Combination of Plasma Neurofilament Light Chain and Mini-Mental State Examination Score Predicts Progression from Mild Cognitive Impairment to Alzheimer’s Disease within 5 Years

Abstract

Background:

Individuals with mild cognitive impairment (MCI) are at high risk of progression to Alzheimer’s disease (AD) dementia, but some remain stable. There is a need to identify those at higher risk of progression to improve patient management and outcomes.

Objective:

To evaluate the trajectory of plasma neurofilament light chain (pNFL) prior to progression from MCI to AD dementia, the performance of pNFL, in combination with the Mini-Mental State Examination (MMSE), as a predictor of progression from MCI to AD dementia and to inform clinicians on the use of pNFL as a predictive biomarker.

Methods:

Participants (n = 440) with MCI and longitudinal follow-up (mean = 4.2 years) from the AD Neuroimaging Initiative dataset were included. pNFL as a marker for neurodegeneration and the MMSE as a cognitive measure were investigated as simple/practical predictors of progression. The risk of progressing from MCI to AD dementia associated with pNFL and MMSE scores was assessed using Cox and logistic regression models.

Results:

The current risk of progression to AD dementia was 37%higher in individuals with high pNFL (> 56 ng/L) compared to those with average pNFL (≤40 ng/L). A combination of baseline pNFL and MMSE could differentiate those who progressed within 5 years (AUC = 0.75) from stable individuals. Including change in MMSE over 6-12 months further improved the model (AUC = 0.84).

Conclusion:

Our findings reveal that combining pNFL with a simple dementia screener (MMSE) can reliably predict whether a person with MCI is likely to progress to AD dementia within 5 years.

Keywords

Alzheimer’s disease clinical progression mild cognitive impairment neurofilament plasma biomarkers

INTRODUCTION

Life expectancy is increasing across the world, leading to an increased prevalence of Alzheimer’s disease (AD) and dementia [1]. Dementia comes at an enormous personal, social, and socioeconomic cost and is expected to affect over 115 million individuals by 2050. There is a need for simple and reliable tools to identify individuals at higher risk of dementia early so that we can better target interventions, slow disease progression, and provide new treatments as they become available. AD accounts for 60 to 80%of all dementia cases and is the focus of this study [2].

Mild cognitive impairment (MCI) is a syndrome characterized by cognitive decline in excess of what is expected for an individual at a given age but that does not significantly interfere with activities of daily life [3]. MCI is usually diagnosed by a clinician using cognitive and behavioral tests after excluding other possible causes. Between 25 to 65%of individuals with MCI progress to AD dementia, with the majority of progressions occurring within 1.5–5 years of MCI diagnosis [3 –6]. Individuals affected have typically lived with MCI for several years before progressing to AD dementia [7].

Identifying which individuals are most at risk of progression to AD dementia would be clinically very useful and would assist in disease management, as well as allow individuals and their families to better plan for the future (see Supplementary Table 4 for details). Currently, psychologists and geriatricians rely on specialized neuropsychological tests to predict whether someone is at higher risk of progressing from MCI to AD dementia [8]. While these tests are effective at predicting progression within 2 years of dementia diagnosis (especially when several tests are used in combination; 56%≤sensitivity≤90%and 70%≤specificity≤87%), they appear to lose sensitivity in the earlier pre-clinical stages and are not quantitative markers of neurobiological deterioration [8].

Since AD pathophysiology begins decades before the clinical onset of the disease [9 –11], imaging, cerebrospinal fluid (CSF), and plasma biomarkers which reflect the underlying AD pathology (amyloid-β (Aβ), hyperphosphorylated tau tangles, and neurodegeneration) are more useful in improving early detection of those at risk than symptomatic assessment alone [12 –18]. CSF assays and positron emission tomography (PET) tracers for Aβ (e.g., fluorbetapir) and hyperphosphorylated tau (e.g., fluortaucipir) can be used to predict clinical outcomes [19 –21], but the relationship between Aβ, tau, and cognitive decline over short timescales is unclear. There is a significant and unpredictable lag between their deposition in the brain and the onset of MCI-AD symptomatology [10].

