The relationships between cerebrospinal fluid neurofilament light chain and hippocampal atrophy with cognitive decline

Abstract

Background:

Identifying predictive biomarkers of cognitive decline is critical for timely intervention in early Alzheimer’s disease and related dementia. Biomarkers such as cerebrospinal fluid (CSF) neurofilament light (NfL), and MRI-based hippocampal atrophy are potential indicators of neurodegeneration, but their long-term predictive value remains unclear.

Objective:

This study examined 20-year longitudinal associations between CSF NfL, MRI-based hippocampal atrophy, and cognitive decline in cognitively normal older adults.

Methods:

A cohort of 279 cognitively normal adults aged ≥55 years was followed from 2003 to 2023 at the Knight ADRC. Participants underwent annual cognitive and neurological assessments, including Clinical Dementia Rating (CDR), CSF NfL quantification, and MRI-based hippocampal volumetry. Cognitive decline was defined as: (1) first progression (CDR ≥ 0.5) and (2) sustained progression (two consecutive CDRs ≥ 0.5). Analyses included Kaplan-Meier survival, Cox proportional hazards models, and linear mixed-effects (LME) models.

Results:

Participants had a mean age of 66.5 years (SD = 6.08); 58.4% were female. Mean follow-up was 11.41 years (SD = 3.5). First progression occurred in 71 participants (25.4%), and sustained progression in 35 (13%). Higher CSF NfL levels were associated with faster time to first (95% CI:0.2–1; p < 0.001) and sustained progression (95% CI:0.46–1; p = 0.008). Cox models showed increased risk of first progression (HR = 1.83; 95% CI: 1.11–3.01; p = 0.018) but not sustained (p = 0.093). LME models showed CSF NfL increase and hippocampal volume decline (p < 0.001) in both outcomes.

Conclusions:

CSF NfL is a strong predictor of cognitive decline and may serve as a screening biomarker for early dementia risk.

Introduction

Neurodegeneration, characterized by the progressive loss of neurons, is an insidious process underlying age-related cognitive decline and diseases like dementia.¹ When neurodegeneration occurs, neurofilament light (NfL) is released into the extracellular space and cerebrospinal fluid (CSF), providing a measurable indicator of axonal damage.² As neuronal damage accumulates, cognitive impairment worsens, progressively interfering with daily functioning.³ Among neurodegenerative diseases, Alzheimer’s disease (AD) accounts for 60%–80% of dementia cases. Vascular dementia comprises 5%–10%, frontotemporal degeneration 3%–10%, and hippocampal sclerosis 3%–13% (0.4%–2% when occurring alone). Dementia with Lewy bodies accounts for approximately 5%, while mixed dementia is present in more than 50% of AD cases. Parkinson’s disease dementia occurs in 3.6% of cases, with 24.5% of individuals with Parkinson’s disease eventually developing dementia.⁴ CSF NfL is elevated in each of these conditions, reflecting its sensitivity to diverse neurodegenerative processes beyond AD. These prevalence estimates highlight the complex and multifaceted nature of dementia, where multiple pathological processes may contribute to cognitive decline.

AD is characterized by the accumulation of amyloid-β (Aβ), a hallmark pathological feature of the disease. However, emerging research suggests that neurodegeneration is not solely driven by amyloid pathology.⁵ Structural brain changes, such as atrophy, vascular damage, and neurofibrillary tangle aggregation in the medial temporal lobe, can occur independently of Aβ deposition.¹ Additionally, hippocampal volume loss has been extensively studied as a key biomarker of neurodegeneration in AD and other dementias, offering crucial insights into disease progression.⁶ Furthermore, increasing evidence suggests that neuroinflammation and irreversible neuronal damage may begin years or even decades before clinical symptoms emerge in individuals with AD and other dementia etiologies.⁷ Despite this, many patients receive a diagnosis only after substantial neurodegeneration has occurred, leading to poor clinical outcomes and limited treatment options.⁸ The limitations of current diagnostic approaches are further exemplified by studies indicating that 16% of patients with a clinical diagnosis of AD are amyloid-negative, calling into question the accuracy of existing diagnostic frameworks.⁹ Similarly, postmortem studies reveal that 14% of individuals diagnosed with probable AD had little to no Aβ plaques, while neuroimaging studies indicate that 15% of clinically diagnosed AD cases are amyloid-negative on positron emission tomography (PET) scans.¹⁰ These findings underscore an urgent need to move beyond an amyloid-centric diagnostic model and explore additional pathways contributing to neurodegeneration.

