam; CDR, Clinical Dementia Rating. CSF analyses and AD-CSF-index calculation Subjects underwent lumbar puncture between 9 12 am and 10 ml of CSF was collected in polypropylene tubes. The samples were centrifuged and stored within the first hour after lumbar puncture in polypropylene tubes at 80 C. Levels of A 1-42, t-tau, and p- tau were measured by enzyme-linked immunosorbent assay (ELISA) kits (Innogenetics, Ghent, Belgium) in three centers, Hospital Clinic and Sant Pau Hospital in Barcelona, and the Salpetiere in Paris, and by Meso Scale Discovery electrochemoluminescence detection (Meso Scale Discovery, MSD, Gaithersburg, MD, http://www.mesoscale.com) in Bonn [19]. The AD-CSF-index was calculated as previously published [18]. In summary from each subject s A 1-42, p-tau, and t-tau values, the index was calculated applying the following formula: AD CSF Index (t tau) = Aβ max Aβ 1 42 Aβ max Aβ min + ttau ttau min ttau max ttau min (1) AD CSF Index (p tau)= Aβ max Aβ 1 42 Aβ max Aβ min + ptau ptau min ptau max ptau min (2) where A max, ttau max, and ptau max represent the 95th percentile of the respective values; A min, ttau min, and ptau min represent the 5th percentile of the distribution values; and A 1-42, ttau, and ptau represent the biomarker values for every individual. Derivation of useful and representative minimum and maximum values of the biomarkers was based on the 5th and 95th percentiles of their respective distributions in two different ways: either calculated separately for every site s data or after pooling the different samples (Table 2). Therefore, four different versions of the AD-CSFindex have been generated as a function of whether t-tau and p-tau values were used and whether intra-site (normalized) or inter-site (pooled) 5th and 95th percentiles have been applied to Equation 1. Pooling MesoScaleDiscovery (Bonn) and ELISA (rest of sites) values together could potentially bias the performance of the AD-CSF-index. Therefore, prior to proceeding with the joint analysis of the data from the four sites, a comparison of the performance of the AD-CSF-index between the two techniques was carried out (Table 3). Statistical analysis Statistical analyses were performed with the SPSS package for Windows (V.17.0) and MatLab 7.7.0. The diagnostic power of the AD-CSF-index in the sample was compared to that of the individual CSF biomarkers and to other composite indices which are commonly used like the ratio tau/a 1-42 and the index described by Hulstaert [8], computed as A 1-42 /(240 + 1.18*tau). The diagnostic power of the different biomarkers and progression indices for discriminating AD patients from controls was evaluated by receiving operating characteristic (ROC) analysis and by logistic regression. The ROC analysis consisted of the calculation of the area under the curve (AUC) for all the CSF biomarkers and progression indices. Additionally, the sensitivity at 85% specificity along with the respective cut-off values were also recorded as this parameter is often presented as a quality criterion. For each index, we assessed with logistic regression the percentage of correctly classified cases, the respective cut-off value and the resulting sensitivity and specificity, and the model fit as measured by Nagelkerke s R 2. In logistic regression analysis, we also included two bivariate models formed by simultaneously entering as covariates A 1-42 and t-tau, and A 1-42 and p-tau.

70 J.L. Molinuevo et al. / Validation of the AD-CSF-Index Table 2 5th and 95th percentiles of the CSF biomarkers at the four sites Site n (Control + AD) A 1-42 (pg/ml) t-tau (pg/ml) p-tau (pg/ml) 5% 95% 5% 95% 5% 95% Bonn a 145 (47 + 98) 218.08 1692.27 15.00 1538.40 15.00 133.52 Paris b 59 (0 + 59) 120.00 417.20 175.00 1177.00 33.00 160.00 St. Pau 59 (12 + 47) 148.00 820.00 120.00 1635.50 32.40 126.90 Clínic 78 (44 + 34) 223.92 885.25 127.95 1829.48 28.87 205.50 Pooled 341 (103 + 238) 128.95 1126.22 86.85 1220.23 22.47 157.37 a Values in mesoscale; b Only AD patients included in this sample. In order to improve the generality of the results of the ROC and logistic regressions analyses, we adopted a resampling scheme. Samples were randomly split: one half was used to derive the corresponding minimum and maximum values for the AD-CSF-index as previously described, and the other half was used to perform the analyses. As this process provides a dispersion of results depending on which cases fall into each split, the process was repeated 1000 times. Median values are reported, as well as 95% confidence intervals (95% CI), calculated as the 5th and 95th percentiles of the resampling distributions. Statistical significance between two particular