Neurodegeneration has better temporal linkage to cognitive decline than Aβ and tau [10, 11], making it a more suitable biomarker for detecting incipient clinical deterioration. It is possible to measure neurodegeneration using imaging techniques such as structural magnetic resonance imaging (MRI) and radiolabelled glucose (FDG-PET). Unfortunately, imaging techniques are costly, not widely available, or easy to interpret for general practitioners and geriatricians.

Neurodegeneration can also be measured using neurofilament light chain (NFL) in CSF or plasma [22 –24]. Briefly, NFL is a major scaffolding protein found in neurons, particularly within large axons. Under normal conditions, the protein is very stable with limited turnover. Increased levels of NFL reflects ongoing neural degeneration (disease activity), and is predictive of current and future degree of disability in a number of neurological disorders [25]. Biomarkers originating from the brain have traditionally been easier to detect in CSF than in plasma (where levels are strongly correlated but much lower) [18, 25] but require an invasive lumbar puncture which must usually be performed in hospital. Recent advances in blood-based assay sensitivity have made it possible to detect low-level biomarkers such as plasma NFL (pNFL) quickly and reliably, with a Spearman correlation between plasma and CSF NFL of 0.59 [23]. In MCI and AD dementia, high pNFL correlates with current and future AD-linked neurobiological alterations such as hippocampal atrophy and cortical thinning, white matter loss, ventricular enlargement, cerebral hypometabolism, Aβ and tau deposition as well as declining performance on dementia screeners such as the Mini-Mental State Examination (MMSE) [23 , 27].

Although pNFL correlates with neurodegeneration severity, the relationship between axonal degeneration and cognitive and behavioral deficits is not straightforward. Individuals with apparently severe neurodegeneration may have limited symptoms, and vice versa. A recent longitudinal study found no relationship between pNFL and cognitive function in cognitively normal (CN) individuals [28], and another found that pNFL correlated with cognitive scores in the MCI population, but not the CN population [29]. These findings are most likely due to individual differences in cognitive reserve whereby certain individuals can sustain higher levels of neurodegeneration without any clinical manifestations [30], and highlight the importance of accounting for the behavioral and cognitive/functional status of an individual before interpreting pNFL. Whereas pNFL likely only captures recent and ongoing axonal degeneration, dementia screeners such as the MMSE are practical indexes of the combined effects of cognitive reserve and past neurobiological damage.

The aim of this study was therefore to evaluate whether pNFL is raised prior to MCI progression to AD dementia, and whether it can be used as a reliable predictor of progression from MCI to AD dementia. Moreover, we sought to investigate the performance of pNFL alone and in combination with MMSE, in predicting progression from MCI to AD within 5 years.

METHOD AND MATERIALS

ADNI study

Data used in this study were obtained from the Alzheimer’s Disease Neuroimaging Initiative (ADNI) database (http://adni.loni.usc.edu) with individuals originating from over 50 sites across the United States and Canada. The ADNI launched in 2003 and currently includes over 2200 participants with longitudinal follow-up, including individuals with normal cogitation (CN), MCI, and early AD dementia. Participants were aged 55–92 years old, had completed at least 6 years of education, were fluent in Spanish or English, and had no significant neurological disease other than MCI or probable AD dementia [31]. Participants were assessed at most every 6 months for up to 8 years. All participating institutions fulfilled ethical requirements and participants provided written informed consent. For the present study, we used data collected between September 2007 and February 2019. This study is reported in accordance with the Standards for Reporting Diagnostic accuracy studies (STARD) guidelines.