In 2018, the National Institute on Aging and the Alzheimer’s Association (NIA-AA) introduced the AT(N) framework, which categorizes AD based on three fundamental biological markers: Aβ (A), tau (T), and neurodegeneration (N).¹ This biomarker-driven paradigm represents a significant shift from traditional symptom-based diagnosis, facilitating early identification of individuals at risk for AD, enabling timely interventions, and promoting precision medicine approaches in dementia research. Although tau biomarkers have been extensively studied and closely linked to cognitive decline in dementia, research specifically focused on neurodegeneration markers remains limited. Our study addresses this gap by investigating the independent prognostic value of neurodegeneration (N), operationalized as CSF NfL and hippocampal atrophy, independently of amyloid and tau status. CSF NfL, a sensitive and broadly applicable biomarker for axonal injury, may capture early neurodegenerative processes preceding overt tau accumulation, highlighting its potential as a critical early indicator of AD-related pathology.¹¹

Given the limited efficacy and affordability of disease-modifying treatments, early detection of neurodegeneration is essential for implementing timely interventions that may slow disease progression. Identifying individuals at risk for cognitive decline allows for the development of personalized management strategies that address modifiable risk factors such as hypertension, diabetes, and lifestyle choices.¹²

This study investigates the long-term relationship between CSF NfL, hippocampal atrophy, and cognitive decline over a 20-year follow-up period. Given their direct association with neuroaxonal injury, we hypothesize that elevated baseline CSF NfL levels and reduced hippocampal volume will independently predict cognitive decline. Furthermore, we assert that CSF NfL has the potential to serve as a screening tool for dementia, not only in AD but also across multiple neurodegenerative conditions, making it a valuable biomarker for both clinical applications and clinical trials.

Methods

Study design and participants

Participants were enrolled in a longitudinal study in the Charles F. and Joanne Knight Alzheimer’s Disease Research Center (Knight ADRC) at Washington University in St Louis and completed annual clinical and cognitive assessments with biomarkers obtained every two to three years. Sociodemographic characteristics, including age, sex, education, and APOE ϵ4 allele, were recorded. Participants had to be cognitively normal (Clinical Dementia Rating CDR^® = 0) at baseline, be 55 years or older, have two or more CSF biomarkers and MRI-based volumetric measurements to be included in this study, have no psychiatric or neurological disorders, be committed to follow-up, and be proficient in English. Exclusions included severe medical conditions limiting life expectancy (e.g., advanced cancer), MRI contraindications (e.g., pacemakers, implants), recent substance abuse, medications affecting cognition, or traumatic brain injury with loss of consciousness. This study received Institutional Review Board (IRB) approval at Washington University in St Louis and adhered to the principles outlined in the Declaration of Helsinki. All participants provided written consent, ensuring the confidentiality of their data and adherence to ethical standards. This report followed the STROBE guidelines for cohort studies.

Biomarker assessments

CSF samples were collected via lumbar puncture following overnight fasting at a standardized time (approximately 8:00 AM) using atraumatic needles. The samples were processed within 2 h to remove cellular debris, aliquoted into polypropylene tubes, and stored at −80°C to preserve biomarker stability. All samples were analyzed in a single batch using a commercial ELISA kit (UMAN Diagnostics, Umeå, Sweden) and a single lot of reagents to minimize assay drift and inter-lot variability. To control for potential confounding due to lot differences over time, a set of previously analyzed bridging samples was re-assayed and used for regression-based normalization. Internal pooled CSF controls and external peptide standards were used in each run to ensure assay reliability, with intra- and inter-assay coefficients of variation maintained below 6%. CSF analytes (Aβ, tau, p-tau, and NfL) were quantified per the manufacturer’s protocols.^13,14 Based on prior early work with CSF biomarkers, a tertile approach was used, with participants classified into high-risk (top 30%) and low-risk (bottom 70%) groups based on baseline CSF NfL distribution.^15–17

