Comparison of the diagnostic accuracy in the context of use of differential diagnostics when discriminating AD from other neurodegenerative disorders
Due to similar clinical manifestations and overlapping brain pathologies, differentiation of AD from other neurodegenerative disorders may prove difficult even with the aid of biomarkers. For example, the symptoms and biomarker patterns observed in patients with dementia with Lewy bodies (DLB) or subcortical vascular dementia (VaD) sometimes closely resemble those of AD, which makes differential diagnosis difficult and decreases the diagnostic accuracy of the core CSF AD biomarkers, especially in the early stages of the disease. Therefore, evidence was gathered on whether adding the CSF Aβ
42/40 ratio to the existing panel of biomarkers could improve the accuracy of the differential diagnosis of AD from other dementia disorders. Here we provide details of 16 studies that have compared the diagnostic accuracy of CSF biomarkers to diagnose ADD versus non-ADDs. These studies demonstrate the usefulness of the CSF Aβ
42/40 ratio for the diagnosis of AD in patients with dementia. Studies with relevant data are also summarized in Table
1.
Table 1
CSF biomarkers to distinguish cases with ADD from cases with non-ADD
| 55 | 68 | 34 | Aβ42 | 158.6 fmol/mL | – | – | – | – |
Aβ42/40 ratio## | 0.078## | 51 | 82 | – | NP |
| 22 | 11 | 35 | Aβ42 | 550 pg/mL | 100 | 80 | 0.923 | – |
Aβ42/40 ratio | 9.75 | 95.2 | 88.4 | 0.944 | NP |
| 69 | 69 | 47 | AD vs controls |
| 16 DLB | | Aβ42 | – | 93 | 87 | 0.949 | – |
| 27 FTD | | Aβ42/40 ratio | – | 93 | 87 | 0.947 | NP |
| 26 VaD | | AD vs non-AD |
| | | Aβ42 | – | 83 | 74 | 0.811 | NP |
Aβ42/40 ratio | – | 85 | 85 | 0.903 | |
| 94 | 166 (MCI) 29 (DD) | 38 | AD vs controls | |
Aβ42 MSD | < 523 | 73 | 89 | 0.88 (0.82–0.93) | – |
Aβ42 MSD/40 ratio | < 0.069 | 93 | 86 | 0.91 (0.86–0.95) | NP |
Aβ42 MSD/38 ratio | < 0.37 | 87 | 82 | 0.89 (0.83–0.93) | NP |
MCI-AD vs MCI |
Aβ42 MSD | < 523 | 67 | 71 | 0.73 (0.66–0.80) | – |
Aβ42 MSD/40 ratio | < 0.069 | 85 | 71 | 0.86 (0.79–0.91) | NP |
Aβ42 MSD/38 ratio | < 0.37 | 88 | 71 | 0.85 (0.79–0.91) | NP |
| 52 | 34 | 42 | AD vs FTD |
Aβ42 | > 464 | 79 | 62 | 0.75 | – |
Aβ42/40 ratio | ≤11.1 | 79 | 76 | 0.85 | n.s. |
Aβ42/38 ratio | ≤2.00 | 88 | 86 | 0.87 | n.s. |
| 80 69 (NP) 11 (AD+CVD) | 75 24 DLB (15 NP) 29 FTD (12 NP) 22 VaD (11 NP) | 30 | Aβ42 | 517 pg/mL | 81 | 59 | 0.747 (0.670–0.827) | – |
Aβ42/40 ratio | 0.057 | 81 | 60 | 0.749 (0.673–0.826) | NP |
| 48 | 127 43 PD 33 PDD | | | | | | | |
51 DLB | 107 | AD vs control |
Aβ42 | 444 ng/L | 94 | 72 | 0.871 (0.811–0.930) | -NP |
Aβ42/40 ratio | 0.125 | 92 | 79 | 0.871 (0.801–0.933) | |
AD vs PDD |
Aβ42 | 449 ng/L | 94 | 61 | 0.805 (0.704–0.905) | – |
Aβ42/40 ratio | 0.150 | 90 | 81 | 0.910 (0.844–0.976) | NP |
AD vs DLB |
Aβ42 | 387 ng/L | 88 | 41 | 0.675 (0.570–0.780) | – |
Aβ42/40 ratio | 0.115 | 90 | 57 | 0.759 (0.664–0.853) | NP |
| | | | AD vs controls |
Aβ42 | 534 pg/mL | 82 | 74 | 0.818 | – |