Participants

All participants (across all ADNI studies) who had a pNFL and a diagnosis of MCI at baseline were considered for inclusion. Participants were retained if they remained MCI over their follow-up or progressed to AD but not if they oscillated/reverted between MCI and AD dementia. Outliers were removed (see below). Individuals with at least two timepoints were labelled as MCI stable (MCIs; n = 299) or MCI progressor (MCIp; n = 141). Individuals were further stratified into stable for at least 5 years (MCI5s; n = 106), or progression to AD dementia within 5 years (MCI5p; n = 140). A subset of these participants had two or more pNFL and MMSE measurements (6–12 months after baseline for MCI5p (n = 66), and any time for MCI5s (n = 68); see Supplementary Methods 4 for details)). Data acquired at or after progression to AD dementia were not considered. In MCI5s, pNFL and MMSE datapoints were only considered if they were verifiably followed by at least 5 years of MCI. Flow of participants can be found in Supplementary Figure 1 and example data in Supplementary Figure 11

Clinical assessment

Detailed diagnostic procedures have been published previously [32]. Briefly, MMSE, Clinical Dementia Rating (CDR), and logical memory II (delayed recall) scores of the Wechsler Memory Scale were used to classify people as cognitively normal, MCI or AD dementia and were administered at all time points. The cognitively normal group were defined as having an MMSE≥24 and CDR = 0. The MCI group met the Petersen/Winblad criteria operationalized as having MMSE≥24, logical memory II score 1SD below the mean and CDR = 0.5 with preserved activities of daily living. Finally, the AD group fulfilled the National Institute of Neurological and Communicative Disorders and Stroke and Alzheimer’s Disease and Related Disorders Association criteria for probable AD [31] with MMSE≤26 and CDR≥0.5. MMSE scores were further considered as broad functional indicators of underlying cumulative neural damage [33].

Plasma NFL

pNFL was measured by operators [24] blinded to clinical diagnoses using and ultrasensitive enzyme-linked immunosorbent assay on a single molecule array platform (Simoa, Quanterix Corp). Intra- and inter-assay coefficients of variation were < 9%[24]. pNFL values are reported in ng/L.

Statistical analyses

Analyses were performed using R (version 4.0.2) under RStudio. pNFL and MMSE distributions were slightly skewed so were log-transformed and scaled. Because log-transformation and scaling did not change any results, untransformed measures were used in the final analyses for ease of interpretation. Age and pNFL were centered on the mean and MMSE on 24 (dementia score). Presence of multivariate outliers was assessed based on Mahalanobis distance using a p < 0.001 threshold. Twenty-five pNFL measurements were detected as outliers and removed. Differences between MCIs and MCIp groups were assessed using t-tests for continuous variables, chi-squared for categorical variables, and bivariate correlations between variables were assessed using a Spearman rank test. The significance level was set at 0.05 for all tests.

The risk of MCI progression to AD dementia associated with pNFL was evaluated using Cox regression analysis, censoring MCI individuals who dropped out at the time of loss to follow-up (Survival package [34]; Supplementary Methods 1). The model is represented using a Kaplan-Meier plot with pNFL and MMSE stratified into low, medium and high based on the median value of the tertiles of their distribution (stratifications were for illustrative purposes only. MMSE and pNFL were modelled as continuous variables). The pNFL and MMSE trajectories antedating MCI progression to AD dementia were evaluated using linear mixed effects with individual-specific random intercepts (lme4 package [35]; Supplementary Methods 2). Longitudinal pNFL and MMSE trajectories in MCI5s and MCI5p groups over the follow-up were modelled and plotted. The ability of baseline pNFL, alone and in combination with MMSE, to predict MCI5p was assessed using logistic regression and receiver operator curve (ROC) analysis (Caret package [36]; Supplementary Methods 3). In addition to baseline measurements, change in pNFL and MMSE was also investigated as a predictor of MCI5p (Supplementary Methods 4). The logistic regression models are visually represented using probability plots with MMSE and MMSE slope stratified into low, medium and high based on the median value of the tertiles of their distribution (stratifications for illustrative purposes only, as above). All models were adjusted for age and sex (because both are known to modulate pNFL and MMSE [24]), and bivariate interactions between age, sex, pNFL, and MMSE were tested. Final models were determined by backwards selection based on parameter significance, starting with high order interactions. Model assumptions of proportional hazards, linearity, normality of residuals and multicollinearity were verified where appropriate.