MRI scans were used for co-registration and anatomical alignment of PET images to ensure accurate localization and quantification of amyloid burden. All structural MR scans (T1w and FLAIR) were acquired on a BioGraph mMR PET-MR 3 T and a Siemens TIM Trio 3 T MRI scanner using a protocol designed to match the ADNI-2 MRI protocol.¹⁸ T1w images were processed with FreeSurfer 5.3 and resampled to 1 × 1 × 1 mm resolution for volumetric segmentation and cortical reconstruction.¹⁹ Thickness and grey matter volumes for 68 cortical regions and volumes for 12 subcortical grey matter structures in left and right hemispheres were derived after quality control of FreeSurfer output through visual inspection and manual editing of the cortical and subcortical segmentation output when necessary. All MRI scans were processed using FreeSurfer version 5.3 (http://surfer.nmr.mgh.harvard.edu/). FreeSurfer performed cortical reconstruction and segmentation of T1-weighted brain images. These steps include correcting for head motion, identifying subcortical and deep gray matter structures, and normalizing image intensity. Each brain was then aligned to a standard atlas based on individual folding patterns and divided into regions based on the structure of the brain’s gyri and sulci. Detailed descriptions of these methods are available in previous publications.²⁰ Hippocampal volumes were normalized for intracranial volume (ICV) using linear regression.^21,22 Participants were classified into two groups: low-volume (bottom 30%) and high-volume (top 70%). Quality control included visual inspection and automated outlier detection.

Cognitive assessments and progression definition

Cognitive status was assessed using the CDR scale, with each evaluation typically conducted 12–18 months apart. Cognitive progression was categorized into two types: first progression, defined as the first shift from CDR 0 to CDR 0.5, and sustained progression, defined as two consecutive CDR evaluations ≥ 0.5.^23,24

A Preclinical Alzheimer Cognitive Composite (PACC) score was calculated using four neuropsychological tests: Category Fluency, the Selective Reminding Test – Free Recall, and Trail Making Tests A and B. Each test was selected to assess key cognitive domains, including verbal fluency, episodic memory, processing speed, and executive function. Raw scores were converted into z-scores, standardized using the mean and standard deviation of a cognitively normal reference population. The final composite score was calculated as the average of these z-scores, providing a comprehensive measure of cognitive performance.²⁵ The Mini-Mental State Examination (MMSE) assesses global cognitive function using 30 items that evaluate orientation, attention, memory, language, and visuospatial abilities.²⁶

Statistical analysis

Baseline characteristics were summarized as means (SDs) and frequencies (%), with t-tests and chi-square tests for group comparisons. Independent two-sample t-tests were used to compare baseline PACC and MMSE scores between progression groups, evaluating whether baseline cognitive performance differed significantly. Separate t-tests were conducted for both the CSF and MRI groups to assess differences between first progressors and non-progressors, as well as sustained progressors and non-progressors. Kaplan-Meier curves estimated cognitive progression probabilities over time, stratified by biomarker-defined groups. Cox proportional hazards models were used to assess the associations between CSF NfL, hippocampal volume, and time to first and sustained progression.

On average, participants contributed 2.1 CSF NfL and 3.4 MRI hippocampal volume measurements during the follow-up period. Due to variability in participant compliance and data availability, CSF and MRI assessments were not consistently paired at each measurement occasion. To address this irregularity, linear mixed-effects models were utilized, as they inherently accommodate unevenly spaced and missing observations over time.

Specifically, linear mixed-effects models examined the longitudinal trajectories of CSF NfL and hippocampal volume over time, assessing the associations between yearly changes in these biomarkers (outcome) and CDR progression groups (non-progressors, first progressors, sustained progressors; exposure). Both models were adjusted for age, sex, education, and APOE ϵ4 allele. Sensitivity analyses were adjusted for additional covariates. Furthermore, stratified analyses were conducted to evaluate whether biomarker-progression associations differed by APOE ϵ4 allele and sex.

To assess potential confounding by amyloid and tau pathology, we conducted sensitivity analyses including Aβ₄₂/Aβ₄₀, tau/Aβ₄₂, and ptau/Aβ₄₂ positivity status as covariates in the Cox proportional hazards models. Multicollinearity among these markers was assessed using the Variance Inflation Factor (VIF), and model specifications were adjusted accordingly. To evaluate whether CSF NfL and MRI-based neurodegeneration predict progression independent of amyloid status, we conducted Cox models restricted to Aβ₄₂/Aβ₄₀-negative participants (n = 204), adjusting for age, sex, race, education, and continuous Aβ₄₂/Aβ₄₀ levels. Separate models were run for first and sustained progression outcomes. All analyses were performed using R version 4.1.1.