tests is established if the respective 95% CIs do not overlap. This criterion can be considered as the resampling equivalent to a non-parametric test for the comparison of the means of two independent samples with a statistical significance threshold of p < 0.05. Between AD patients and controls, the effect size achieved by the different biomarkers and progression indices was assessed by calculating their respective Cohen s d and Student s t values. Higher absolute Cohen s d and Student s t values are indicative of higher effect sizes and, thus, higher statistical power in between-group analyses. Additionally in the control group, a correlation analysis was performed between the CSF biomarkers and progression indices versus Age as the main risk factor for developing AD. Pearson s r was computed for all CSF biomarkers and composite indices. RESULTS Descriptive statistics Mean and standard deviations of the quantifications of the CSF biomarkers for the four different sites are shown in Table 4 and Fig. 1. The investigation of statistically significant differences among these values falls beyond the scope of this article and are only shown for descriptive purposes. As expected, A 1-42 values of the Bonn site were higher than those of the other sites for controls and AD cases. This result parallels a previous analysis, where the detection by the electrochemoluminescence based MesoScaleDiscovery system showed a tendency toward higher levels as compared to ELISA detection [19]. While the exact reason for this phenomenon remains unclear, one may hypothesize that the electrochemoluminescence detection may be more sensitive than an ELISA based detection. Four different versions of the AD-CSF-index were calculated, as indicated in Equation 1, by using t- tau or p-tau and either the site-normalized or pooled minimum and maximum values in Table 2. The performance of the AD-CSF-index was independent of the method for quantification of the CSF biomarkers. We compared the performance of the indices between the site using MesoScaleDiscovery (Bonn) and the other sites using ELISA. Qualitatively, both methods for the quantification of the CSF biomarkers achieved a very similar performance (Table 3). These results support the appropriateness of analyzing pooled ELISA and MesoScaleDiscovery derived data for a unified analysis in the sample. Comparison of discrimination power The ROC analysis (Fig. 3) showed that the normalized versions of the AD-CSF-index clearly presented the best AUC values among all tested diagnostic indices. Indeed their AUC was significantly higher (p < 0.05) than that of A 1-42 and p-tau biomarkers (Table 5). Regarding the comparison between intra-site normalized and inter-site pooled versions of the AD- CSF-index, the first performed better than the pooled versions, as expected. However, these differences did not reach statistical significance. Nevertheless, the latter showed a similar performance than tau over A 1-42 ratios or the Hulstaert indices (Table 5). Again, no statistically significant differences could be detected between them. The logistic regression analysis confirmed this pattern as the normalized versions of the AD-CSF-index

J.L. Molinuevo et al. / Validation of the AD-CSF-Index 71 Table 3 Comparison of the performance of ELISA and MesoScale derived AD-CSF-indices calculated using 5th and 95th percentiles of the pooled and individual distribution values Index Technique ROC Analysis Logistic regression AUC Sens @ 85% Spec [%] Cut-off corr class [%] Model Fit Cut-off Max/min values derived from pooled distribution values (Max: 95th percentile; Min: 5th percentile) AD-CSF-index (t-tau) MesoScale 0.939 93.8 0.5066 89.1 0.690 0.4302 ELISA 0.949 94.2 0.8804 90.4 0.731 0.9159 AD-CSF-index (p-tau) MesoScale 0.929 89.7 0.4750 87.0 0.674 0.3748 ELISA 0.939 94.7 0.8943 89.3 0.582 0.9193 Max/min values derived from respective distribution values (Max: 95th percentile; Min: 5th percentile) AD-CSF-index (t-tau) MesoScale 0.938 94.8 0.7788 89.9 0.6915 0.7379 ELISA 0.953 96.2 0.7701 90.1 0.7482 0.8041 AD-CSF-index (p-tau) MesoScale 0.935 90.7 0.8141 89.9 0.6884 0.7861 ELISA 0.943 95.2 0.7477 88.7 0.6917 0.7804 1500.00 Diagnosis AD CON 1000.00 Errorbars indicate Standard Deviations Abeta 500.00 0.00 bonn clinic paris stpau 1500.00 150.00 1000.00 100.00 ttau 500.00 ptau 50.00 0.00 0.00 bonn clinic paris stpau bonn clinic paris stpau Fig. 1. Mean (SD) values of the CSF biomarkers (in pg/ml) obtained at the four different sites. presented the highest percentage of correctly classified cases and the best model fits and performed significantly better than all the individual biomarkers and the Hulstaert (p-tau). In addition, the AD-CSFindex (t-tau) did perform significantly better than the p-tau/a 1-42 ratio. With regards to the rest of the