RESULTS

Participants’ characteristics are presented in Table 1. During a mean (SD) follow-up of 4.6 years (1.8), 141 (32%) progressed to AD dementia. At baseline, MCIp were 1.52 years older than MCIs, with no differences in sex and education between the groups. These results were similar within MCI5s and MCI5p groups, whose demographic characteristics are presented in Supplementary Tables 1 and 2.

Table 1

Baseline demographics. This table pertains to the sample used in the Cox survival model. This is the largest and thus most representative sample of our study. A subset of these individuals was appropriate and retained for our mixed effects and logistic regression modelling (see methods, Supplementary Figure 1, Supplementary Tables 1 and 2). AD, Alzheimer’s disease; MCIs, mild cognitive impairment stable; MCIp, mild cognitive impairment progressors to AD dementia; MMSE, Mini-Mental Score Examination; NFL, neurofilament light chain

	MCIp (N = 141)	MCIs (N = 299)	Total (N = 440)	p
Age at baseline, y				0.048¹
Mean (SD)	74.51 (7.36)	72.99 (7.60)	73.48 (7.55)
Range	56.75 –88.83	55.33 –91.50	55.33 –91.50
Sex				0.529²
Female	62 (44.0%)	122 (40.8%)	184 (41.8%)
Male	79 (56.0%)	177 (59.2%)	256 (58.2%)
Education, y				0.865¹
Mean (SD)	16.04 (2.73)	15.99 (2.87)	16.01 (2.82)
Range	9.00 –20.00	8.00 –20.00	8.00 –20.00
MMSE score at baseline				< 0.001¹
Mean (SD)	27.38 (1.82)	28.19 (1.69)	27.93 (1.77)
Range	21.00 –30.00	22.00 –30.00	21.00 –30.00
plasma NFL at baseline (ng/L)				< 0.001¹
Mean (SD)	47.59 (20.01)	36.58 (16.52)	40.11 (18.43)
Range	16.00 –128.90	10.50 –124.40	10.50 –128.90
CDR score				0.008¹
N-Miss	0	1	1
Mean (SD)	0.51 (0.10)	0.48 (0.10)	0.49 (0.10)
Range	0.00 –1.00	0.00 –0.50	0.00 –1.00
Logical memory II (delayed recall)				< 0.001¹
Mean (SD)	5.18 (4.00)	8.18 (3.00)	7.22 (3.63)
Range	0.00 –18.00	0.00 –20.00	0.00 –20.00
Time to AD dementia onset, y
N-Miss	0	299	299
Mean (SD)	2.85 (2.12)	NA	2.85 (2.12)
Range	0.08 –8.50	NA	0.08 –8.50
Follow-up time, y				0.243¹
Mean (SD)	4.38 (2.15)	4.10 (2.40)	4.19 (2.33)
Range	0.58 –9.17	0.25 –10.08	0.25 –10.08

¹Linear Model ANOVA. ²Pearson’s Chi-squared test.

Plasma NFL and MMSE

At baseline, pNFL was 11.01 ng/L (30%) higher and MMSE was 0.81 (2.9%) points lower in MCIp compared to MCIs (Table 1). pNFL and MMSE were inversely correlated. pNFL correlated with age, but not with sex or education. MMSE correlated with education and inversely with age but not with sex (Supplementary Figure 2). These results were similar within MCI5s and MCI5p groups.