Aβ40/42 ratio | 8.3 | 59 | 81 | 0.719 | NP |
AD vs FTD | | | | | |
Aβ42 | 538 pg/mL | 70 | 82 | 0.791 | – |
Aβ40/42 ratio | 5.4 | 59 | 87 | 0.778 | NP |
| 367 (AD+ non-AD) | – | 0 | AD vs non-AD |
Aβ42 | 737–836 pg/mL | 78 | 79 | 0.81 | – |
Aβ42/40 ratio | 0.050–0.082 | 73 | 78 | 0.81 | NP |
| 100 | 50 | 50 | AD vs controls |
50 (AD) | 17 (DLB) | | Aβ42 | < 722 pg/mL | 98.0 | 74.0 | 0.874 | – |
50 (MCI-AD) | 17 (FTD) | | Aβ42/40 ratio | < 0.1099 | 85.7 | 78.0 | 0.881 | NP |
| 16 (VaD) | | Aβ42/38 ratio | < 0.269 | 81.6 | 82.0 | 0.858 | NP |
| | | AD vs non-AD |
Aβ42 | < 694 pg/mL | 95.9 | 40.0 | 0.686 | – |
Aβ42/40 ratio | < 0.1215 | 93.9 | 50.0 | 0.782 | NP |
Aβ42/38 ratio | < 0.2730 | 81.6 | 68.0 | 0.804 | NP |
| 70 | 55 | 15 | Pro-AD vs pro-DLB |
31 (AD-d) | 20 (DLB-d) | | Aβ42 | ≤ 730 ng/L | 84.6 | 71.4 | 0.84 (0.74–0.92) | – |
39 (pro-AD) | 35 (pro-DLB) | | Aβ42/40 ratio | ≤ 0.0529 | 88.9 | 100 | 0.95 (0.83–0.99) | NP |
| | | AD-d vs DLB-d |
Aβ42 | ≤ 504 ng/L | 67.7 | 80 | 0.77 (0.63–0.88) | – |
Aβ42/40 ratio | ≤ 0.0799 | 92.3 | 88.9 | 0.86 (0.64–0.97) | NP |
| Cohort 2 | Cohort 2 | Cohort 2 | AD vs MCI |
75 (AD) | 62 (MCI) | 53 | Aβ42 | – | – | – | 0.817 (0.743–0.890) | – |
35 (MCI-AD) | 34 (VaD) | Cohort 3 | Aβ42/40 ratio | – | – | – | 0.879 (0.823–0.936) | < 0.028 |
Cohort 3 | 47 (DLB/PDD) | 328 | Aβ42/38 ratio | – | – | – | 0.856 (0.790–0.923) | < 0.222 |
137 | 33 (FTD) | | AD vs DLB/PDD |
| Cohort 3 | | Aβ42 | – | – | – | 0.583 (0.476–0.690) | – |
| 35 (DLB/PDD) | | Aβ42/40 ratio | – | – | – | 0.792 (0.707–0.877) | < 0.001 |
| 128 (PD) | | Aβ42/38 ratio | – | – | – | 0.796 (0.710–0.883) | < 0.001 |
| | | AD vs VaD |
Aβ42 | – | – | – | 0.698 (0.580–0.816) | – |
Aβ42/40 ratio | – | – | – | 0.880 (0.814–0.946) | < 0.001 |
Aβ42/38 ratio | – | – | – | 0.860 (0.786–0.935) | < 0.001 |
AD vs non-AD | |
Aβ42 | – | – | – | 0.720 (0.651–0.788) | – |
Aβ42/40 ratio | – | – | – | 0.863 (0.813–0.912) | < 0.001 |
Aβ42/38 ratio | – | – | – | 0.863 (0.813–0.913) | < 0.001 |
| 342 | 562 | 0 | AD vs non-AD |
Cohort 1 | Cohort 1 | | Cohort 1 | |
124 | 276 | | Aβ42 | 500 pg/mL | – | – | 0.78 (0.734–0.818) | – |
Cohort 2 | Cohort 2 | | Aβ42/40 ratio | 0.1 | – | – | 0.90 (0.865–0.926) | < 0.0001 |
218 | 286 | | Cohort 2 | |
Aβ42 | 700 pg/mL | – | – | 0.60 (0.553–0.641) | – |
Aβ42/40 ratio | 0.05 | – | – | 0.77 (0.728–0.803) | < 0.0001 |
In a study of patients with ADD, normal controls, patients with non-ADD and patients with other neurological diseases, Shoji et al. [
12] found that the ADD group had a significantly higher level of tau than the normal control group (
p < 0.001), but Aβ
40 levels did not show any significant differences between the groups. The reduction of Aβ
42 levels in AD also resulted in a significant increase in the Aβ
40/42 ratio (note that the ratio reported in this study has Aβ
40 in the numerator, in contrast to most of the other studies summarized in the current paper) as an improved marker. The authors therefore concluded that the Aβ ratio is another important marker for AD.
