Cerebrospinal Fluid Biomarkers for the Differential Diagnosis between Alzheimer’s Disease and Frontotemporal Lobar Degeneration: Systematic Review,HSROC Analysis,and Confounding Factors

INTRODUCTION

Differential diagnosis in dementia is at present one of the main challenges in clinical practice and research due to heterogeneity and overlap between neurodegenerative diseases. Alzheimer’s disease (AD) is the most frequent neurodegenerative disease and the most common cause of dementia [1, 2]. AD is characterized by extracellular deposits of amyloid-β (Aβ) peptides in senile plaques and intracellular aggregates of tau protein in neurofibrillary tangles [3, 4]. The typical clinical presentation of AD includes impairment in episodic memory with cognitive decline involving other functions as well as changes in mood and behavior as the disease progresses [3]. With the emergence of new treatments aimed to modify the course of the disease, an accurate differential diagnosis of AD becomes crucial.

Frontotemporal lobar degeneration (FTLD) is the most common cause of dementia under the age of 60 [5]. FTLD is the umbrella term encompassing a heterogeneous group of well-defined clinical syndromes, including (1) three frontotemporal dementia (FTD) variants: behavioral variant frontotemporal dementia (bvFTD), progressive non-fluent aphasia, and semantic dementia; (2) progressive supranuclear palsy syndrome; and (3) corticobasal syndrome [6, 7]. The Lund-Manchester groups have actively guided the diagnostic criteria since 1994 [8–10]. These criteria have been widely used although alternative criteria also exist [11], as well as recent refinements [12–15]. Despite well-articulated clinical diagnostic criteria for FTLD, differential diagnosis between FTLD and AD is sometimes difficult. Pathological studies show incorrect FTLD diagnosis in up to 28% of the cases [16, 17]. This occurs especially in the early phase of both diseases as well as in atypical forms (e.g., AD with significant frontal involvement, or FTLD with significant memory deficit) [18, 19].

AD pathophysiology can now be studied in vivo thanks to certain biomarkers in the cerebrospinal fluid (CSF). Currently validated core CSF biomarkers of AD are Aβ₄₂, total tau (t-tau), and phosphorylated tau (p-tau) proteins [20]. A large body of evidence shows that AD patients present decreased levels of Aβ₄₂ and increased levels of t-tau and p-tau compared to healthy controls [21–23]. This has led to their inclusion in the revised diagnostic criteria for AD as supportive feature of the clinical diagnosis [20]. However, findings on the potential of the CSF biomarkers to discriminate AD from other dementias are less conclusive. Although recent studies comparing AD and FTLD suggest increased t-tau and p-tau concentrations and reduced Aβ₄₂ levels in AD, results show substantial heterogeneity across studies and inconsistent findings [24–26].

To our knowledge, three meta-analysis has previously evaluated the role of CSF tau and Aβ₄₂ in the differential diagnosis between AD and FTLD. Tang et al. [25, 26] analyzed mean differences in the concentrations of p-tau and Aβ₄₂, but not their diagnostic performance. Van Harten et al. [24] only analyzed t-tau and p-tau, obtaining 74% sensitivity and 74% specificity for t-tau, and 79% sensitivity and 83% specificity for p-tau. However, these results were obtained by calculating summary points from separate meta-analyses of sensitivity and specificity, a statistical approach that does not account for the covariation between them, or for the variability in the diagnostic cut-off values across studies [27, 28]. Furthermore, new articles have been published since then. The usefulness of CSF biomarkers for the differential diagnosis between AD and FTLD is still controversial and a topic of great interest at present. Therefore, the objective of this study was to perform an updated systematic review on the current evidence on the diagnostic performance of CSF biomarkers t-tau, p-tau, and Aβ₄₂ (as well as their ratios and combinations), in the discrimination between AD and FTLD patients. We used a more theoretically motivated statistical approach, the Hierarchical Summary Receiver Operating Characteristic (HSROC) model [29], that is deemed more appropriate when considerable variation is present in cut-offs for the discrimination between diagnostic groups. We also aimed to explore relevant confounding factors of CSF biomarkers’ diagnostic performance.