Results

Baseline characteristics and biomarker-based stratification

The study included 279 cognitively normal individuals (mean age: 66.5 years [SD, 6.08]). The cohort was predominantly White (91.1%, n = 254), with 8.9% (n = 25) identified as Black or African American based on self-report. The age distribution did not differ significantly between African American (Mean: 64.6 years, SD: 6.00) and White (Mean: 66.7 years, SD: 6.07) groups (p = 0.11). The cohort consisted of 58.4% females (n = 163) and 41.6% males (n = 116), with similar proportions across racial groups. 4 out of 35 sustained progressors and 15 out of 71 first-time progressors reverted to CDR 0 at their last follow-up. The mean follow-up duration for CDR assessment was 11.41 years (SD, 4.40), with longer follow-up in first progressors (12.90 years [SD, 3.51], p < 0.001) and sustained progressors (13.67 years [SD, 3.16], p < 0.01) compared to non-progressors (Table 1). For CSF NfL assessments, the mean follow-up was 7.36 years (SD, 3.98). First progressors were followed for 7.77 years (SD, 3.40), while sustained progressors had a longer follow-up of 8.96 years (SD, 3.35), P < 0.05. The mean follow-up for MRI assessments was 8.01 years (SD, 4.10). First progressors had 8.01 years (SD, 3.86), while sustained progressors had 9.13 years (SD, 3.70), a non-significant increase.

Table 1.

Baseline characteristics stratified by cognitive decline progression using the Clinical Dementia Rating scale.

	Total (N = 279)	First Progressiona (N = 71)	Sustained Progressionb (N = 35)
Age (y)
Mean (SD)	66.5 (6.08)	68.6 (5.74) ***	69.0 (5.77) ***
Sex
Female	163 (58.4%)	37 (52.1%)	17 (48.6%)
Male	116 (41.6%)	34 (47.9%)	18 (51.4%)
Years of Education
Mean (SD)	16.1 (2.55)	15.8 (2.84)	16.3 (2.54)
APOE ϵ4 Status
Control	183 (65.6%)	39 (54.9%)	14 (40.0%)
Case	96 (34.4%)	32 (45.1%) *	21 (60.0%) ***
Follow-Up Measure
CDR (y)	11.41 (4.40)	12.90 (3.51) ***	13.67 (3.16) **
CSF NfL (y)	7.36 (3.98)	7.77 (3.40)	8.96 (3.35) *
MRI (y)	8.01 (4.10)	8.01 (3.86)	9.13 (3.70)
MMSE Score Baseline
Mean (SD)	29.2 (1.20)	28.9 (1.63) *	28.6 (2.08) *
PACC Score Baseline
Mean (SD)	−0.0855 (0.663)	−0.209 (0.713)	−0.324 (0.833)
CSF NfL (pg/mL)
Mean (SD)	705 (269)	803 (287) ***	836 (323) ***
Hippocampal Volume
Mean (SD)	7710 (853)	7480 (976) **	7350 (1110) **
${\text{Aβ}}_{\text{42}}/{\text{Aβ}}_{40}^{\text{c}}$
Positive	75 (26.9%)	39 (54.9%) ***	19 (54.3%) ***
$\text{tau}/{\text{Aβ}}_{42}^{\text{c}}$
Positive	69 (24.7%)	32 (45.1%) ***	15 (42.9%) ***
$\text{ptau}/{\text{Aβ}}_{42}^{\text{c}}$
Positive	67 (24.0%)	36 (50.7%) ***	18 (51.4%) ***
Diagnosis
Uncertain dementia	NA	42	14
AD Dementia	NA	17	18
MCI	NA	6	NA
Non-AD Dementia	NA	6	2
Mixed Dementia	NA	NA	1

^***p < 0.001

^**p < 0.01

^*p < 0.05, compared to Non-Progressor (CDR = 0).

CDR: Clinical Dementia Rating; CSF: cerebrospinal fluid; NfL: neurofilament light chain; Aβ: amyloid-β; APOE4: apolipoprotein E ϵ4 allele; p-tau: phosphorylated tau; MRI: magnetic resonance imaging; AD: Alzheimer’s disease; MCI: mild cognitive impairment; PACC: Preclinical Alzheimer Cognitive Composite; MMSE: Mini-Mental State Examination.

^aFirst progressors are defined as participants with at least one follow-up visit where CDR increased from 0 to ≥ 0.5.

^bSustained progressors are a subset of first progressors who had two or more consecutive visits with CDR ≥ 0.5.

^cBiomarker Cut-Offs for Positivity: Aβ₄₂/Aβ₄₀ < 0.0673; p-Tau/Aβ₄₂ > 0.0649; t-Tau/Aβ₄₂ > 0.488.

