Baseline CSF tau and short‐term clinical change during lecanemab therapy: A prospective real‐world cohort study in Japan

Abstract

INTRODUCTION

Real‐world prognostic evidence for cerebrospinal fluid (CSF) biomarkers during lecanemab therapy remains limited. We examined whether baseline CSF phosphorylated tau 181 (pTau181), total tau, and amyloid beta (Aβ) 42/Aβ40 were associated with 6‐month clinical change.

METHODS

This prospective single‐center cohort included 50 patients with early Alzheimer's disease spectrum treated with lecanemab. CSF biomarkers were measured using the Lumipulse G system. The primary outcome was 6‐month change in Clinical Dementia Rating Sum of Boxes (CDR‐SB). Models were adjusted for age, sex, and baseline Mini‐Mental State Examination; sensitivity analyses included APOE ε4 status.

RESULTS

CDR‐SB worsening occurred in 19 participants (38.0%). Higher pTau181 (β per SD = 0.408; 95% CI: 0.106 to 0.709; p = 0.0091) and total tau (β = 0.341; 95% CI: 0.032 to 0.651; p = 0.0314) were associated with worsening. Aβ42/Aβ40 was not.

DISCUSSION

Baseline CSF pTau181 and total tau were associated with 6‐month clinical trajectories, supporting cautious cohort‐level risk stratification.

Highlights

• Baseline CSF pTau181 predicted 6‐month CDR‐SB worsening during lecanemab.

• Baseline total tau also predicted short‐term CDR‐SB worsening.

• Aβ42/Aβ40 was not associated with 6‐month clinical change.

• Apolipoprotein E (APOE)‐adjusted analyses supported the main tau findings.

• Findings require validation in larger controlled cohorts.

1. INTRODUCTION

Lecanemab reduced clinical decline versus placebo over 18 months in the phase 3 CLARITY‐AD trial, establishing anti‐amyloid protofibril therapy as a disease‐modifying option for early Alzheimer's disease (AD). ¹ However, trial populations, visit schedules, and inference windows differ from routine clinical practice, where clinicians must counsel patients and families using short‐interval information soon after treatment initiation. Real‐world prognostic evidence anchored to fluid biomarkers remains limited, particularly during the first 6 months of treatment.

Tau burden is a plausible determinant of early clinical trajectories under amyloid‐lowering therapy. In TRAILBLAZER‐ALZ 2, participants with low‐to‐intermediate tau burden derived greater clinical benefit from donanemab than those with high tau burden, supporting the concept that downstream tau pathology modulates near‐term response to anti‐amyloid treatment. ² In clinical practice, tau positron emission tomography (PET) is not widely available, whereas CSF and blood biomarkers are more feasible for patient stratification.

Automated cerebrospinal fluid (CSF) assays allow standardized measurement of amyloid and tau biomarkers. The Aβ42/Aβ40 ratio improves concordance with amyloid PET compared with amyloid beta (Aβ) 42 alone, and Lumipulse‐based Aβ42/Aβ40 cutoffs have been clinically validated. ³ , ⁴ , ⁵ International pre‐analytical guidance further supports standardized CSF handling for routine amyloid and tau measurements. ⁶ These developments provide a practical basis for evaluating CSF biomarkers as predictors of early clinical trajectories in patients treated with lecanemab.

We hypothesized that higher baseline CSF phosphorylated tau 181 (pTau181) and total tau and lower Aβ42/Aβ40 would be associated with greater 6‐month worsening in Clinical Dementia Rating Sum of Boxes (CDR‐SB) among patients receiving lecanemab, with tau biomarkers showing stronger prognostic associations than the amyloid ratio.

RESEARCH IN CONTEXT

1. Systematic review: We reviewed pivotal anti‐amyloid therapy trials, CSF biomarker validation studies, and recent real‐world lecanemab reports. Prior trials established clinical efficacy over 18 months, and tau‐burden analyses suggest that downstream tau pathology influences anti‐amyloid treatment trajectories. However, real‐world evidence linking baseline CSF tau markers to short‐term clinical change during lecanemab therapy remains limited.

2. Interpretation: In this single‐center real‐world cohort, higher baseline CSF pTau181 and total tau were associated with greater 6‐month CDR‐SB worsening, whereas Aβ42/Aβ40 showed weaker prognostic signal. The modest AUC values support cautious cohort‐level risk stratification, not individual prediction or treatment selection.

