Demographic data:Charts of 190 patients were reviewed, including 98 men (51.6 %) and 92 women (48.4 %). The mean age of the patients at the time of enrolment was 71 (+/- 9) years. The youngest patient was 52 years old, the oldest 92 years old. 49 patients (25.8 %) were younger than 65 years at the time of inclusion and thus fulfilled the diagnostic criterion for pre-senile dementia. 113 patients (59.5 %) were finally diagnosed with AD, 21 (11.1 %) with FTD and only 5 (2.6 %) were considered to have VaD. 51 patients (26.8%) could not be assigned to any of these diagnosis groups. In these patients, the diagnosis remained either vague (e.g. "cognitive deficits", "dementia development") or no clear distinction was possible. However, as described above, the analysis of the patients was independent of the clinical diagnoses.
We aimed to assess whether the stage of the illness had an influence on the CSF parameters and the neuropsychological test results. Therefore, the patients were divided into two groups, based on the median of the time span between symptom onset and diagnosis: Patients in whom the symptoms had appeared no more than twelve months ago (Group A, 56.3%) and those who reported a cognitive impairment that had been present for more than one year (Group B, 43.7%). The patients in group A were slightly older (M = 72 ± 9 years) than those in group B (69 ± 8.0 years). In addition, the proportion of women in group A (55.1 %) was significantly higher than in group B (39.8 %).
Associations between CSF parameters and neuropsychological testing:The CSF parameters showed that Aβ1-42 had the strongest relation with the results of the neuropsychological tests. Reduced CSF concentrations of this biomarker were particularly associated with poor results in the memory subtests. Strong relations were found in the Word List Learning, Word List Delayed Recall and Word List Recognition tests (Figure 1): In the Word List Learning test, patients with normal Aβ1-42 values performed significantly better by an average of almost one SD than those with pathologically low Aβ1-42 concentrations (Figure 1). Furthermore, there was a significant, albeit weak, correlation between the results of this test and the Aβ1-42 concentration in the CSF (Table 1). In addition, Aβ1-42 was the most important predictor for the independent variable Word List Learning calculated using multiple linear regression, ahead of Aβ quotient and pTau (R² = 0.17, p < 0.001).
A similar result was obtained for the Word List Delayed Recall test. This verbal episodic memory test assesses the ability to retain newly learnt content in the long term. In AD, this part of the CERAD-Plus test battery is often the first to show abnormalities [23]. Here, too, there was a significant mean difference between patients with normal and pathological Aβ1-42 values, although this was less pronounced than with Word List Learning at a comparable effect size (Figure 1). The results from Word List Delayed Recall and the Aβ1-42 concentration showed equally weak but significant correlations (Table 1). Again, the Aβ1-42 was the most important predictor in the prediction model (beside Aβ quotient and pTau) with a moderate variance explanation (R² = 0.15, p = 0.001).
The greatest mean difference was observed in Word List Recognition: Patients with normal Aβ1-42 concentrations performed more than one SD better than patients with low Aß1-42 concentrations (Figure 1). The correlation between the two parameters was again weak, but robust (Table 1). In the multiple linear regression, Aβ1-42 was the only significant predictor for Word List Recognition (R² = 0.07, p = 0.021).
In addition to the specific memory subtests, other subtests of the CERAD-Plus battery also showed significant relations with Aβ1-42 as follows. For the Boston Naming Test, a significant difference between normal and pathological Aβ1-42 concentrations was detectable (ΔM = 0.87, p < 0.001, d = 0.55, Figure 2). Both parameters showed a weak but significant correlation with each other (ρ = 0.203, p = 0.005, Table 1).
The largest mean difference between patients with pathological and normal Aβ1-42 concentrations was found for the MMST (Figure 2). This test also showed a significant, albeit weak, correlation with Aβ1-42 (Table 1). A small mean difference with only a little effect size was found in Constructional Praxis Savings test (Figure 2). The weakest, but still significant correlation to Aβ1-42 was also detectable here (Table 1). For the Constructional Praxis and CDT, multiple linear regression did not reveal any relevant relations with Aβ1-42 or other CSF parameters. However, we found significant mean differences and correlations between Aβ1-42 and CDT, which were clearly less pronounced than the relations of A1-42 to the tests mentioned above. While the Aβ quotient was considered in some studies as a possible and even superior alternative to Aβ1-42 [24, 25], the correlations between this parameter and the neuropsychological test results in this study were far less pronounced than for Aβ1-42. The same applied to pTau, for which correlations and mean differences with the memory subtests could also be delineated, but these were weaker than for Aβ1-42 (Supplementary Table 3).For the biomarkers tTau, NSE and S100B there were no relevant correlations or mean differences with the neuropsychological test results (Supplementary tables 4, 5, 6).
Influence of the stage of illness: When the stage of illness was included in the analysis, significantly stronger relations emerged, particularly for Aβ1-42, in the patients with a symptom onset of no more than twelve months (Group A) than in those who already had dementia symptoms for more than a year (Group B): For the subtests Word List Learning, Word List Delayed Recall, MMST, Boston Naming Test and Constructional Praxis Savings in Group A, larger mean differences were recorded between patients with normal and pathological CSF Aβ1-42 concentrations compared to the total sample. Effect sizes of the respective mean differences, which were mostly moderate in the total sample, was now high for group A in the Word List Learning, Word List Recognition and Boston Naming Test Tests. For Group B there was no difference between normal and pathological Aβ1-42 concentrations for any of the above-mentioned tests (Figure 3).
A similar result was found for the Spearman correlations between the neuropsychological test results and the CSF Aβ1-42 concentration (Table 2). While these were weak in the overall sample for all subtests considered, there were moderate correlations in Group A for the Word List Learning, Word List Delayed Recall, MMST and Boston Naming Test. No correlations were detected for these tests in Group B. The exception here was the Word List Delayed Recall test, for which a weak but significant correlation (ρ = 0.243) was detectable, although this was weaker than the moderate correlation in Group A (ρ = 0.337, Table 2).