Participants who progressed had a higher mean age (first progression: 68.6 years [SD, 5.74]; sustained progression: 69.0 years [SD, 5.77] compared with non-progressors (both p < 0.001). Sex distribution was comparable (first progression: 52.1% female; sustained progression: 48.6% female). The mean year of education was higher in the sustained progression group (16.3 years [SD, 2.54]) than in the first progression group (15.8 years [SD, 2.84]), but not statistically significant. APOE ϵ4 carrier status differed between groups, where 45.1% of the first progression group and 60.0% of the sustained progression group carried at least one ϵ4 allele (p < 0.05 and p < 0.001, respectively).

The mean time to first progression (CDR ≥ 0.5) was 9.03 years (SD = 4.61), while the mean time to sustained progression (defined as two consecutive CDR ≥ 0.5 assessments) was 9.89 years (SD = 4.16). Supplemental Figures 1A and 1B display individual trajectories of CDR Global Score over time for first and sustained progressors, respectively. Supplemental Figures 2A and 2B present corresponding trajectories using the CDR Sum of Boxes. Together, these figures illustrate longitudinal patterns of cognitive change, clearly distinguishing between transient and sustained progression.

Baseline cognitive performance varied by progression status. The mean (SD) MMSE score was 28.9 (1.63) in first progressors and 28.6 (2.08) in sustained progressors. Similarly, the PACC baseline score was −0.209 (0.713) in first progressors and −0.324 (0.833) in sustained progressors. At baseline, PACC scores were negatively correlated with CSF NfL levels (r = –0.32, p < 0.001) and positively correlated with normalized hippocampal volume (r = 0.24, p < 0.001). Similar but weaker associations were observed between MMSE and both biomarkers (CSF NfL: r = –0.17, p = 0.002; hippocampal volume: r = 0.14, p = 0.008). These associations are visualized in Supplemental Figures 5A and 5B.

Baseline CSF NfL concentrations were higher among individuals with first progression (803 pg/mL [SD, 287]; p < 0.001) and sustained progression (836 pg/mL [SD, 323]; p < 0.001) compared to non-progressors. Normalized Hippocampal volume was lower in progressors (first progression: 7480 mm³ [SD, 976]; sustained progression: 7350 mm³ [SD, 1110]; both p < 0.01).

Aβ₄₂/Aβ₄₀ positivity was 54.9% in the first progression group and 54.3% in the sustained progression group (both p < 0.001). Similarly, tau/Aβ₄₂ positivity was seen in 45.1% and 42.9% of the first and sustained progression groups, respectively (both p < 0.001). Lastly, ptau/Aβ₄₂ positivity was detected in 50.7% of the first progression group and 51.4% of the sustained progression group (both p < 0.001).

Survival and progression analyses

Kaplan-Meier survival analysis showed that individuals with elevated CSF NfL levels had a faster time to first progression (95% CI, 0.2–1, p = 0.0003) and sustained progression (95% CI, 0.46–1, p = 0.0075). Lower hippocampal volume was associated with an increased risk of first progression (95% CI, 0.13–1, p = 0.018) and sustained progression (95% CI, 0.23–1, p = 0.038), though the effect size was smaller compared to CSF NfL (Figure 1). Cox proportional hazards models, adjusting for age, sex, education, and APOE ϵ4 status, indicated that elevated CSF NfL was associated with a higher risk of first progression (HR, 1.83; 95% CI, 1.11–3.01; p = 0.018) but was not significantly associated with sustained progression (HR, 1.83; 95% CI, 0.90–3.69; p = 0.093). Hippocampal atrophy was not a significant predictor of first progression (HR, 1.30; 95% CI, 0.78–2.17; p = 0.31) or sustained progression (HR, 1.39; 95% CI, 0.68–2.84; p = 0.37).

Figure 1.

Survival curves for CSF NfL and MRI-based progression.

Longitudinal biomarker trajectories

Linear mixed-effects models indicated that CSF NfL levels increased at a significantly faster rate in progressors compared to non-progressors (β = 47.06; 95% CI, [39.14 to 54.98]; p < 0.001). The rate of CSF NfL accumulation was greater in sustained progressors (β=35.68; 95% CI, [31.35 to 40.01]; p < 0.001) compared to first progressors. Similarly, hippocampal volume declined more steeply in both first progressors (β=−109.14; 95% CI, [−118.31 to −99.97]; p < 0.001) and sustained progressors (β=−64.34; 95% CI, [−69.53 to −59.15]; p < 0.001) compared to non-progressors (Table 2).