3. Future directions: Larger multicenter cohorts with longer follow‐up, standardized recording of treatment interruptions and amyloid‐related imaging abnormalities, registry‐based comparators, and decision‐curve analyses are needed before CSF tau thresholds can guide clinical decisions.

2. METHODS

Design and participants: We conducted a prospective single‐center observational cohort study at a dementia specialty center in Japan. The study included 50 outpatients who initiated lecanemab therapy between January 2024 and March 2025. Eligible participants were clinically diagnosed within the early AD spectrum, including mild cognitive impairment (MCI) due to AD or mild AD dementia, according to established National Institute on Aging‐Alzheimer's Association (AA) criteria. ⁷ , ⁸ Because the study population represented the treated early AD clinical spectrum and the sample size was limited, participants were analyzed together in the primary models. Diagnosis‐stratified models were not performed because subgroup estimates would have been unstable. Baseline clinical severity was instead accounted for by adjustment for baseline Mini‐Mental State Examination (MMSE).

All participants had biomarker evidence of amyloid pathology. Inclusion required CSF Aβ42/Aβ40 ≤ 0.067 on the Lumipulse platform, consistent with prior Lumipulse validation studies. ⁵ Patients with other major neurological, psychiatric, or systemic conditions judged to substantially affect cognition or treatment eligibility were excluded according to routine clinical practice for lecanemab initiation.

CSF collection and biomarker assays: CSF was collected under a standardized pre‐analytical protocol aligned with AA guidance. ⁶ CSF samples were centrifuged at 1,900 × g at 4°C and stored at −80°C until analysis. CSF pTau181, total tau, Aβ42, and Aβ40 were quantified using the Lumipulse G system (Fujirebio, Tokyo, Japan). The Aβ42/Aβ40 ratio was computed in assay. The platform has shown high precision and strong concordance with amyloid PET, particularly for the Aβ42/Aβ40 ratio. ³ , ⁴ , ⁵

Lecanemab treatment: Lecanemab was administered intravenously at 10 mg/kg every 2 weeks according to the approved dosing regimen. Clinical monitoring and safety assessments were performed in routine care. Detailed comorbidity burden, vascular imaging markers, concomitant medications, and treatment interruptions were not systematically quantified for the present analysis.

Outcomes: The primary outcome was 6‐month change in CDR‐SB, defined as CDR‐SB at 6 months minus baseline CDR‐SB. CDR‐SB worsening was defined as ΔCDR‐SB > 0. Secondary clinical measures included 6‐month MMSE and ΔMMSE.

Statistical analysis: Baseline characteristics and 6‐month outcomes were summarized for all participants. Because MCI/AD subgroup assignment was not used as a primary analytic factor, descriptive tables were presented for the full cohort only. Continuous variables are reported as mean ± SD and categorical variables as n/N (%).

Primary ordinary least‐squares (OLS) models tested associations between each baseline CSF biomarker and ΔCDR‐SB, adjusted for age, sex, and baseline MMSE. Biomarkers were standardized to allow effect estimates per 1‐SD increase. Logistic regression models evaluated CDR‐SB worsening (ΔCDR‐SB > 0) with the same covariates. Because prior evidence supports higher tau and lower Aβ42/Aβ40 as risk markers for clinical worsening, directional one‐sided p values were used for pre‐specified directional hypotheses; two‐sided p values and 95% confidence intervals (CIs) are also reported to support transparent interpretation. ¹ , ² , ⁹ , ¹⁰ , ¹¹ , ¹² Receiver operating characteristic (ROC) curves and area under the curve (AUC) were calculated from model‐predicted probabilities. Youden's J was used to derive descriptive thresholds, and DeLong's method was used for ROC comparisons. ¹³ , ¹⁴ , ¹⁵

Apolipoprotein E (APOE) genotyping was available in 43 of 50 participants. To address missing APOE data, participants with available versus missing APOE genotyping were compared across baseline clinical variables, CSF biomarkers, and 6‐month outcomes. Sensitivity analyses were then performed among APOE‐available participants by adding APOE ε4 carrier status to the primary OLS and logistic regression models. These models were interpreted as sensitivity analyses because of the reduced sample size and limited number of CDR‐SB‐worsening events.

To evaluate whether binary Youden‐derived thresholds oversimplified biomarker–outcome relationships, exploratory three‐range analyses were performed. Baseline pTau181 and total tau were divided into tertiles, and mean ΔCDR‐SB was summarized across low, intermediate, and high tertile groups. Linear trend across tertiles was assessed using an ordinal tertile variable. Because externally validated three‐zone CSF tau thresholds for early lecanemab outcomes are not established, these tertile analyses were considered exploratory.