MATERIAL AND METHODS

RESULTS

The selection flow is shown in Fig. 1. Key characteristics and sensitivity-specificity results of the 30 selected studies [41–70] are displayed in Table 1, and methodological quality in Table 2. Eleven studies specified the FTLD variant included, but only one reported biomarkers’ diagnostic performance separately for each variant [51], and other two included only bvFTD patients [45, 50]. Subgroup analyses could only be performed for age, disease duration, global cognitive impairment (i.e., MMSE), postmortem diagnostic confirmation, and cut-off value (internal versus external). For continuous variables, subgroups were defined by the median value of the specific confounder (see Table 3). For FTLD diagnosis, a sensitivity analysis was performed excluding studies using other criteria than Neary et al. [8]. Table 3 shows mean values for the five parameters of the HSROC models and their 95% confidence intervals (CI).

DISCUSSION

In this study we provide an updated and comprehensive meta-analysis on the diagnostic performance of core CSF biomarkers of AD for the differential diagnosis between AD and FTLD. Our results on t-tau and p-tau are almost identical to those obtained in a previous meta-analysis [24]. Results replication is enhanced by the fact that we followed a more appropriate statistical approach (HSROC model) and included several studies that have been recently published. Moreover, we extended this previous study by analyzing Aβ₄₂, a relevant CSF biomarker in AD. By doing so, our findings revealed that only by combining p-tau and Aβ₄₂ results achieved clinical usefulness. Another contribution is that, since the performance of CSF biomarkers’ is influenced by different confounding factors such as age [31], several subgroup meta-analyses were conducted in order to identify for which specific kind of patients these CSF biomarkers might be useful in clinicalpractice.

When using the CSF biomarkers separately, none of them achieved optimal diagnostic performance, as defined by sensitivity and specificity values ≥80% [39]. P-tau showed the best performance reaching 84% sensitivity and 78% specificity. The fact that p-tau arose as the biomarker with best diagnostic performance could be explained by the fact that p-tau is not only a marker of axonal damage and neuronal degeneration, as t-tau, but it is more closely related to AD pathophysiology and the formation of neurofibrillary tangles [71, 72]. Moreover, CSF concentrations of p-tau in FTLD are more comparable to concentrations in controls than to concentrations in AD patients [24, 26]. Combinations were needed to achieve optimal diagnostic performance. The p-tau/Aβ₄₂ ratio showed optimal results, with 87% sensitivity and 80% specificity. Negative likelihood ratios indicated that normal values in the p-tau/Aβ₄₂ ratio may help to rule-out AD. This better performance of the ratio in comparison with the CSF biomarkers alone is likely explained by the fact that two aspects of the AD pathology, i.e., plaques (Aβ₄₂), and neurodegeneration (tau), are captured by the ratio.

A novelty in this study is the evaluation of different confounders of CSF biomarkers’ diagnostic performance. To our knowledge, no previous meta-analyses have performed such kind of analyses in AD versus FTLD, despite the well-known influence of external factors on biomarkers’ performance [31]. We a priori defined a comprehensive list of confounders. However, due to the limited data available, subgroup analyses could only be performed for age, disease duration, global cognitive impairment, and cut-off values. Below, we provide several recommendations to overcome this issue in future studies. Although we did not find any evidence of a statistically significant effect of these confounders on biomarkers’ accuracy, the likelihood ratios indicated moderate increase in probability and several interesting trends were observed. Regarding age, the p-tau/Aβ₄₂ ratio showed optimal diagnostic performance in studies including younger samples (mean age <70.3 years). Sensitivity was very high (91%) in studies including older samples (mean age ≥70.3 years), however at the cost of low specificity (71%). Sensitivity values are usually higher in older populations due to the fact that age-related brain damage is added to the disease-related brain damage [73, 74]. We previously showed higher sensitivity values of the CSF biomarkers in MCI patients older than 70 years compared to MCI patients younger than 70 years [31]. Mattsson and co-workers also found increased sensitivity in MCI patients older than 65 years compared to MCI patients younger than 65 years [75]. Greater AD pathology in older FTLD patients may also be an explanation of worse specificity in comparison with sensitivity, given the high percentage of misdiagnosis in 28% of FTLD patients [16, 17].