Table 2.

Linear Mixed Effects (LME) model results for CSF and MRI Biomarkers.

Variable	CSF NfL: First Progressionb	CSF NfL: Sustained Progressionc	MRI: First Progression	MRI: Sustained Progression
Visit Duration (CSF/MRI)	47.06 (4.04, 11.66) ***	35.68 (2.21, 16.13) ***	−109.14 (4.68, −23.30) ***	−64.34 (2.65, −24.29) ***
Non-Progressor	−47.78 (46.50, −1.03)a	NA	137.84 (110.37, 1.25)	NA
Sustained Progression	NA	34.22 (61.39, 0.56)	NA	−258.10 (143.71, −1.80)
Education	11.64 (7.56, 1.54)	9.69 (7.57, 1.28)	21.16 (19.10, 1.11)	28.24 (18.92, 1.49)
Sex	−121.38 (39.63, −3.06) **	−123.60 (39.74, −3.11) **	−88.03 (100.12, −0.88)	−88.19 (99.27, −0.89)
PACC Baseline	−67.70 (29.94, −2.26) *	−64.83 (30.13, −2.15) *	131.42 (76.07, 1.73)	119.32 (75.68, 1.58)
Age Baseline	22.69 (3.27, 6.95) ***	22.84 (3.26, 7.01) ***	−61.42 (8.24, −7.45) ***	−61.99 (8.13, −7.62) ***
Visit Duration (CSF): Non-Progressord	−11.04 (4.68, −2.36) **	NA	NA	NA
Visit Duration (CSF): Sustained Progressiond	NA	19.40 (5.54, 3.51) ***	NA	NA
Visit Duration (MRI): Non-Progressor	NA	NA	49.15 (5.46, 9.01) ***	NA
Visit Duration (MRI): Sustained Progression	NA	NA	NA	−53.38 (6.63, −8.05) ***

^***p < 0.001

^**p < 0.01

^*p < 0.05.

CSF: cerebrospinal fluid; MRI: magnetic resonance imaging; PACC: Preclinical Alzheimer Cognitive Composite.

^aValues in parentheses indicate (Standard Error, t-value).

^bFirst Progression: First transition from CDR = 0 (cognitively normal) to CDR = 0.5 (mild cognitive impairment).

^cSustained Progression: Two consecutive CDR evaluations ≥ 0.5.

^dInteraction terms represent combined effects of Visit Duration and progression status (e.g., Non-Progressor or Sustained Progression).

Supplemental Figures 3A and 3B depict the longitudinal trajectories of CSF NfL levels comparing non-progressors with first progressors and sustained progressors, respectively. Supplemental Figures 4A and 4B present the corresponding trajectories for normalized hippocampal volume. These stratified visualizations provide clear insights into biomarker changes over time and complement the linear mixed-effects model results presented in Table 2.

Sensitivity analyses

Stratification by APOE ϵ4 status showed that CSF NfL was not a significant predictor of first progression in ϵ4 carriers (95% CI: −32.65,123.69, p = 0.249) or sustained progression (95% CI: −38.86,119.34, p = 0.314). Similarly, MRI hippocampal volume was not significantly associated with APOE ϵ4 status in either the first progression (95% CI: −200.28,193.06, p = 0.971) or sustained progression (95% CI: −179.28,214.78, p = 0.858) model. Additionally, baseline PACC scores were significantly lower in progressors compared to non-progressors across CSF and MRI data.

In models adjusting for Aβ₄₂/Aβ₄₀, tau/Aβ₄₂, and ptau/Aβ₄₂ positivity, the hazard ratios for CSF NfL (HR = 1.43, p = 0.18) and MRI Hippocampal Volume (HR = 1.48, p = 0.15) were attenuated, though model concordance improved (from 0.691 to 0.756 and 0.678 to 0.76, respectively). However, high multicollinearity was observed among the AD biomarkers (VIFs ranging from 3 to 9.7), which limited model interpretability. Therefore, all three were not included simultaneously in the final models.

