Predictive value for cerebrospinal fluid Alzheimer's disease profile of different measures of verbal episodic memory in patients with MCI

Neuropsychological evidence of memory impairment represents the main feature of the clinical onset of typical Alzheimer’s disease (AD). Rey’s Auditory Verbal Learning Test (RAVLT) and Logical Memory (LM) are two tests both assessing verbal episodic memory, widely used in clinical practice. Our aim was to investigate the added value of their combined use in predicting cerebrospinal fluid (CSF) AD biomarkers positivity in a retrospective consecutive series of patients with mild cognitive impairment (MCI). 169 MCI patients were included. For all of them neuropsychological assessment and CSF analysis were available. According to CSF A/T/(N) profile, 109 were defined as MCI due to AD (A+T+), and 60 were non-AD MCI (A−T−). Logistic regression model and receiver-operating characteristic (ROC) curves were analyzed to evaluate the discriminatory power of single and combined sub-measures between AD and non-AD patients. The combination of RAVLT-del with LM could acceptably discriminate the two groups (AUC: 0.69, CI 95% 0.617–0.761, sens: 0.75, spec. 0.58, p < 0.001), while the single tests did not show sufficient discriminative performance. Our study shows that the combination of RAVLT delayed recall with LM better predicts the biological AD diagnosis (A+T+), showing a good discriminative power between MCI-AD from non-AD MCI. Since RAVLT and LM assess different components of verbal episodic memory, they should be considered as complementary, rather than interchangeable, tests.

converted to dementia displayed a significant forgetting in both tests, showing no benefit of semantically organized structure of the story 18,38 .
In sum, some evidences indicate that the combination of list learning with story recall can be of help in detecting episodic memory impairment.However, the predictive value of these neuropsychological measures can be assessed if available the pathophysiological causes of MCI, with special interest on MCI due to AD.The aim of the present study is to investigate the role of the sub-measures of these two memory tests in predicting the CSF AD biomarkers positivity in a retrospective, consecutive cohort of well-characterized MCI patients, referring to our Memory Clinic for diagnostic assessment.

Study population
Our population consisted of 169 patients referring to the Centre for Memory Disturbances, Section of Neurology, Perugia University Hospital from 2016 to 2022.Each patient underwent clinical evaluation, comprehensive neuropsychological assessment and lumbar puncture for CSF profiling according to the A/T/(N) system 13 .All patients with Mini-Mental State Examination adjusted score above the cut-off (≥ 23.8) 41 , impairment of at least one cognitive domain defined as Equivalent Score (ES) = 0, and a Clinical Dementia Rating scale global score = 0.5 were classified as MCI.

Neuropsychological testing
Each patient underwent a baseline comprehensive neuropsychological assessment, including: Mini-Mental State Examination (MMSE) 41 for assessing global cognition; Trail Making Test (TMT) part A and B 42 and the Frontal Assessment Battery (FAB) 43 for assessing attention and executive functions; the Rey's Auditory Verbal Learning Test (RAVLT, with sub-measures of immediate and delayed recall, true and false recognitions) 32 and the Short story test of "Anna Pesenti" (immediate and delayed recall) 34 for the assessment of verbal episodic memory; the digit span forward and backward 44 for the assessment of short-term verbal memory and working memory; 1-min phonemic fluency 32 and category word fluency 34 for assessing language; copy of drawings with and without landmarks from the Mental Deterioration Battery (MDB) 32 and the Clock Drawing Test (CDT) 45 for the assessment of visuo-constructional abilities; the Raven Progressive Matrices (MP'47) 46 for the assessment of logical-perceptual reasoning.The neuropsychological assessment also included the Clinical Dementia Rating Scale (CDR) 47 for staging functional decline.MCI subtype was classified according to neuropsychological profile observed as follows: amnestic-single-domain MCI (a-sd MCI) if only the memory domain was impaired; amnestic-multidomain MCI (a-md MCI) if memory and at least another domain were involved; if memory domain was spared but other domains were impaired, patients were classified as non-amnestic MCI (na-MCI), including both the non-amnestic single domain MCI (na-sd-MCI) and non-amnestic multi-domain MCI (na-md-MCI) subtype 6 .

Sub-measures of verbal episodic memory tests
Among the neuropsychological tests, we analyzed the scores obtained at RAVLT and LM.The immediate recall of RAVLT (RAVLT-imm) consists in the total number of words recalled during the 5 learning trials (0-75).The delayed recall of RAVLT (RAVLT-del) consists in the number of words spontaneously recalled after the 15-min interval (0-15).The True and False Recognitions (RAVLT-true, RAVLT-false) are scored as the number of true hits (0-15) and false alarms (0-31) during the recognition trial.The short story recall (LM) total score was calculated as the average between the immediate and the 10-min delayed recall (0-28).