Regarding global cognitive impairment, normal CSF levels of p-tau may help to rule-out AD in patients with lower MMSE score (mean <19.3). The opposite was observed for the t-tau/Aβ₄₂ ratio, where normal values of the ratio may help to rule-out AD in patients with higher MMSE score (mean >19.2). To our knowledge, previous meta-analyses have not evaluated CSF biomarkers’ diagnostic performance in relation to disease severity. However, greater brain structural changes and alterations in brain connectivity, as markers of disease severity, are associated with more pathological CSF biomarker levels [76–78]. Therefore, it seems reasonable to expect higher diagnostic performance in more severely impaired patients.

Specific subgroup analyses could not be performed for other confounders due to insufficient information available. Nonetheless, clinical-pathophysiological heterogeneity of FTLD as well as postmortem diagnostic confirmation is of relevance and results from individual studies deserve some discussion. An important candidate to influence CSF biomarkers’ diagnostic performance is the clinical and pathophysiological heterogeneity of the FTLD spectrum. The different clinical FTLD syndromes, i.e., FTD (including bvFTD, progressive non-fluent aphasia, and semantic dementia), progressive supranuclear palsy syndrome, and corticobasal syndrome, could imply different profiles of CSF biomarkers. However, data reported in the articles did not enable subgroup analyses. Unsatisfactory results were obtained in the few studies including or reporting data separately for bvFTD [45, 50], except in de Souza et al. [51]. This last study showed that Aβ₄₂ was less sensitive but more specific when AD was compared with bvFTD than when compared with semantic dementia. The opposite occurred for tau. Moreover, progressive non-fluent aphasia could not clearly be differentiated from AD. Grossman et al. [53] showed that semantic dementia was the variant with more AD-like CSF biomarkers profile, and significantly different from the other FTLD variants. Regarding pathophysiological heterogeneity in FTLD, three subtypes have been described depending on the main protein deposits [14, 79]: tau, TDP-43 (TAR DNA-binding protein 43), and FUS (tumor associated protein fused in sarcoma). There is a strong correspondence between the clinical diagnosis of semantic dementia and the TDP-43 subtype [7, 81], and both progressive supranuclear palsy and corticobasal syndromes with the tau subtype [82, 83]. However, the clinical diagnosis of bvFTD has not been clearly associated with any of these [7]. Hence, results from both primary studies and the current meta-analysis should thus be interpreted in this context of FTLD heterogeneity. Some recommendations are given below.

Another source of heterogeneity is the frequent use of clinical diagnosis as reference standard instead of postmortem confirmed diagnosis. Previous research has evidenced considerable variability in false positive rates of FTLD clinical diagnosis when pathologically confirmed [84–86]. Irwin et al. [56] found that clinical diagnosis yielded 67% sensitivity and 87% specificity when predicting postmortem confirmed diagnosis. Toledo et al. [69] reported that the use of clinical diagnosis instead of pathologically confirmed diagnosis underestimates sensitivity and specificity values in about 10–20% for CSF tau and Aβ biomarkers. However, studies with pathological diagnosis are still scarce and normally include small samples. In this review, only eight studies included samples with postmortem diagnostic confirmation, some of them overlapped, and they did not evaluate the same biomarkers. However, the four studies that evaluated the ratio t-tau/Aβ₄₂ obtained an excellent pooled result (88% sensitivity and 95% specificity). Nonetheless, CSF biomarkers’ diagnostic performance in pathologically confirmed AD still remains uncertain. It must be taken into account that most of the interpretations from primary studies on the potential mechanisms involved in FTLD are derived from clinically diagnosed patients.