We also conducted a sensitivity analysis restricted to Aβ₄₂/Aβ₄₀-negative individuals to assess whether CSF NfL and hippocampal volume remained predictive of progression. For sustained progression, CSF NfL positivity was associated with a 3.76 times increased risk (HR = 3.76; 95% CI: 1.22–11.54; p = 0.021), and low hippocampal volume was associated with a 3.60 times increased risk (HR = 3.60; 95% CI: 1.21–10.72; p = 0.021). Similar associations were observed for first progression, though the NfL result did not reach statistical significance.

Discussion

This study highlights the role of CSF NfL as an early indicator of neurodegeneration and cognitive decline in individuals aged 55 and older. Our longitudinal findings demonstrate that higher CSF NfL levels are significantly associated with an increased risk of cognitive decline, as indicated by first progression. Additionally, individuals with elevated CSF NfL levels were more likely to experience sustained progression, defined as two consecutive assessments indicating a sustained incident of cognitive decline. These findings confirm the potential of CSF NfL as a biomarker for identifying individuals at risk for cognitive impairment, supporting its utility as a screening tool for neurodegenerative diseases beyond AD.

Our study also evaluated the relationship between longitudinal hippocampal volume and cognitive decline. While hippocampal atrophy alone did not strongly predict first progression, a steeper rate of hippocampal volume reduction was observed among individuals who experienced sustained progression. These findings suggest that while hippocampal atrophy is a hallmark of neurodegeneration, its predictive value for early-stage cognitive decline may be less pronounced than CSF NfL. A prior study reported similar trends, indicating that structural changes in the hippocampus may occur later in disease progression, whereas elevated CSF NfL reflects ongoing neuroaxonal injury at earlier stages.²⁷ This underscores the significance of CSF NfL in detecting neurodegeneration before substantial structural brain changes become evident.

In addition to hippocampal atrophy, recent studies suggest that white matter atrophy may be an even more sensitive biomarker of cognitive decline. These findings underscore the potential value of expanding neurodegeneration metrics beyond the hippocampus to include white matter integrity, especially when used in conjunction with fluid biomarkers such as CSF NfL.^28,29

In our stratified analysis, CSF NfL was not a significant predictor of progression among APOE ϵ4 carriers, which may reflect the complex influence of APOE-related mechanisms, such as neuroinflammation and blood-brain barrier disruption, on NfL dynamics. These findings suggest that although CSF NfL is considered an amyloid-independent marker, its predictive utility may be modulated by genetic risk factors, such as APOE ϵ4, warranting further mechanistic investigation.³⁰ Notably, the associations between CSF NfL, hippocampal atrophy, and clinical progression remained robust in participants who were Aβ₄₂/Aβ₄₀-negative, supporting the hypothesis that these markers serve as early indicators of disease progression independent of amyloid status and may help detect non-amyloid pathways of neurodegeneration.

Our findings align with growing evidence demonstrating that CSF NfL is a risk factor for mild cognitive impairment (MCI) regardless of amyloid status.^31,32 Unlike amyloid and tau biomarkers, which are primarily associated with AD-specific pathology,³¹ CSF NfL remains strongly linked to MCI even when CSF Aβ₄₂ levels are not reduced.³¹ This underscores its role as a marker of neurodegeneration that extends beyond the amyloid cascade, making it a clinically relevant biomarker for various neurodegenerative diseases, including frontotemporal dementia, vascular dementia, and Lewy body dementia.^33,34 Notably, the association between CSF NfL and cognitive impairment is observed both in individuals on the AD trajectory and in those without amyloid-related pathology.³¹ This emphasizes its potential as a non-amyloid pathway biomarker in neurodegenerative processes and supports its relevance in dementia screening across diverse subtypes.

Our findings are consistent with prior studies demonstrating that elevated CSF NfL levels predict cognitive decline and progression from MCI to AD dementia. Importantly, we extend this evidence to cognitively normal older adults, showing that higher CSF NfL levels are associated with increased risk of progression, including in amyloid-negative individuals. These results underscore the value of CSF NfL as a dynamic and broadly applicable biomarker of neurodegeneration, supporting its potential utility in early risk stratification across the AD continuum.^27,35 Importantly, elevated CSF NfL predicts the risk of transition to MCI even in cognitively healthy individuals.³¹ CSF NfL levels increase regardless of AD stage/severity.³⁶ In contrast, the rate of increase is significantly higher in individuals with MCI who later progress to AD dementia compared to those with stable MCI.^37–40 These findings support CSF NfL’s role as a dynamic biomarker of neurodegeneration, independent of amyloid pathology, further establishing its clinical relevance for dementia screening and risk assessment.