Analyses were performed using Python 3.11 with pandas, numpy, statsmodels, scikit‐learn, scipy, and matplotlib.

Ethics: The study was approved by the ethics committees of Kobe University Graduate School of Health Sciences (Approval No. 1073) and Hyogo Prefectural Rehabilitation Hospital at Nishi‐Harima (Approval No. 2107). All participants or their legal representatives provided written informed consent.

3. RESULTS

Participant characteristics: A total of 50 participants were included in the analysis. Baseline demographic, clinical, and CSF biomarker characteristics are summarized in Table 1. Mean age was 74.7 ± 4.2 years, and 18 participants were male. Baseline MMSE was 24.54 ± 1.96, baseline CDR‐SB was 2.33 ± 1.17, and baseline Alzheimer's Disease Assessment Scale Cognitive Subscale (ADAS‐Cog) was 9.69 ± 3.24. APOE genotyping was available in 43 participants; among them, 29 (67.4%) were APOE ε4 carriers. Baseline CSF Aβ42/Aβ40 was 0.047 ± 0.009, total tau was 637.7 ± 283.7 pg/mL, and pTau181 was 95.7 ± 46.0 pg/mL.

TABLE 1.

Baseline demographic, clinical, and CSF biomarker characteristics.

Variable	All participants (n = 50)
Age, years	74.7 ± 4.2
Sex, male/female	18/32
Baseline MMSE	24.54 ± 1.96
Baseline CDR‐SB	2.33 ± 1.17
Baseline ADAS‐Cog	9.69 ± 3.24
APOE ε4 carrier, n/N (%)	29/43 (67.4%)
Aβ42/Aβ40	0.047 ± 0.009
Total tau, pg/mL	637.7 ± 283.7
pTau181, pg/mL	95.7 ± 46.0

Six‐month clinical outcomes: Six‐month clinical outcomes are shown in Table 2. Mean CDR‐SB at 6 months was 2.31 ± 1.17, and mean ΔCDR‐SB was −0.02 ± 1.14. Nineteen of 50 participants (38.0%) showed CDR‐SB worsening, defined as ΔCDR‐SB > 0. Mean MMSE at 6 months was 23.14 ± 2.83, and mean ΔMMSE was −1.40 ± 2.44. These results indicate that overall group‐level CDR‐SB change over 6 months was small, while individual trajectories varied.

TABLE 2.

Six‐month clinical outcomes.

Outcome	All participants (n = 50)
CDR‐SB at 6 months	2.31 ± 1.17
MMSE at 6 months	23.14 ± 2.83
ΔCDR‐SB (6 months – baseline)	−0.02 ± 1.14
ΔMMSE (6 months – baseline)	−1.40 ± 2.44
CDR‐SB worsening (Δ > 0), n/N (%)	19/50 (38.0%)

Associations between baseline CSF biomarkers and 6‐month CDR‐SB change: Unadjusted scatter plots of baseline CSF biomarkers versus ΔCDR‐SB are shown in Figure 1. In primary OLS models adjusted for age, sex, and baseline MMSE, higher baseline pTau181 and total tau were associated with greater 6‐month CDR‐SB worsening (Table 3). Per 1‐SD increase, pTau181 showed a standardized β of 0.408 (95% CI: 0.106 to 0.709; two‐sided p = 0.0091; directional one‐sided p = 0.0046). Total tau showed a standardized β of 0.341 (95% CI: 0.032 to 0.651; two‐sided p = 0.0314; directional one‐sided p = 0.0157). Aβ42/Aβ40 was not significantly associated with ΔCDR‐SB (standardized β = 0.094; 95% CI: −0.239 to 0.426; two‐sided p = 0.573).

FIGURE 1.

Baseline CSF biomarkers and 6‐month change in CDR‐SB. (A) Aβ42/Aβ40, (B) total tau, and (C) pTau181 are plotted against ΔCDR‐SB (6 months – baseline). Solid lines indicate least‐squares fits, and shaded bands indicate 95% confidence intervals.

TABLE 3.

Primary biomarker models for 6‐month CDR‐SB change and worsening.