Lewczuk et al. [
13] measured concentrations of Aβ
42, Aβ
40 and total tau (T-tau) in order to compare their accuracy in discriminating patients with ADD, non-ADD and control subjects. The results showed that concentrations of Aβ
42 were decreased (
p < 0.001) and of T-tau were increased (
p < 0.001) in ADD patients, while Aβ
40 concentrations did not differ significantly among the groups. For all groups when the Aβ
42/40 ratio was used, more patients were classified correctly, compared to when the Aβ
42 concentration alone was used (94 vs 86.7% when comparing ADD to controls, 90 vs 85% when comparing ADD to non-ADD and 90.8 vs 87% when comparing ADD to non-ADD plus controls). The improvement of the diagnostic accuracy reported in this study was not significant, probably due to the small numbers of subjects and a clear ceiling effect (a relatively high number of patients were already correctly classified using Aβ
42 alone).
Gabelle et al. [
14] evaluated the value of individual and combined measurements of CSF biomarkers. They found that both Aβ
40/42 and Aβ
38/42 ratios were significantly altered in AD. They also found that the Aβ
40/38 ratio was the only one that differentiated clearly control subjects from FTD subjects, while not being significant between AD and FTD. In the ROC curves, they found that for FTD versus AD diagnosis, the best AUCs for amyloid biomarkers were the Aβ
38/42 ratio and the Innogenetics Aβ/Tau index (IATI) (AUCs = 0.87). However, the Aβ
40/42 ratio or Aβ
42 alone had very close and statistically undifferentiated AUC values. The authors concluded that the Aβ
38/42, Aβ
40/42 and the IATI ratios were also better than individual biomarkers to identify AD therefore justifying their clinical relevance.
In another study carried out by Wiltfang et al. [
15], the authors found that alterations of Aβ
42 concentrations might not only result from AD pathology but may also be related to total Aβ peptide concentrations. In such cases, healthy individuals with relatively low total Aβ might be misdiagnosed as having ‘pathologically low’ Aβ
42 concentrations, and vice versa, AD subjects with high total Aβ might be misinterpreted as normal Aβ
42 carriers. It was therefore concluded that the analysis of CSF Aβ
42 alone (i.e. without Aβ
40) might lead to misinterpretation of the neurochemical dementia diagnostics outcome in subjects with constitutively high or low concentrations of total Aβ peptides. Consequently, the authors conclude that the CSF Aβ
42/40 ratio can possibly improve the reliability of the neurochemical dementia diagnosis.
A study by Slaets et al. [
16] compared the use of different biomarkers for the diagnosis of AD. Addition of the CSF Aβ
42/40 ratio to the existing panel of biomarkers, Aβ
42, Aβ
40 and tau phosphorylated at threonine 181 (P-tau
181) was compared to the panel without the addition of the ratio. The results showed that the CSF Aβ
42/40 ratio was significantly decreased in AD patients (0.043 ± 0.021) compared to non-AD patients (0.064 ± 0.027;
p < 0.001) and controls (0.053 ± 0.023;
p < 0.001). Following receiver operating characteristic (ROC) analysis, the optimal cut-offs discriminating the groups were defined as the values maximising the sum of the sensitivity and the specificity. Although the difference between the areas under the ROC curves (AUC) for Aβ
42 and Aβ
42/40 turned out to be insignificant, the diagnostic accuracy of the decision tree that contained Aβ
42, Aβ
40, P-tau
181 and the Aβ
42/40 ratio was significantly better than the diagnostic accuracy of the decision tree without Aβ
40 and the Aβ
42/40 ratio (80% vs 74%;
p < 0.001). The authors concluded that there was no difference in CSF Aβ
40 levels found between AD and non-AD patients, but that adding CSF Aβ
40 and the CSF Aβ
42/40 ratio to a biomarker-based decision tree, might have an added value for discriminating AD from non-AD patients in cases with intermediate CSF P-tau
181 values.