72 J.L. Molinuevo et al. / Validation of the AD-CSF-Index Fig. 2. Distribution of the A 1-42 and p-tau values and coded by referring site and diagnosis. Table 4 Mean (SD) values of the CSF biomarkers obtained at the four different sites Bonn a Paris St. Pau Clínic mean (SD) mean (SD) mean (SD) mean (SD) A 1-42 (pg/ml) AD 592.06 (285.47) 246.90 (96.63) 347.91 (173.74) 315.85 (66.96) Control 1170.72 (470.50) 645.04 (221.80) 585.96 (224.32) t-tau (pg/ml) AD 754.32 (604.66) 628.37 (311.99) 689.60 (360.77) 863.05 (467.33) Control 219.90 (242.29) 184.38 (53.48) 371.78 (325.01) p-tau (pg/ml) AD 76.94 (33.31) 88.77 (36.23) 85.82 (29.55) 114.93 (49.56) Control 39.27 (23.49) 44.73 (10.14) 64.93 (34.95) a Values in mesoscale. SD, standard deviation. indices, the pooled AD-CSF-index (t-tau) and the Hulstaert (t-tau) achieved similar results and significantly outperformed the individual CSF values. Finally, the pooled AD-CSF-index (p-tau) showed a statistically significant better performance than A 1-42 and p-tau, but not t-tau. In terms of overall classification accuracy, the optimal cutoff points were again very stable, as they ranged from 0.73 to 0.90 for the four versions of the AD-CSF-index (Table 6). The effect size measurements for discriminating between control and AD groups are shown in the Supplementary Table 1 (available online: http:// dx.doi.org/10.3233/jad-130203). All four versions of the index were significantly (p < 0.001) higher in

J.L. Molinuevo et al. / Validation of the AD-CSF-Index 73 Fig. 3. ROC curves for all tested AD progression indices. Please refer to Table 5 for the statistical performance of the different indices. Table 5 Diagnostic power of the AD progression indices based on CSF biomarkers as studied by ROC analysis Index AUC (CI 95%) Sensitivity at 85% specificity [%] Cutoff Value for 85% specificity A 1-42 0.858 (0.825 0.893) 74.2 468.8 p-tau 0.832 (0.796 0.870) 72.2 62.90 t-tau 0.862 (0.830 0.897) 80.3 355.0 p-tau/a b 1-42 0.910 (0.885 0.938) 87.2 0.0987 t-tau/a a,b,c 1-42 0.925 (0.900 0.950) 90.4 0.5913 Hulstaert (p-tau) 0.888 (0.856 0.920) 83.2 1.651 Hulstaert (t-tau) a,b,c 0.934 (0.911 0.957) 90.8 0.8368 AD-CSF-index pooled (p-tau) b 0.901 (0.872 0.933) 86.7 0.9256 AD-CSF-index pooled (t-tau) a,b,c 0.923 (0.897 0.949) 90.1 0.9088 AD-CSF-index (p-tau) a,b,c,d 0.945 (0.922 0.966) 94.1 0.7452 AD-CSF-index (t-tau) a,b,c,d 0.943 (0.920 0.967) 94.7 0.7807 a AUC significantly higher (p < 0.05) than A 1-42 ; b AUC significantly higher (p < 0.05) than p-tau; c AUC significantly higher (p < 0.05) than t-tau; d AUC significantly higher (p < 0.05) than Hulstaert (p-tau). AD patients than in controls (pooled (p-tau): control = 0.6544, AD = 1.2870; pooled (t-tau): control = 0.6083, AD = 1.2682; normalized (p-tau): control = 0.5845, AD = 1.2822; normalized (t-tau): control = 0.5204, AD = 1.2272) as well as the rest of progression indices and CSF biomarkers except for the A 1-42 values and Hulstaert indices which were significantly lower. However, the AD-CSF-index presented with the highest effect sizes of all the studied indices. The AD-CSF-index values also presented the strongest association with the age of controls among all tested methods: AD-CSF-index (p-tau) r = 0.375 (p < 0.001) and AD-CSF-index (t-tau) r = 0.339 (p < 0.001) (Supplementary Table 2). DISCUSSION The results from this study show that the AD- CSF-index represents a novel approach, combining normalized CSF determined A and t-tau or p- tau values, for the biological diagnosis of AD. The AD-CSF-index presents an optimal AUC with high sensitivity and specificity and seems to be a simple and intuitive way to interpret AD CSF biomarkers results. One of the limitations of CSF biomarkers is that the values obtained in one laboratory may be different from those obtained in another lab. Even when using the same assay, considerable variability in absolute concentrations of AD biomarkers has been found

