Cerebrospinal Fluid β-Amyloid1–42 Levels in the Differential Diagnosis of Alzheimer’s Disease—Systematic Review and Meta-Analysis

Objectives The purpose of this study was to carry out systematic review of the literature and meta-analysis to evaluate the diagnostic utility of cerebrospinal fluid (CSF) levels of the 42 amino acid form of amyloid-beta (Aβ1–42) as a biomarker for differentiating Alzheimer’s disease (AD) from non-AD dementia. Methods Design. Systematic literature review was used to evaluate the effectiveness of the Aβ for the diagnosis of AD. The Scottish Intercollegiate Guidelines Network (SIGN) tool was used to evaluate independently the quality of the studies. Data sources. The literature review covered from January 1, 2004, to October 22, 2013, and searched eight domestic databases including Korea Med and international databases including Ovid-MEDLINE, EMBASE, and Cochrane Library. Data Extraction and Synthesis. Primary criteria for inclusion were valid studies on (i) patients with mild cognitive impairment with confirmed or suspected AD and non-AD dementia, and (ii) assessment of Aβ1–42 levels using appropriate comparative tests. Results A total of 17 diagnostic evaluation studies were identified in which levels of CSF Aβ1–42 were assessed. Meta-analysis was performed on 11 robust studies that compared confirmed AD (n = 2211) with healthy individuals (n = 1030), 10 studies that compared AD with non-AD dementias (n = 627), and 5 studies that compared amnestic mild cognitive impairment (n = 1133) with non-amnestic type subjects (n = 1276). Overall, the CSF Aβ1–42 levels were reduced in AD compared to controls or non-AD dementia. The effectiveness of test was evaluated for diagnostic accuracy (pooled sensitivity, 0.80 (95% CI 0.78–0.82); pooled specificity, 0.76 (95% CI 0.74–0.78). Conclusions Reduced CSF Aβ1–42 levels are of potential utility in the differential diagnosis of AD versus non-AD dementias and controls. Diagnostic accuracy was high in AD versus healthy controls. However, differential diagnosis for MCI or non-AD might be evaluated by other biomarkers.


Introduction
A substantial proportion of current therapeutic development in AD focuses on therapies targeting the Aβ peptide or Aβ aggregates, the core pathology of AD [1,2]. However, large-scale clinical trials of Aβ removal by immunological or pharmacologic means have yielded no reproducible benefits [2]. There are two routes to resolve this dilemma. First, anti-Aβ therapies (and perhaps anti-tau therapies) might be conducted on minimally affected individuals (secondary prevention in stages 1/2). A second strategy is to develop therapies that are likely to be of benefit in symptomatic patients (i.e., in a preclinical stage 3 or prodromal AD) [2]. Therefore, further development of AD therapeutics will require the establishment of biomarkers that accurately reflect the progression of AD pathology, thereby permitting early diagnosis of AD and facilitating drug trials selectively targeting the early predementia stages of the disease [3].
The sampling of cerebrospinal fluid (CSF) represents the most direct and convenient methods to study the biochemical changes occurring in the central nervous system. Aβ  , tau, and phosphorylated forms of tau have emerged as attractive diagnostic and prognostic CSF biomarkers for ongoing AD research [4,5]. Decreased CSF Aβ  has been proposed as an useful diagnostic tool for AD [4]. It has been reported that the mean level of Aβ  in the CSF are reduced to around 50% in subjects with AD relative to age-matched controls against initial prediction [4], and diagnosis of AD has evolved towards separate categories of preclinical and overt dementia based on levels of CSF Aβ 1-42 [6]. However, CSF Aβ 1-42 levels have been reported to fluctuate over time in a cohort of old and young individuals [7], and no absolute threshold has been identified that would differentiate between mild cognitive impairment and AD in mildly symptomatic individuals [8].
In the present study we aimed to review systematically the reported association between CSF Aβ 1-42 and AD with a view to evaluating the clinical usefulness of CSF Aβ  in the differential diagnosis of AD versus non-AD cognitive impairment. criteria, (ii) review studies selected for inclusion, (iii) overview data extraction, (iv) refine and validate the conclusions of the study.  Table). All 369 abstracts were reviewed using a combination of the search terms. The Patients-Intervention-Comparators-Outcomes (PICO) and search strategy was drafted. Study groups included patients with suspected mild cognitive impairment and/or AD, and study selection focused on reports that included measurements of Aβ levels. The reference standard was clinical diagnosis with medical results being followed up for more than 1 year. Literature searches using MEDLINE and EMBASE are summarized in S1

Quality of documents
The quality assessment tool selected for literature selection was adopted from the UK Scottish Intercollegiate Guidelines (SIGN) 'Methodology Checklist' (2004 March). SIGN is a systematic evaluation tool for the quality of the original literature and divides reports into randomized controlled trials, cohort studies, case-control studies, diagnostic assessments, and economic evaluation studies. Most of the literature on health technology assessment comprises non-randomized clinical trials and observational studies, and selection criteria were adapted accordingly (Table 1). Each stage of categorization was performed independently by two evaluators; their joint recommendations graded reports as summarized in Table 2. The present study excluded 'The Swedish Council on Technology Assessment in Health Care Study' in view of limitations as follows: (i) the study did not fulfil PICO standards; (ii) database searching was based on the references of pre-selected literature; (iii) the study included diverse controls ranging from non-AD dementias to other psychiatric or neurological disorders.