Lumbar puncture and CSF analysis
Lumbar puncture (LP) was performed according to international guidelines 48 .Briefly, 10-12 mL of CSF were collected in sterile polypropylene tubes and centrifuged at room temperature for 10 min (2000 × g).Aliquots (0.5 mL) were frozen at − 80 °C.CSF analysis included routine chemical-physical parameters (glucose and total proteins) and cell count.Blood-contaminated samples, i.e., more than 50 red cells/μL were excluded from the analysis.Aβ40, Aβ42, t-Tau and p-Tau were analyzed using Lumipulse G600-II fully automated chemiluminescent enzyme immunoassay system in our Lab. of Clinical Neurochemistry.The CSF A/T/(N) profile was subsequently considered for all patients, and A+/T+ profile was considered as CSF AD-like profile.According to the cut-off values calculated in our Lab 49 , "A+" corresponds to a CSF Aβ42/40 ratio < 0.072, and "T+" corresponds to a CSF phospho-tau > 50 pg/mL.

Ethical declaration
Since 2008, CSF collection for the early diagnosis of Alzheimer's disease is routinely performed in our Center, as approved by the local Ethics Committee (CER Umbria, Protocol N° 19369/08/AV, registry N. 1287/08, date: 9 October 2008).All patients gave their written informed consent for the study participation.The research was performed in accordance with the Declaration of Helsinki.

Statistical analyses
Statistical analyses were conducted via Statistical Package for the Social Sciences (IBM SPSS).We first analyzed the differences between the two cohorts in age, MMSE, RAVLT and Short story sub-measures.
Wilcoxon Mann-Whitney U test was used to determine the power of each RAVLT and Short story submeasure to discriminate between patients in the AD group (MCI-AD) from the subgroup with negative biomarkers profile (non-AD MCI).
ROC curves were generated to estimate the power of each RAVLT and Short story sub-measure to discriminate between patients with MCI-AD from non-AD MCI.
A stepwise logistic regression model was used to investigate the power of combined RAVLT and Short story sub-measures to discriminate between patients with MCI-AD from non-AD MCI.
We also investigated the predictive value of the total impairment in multiple sub-measures, calculated as cumulative score derived by the number of memory sub-measures impaired (0-5), in which a score of 1 is attributed to performances below current cutoffs and 0 to measures within normal range, and the biological diagnosis of AD based on CSF biomarkers.This index has been proposed as exploratory analysis that holds together all the sub-measures of the two episodic memory tests (RAVLT-imm, RAVLT-del, RAVLT-True, RAVLT-False, LM).A p-value < 0.05 was considered statistically significant.All analyses were performed using IBM SPSS V.25.0.0.

Demographical, clinical and CSF features
Demographic data, clinical features, and biomarkers values are listed in Table 1.

Whole MCI group
The mean age of the whole MCI group (n.169; 73 M, 96 F) was 72.34 ± 5.35 years, with mean years of education of 10.56 ± 4.3.Concerning the clinical subtype of MCI patients, the majority of them showed an amnestic phenotype (n.140, 83%).Among aMCI, the wide majority (113/140, 80%) showed multiple domain, being classified as md-a-MCI.This subgroup represented the most prevalent in the whole cohort (67%).

MCI-AD vs. non-AD MCI
Based on CSF profile, 109 were classified as MCI due to AD (A+T+ , 64.5%), and 60 as non-AD MCI (A−T−, 35.5%).No differences were found between the two groups in terms of age, education and gender.With respect to the distribution of the clinical subtypes (amnestic/non amnestic, single/multiple domain), we found that the great majority of MCI-AD (91%), and more than half of the non-AD MCI group (67%) showed memory impairment.
By using the Wilcoxon Mann-Whitney U test, we found that MCI-AD patients showed worse scores on MMSE, RAVLT-imm, RAVLT-del, RAVLT-false and LM with respect to non-AD MCI (p < 0.05, see Table 1).

Predictive value of neuropsychological measures
We used a stepwise logistic regression model to investigate the predictive value of the memory tests sub-measures for AD diagnosis.RAVLT-del and LM reached statistical significance in predicting the AD diagnosis (A+/T+) www.nature.com/scientificreports/(RAVLT-del: p = 0.0091; LM: p = 0.05).The combination of both measures showed the best performance in discriminating the two groups (AUC: 0.69, CI 95% 0.617-0.761,p < 0.001) (see Fig. 1).We also applied stepwise logistic regression model to evaluate the association of each memory test with any single biomarker.We thus found that positivity of amyloidosis (A+) is predicted by RAVLT-false (p = 0.04) and LM (p = 0.005).The combination of LM and RAVLT-false gives the best predictive value (p < 0.001).RAVLT-del predicts positivity of tauopathy (T+, p = 0.004) and neurodegeneration (N+, p = 0.003).