74 J.L. Molinuevo et al. / Validation of the AD-CSF-Index Table 6 Diagnostic power and cut-off values of the AD progression indices based on CSF biomarkers as determined by logistic regression Index Cases Model fit Cut-off Sensitivity Specificity correctly classified [Nagelkerke s R 2 ] value at cut-off at cut-off [%] (CI 95%) (CI 95%) value [%] value [%] A 1-42 79.22 (74.51 82.94) 0.4239 (0.3395 0.5190) 625.6 87.0 67.0 p-tau 78.04 (74.12 81.37) 0.3688 (0.2718 0.4826) 53.12 81.6 72.3 t-tau 81.96 (78.43 85.10) 0.4428 (0.3452 0.5541) 346.8 81.6 83.3 A 1-42 & p-tau (bivariate model) 81.98 (78.54 85.19) 0.5759 (0.4911 0.6752) 88.3 80.0 A 1-42 & t-tau (bivariate model) b 85.24 (81.58 88.74) 0.6335 (0.5531 0.7221) 88.0 82.5 p-tau/ A 1-42 84.71 (81.18 88.24) 0.5553 (0.4802 0.6621) 0.1141 83.7 86.8 t-tau/ A 1-42b 85.49 (82.35 89.02) 0.5968 (0.5050 0.7096) 0.7745 84.1 87.9 Hulstaert (p-tau) 83.14 (79.61 86.47) 0.5249 (0.4404 0.6208) 1.982 88.0 75.9 Hulstaert (t-tau) a,b,c,d 88.24 (85.49 91.18) 0.6910 (0.6212 0.7640) 0.958 92.9 81.4 AD-CSF-index pooled (p-tau) a,b 86.27 (83.14 89.02) 0.5772 (0.4986 0.6596) 0.9006 88.3 82.8 AD-CSF-index pooled (t-tau) a,b,c 88.24 (85.10 90.98) 0.6335 (0.5530 0.7261) 0.9012 90.6 84.3 AD-CSF-index (p-tau) a,b,c,d,e 90.20 (87.45 90.98) 0.7094 (0.6311 0.8013) 0.7333 93.9 84.1 AD-CSF-index (t-tau) a,b,c,d,e 90.59 (88.04 93.33) 0.7280 (0.6492 0.8019) 0.7996 94.2 85.9 a Percentage of cases correctly classified significantly higher (p < 0.05) than A 1-42 ; b Percentage of cases correctly classified significantly higher (p < 0.05) than p-tau; c Percentage of cases correctly classified significantly higher (p < 0.05) than t-tau; d Percentage of cases correctly classified significantly higher (p < 0.05) than A 1-42 & p-tau (bivariate model); e Percentage of cases correctly classified significantly higher (p < 0.05) than Hulstaert (p-tau). due to pre-analytical and analytical factors [19]. Furthermore, several studies have shown that the CSF results obtained from different analytical platforms provide very different values [14, 20]. The lack of standardized cut-off values for the diagnosis of AD hampers widespread application of CSF biomarkers for diagnostic purposes in the clinical setting [21]. The development of an index that normalizes the CSF biomarker results among different techniques, laboratories, or populations and that simultaneously captures the pathophysiological relevant changes of CSF A and tau levels may be a useful start for harmonizing AD CSF biomarker interpretation. In this study, we demonstrated that the AD-CSFindex resulted in similar AUC and diagnostic accuracy values although biomarker data were obtained through ELISA and electrochemoluminescence detection. Although the raw A values coming from the ELISA and mesoscale platform were significantly different, the AD-CSF-index was able to normalize them. Its performance with the luminex X-MAP platform is still to be determined. The AD-CSF-index constitutes a valuable alternative to other indices or diagnostic approximations previously described [8, 22 24]. The AD-CSF-index yields a higher diagnostic power and provides stable cut-off points associated with particular sensitivity/specificity values. With respect to diagnostic power, the normalized versions of the AD-CSF-index and particularly the version implementing p-tau values, showed significantly higher diagnostic power than the individual biomarkers or the rest of studied indices. Even the pooled versions were among the best of all the tested indices in our comparisons. Regarding the stability of the cut-off points, although far from reaching a common cut-off point, the AD-CSF-index performed consistently, since cutoff values of approximately 0.75 provided the lowest achievable overall classification errors and cut-off points of about 0.95, specificities around 85%. Whenever pre-analytical and analytical conditions will be fully harmonized, the AD-CSF-index could be a simple way to establish a cut-off point in order to differentiate controls from AD patients. In that sense, there is an urgent need to standardize pre-analytical and analytical factors and to provide reference materials, so that analytical methods and platforms would be harmonized. Another advantage is that the AD-CSF-index adapts to specific samples by discarding the lowest and highest values into them. In any case, these discarded values are less reliable as, by definition, they tend to fall near the limits of the linear segment of the calibration curves of the biomarker quantification techniques. With respect to other indices [8, 15], the normalization strategy adopted in the AD-CSF-index makes it less sensitive to changes in the absolute values of the biomarkers which are known to significantly vary across quantification techniques. Even in applications in which only a few samples are available, our validation shows that the AD-CSF-index is fairly insensitive to changes in the minimum and maximum values