Data from the ADNI cohort have demonstrated that higher CSF NfL levels are associated with a more rapid cognitive decline in individuals with MCI.^32,35 Furthermore, elevated CSF NfL has been linked to shorter survival in patients with AD dementia,^31,41 suggesting that NfL reflects the severity and progression of neurodegeneration rather than a specific pathology.^{27,31–33,42} These findings support the notion that CSF NfL reflects the intensity of ongoing neurodegenerative processes, rather than being specific to a single pathology and establishing CSF NfL as an ideal candidate for dementia screening independent of AD.

CSF NfL as a screening tool for neurodegeneration has clinical utility for identifying individuals at risk for dementia and tracking early-stage neuronal damage. Unlike biomarkers primarily associated with AD pathology, CSF NfL’s ability to detect early neuronal damage makes it a valuable tool not only for screening and risk assessment but also for clinical settings and clinical trials, where it can aid in patient stratification, disease monitoring, and evaluating treatment efficacy. Building on this foundation, our findings provide a basis for developing scalable and accessible biomarker-based screening strategies by exploring blood-based biomarkers, such as plasma NfL. Plasma NfL has shown potential in identifying asymptomatic individuals at risk for AD dementia, detecting neurodegenerative changes beyond amyloid pathology, monitoring disease progression, and improving diagnostic accuracy.^43–52 As research advances, integrating blood-based biomarkers could prove invaluable, given the ease of acquisition with measures that could enhance early detection and risk assessment across diverse neurodegenerative conditions, further expanding the utility of biomarker-based screening in clinical practice.

This study has several strengths. First, the longitudinal design with a 20-year follow-up period allowed for a comprehensive, uniform assessment of clinical, neurological, and cognitive trajectories. Second, we quantified NfL and hippocampal volume using a validated and rigorous methodological process in multiple follow-ups in the cohort.^53,54 Third, the stratified analyses provided insights into potential modifying factors such as APOE ϵ4 status and sex differences. However, we acknowledge several limitations. The models adjusted for key covariates and unmeasured confounders such as medications, comorbidities, family history, social determinants of health, and behavior may still influence outcomes. Second, our study is limited by the lack of racial and ethnic diversity in the cohort, which constrains the generalizability of our findings. Further validation of CSF NfL as a screening tool is needed in more diverse populations. Ongoing and planned research efforts should focus on examining fluid biomarkers, including plasma NfL, in racially and ethnically diverse cohorts to address this gap and promote equity in the early detection of dementia. Third, while CSF NfL showed strong predictive value, its clinical applicability may be constrained by the invasive nature of lumbar puncture procedures. Given established ethnoracial disparities in dementia risk and biomarker profiles,^55–58 future studies should prioritize diverse participant samples to improve generalizability. Additionally, establishing clinically relevant CSF and plasma NfL cutoffs is crucial for screening neurodegenerative conditions. Exploring the predictive power of plasma NfL in a generalizable population could further clarify its role in disease risk assessment and progression monitoring. Incorporating machine learning approaches may also enhance biomarker-based risk prediction models, improving accuracy in identifying individuals at risk for cognitive decline.

Our study provides compelling evidence that CSF NfL is a robust screening tool for dementia, independent of amyloid pathology, outperforming hippocampal volume in its prognostic utility. Although both CSF NfL and hippocampal atrophy were associated with cognitive decline, CSF NfL demonstrated a more robust and consistent relationship with progression. This finding may reflect the broader scope of neuroaxonal damage captured by NfL, compared to the more localized degeneration reflected in hippocampal volume loss. Nonetheless, hippocampal atrophy remains a key biomarker of AD progression, and its inclusion adds anatomical specificity to our multimodal assessment of neurodegeneration. Together, these markers may offer complementary insights when used in tandem for early detection strategies. These results contribute to the growing body of research supporting biomarker-driven dementia risk stratification and early intervention strategies. Understanding the trajectory of cognitive decline in relation to biomarker changes is crucial for advancing precision medicine in neurodegenerative diseases. By integrating multimodal biomarker assessments within a longitudinal study, this research will improve risk assessment and monitoring of individuals at risk for dementia. These findings underscore the value of CSF NfL as a biomarker for neurodegeneration across various neurodegenerative conditions, supporting its use in clinical practice and trials to enhance early diagnosis and treatment strategies.

Supplementary Material

Supplemental material for this article is available online.

Data availability statement

The data that support the findings of this study are available from the corresponding author upon reasonable request.