Discussion
The aim of the present study was to investigate the predictive value of two traditional tests assessing verbal episodic memory (the RAVLT and the short story recall, LM) and their distinct sub-measures on CSF AD profile in a retrospective, consecutive cohort of MCI patients including MCI AD and non-AD MCI.As expected, the two subgroups significantly differed in the mean scores on MMSE, RAVLT-imm, RAVLT-del, RAVLT-false and LM.
In our cohort, 93% of patients with MCI due to AD showed deficits in memory domain.Of them, 18% showed single domain amnestic MCI, vs. 82% showing multi-domain amnestic MCI.These results are in line with previous evidence indicating that multi-domain amnestic profile is much more common than single-domain amnestic phenotype [50][51][52] .
Our main finding is that the combination of two distinct verbal episodic memory tests reached the highest sensitivity in predicting CSF AD profile, i.e.A+/T+.Specifically, the RAVLT-del and LM better predicted CSF AD profile, with a fair discriminative power between MCI-AD and non-AD MCI.Such results are in line with previous evidence of a specific neuropsychological pattern of episodic memory impairment in patients with MCI due to AD. Subjects with a "pure" amnestic syndrome of hippocampal type showed a consistent rate of forgetting in a word list and in a story recall, as reflected by impaired delayed recall at both tests 18,53 .However, poor delayed recall per se is not specific for AD and the use of a single cognitive measure may lead to misclassification of MCI.In fact, a single-measure/single-test approach, as well as the use of a single test to assess a cognitive domain, has been largely criticized as it may lead either to under-or over-estimation of cognitive decline 8 .
Free and Cued Selective Reminding Test (FCSRT) is another widely used tool for the assessment of hippocampal memory, based on encoding specificity paradigm 14,24 .Total (free + cued) recall of FSCRT has been found to be associated with anterior MTL atrophy 24 and CSF AD profile 54 .Due to the intrinsic paradigm used, the FCSRT is not properly an episodic memory test 24 .These findings support the importance, in clinical practice, of the combined use of memory tests giving complementary information, such as LM and RAVLT.When considering the association of neuropsychological measures with the single CSF biomarkers, we observed different behaviors.The delayed recall of a word list correlated with CSF p-Tau, while the short story recall was found to be associated with CSF Aβ42/40.Previous studies investigating the association of specific measures of episodic memory with CSF AD biomarkers reported contradictory results [55][56][57][58][59][60][61] .
The tendency to produce false positives during recognition tasks is a key feature of the amnestic syndrome of hippocampal type underlying the episodic memory deficits in AD.Amnestic MCI patients showing impaired free delayed recall associated with a recognition deficit may have a more profound consolidation deficit ("encoding/ storage" pattern), distinct from those with impaired delayed recall with spared recognition ("retrieval" deficit) 50 .Such syndrome characterizing the clinical onset of typical AD encompasses the lack of benefit by cue associated with tendency to false positives in recognition word list tasks.False recognitions are commonly defined as false memories or confabulations, a genuine memory dysfunction.However, other explanations could be taken into account.A recent paper clarified the distinction between false recall, as tendency to produce self-generated wrong responses, and false alarms, induced by external stimuli in specific contexts.Some evidences support the role of prefrontal regions, as ventrolateral prefrontal cortex, inferior frontal gyrus and anterior cingulate cortex, to avoid false alarms during cognitive tasks.Such regions offer specific contributions to executive processes involved in this behavior: inhibitory control and suppression of inappropriate information, strategic encoding, sustained attention, updating and control over interference driven by working memory.In this perspective, false recognitions can be interpreted as inaccurate commission of a response 30 .
Overall, our data have implication for routine clinical practice.A thorough neuropsychological assessment combining multiple cognitive measures is required to correctly define MCI, in order to identify those subjects requiring biomarkers assessment.In our cohort, the combined use of RAVLT and LM sub-measures discriminated patients with MCI due to AD from patients with MCI not related to AD.In the era of forthcoming diseasemodifying therapies, which have the best chance to be effective the earlier they are applied, it is mandatory to identify AD patients in the prodromal phase of disease.
The present study has some limitations.It represents a retrospective investigation in a cohort of MCI patients subgrouped, according to the CSF profile, in MCI-AD (A+/T+ , n. 109) and non-AD MCI (A-/T-, n. 60).As expected, approx.2/3 of patients referred to our Memory Clinic due to memory complaints, were affected by AD.We did not carry out specific statistical analysis according to single/amnestic or multi-domain MCI pattern.Only some quantitative measures from the two tests assessing verbal episodic memory were considered, in order to evaluate their predictive value on CSF AD profile; thus, qualitative parameters (e.g.primacy and recency effect, retroactive and proactive interference) were not included in the analysis.
In conclusion, in our study, the use of two specific tests of verbal episodic memory (RAVLT and LM) contributed to highlight the limits of the one-test approach for the correct evaluation of episodic memory deficits in MCI patients.Most importantly, the combined use of RAVLT and LM showed the best capacity to predict CSF AD profile.These results further support previous findings about the complementary and not interchangeable nature of the most widely used tests for episodic memory 18,40 .
In clinical practice, MCI patients showing altered delayed recall sub-measures at RAVLT and LM should undergo CSF analysis in order to rule out the presence of Alzheimer's disease.

Figure 1 .
Figure 1.Comparison between ROC curves generated to discriminate between patients with an A+/T+ CSF profile and non-AD patients: RAVLT-del alone, Short story recall (LM) alone and the one generated with the combined use of both sub-measures.