J.L. Molinuevo et al. / Validation of the AD-CSF-Index 75 applied into Equation 1 [18]. Moreover, Table 1 provides reference values that can be used for research groups using MesoScale or ELISA CSF determination methods. Another possibility in order to create a formula for generalized use consists of selecting the minimum and maximum quantifiable values from the calibration curve provided by the quantification method or by the laboratory performing the analysis. These values are routinely estimated as a part of the Good Laboratory Practices mandatory for this kind of analytic procedures. Its ease of use together with the excellent AUC suggests that the AD-CSF-index may help to offer CSF analysis its optimal biomarker value [25]. From a pathophysiological perspective, the AD- CSF-index represents a simple way of determining the progression of the biological continuum of AD [26]. Currently, the pathophysiological process of AD and its clinical symptomatology are best conceptualized as a continuum from completely asymptomatic individuals with biomarker evidence suggestive of AD to predementia or prodromal AD patients and ending with a full blown dementia picture [27 29]. In that sense, since the AD-CSF-index is calculated through the addition of normalized CSF determined A and t-tau or p-tau values, it may be able to capture the two main CSF biochemical changes that characterize the AD continuum [26 29]. Actually, the potential of the AD-CSF-index as a descriptor of the AD continuum for research purposes is supported by our results, as they reached the highest effect sizes between AD patients and controls and, in the latter group, the tightest association to age. Therefore, this approximation may be a simple and useful way not only to diagnose the patient but also to determine where in the continuum the patient is located. Nowadays, there is also increasing interest in determining the shape of the trajectories together with the relationship existing among the major AD biomarkers [30]. Since the AD-CSF-index captures the variations of both amyloid and tau, it could be possible to correlate it with the changes observed with other structural markers, such as MRI. In that sense, the AD-CSF-index could be used as a longitudinal, biological, and chronological marker of the CSF alterations ranging from the preclinical stage to mild dementia. Nevertheless, further studies describing the behavior of the AD-CSF-index versus neuropsychological tests and neuroimaging techniques are required to fully confirm this conclusion. Since the publication of the new AD diagnostic criteria in 2007 [4], there has been a lot of discussion on its applicability in a clinical setting [5, 31, 32]. One of the argued drawbacks is the reliability of the different markers [33] and its capacity to differentiate cognitively-preserved elderly from AD patients. It is currently well known that cognitively-preserved elderly may show signs of brain amyloidosis, detected through CSF [34] or A PET imaging [35]. Although its meaning is under debate [36, 37], it is clear that a positive AD CSF profile in order to detect AD with high specificity should include both amyloid and a neuronal injury marker, such as t-tau. This is of remarkable importance if we want to increase the specificity of the diagnosis, and that is the reason for the greater specificity of the AD-CSF-index as compared to A. Furthermore, the A levels by themselves are of little value to determine the proximity of dementia onset, since they may be fully changed even 10 years before dementia onset [38]. We are aware that the study presents several limitations. The study has a cross-sectional design, so its predictive capacity in people with isolated memory or other cognitive deficits that may progress to dementia cannot be addressed. We have also limited the study to two of the existing analytical platforms, so its behavior with Luminex X-MAP driven data is unknown. Further studies should address these questions, together with the association of the index with existing structural changes along the continuum. In summary, the AD-CSF-index represents a novel approximation, combining normalized CSF determined A and t-tau or p-tau values, for the biological diagnosis of AD. The AD-CSF-index presents an optimal AUC with high sensitivity and specificity and seems to be a simple and intuitive way to interpret AD CSF biomarkers results. ACKNOWLEDGMENTS Consolider-Ingenio 2010 (CSD 2010-00045 PI: José L Molinuevo); This publication is part of the BIOMARKAPD project within the EU Joint Programme for Neurodegenerative Diseases (JPND) funded by the ISCIII. FIS-Fondo europeo de desarrollo regional, una manera de hacer Europa (PI11/01071 IP: Lorena Rami); proyecto IMSERSO (197/2011 IPD: Lorena Rami); Mapfre company grant (IP: Lorena Rami); programa de investigadores del sistema nacional Miguel Servet (CP 08/00147; IP: Lorena Rami) and the economic support of VALLVERD, S.A Authors disclosures available online (http://www.jalz.com/disclosures/view.php?id=1702).