Biomarker	OLS β per SD (95% CI)	OLS p	Logistic OR per SD (95% CI)	Logistic p	AUC	Youden cutoff
pTau181	0.408 (0.106 to 0.709)	0.0091	2.50 (1.19 to 5.24)	0.0153	0.730	85.45 pg/mL
Total tau	0.341 (0.032 to 0.651)	0.0314	2.20 (1.11 to 4.39)	0.0246	0.715	520.5 pg/mL
Aβ42/Aβ40	0.094 (−0.239 to 0.426)	0.5734	1.58 (0.84 to 2.96)	0.1573	0.633	0.0575

Logistic models for CDR‐SB worsening: In logistic regression models for CDR‐SB worsening, pTau181 and total tau showed modest discrimination (Table 3; Figure 2). Per 1‐SD increase, pTau181 was associated with higher odds of worsening (odds ratio [OR] 2.50; 95% CI: 1.19 to 5.24; p = 0.0153), with an AUC of 0.730. Total tau was also associated with worsening (OR 2.20; 95% CI: 1.11 to 4.39; p = 0.0246), with an AUC of 0.715. The Aβ42/Aβ40 ratio showed weaker discrimination (AUC = 0.633). ROC‐derived Youden cutoffs were 85.45 pg/mL for pTau181, 520.5 pg/mL for total tau, and 0.0575 for Aβ42/Aβ40. These thresholds are descriptive and internally derived.

FIGURE 2.

ROC curves for 6‐month CDR‐SB worsening using primary models. Curves show discrimination for baseline pTau181, total tau, and Aβ42/Aβ40 in logistic regression models adjusted for age, sex, and baseline MMSE. AUCs are shown in the legend.

Threshold‐based trajectories: Threshold‐based group trajectories are shown in Figure 3. Participants with baseline total tau above 520.5 pg/mL and those with pTau181 above 85.45 pg/mL showed greater mean 6‐month CDR‐SB worsening than participants below these cutoffs. Because these thresholds were derived within the present cohort, they should not be interpreted as validated clinical decision thresholds.

FIGURE 3.

Threshold‐based group trajectories from baseline to 6 months. Patients were stratified by ROC‐derived cutoffs for total tau (520.5 pg/mL) and pTau181 (85.45 pg/mL). Lines show mean ± SEM of ΔCDR‐SB with zero as baseline reference.

APOE missingness and APOE‐adjusted sensitivity analyses: Participants with available APOE genotyping (n = 43) and those with missing genotyping (n = 7) did not show apparent differences in baseline severity, CSF biomarker levels, or 6‐month outcomes (Table S1). ΔCDR‐SB was −0.01 ± 1.18 in APOE‐available participants and −0.07 ± 0.93 in participants with missing APOE data (p = 0.883). CDR‐SB worsening occurred in 17 of 43 participants (39.5%) with APOE data and two of seven participants (28.6%) with missing APOE data (p = 0.695). In APOE‐adjusted sensitivity analyses, pTau181 and total tau remained associated with ΔCDR‐SB and CDR‐SB worsening, whereas Aβ42/Aβ40 did not show a significant association (Table S2; Figure S1).

Exploratory three‐range analyses are summarized in Table S3 and Figure S2. For total tau, mean ΔCDR‐SB increased across low, intermediate, and high tertiles (−0.50, −0.06, and +0.50, respectively), with a significant ordinal trend (β per tertile = 0.500; trend p = 0.0090). For pTau181, mean ΔCDR‐SB was −0.18, −0.28, and +0.38 across low, intermediate, and high tertiles, respectively, without a statistically significant monotonic trend (β per tertile = 0.279; trend p = 0.1543). These analyses suggest that total tau may show a more graded relationship with short‐term clinical change, while pTau181 showed greater worsening in the highest tertile. The small tertile group sizes limit inference.

4. DISCUSSION

In this prospective real‐world cohort of patients receiving lecanemab, higher baseline CSF pTau181 and total tau were associated with greater heterogeneity in 6‐month CDR‐SB trajectories after adjustment for age, sex, and baseline MMSE. The overall mean CDR‐SB change was small, but 38.0% of participants showed worsening by ΔCDR‐SB > 0. These findings support CSF tau as a candidate marker for early cohort‐level risk stratification among treated patients, while the internally derived AUCs around 0.7 do not support stand‐alone individual‐level prediction or treatment selection.

The findings presented here should be interpreted as prognostic associations within a treated cohort, not as evidence that CSF tau modifies lecanemab efficacy. CLARITY‐AD used an 18‐month primary endpoint and demonstrated group‐level treatment benefit versus placebo over that period. ¹ By contrast, the analysis presented here evaluated a 6‐month window without an untreated control group. Therefore, the observed biomarker‐associated trajectories cannot determine whether lecanemab altered the natural short‐term course in any biomarker‐defined subgroup.