Data Extraction
Because documents put forward for evaluation comprised more than one type of study, data extraction was repeated several times and analyzed by two evaluators. Selection and categorization were performed in consultation with other researchers who advised on problem resolution. The data were then categorized according to type of data, study characteristics, and the

1+ +
High-quality meta-analyses, systematic reviews of RCTs, or RCTs with a very low risk of bias 1+ Well-conducted meta-analyses, systematic reviews, or RCTs with a low risk of bias 1-Meta-analyses, systematic reviews, or RCTs with a high risk of bias 2+ + High-quality systematic reviews of case-control or cohort studies High-quality case-control or cohort studies with a very low risk of confounding or bias and a high probability that the relationship is causal 2+ Well-conducted case-control or cohort studies with a low risk of confounding or bias and a moderate probability that the relationship is causal 2-Case-control or cohort studies with a high risk of confounding or bias and a significant risk that the relationship is not causal

Statistical Analyses
Funnel plot was used to address publication bias. Sensitivity testing was also conducted to assess the magnitude of publication bias, which was determined using a fail-safe number, defined as the minimum number of patients with non-significant findings that are needed to overturn the conclusion of a meta-analysis [13][14][15]. Larger fail-safe numbers indicate that the results are less prone to publication bias. For each outcome we tested the heterogeneity of results across the studies using "I 2 ". If significant heterogeneity was observed (p<.10), a random effects model-which assigns a weight to each study based on individual study variance as well as between study variance-was used to pool the results together. Also Mann-Whitney test was used to compare numerical values of β-amyloid levels between different reports in same disease categories (χ 2 ). Confidence intervals were determined using the means and standard deviations reported in each document. Meta-analysis was performed to assess the overall diagnostic accuracy of the pooled reports based on the random effects model. In addition, the fail-safe Number was calculated manually with EXCEL, suggested by Corwin [16]. SPSS (Statistical Package for the Social Sciences) 21.0 (SPSS/IBM Inc, New York) was used to recalculate the reported the χ 2 values. Revman 5.0 Meta DiSc 1.4 version (Hospital Universtario Ramony Cajal, Madrid, Spain) was subsequently used for meta-analysis of the entire dataset.

Results
Following systematic analysis of the literature and retrieval of primary data, meta-analysis was performed on eleven robust studies that compared Aβ 1-42 levels in AD (n = 2211) with healthy individuals (n = 1030), 10 studies that compared AD with non-AD dementias (n = 627), and five studies that compared a-MCI (amnestic mild cognitive impairment) (n = 1133) with na-MCI (non-amnestic mild cognitive impairment) subjects (n = 1276). The present evaluation is therefore based on the results of 17 published studies (Fig. 1). The primary documents and the extracted data are listed in Table 3. All selected paper used ELISA Kit of Innotest kind as a test tools, despite not limited to scan tool and the type of the selected documents. Range of test was 125~2000pg/mL, respectively and threshold was varied from 290 to 679pg/mL according to each document.

Results of systematic literature review
The diagnostic efficacy of CSF Aβ 1-42 in AD and healthy controls was reported in eleven documents. CSF Aβ 1-42 levels in AD ranged from 194±88.7 to 545±230 pg/ml, whereas levels in the healthy control group ranged from 383.5±101.8 to 1020±230 pg/ml (p <.001) (Figs. 2, 3 and  Table 3).
Additionally, a sub-analysis according to age and MMSE was performed to determine the cause of the heterogeneity within the effect size of the difference between AD and non-AD. There were no significant findings (S1 Fig.).
The diagnostic accuracy of CSF Aβ 1-42 levels in a-MCI versus AD, and a-MCI versus healthy controls, was only reported in one document and meta-analysis could therefore not be performed.