76 J.L. Molinuevo et al. / Validation of the AD-CSF-Index SUPPLEMENTARY MATERIAL Supplementary material can be found here: http:// dx.doi.org/10.3233/jad-130203 REFERENCES [1] McKhann GM, Knopman DS, Chertkow H, Hyman BT, Jack CR Jr., Kawas CH, Klunk WE, Koroshetz WJ, Manly JJ, Mayeux R, Mohs RC, Morris JC, Rossor MN, Scheltens P, Carrillo MC, Thies B, Weintraub S, Phelps CH (2011) The diagnosis of dementia 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 7, 263-269. [2] Shaw LM, Vanderstichele H, Knapik-Czajka M, Clark CM, Aisen PS, Petersen RC, Blennow K, Soares H, Simon A, Lewczuk P, Dean R, Siemers E, Potter W, Lee VM, Trojanowski JQ; Alzheimer s Disease Neuroimaging, Initiative (2009) Alzheimer s Disease Neuroimaging Initiative. Cerebrospinal fluid biomarker signature in Alzheimer s disease neuroimaging initiative subjects. Ann Neurol 65, 403-413. [3] Engelborghs S, De Vreese K, Van de Casteele T, Vanderstichele H, Van Everbroeck B, Cras P, Martin JJ, Vanmechelen E, De Deyn PP (2008) Diagnostic performance of a CSFbiomarker panel in autopsy-confirmed dementia. Neurobiol Aging 29, 1143-1159. [4] Dubois B, Feldman HH, Jacova C, Dekosky ST, Barberger- Gateau P, Cummings J, Delacourte A, Galasko D, Gauthier S, Jicha G, Meguro K, O brien J, Pasquier F, Robert P, Rossor M, Salloway S, Stern Y, Visser PJ, Scheltens P (2007) Research criteria for the diagnosis of Alzheimer s disease: Revising the NINCDS-ADRDA criteria. Lancet Neurol 6, 734-746. [5] Rami L, Solé-Padullés C, Fortea J, Bosch B, Lladó A, Antonell A, Olives J, Castellví M, Bartres-Faz D, Sánchez- Valle R, Molinuevo JL (2012) Applying the new research diagnostic criteria: MRI findings and neuropsychological correlations of prodromal AD. Int J Geriatr Psychiatry 27, 127-134. [6] Shaw LM, Vanderstichele H, Knapik-Czajka M, Clark CM, Aisen PS, Petersen RC, Blennow K, Soares H, Simon A, Lewczuk P, Dean R, Siemers E, Potter W, Lee VM, Trojanowski JQ; Alzheimer s Disease Neuroimaging Initiative. (2009) Cerebrospinal fluid biomarker signature in Alzheimer s disease neuroimaging initiative subjects. Ann Neurol 65, 403 413, [7] Sunderland T, Linker G, Mirza N, Putnam KT, Friedman DL, Kimmel LH, Bergeson J, Manetti GJ, Zimmermann M, Tang B, Bartko JJ, Cohen RM (2003) Decreased -amyloid1-42 and increased tau levels in cerebrospinal fluid of patients with Alzheimer disease. J Am Med Assoc 289, 2094-2103. [8] Hulstaert F, Blennow K, Ivanoiu A, Schoonderwaldt HC, Riemenschneider M, De Deyn PP, Bancher C, Cras P, Wiltfang J, Mehta PD, Iqbal K, Pottel H, Vanmechelen E, Vanderstichele H (1999) Improved discrimination of AD patients using beta-amyloid(1-42) and tau levels in CSF. Neurology 52, 1555-1562. [9] Sunderland T, Linker G, Mirza N, Putnam KT, Friedman DL, Kimmel LH, Bergeson J, Manetti GJ, Zimmermann M, Tang B, Bartko JJ, Cohen RM (2003) Decreased beta-amyloid1-42 and increased tau levels in cerebrospinal fluid of patients with Alzheimer disease. JAMA 289, 2094-2103. [10] Blennow K, Vanmechelen E (1998) Combination of the different biological markers for increasing specificity of in vivo Alzheimer s testing. J Neural Transm Suppl 53, 223-235. [11] Shoji M, Matsubara E, Kanai M, Watanabe M, Nakamura T, Tomidokoro Y, Shizuka M, Wakabayashi K, Igeta Y, Ikeda Y, Mizushima K, Amari M, Ishiguro K, Kawarabayashi T, Harigaya Y, Okamoto K, Hirai S (1998) Combination assay of CSF tau, A beta 1-40 and A beta 1-42(43) as a biochemical marker of Alzheimer s disease. J Neurol Sci 158, 134-140. [12] Maddalena A, Papassotiropoulos A, Müller-Tillmanns B, Jung HH, Hegi T, Nitsch RM, Hock C (2003) Biochemical diagnosis of Alzheimer disease by measuring the cerebrospinal fluid ratio of phosphorylated tau protein to betaamyloid peptide42. Arch Neurol 60, 1202-1206. [13] Riemenschneider