Risk of progressing to AD dementia associated with baseline pNFL and MMSE

Associations between pNFL, MMSE, and risk of progression to AD dementia are reported in Table 2 and represented as a Kaplan-Meier plot in Fig. 1. High pNFL (> 56 ng/L) was associated with a 37%, and low MMSE (< 26) with a 77%increased risk compared to the whole sample mean (pNFL = 40.1 ng/L and MMSE = 28). All else being equal, every additional 1 ng/L above 40 ng/L in pNFL was associated with a 2%increased risk of progression. Similarly, every 1-point MMSE below 24 was associated with a 33%increased progression risk. An interaction between pNFL and MMSE (HR = 1.00–1.01) was detected, meaning that the risk of progression increased more strongly in individuals with high pNFL and low MMSE (Supplementary Figure 3).

Table 2

Summary of model results. Hazard ratios pertain to Cox regression and adjusted odds ratio to logistic regression. All models indicate that high pNFL and low MMSE are associated with an increased risk of MCI progression to AD dementia.CI, confidence interval; SE, standard error; MMSE, Mini-Mental State Examination; NFL, neurofilament light chain

	Cox regression model	Baseline biomarkers logistic regression model	Baseline and change in biomarkers logistic regression model
Intercept
β coefficient (SE)		3.37 (1.75)	2.54 (2.81)
Hazard Ratio or adjusted Odds Ratio (95%CI)		29.20 (1.01 –971.46)	12.65 (0.05 –3688.92)
Z value		1.93	0.90
p-value		0.05	0.37
Age (y)
β coefficient (SE)	–0.03 (0.01)	–0.03 (0.02)	–0.00 (0.04)
Hazard Ratio or adjusted Odds Ratio (95%CI)	0.97 (0.95 –1.00)	0.97 (0.93 –1.02)	1.00 (0.93 –1.07)
Z value	–2.14	–1.19	–0.11
p-value	0.03	0.24	0.92
Sex [male]
β coefficient (SE)	0.09 (0.17)	0.07	0.02 (0.45)
Hazard Ratio or adjusted Odds Ratio (95%CI)	1.10 (0.78 –1.54)	1.07 (0.60 –1.90)	1.02 (0.42 –2.49)
Z value	0.52	0.22	0.04
p-value	0.60	0.82	0.97
MMSE score (points)
β coefficient (SE)	–0.29 (0.0381)	–0.32	–0.68 (0.16)
Hazard Ratio or adjusted Odds Ratio (95%CI)	0.75 (0.70 –0.81)	0.72 (0.61 –0.85)	0.51 (0.36 –0.68)
Z value	–7.54	–3.80	–4.16
p-value	4.65E-14	1.48E-4	3.16e-05
Plasma NFL (ng/L)
β coefficient (SE)	0.02 (0.00)	0.05	0.05 (0.02)
Hazard Ratio or adjusted Odds Ratio (95%CI)	1.02 (1.01 –1.03)	1.05 (1.03 –1.07)	1.06 (1.02 –1.10)
Z value	5.53	4.52	2.96
p-value	3.28E-08	6.07e-06	3.07E-3
MMSE^* Plasma NFL
β coefficient (SE)	0.00 (0.00)
Hazard Ratio or adjusted Odds Ratio (95%CI)	1.00 (1.00 –1.01)
Z value	2.17
p-value	0.03	ns	ns
MMSE change (points/year)
β coefficient (SE)			–0.58 (0.15)
Hazard Ratio or adjusted Odds Ratio (95%CI)			0.56 (0.41 –0.74)
Z value			–3.82
p-value			1.32E-4
Plasma NFL change (ng/L/year)
β coefficient (SE)			0.02 (0.02)
Hazard Ratio or adjusted Odds Ratio (95%CI)			1.02 (0.99 –1.06)
Z value			0.97
p-value			0.33

ns: non-significant; text in bold indicates values below the statistical threshold.