Discussion
In this study we have evaluated the clinical utility of CSF Aβ 1-42 levels in the diagnosis of AD versus healthy controls and non-AD dementias. Data retrieved from systematic literature review did not identify threshold CSF Aβ 1-42 levels that can distinguish between healthy controls and subjects with AD because there was highly significant heterogeneity and the ranges frequently overlapped. The fact that there is not a threshold, in other words a cut off, which can distinguish AD from healthy controls, as well from the other categories analyzed should be highlighted and it is a result of the meta-analysis along with those reported. However, this meta-analysis confirms that, overall, CSF Aβ 1-42 levels in AD are significantly lower than in healthy controls. Although meta-analysis was unable to differentiate reliably between a-MCI and healthy controls, several reports have attested to the clinical utility of CSF Aβ 1-42 levels in MCI. Maruyama et al. reported that CSF Aβ 1-42 levels did not differ significantly between the healthy control group and MCI [17]. Another study showed the values of CSF Aβ 1-42 were significantly lower in the progressive MCI group than in the control subjects and the stable MCI group [18]. CSF Aβ 1-42 concentration has a high diagnostic accuracy for correct allocation of AD patients in case-control studies and, together with CSF tau levels, can predict incipient AD in patients  with MCI [19]. Values of CSF Aβ 1-42 differed according to sample state (fresh versus frozen samples), but overall values were lower in AD patients than in MCI patients [20]. However, a threshold value discriminating between a-MCI and healthy controls could not be established. Instead, other studies have employed the ratio of CSF Aβ 1-42 to either Aβ 1-40, total tau, or phosphorylated tau as a potential measure of the evolution of MCI to AD [19,[21][22][23][24].
In the present analysis there were significant differences between the a-MCI and na-MCI groups. CSF Aβ 1-42 levels were lower in a-MCI (range 172.6±53.5 to 622.9±275.6 pg/ml) than in na-MCI (range 228.0±37.35 to 789.9±38.12pg/ml), and the pooled MD between groups was significant (pooled MD, 59.77 pg/ml). However, there was highly significant heterogeneity (I 2 = 66%), and calculated diagnostic accuracy for MCI alone gave SN and SP values, respectively, of 0.52-0.83 and 0.50-0.84.
Significant discriminatory power was also seen in AD versus non-AD dementia. CSF Aβ 1-42 levels in AD (range 194.0±88.7 to 545.0±230.0 pg/ml) were significantly below those reported in non-AD dementia (range 184.5±121.0 to 800.0 ± 174.0 pg/ml). The pooled MD value between groups was significantly lower in AD (pooled MD, 187.21 pg/ml). However, there was also significant heterogeneity (I 2 = 66%), and the calculated diagnostic accuracy of AD versus non-AD dementia gave SN and SP values, respectively, of 0.71-0.91 and 0.44-0.82. These findings may be summarized as follows. First, in patients with probable AD, CSF Aβ 1-42 levels are of value in differential diagnosis of AD from other dementias and from healthy controls. The mean concentration of Aβ 1 -42 in the CSF is significantly reduced by around 50%, in subjects with AD relative to age-matched controls [4,25]. There are debates about whether the Aβ 1-42 alone is useful or not in differentiating AD from non-AD dementias including frontotemporal dementia, vascular dementia, and dementia with Lewy bodies (DLB). Because concurrent presence of fibrillar Aβ deposits occurs in the majority of patients with DLB, it is possible that the reduced Aβ 1 -42 levels in the CSF have also been documented in patients with other dementia [4]. However, meta-analytic study indicates that CSF Aβ 1-42 can serve as a diagnostic and surrogate biomarker for Aβ deposition in the brain [26]. Second, the ranges of Aβ 1-42 levels partially overlap between AD and a-MCI, and it is therefore not possible to establish a cut-off value that discriminates between the two groups. Moreover, it is possible that a-MCI is an extension of AD pathology, and it has been suggested that a-MCI might be redefined to as a-MCI due to AD [6]. There might be the following several reasons; Some outstanding prospective CSF studies in MCI subjects would be particularly useful to add strength to this claim [27,28]. However, we decided to enroll papers published since 2004, because the criteria for MCI were revised to encompass other patterns of cognitive impairment in addition to memory loss [29]. In this paper we analyzed CSF results according to a-MCI and na-MCI. The other is considerable intra-or inter-laboratory variability of CSF analyses, which may influence the diagnostic classification of dementia according to results of CSF [30]. The intra-and inter-laboratory variability in CSF results from differences in pre-analytical and analytical procedures, lot-to-lot variation of analytical kits, freezing conditions and storage time [31][32][33]. It is necessary for research community to overcome this confusing situation that CSF variability was largest for Aβ  .
In summary, this meta-analysis establishes that reduced Aβ 1-42 levels are of diagnostic utility in AD, and relatively high CSF levels of Aβ 1-42 are indicative of non-AD pathology (e.g., na-MCI, non-AD dementias). However, CSF Aβ 1-42 levels alone are insufficient for reliable differential diagnosis of AD. Further research on the use of combinations of biomarkers, for example Aβ 1-42 levels in conjunction with other markers (e.g., total Aβ, Aβ 1-40 , tau, phosphorylated tau), will be necessary in order to develop CSF biochemical measurements permitting reliable diagnosis of AD versus other non-AD cognitive impairments.
Supporting Information S1 Fig. Sub-group analysis by age and MMSE in the groups of AD and non-AD. A sub-analysis according to age and MMSE has performed to determine the cause of the heterogeneity within the effect size of the difference between AD and non-AD. There were no significant findings. Abbreviations: AD, Alzheimer's disease; non-AD, non-AD dementia. (TIF) S1 PRISMA Checklist. For meta-analyses and systematic reviews, a PRISMA checklist.