M, Lautenschlager N, Wagenpfeil S, Diehl J, Drzezga A, Kurz A (2002) Cerebrospinal fluid tau and betaamyloid 42 proteins identify Alzheimer disease in subjects with mild cognitive impairment. Arch Neurol 59, 1729-1734. [14] Fagan AM, Shaw LM, Xiong C, Vanderstichele H, Mintun MA, Trojanowski JQ, Coart E, Morris JC, Holtzman DM (2011) Comparison of analytical platforms for cerebrospinal fluid measures of -amyloid 1-42, total tau, and p-tau181 for identifying Alzheimer disease amyloid plaque pathology. Arch Neurol 68, 1137-1144. [15] Fagan AM, Roe CM, Xiong C, Mintun MA, Morris JC, Holtzman DM (2007) Cerebrospinal fluid tau/beta-amyloid(42) ratio as a prediction of cognitive decline in nondemented older adults. Arch Neurol 64, 343-349. [16] Visser PJ, Verhey F, Knol DL, Mintun MA, Morris JC, Holtzman DM (2009) Prevalence and prognostic value of CSF markers of Alzheimer s disease pathology in patients with subjective cognitive impairment or mild cognitive impairment in the DESCRIPA study: A prospective cohort study. Lancet Neurol 8, 619-627. [17] Craig-Schapiro R, Kuhn M, Xiong C, Pickering EH, Liu J, Misko TP, Perrin RJ, Bales KR, Soares H, Fagan AM, Holtzman DM (2011) Multiplexed immunoassay panel identifies novel CSF biomarkers for Alzheimer s disease diagnosis and prognosis. PLoS One 6, e18850. [18] Molinuevo JL, Gispert JD, Pujol J, Rojas S, Lladó A, Balasa M, Antonell A, Sánchez-Valle R, Rami L (2012) A new approach to the Alzheimer s disease diagnosis with biomarkers: Description of the AD-CSF-index. Rev Neurol 54, 513-522. [19] Mattsson N, Andreasson U, Persson S, Arai H, Batish SD, Bernardini S, Bocchio-Chiavetto L, Blankenstein MA, Carrillo MC, Chalbot S, Coart E, Chiasserini D, Cutler N, Dahlfors G, Duller S, Fagan AM, Forlenza O, Frisoni GB, Galasko D, Galimberti D, Hampel H, Handberg A, Heneka MT, Herskovits AZ, Herukka SK, Holtzman DM, Humpel C, Hyman BT, Iqbal K, Jucker M, Kaeser SA, Kaiser E, Kapaki E, Kidd D, Klivenyi P, Knudsen CS, Kummer MP, Lui J, Lladó A, Lewczuk P, Li QX, Martins R, Masters C, McAuliffe J, Mercken M, Moghekar A, Molinuevo JL, Montine TJ, Nowatzke W, O Brien R, Otto M, Paraskevas GP, Parnetti L, Petersen RC, Prvulovic D, de Reus HP, Rissman RA, Scarpini E, Stefani A, Soininen H, Schröder J, Shaw LM, Skinningsrud A, Skrogstad B, Spreer A, Talib L, Teunissen C, Trojanowski JQ, Tumani H, Umek RM, Van Broeck B, Vanderstichele H, Vecsei L, Verbeek MM, Windisch M, Zhang J, Zetterberg H, Blennow K (2011) The Alzheimer s Association external quality control program for cerebrospinal fluid biomarkers. Alzheimers Dement 7, 386-395.e6. [20] Schipke CG, Prokop S, Heppner FL, Heuser I, Peters O (2011) Comparison of immunosorbent assays for the quantification

J.L. Molinuevo et al. / Validation of the AD-CSF-Index 77 of biomarkers for Alzheimer s disease in human cerebrospinal fluid. Dement Geriatr Cogn Disord 31, 139-145. [21] Hort J, O Brien JT, Gainotti G, Pirttila T, Popescu BO, Rektorova I, Sorbi S, Scheltens P, Scientist Panel on EFNS, Dementia (2010) EFNS guidelines for the diagnosis and management of Alzheimer s disease. Eur J Neurol 17, 1236-1248. [22] Galasko D, Chang L, Motter R, Clark CM, Kaye J, Knopman D, Thomas R, Kholodenko D, Schenk D, Lieberburg I, Miller B, Green R, Basherad R, Kertiles L, Boss MA, Seubert P (1998) High cerebrospinal fluid tau and low amyloid beta42 levels in the clinical diagnosis of Alzheimer disease and relation to apolipoprotein E genotype. Arch Neurol 55, 937-945. [23] Lewczuk P, Zimmermann R, Wiltfang J, Kornhuber J (2009) Neurochemical dementiadiagnostics: A simple algorithm for interpretation of the CSF biomarkers. J Neural Transm 116, 1163-1167. [24] Li G, Sokal I, Quinn JF, Leverenz JB, Brodey M, Schellenberg GD, Kaye JA, Raskind MA, Zhang J, Peskind ER, Montine