Nutu et al. [
17] evaluated whether the CSF Aβ
42/40 ratio could be used for differentiating AD from DLB and Parkinson’s Disease Dementia (PDD). The primary finding of this study was that the CSF Aβ
42/40 ratio increased discrimination of AD from PDD and DLB compared with either of the two Aβ biomarkers individually. Of note, in this study, Aβ
40 was significantly higher in AD. Furthermore, the authors concluded that use of the Aβ
42/40 ratio could improve the differentiation of AD from PDD and DLB.
In a study by Baldeiras et al. [
18], the added value of another CSF Aβ-peptide (Aβ
40), along with the core CSF markers T-tau, P-tau
181 and Aβ
42, in the discrimination between two large dementia groups of FTD (
n = 107), AD (
n = 107) and non-demented subjects (
n = 33) was evaluated. The authors found that their data ‘taken together’ indicated that ‘although CSF Aβ
40 has no added value in the distinction between AD and FTD patients, it might be useful in the discrimination between AD and FTD patients from non-demented controls, and therefore could be considered in patients diagnostic work-up’.
In a prospective study of subjects with cognitive disorders at three French memory centres (Paris-North, Lille and Montpellier; the PLM study), Dumurgier et al. [
19] assessed whether the use of the Aβ
42/40 ratio would reduce the frequency of indeterminate CSF profiles. They found that, on the basis of local optimum cut-offs for Aβ
42 and P-tau
181, 22% of patients had indeterminate CSF profiles. The systematic use of the Aβ
42/40 ratio instead of Aβ
42 levels alone decreased the number of indeterminate profiles (17%;
p = 0.03), but it failed to improve the classification of subjects (NRI = − 2.1%;
p = 0.64). In contrast, use of the Aβ
42/40 ratio instead of Aβ
42 levels alone in patients with a discrepancy between P-tau
181 and Aβ
42 led to a reduction by half of the number of indeterminate profiles (10%;
p < 0.001) and was also in agreement with clinician diagnosis (NRI = 10.5%;
p = 0.003). The authors therefore concluded that in patients with a discrepancy between CSF P-tau
181 and CSF Aβ
42, the assessment of the Aβ
42/40 ratio led to a reliable biological conclusion in over 50% of cases that agreed with a clinician’s diagnosis.
Sauvee et al. [
20] investigated whether the CSF Aβ
42/40 ratio could be used to improve the accuracy of diagnostically relevant conclusions in patients with ambiguous CSF Aβ
42 or tau results. They found that one third of the biomarker profiles of patients with atypical dementia were ambiguous. The addition of the CSF Aβ
42/40 ratio increased the proportion of interpretable profiles from 69 to 87%, without changing the conclusion when the usual biomarkers (Aβ
42 and P-tau
181) were concordant. The authors therefore concluded that their results support the use of the Aβ
42/40 ratio in addition to the usual CSF AD biomarkers for patients with ambiguous profiles. They added that this method could be specifically directed to this population (i.e. patients with ambiguous results) in order to improve the level of certainty for clinical routine practice.
Lewczuk et al. [
21] also compared the diagnostic accuracies of the CSF Aβ
42/40 ratio and CSF Aβ
42 alone. Analysis of Aβ
42 gave a sensitivity and specificity of 69.3% and 88.9%, respectively, whereas the Aβ
42/40 ratio showed significantly improved performance with sensitivity and specificity of 93.3% and 100%, respectively. Thus, the authors concluded that the CSF Aβ
42/40 ratio concentration shows significantly better diagnostic performance compared to the CSF Aβ
42 concentration alone. It should be noted, however, that this study must not be interpreted as providing absolute values of the diagnostic accuracies, but their relative comparison.
In another study including various CSF biomarkers, Spies et al. [
22] investigated the CSF Aβ
42/40 ratio under the assumption that Aβ
40 closely represents the total cerebral Aβ load. They found that the Aβ
42/40 ratio improves differentiation of AD patients from VaD, DLB and non-ADD patients when compared to Aβ
42 alone. Furthermore, they found that the Aβ
42/40 ratio is a more easily interpretable alternative to the combination of Aβ
42, P-tau
181 and T-tau when differentiating AD from either frontotemporal dementia (FTD) or other non-ADDs. Since they found different Aβ
40 concentrations in the various dementia groups, the authors also added that it can be debated if the Aβ
42/40 ratio is a good representation of the Aβ
42 fraction of the total Aβ load and thus eliminates inter-individual differences in total Aβ concentrations.
A study of patients with AAD, non-ADDs (DLB, FTD, VaD), MCI due to AD and non-demented controls found that the CSF Aβ
42/40 ratio improved the diagnostic performance of Aβ
42 in most differential diagnostic situations. Struyfs et al. [
23] also found that the Aβ
42/40 ratio was the best biomarker to distinguish between AD-MCI and FTD.