Variability in cut-off values to discriminate between diagnostic groups is common in the field. Cut-offs are usually obtained from the same sample in which sensitivity and specificity values are calculated [60, 87]. Other studies have used external cut-off values [55, 67]. We did not find statistically significant differences between the two methods. Besides potential differences in the samples evaluated, technical factors regarding the analytical platforms used to measure the CSF biomarkers could also introduce bias in the evaluation of their diagnostic performance. Results from a multicenter study comparing the analytical accuracy of CSF t-tau measurements show an inter-center coefficient of variation for t-tau levels of 16% [88]. Currently, efforts are being made to harmonize biomarker methods and research protocols with the hope of making results comparable among centers [89, 90].

Some limitations should also be discussed. The assessment of methodological quality reflected possible patient selection bias in eight of the included studies. This was related to inclusion of not completely consecutive or random samples and not perfect avoidance of inappropriate exclusions. In particular, patients were normally selected from specialized centers on the basis of availability of CSF data, a procedure not always performed in all incoming patients. Further, underreporting of data was observed. Most of the studies did not report information on confounders other than age, disease duration, global cognitive impairment, clinical diagnostic criteria for FTLD, and cut-offs values for discriminating between diagnostic groups. Further, multivariate analysis on the effect of these confounding factors was not possible. This was supplied by performing subgroup analyses. Although these analyses provided relevant information they need to be considered with caution due to aspects such as the use of univariate approach or use of arbitrary thresholds (medians) to define the subgroups. Beta CIs were wide in some analyses, and the assumption of symmetry (equal accuracy for different thresholds) must thus be considered with caution. Another limitation is that MMSE is commonly used as a measure of global cognition and disease severity in AD. However, other instruments such as the Frontotemporal Dementia Rating Scale [91] are more appropriate for the same purpose in FTLD, which could vary the results of the subgroup analyses on MMSE. Finally, FTLD is a heterogeneous disorder both clinically and pathophysiologicaly. Primary studies do not normally specify the FTLD subtype included or provide CSF biomarkers values for combined groups, therefore it was not possible to conduct separate meta-analyses for the different subtypes. This is an important limitation and future primary studies should provide more information in thisregard.

CONCLUSIONS

At present, differential diagnosis between AD and FTLD is one of the main challenges and great controversy exists on the clinical utility of core CSF biomarkers. A previous meta-analysis showed unsatisfactory sensitivity and specificity values for CSF t-tau, and better but still suboptimal results for p-tau. Moreover, Aβ₄₂ was not investigated [24]. By using a more appropriated statistical technique and including later studies, we have replicated those results on t-tau and p-tau, and offered meta-analytic evidence on the optimal performance of the p-tau/Aβ₄₂ ratio in the differential diagnosis between AD and FTLD. In particular, the p-tau/Aβ₄₂ ratio may be especially indicated for young AD and FTLD patients, where differential diagnosis is indeed more frequent in the clinical practice. Moreover, depending on the level of global cognitive impairment, the use of p-tau may be preferable (high MMSE scores); otherwise the t-tau/Aβ₄₂ ratio could be more appropriate (low MMSE scores).

Based on the findings of the current study and the systematic review performed, the following bullet points are meant to conclude on the current state of the topic and propose recommendations for future research:

The reviewed core CSF biomarkers are currently used in some centers for the differential diagnosis of AD [92]. If methodological limitations are overcome in the future, the use of CSF biomarkers may be generalized to other hospitals and be definitively implemented as part of the clinical routine.

To advance in these aspects is expected to make it possible to definitively identify those conditions where the CSF biomarkers are useful both for clinical practice and research.