TJ (2007) CSF tau/abeta42 ratio for increased risk of mild cognitive impairment: A follow-up study. Neurology 69, 631-639. [25] The Ronald and Nancy Reagan Research Institute of the Alzheimer s Association and the National Institute on Aging Working Group (1998) Consensus report of the Working Group on: Molecular and Biochemical Markers of Alzheimer s Disease. Neurobiol Aging 19, 109-116. [26] Jack CR Jr., Knopman DS, Jagust WJ, Shaw LM, Aisen PS, Weiner MW, Petersen RC, Trojanowski JQ (2010) Hypothetical model of dynamic biomarkers of the Alzheimer s pathological cascade. Lancet Neurol 9, 119-128. [27] Dubois B, Feldman HH, Jacova C, Cummings JL, Dekosky ST, Barberger-Gateau P, Delacourte A, Frisoni G, Fox NC, Galasko D, Gauthier S, Hampel H, Jicha GA, Meguro K, O Brien J, Pasquier F, Robert P, Rossor M, Salloway S, Sarazin M, de Souza LC, Stern Y, Visser PJ, Scheltens P (2010) Revising the definition of Alzheimer s disease: A new lexicon. Lancet Neurol 9, 1118-1127. [28] Sperling RA, Aisen PS, Beckett LA, Bennett DA, Craft S, Fagan AM, Iwatsubo T, Jack CR Jr, Kaye J, Montine TJ, Park DC, Reiman EM, Rowe CC, Siemers E, Stern Y, Yaffe K, Carrillo MC, Thies B, Morrison-Bogorad M, Wagster MV, Phelps CH (2011) 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 7, 280-292. [29] Rami L, Fortea J, Bosch B, Solé-Padullés C, Lladó A, Iranzo A, Sánchez-Valle R, Molinuevo JL (2011) Cerebrospinal fluid biomarkers and memory present distinct associations along the continuum from healthy subjects to AD patients. J Alzheimers Dis 23, 319-326. [30] Jack CR Jr., Vemuri P, Wiste HJ, Weigand SD, Lesnick TG, Lowe V, Kantarci K, Bernstein MA, Senjem ML, Gunter JL, Boeve BF, Trojanowski JQ, Shaw LM, Aisen PS, Weiner MW, Petersen RC, Knopman DS; Alzheimer s Disease Neuroimaging Initiative (2012) Shapes of the trajectories of 5 major biomarkers of Alzheimer disease. Arch Neurol 69, 856-867. [31] Bouwman FH, Verwey NA, Klein M, Kok A, Blankenstein MA, Sluimer JD, Barkhof F, van der Flier WM, Scheltens P (2010) New research criteria for the diagnosis of Alzheimer s disease applied in a memory clinic population. Dement Geriatr Cogn Disord 30, 1-7. [32] Galluzzi S, Ghidoni R, Paghera B, Amicucci G, Bonetti M, Zanetti O, Cotelli M, Gennarelli M, Frisoni GB; Translational Outpatient Memory Clinic Working Group (2010) The new Alzheimer s criteria in a naturalistic series of patients with mild cognitive impairment. J Neurol 257, 2004-2014. [33] Foster NL (2007) A new framework for the diagnosis of Alzheimer s disease. Lancet Neurol 6, 667-669. [34] Antonell A, Fortea J, Rami L, Bosch B, Balasa M, Sánchez- Valle R, Iranzo A, Molinuevo JL, Lladó A (2011) Different profiles of Alzheimer s disease cerebrospinal fluid biomarkers in controls and subjects with subjective memory complaints. J Neural Transm 118, 259-262. [35] Mintun MA, Larossa GN, Sheline YI, Dence CS, Lee SY, Mach RH, Klunk WE, Mathis CA, DeKosky ST, Morris JC (2006) [11C]PIB in nondemented population: Potential antecedent marker of Alzheimer disease. Neurology 67, 446-452. [36] Hulette CM, Welsh-Bohmer KA, Murray MG, Saunders AM, Mash DC, McIntyre LM (1998) Neuropathological and neuropsychological changes in normal aging: Evidence for preclinical Alzheimer disease in cognitively normal individuals. J Neuropathol Exp Neurol 57, 1168-1174. [37] Molinuevo JL, Sánchez-Valle R, Lladó A, Fortea J, Bartrés- Faz D, Rami L (2011) Identifying earlier Alzheimer s disease: Insights from the preclinical and prodromal phases. Neurodegener Dis 10, 158-160. [38] Buchhave P, Minthon L, Zetterberg H, Wallin AK, Blennow K, Hansson O (2012) Cerebrospinal fluid levels of -amyloid 1-42, but not of tau, are fully changed already 5 to 10 years before the onset of Alzheimer dementia. Arch Gen Psychiatry 69, 98-106.