Similarly, Janelidze et al. [
24] also found that the CSF Aβ
42/40 ratio, as well as the CSF Aβ
42/38 ratio, was ‘superior biomarkers of AD pathology compared with Aβ
42 alone’. Using three commercially available CSF biomarker immunoassays, this study found that the CSF Aβ
42/40 and Aβ
42/38 ratios improved differentiation of AD from non-ADs, especially when separating AD from DLB/PDD and VaD.
The authors point to several potential explanations for the improved accuracy when using the CSF Aβ42/40 and Aβ42/38 ratios instead of Aβ42. They suggest that it might be that subcortical pathologies not specific to AD, such as WMLs and alpha-synuclein pathology, cause reduced levels of all CSF Aβ species, including Aβ42. A second explanation for the improved diagnostic accuracy of the Aβ42/40 and Aβ42/38 ratios could be that differences in the overall production and clearance of Aβ probably contribute to inter-individual variability in total CSF Aβ levels. This is supported by the present finding that in CSF Aβ42 correlates with Aβ38 and Aβ40 even in healthy controls. Consequently, when detecting Aβ42 brain pathology with CSF Aβ42, using ratios to Aβ40 or Aβ38 might correct for inter-individual differences in total Aβ levels.
In another study, which evaluated the differential diagnosis of DLB and AD, once again, the CSF Aβ
42/40 ratio was found to aid diagnosis. The study by Bousiges et al. [
25] found that the Aβ
42/40 ratio remained unchanged in DLB patients between the prodromal and demented stages, contrary to what was observed in AD. The Aβ
42/40 ratio therefore makes it possible to distinguish between the two pathologies ‘at a time when the differential diagnosis is difficult’.
Finally, in a study by Lehmann et al. [
26], the Aβ
42/40 ratio was added to a previously described PLM scale (Paris-Lille-Montpellier study), which combines Aβ
42, T-tau and P-tau
181 biomarkers, in order to evaluate an optimized PLM
R scale (PLM ratio scale). Nine hundred and four participants (342 AD and 562 non-AD) were studied, in two chronologically different cohorts (400 Mtp-1 and 504 Mtp-2). For AD patients, the mean CSF Aβ
42 and CSF Aβ
42/40 ratio was 553 ± 216 pg/mL and 0.069 ± 0.022 pg/mL in Mtp-1 and 702 ± 335 pg/mL and 0.045 ± 0.020 pg/mL in Mtp-2. The distribution of AD and non-AD differed between the PLM and the PLM
R scales (
p < 0.0001). The percentage AD well-classified (class 3) increased with PLM
R from 38 to 83% in Mpt-1 and from 33 to 53% in Mpt-2. A sharp reduction of the discordant profiles going from 34 to 16.3% and from 37.5 to 19.8%, for Mtp-1 and Mtp-2 respectively, was also observed. The authors concluded that the integration of the Aβ
42/40 ratio in the PLM
R scale resulted in an easy-to-use tool which reduced the discrepancies in biologically doubtful cases and increased the confidence in the diagnosis.
In order to try to assess what is the overall impact on diagnosis, an estimation was made of what the actual percentage of patients that are misdiagnosed by Aβ
42 alone and that become correctly classified with the Aβ
42/40 ratio. Assuming normal distribution of Aβ
40 across the population of interest [
15], a very conservative estimation is that neglecting Aβ
40 (which is equivalent to applying Aβ
42 as the sole CSF biomarker of amyloidosis) leads to misdiagnoses of ca. 5–10% of cases. This is further confirmed by Baldeiras et al. [
27], who found an increase in the proportion of interpretable CSF profiles from 61 to 75% (i.e. ca. 20% of the baseline value) of the MCI patients. Also, Dorey et al. [
28] report that determining CSF Aβ
40 concentrations corrected diagnosis in AD patients with non-pathological CSF Aβ
42 levels in 76.2% of cases using the CSF Aβ
42/40 ratio.
In summary, the accumulation of evidence clearly points to the usefulness of the CSF Aβ42/40 ratio for the diagnosis of AD in patients with dementia. The CSF Aβ42/40 ratio is also better than CSF Aβ42 alone at distinguishing AD dementia from non-AD dementias, not only from controls. The evidence therefore strongly suggests that the CSF Aβ42/40 ratio, rather than CSF Aβ42 alone, should be used in the clinical work-up of AD, as a way to improve the diagnostic accuracy for distinguishing ADD from other dementia disorders.