Our results are consistent with the broader concept that downstream tau burden shapes clinical trajectories in the anti‐amyloid treatment era. In TRAILBLAZER‐ALZ 2, tau burden influenced clinical benefit from donanemab. ² The CSF‐based findings discussed here extend this concept to routine lecanemab practice, where tau PET is often unavailable. However, the findings do not establish an actionable CSF tau cutoff. The ROC‐derived thresholds were internally derived and require external validation before clinical use.

The findings also align with a recent real‐world lecanemab report from Kanazawa University, which found that higher phosphorylated tau levels were associated with cognitive decline and amyloid‐related imaging abnormalities during lecanemab treatment. ¹⁶ Although the outcome definitions and sampling frame differ from those in the present study, both reports support the relevance of baseline tau‐related biomarker burden when interpreting early clinical and safety trajectories during lecanemab therapy. The convergence should be considered preliminary because both datasets are observational and susceptible to selection, monitoring, and treatment‐interruption biases.

pTau181 and total tau should nevertheless be interpreted differently. pTau181 is more closely related to AD‐type tau phosphorylation, whereas total tau is a broader and less disease‐specific marker of neuronal injury and neurodegeneration. Therefore, the total tau association may reflect baseline neuronal injury burden, vulnerability to clinical decline, or co‐pathology rather than AD‐specific tau pathology alone. The stronger ordinal trend in the total tau tertile analysis is biologically plausible but exploratory.

APOE‐adjusted sensitivity analyses addressed an important reviewer concern. APOE ε4 status was available in 43 participants, and the associations of pTau181 and total tau remained directionally consistent after adding APOE ε4 carrier status to the models. However, these models had fewer observations and 17 worsening events, so estimates are less stable than those from the primary models. APOE missingness comparisons did not show apparent baseline or outcome differences, but the missing subgroup included only seven participants.

The exploratory three‐range analyses were added because binary biomarker thresholds may oversimplify risk relationships. Total tau showed a graded tertile association with ΔCDR‐SB, whereas pTau181 showed greater worsening in the highest tertile without a significant monotonic trend. These findings support reporting biomarker distributions rather than relying only on single cutoffs. They do not establish validated low‐, intermediate‐, or high‐risk clinical categories.

Several limitations should be noted. First, the sample size was modest, and all analyses were performed in a single center. Second, the study lacked an untreated control group, which precludes causal inference about treatment efficacy or effect modification by biomarkers. Third, the follow‐up period was limited to 6 months, whereas pivotal treatment effects in randomized trials were assessed over longer intervals. ¹ Fourth, detailed comorbidity burden, vascular imaging markers, concomitant medications, treatment interruptions, and amyloid‐related imaging abnormalities were not systematically quantified for the present analysis. Fifth, ROC‐derived thresholds and tertile categories were internally derived and require external validation. Finally, APOE‐adjusted analyses were limited by missing genotyping and low event counts.

In conclusion, baseline CSF pTau181 and total tau were associated with heterogeneity in 6‐month clinical trajectories among patients receiving lecanemab in routine practice, whereas Aβ42/Aβ40 showed weaker prognostic signal within this amyloid‐positive cohort. CSF tau may support cautious, cohort‐level early risk stratification and monitoring, but the data discussed here do not justify stand‐alone individual prediction, treatment selection, or causal claims about lecanemab efficacy. Longer follow‐up, external validation, untreated or registry‐based comparators, and decision‐curve analyses are needed before clinical implementation.

CONFLICT OF INTEREST STATEMENT

Ryuichi Takahashi received lecture honoraria from Eisai Co., Ltd. and Eli Lilly Japan K.K.; payments were made to him. Eiji Mizuta received speaker bureau honoraria from Eisai Co., Ltd.; payments were made to him. Hisatomo Kowa received honoraria for presentations from Eisai Co., Ltd.; payments were made to him. Tetsuo Kashibayashi received lecture honoraria from Eisai Co., Ltd. and Eli Lilly Japan K.K.; payments were made to him. Tetsuo Kashibayashi also serves as an unpaid councilor of the Japanese Psychogeriatric Society. Akira Hashiramoto, Jun Fujita, and Kohsuke Yoshida declare no relationships, activities, or interests related to the content of this manuscript. Author disclosures are available in the Supporting Information.

CONSENT STATEMENT

All participants or their legal representatives provided written informed consent.

Supporting information

Supporting Information

Supporting Information

Supporting Information

Supporting Information