Fig. 1

Plasma NFL and MMSE scores predict risk of progressing from MCI to AD dementia. This Kaplan-Meier figure displays Cox model estimates of the rate of progression to AD dementia in individuals with MCI who have various plasma NFL levels and MMSE scores. For illustrative purposes, NFL and MMSE scores were categorized in low, medium, and high based on the median value of the lower, middle, and upper tertiles of their distributions. Individuals with high plasma NFL and low MMSE were most at risk of progressing to AD dementia, whereas individuals with low plasma NFL and high MMSE were least at risk. MCI, mild cognitive impairment; MMSE, Mini-Mental State Examination; NFL, neurofilament light chain.

Change in pNFL and MMSE over the 5-year follow-up

Changes in pNFL and MMSE over the 5-year follow-up within the MCI5s and MCI5p groups are presented in Supplementary Table 3 and Fig. 2. After controlling for age and sex, no change in pNFL or MMSE was detected in MCI5s over a median period of 2.0 years. However, in the MCI5p group mean pNFL increased by 5.9%/year (from baseline), while MMSE decreased by 1.1%/year leading up to diagnosis over a median period of 2.1 years. For an average MCI5p individual, MMSE was 2 points lower and pNFL was 15 ng/L higher at AD dementia onset compared to MCI5s individuals who remained stable for at least another 5 years. pNFL and MMSE started to significantly differ between MCI5p and MCI5s at least 2 years prior to these endpoints (Fig. 2). There were no significant bivariate interactions.

Fig. 2

Plasma NFL levels are raised (A) and MMSE scores are reduced (B) years before a clinical diagnosis of AD dementia is made. Models were age and sex adjusted. The stable MCI group is included for comparison. For the stable group, the predictor variable (x-axis) is time to endpoint (with endpoint being the time at which we no longer had data to verify whether the patient was stable for 5 years). MCI, mild cognitive impairment; MMSE, Mini-Mental State Examination; NFL, neurofilament light chain.

Predicting progression to AD dementia within 5 years using baseline biomarkers

In order to determine MCI progression to AD dementia within 5 years, we tested the predictive value of baseline pNFL and MMSE (Table 2 and Fig. 3A). Every additional 1 ng/L of pNFL above 40 ng/L was associated with 5%increased odds, and every additional point in MMSE above 24 was associated with 39%decreased odds of MCI5p. The AUC of the full model was 0.75 (0.69–0.81), with a sensitivity of 0.69 (0.48–0.80), specificity of 0.76 (0.62–0.93), and overall correct classification of 0.71 (0.65–0.77) (Fig. 3). DeLong comparison of AUCs revealed that the full model performed significantly better than the models that contained pNFL or MMSE in isolation (Z = 2.63; p < 0.01). The pNFL only and MMSE only models were comparable in performance to each other (AUC 0.69 and 0.68). At a specificity of 76%(the optimal Youden test specificity), including pNFL alongside the MMSE helped identify 1 more MCI5p every 11 tests compared to the model without pNFL (Supplementary Figure 4A). Performance of all models was significantly improved when considering shorter progression timeframes (i.e., 0–4, 0–3, 0–2, 0-1 years) (Supplementary Figure 6). No significant bivariate interactions were detected.

Fig. 3

Combining plasma NFL and MMSE score improves classification accuracy relative to either measure used in isolation. Receiver operating characteristic (ROC) curve shows performance of different models at differentiating between MCI individuals who remain stable from those who progress to AD within 5 years (A). Sensitivities, specificities, and accuracies as a function of the MCI progression probability threshold (cut-off) used to separate stable MCI from those progressing to AD dementia (B). Clinicians wishing to maximize test sensitivity (at the cost of specificity) will opt for a low MCI progression probability threshold (as calculated in panel D), whereas clinicians wishing to maximize test specificity (at the cost of sensitivity) will opt for a high MCI progression probability threshold. High MCI progression probability scores may warrant accelerated follow-up, and greater consideration should be given to the likelihood of earlier progression to AD. Probability of progressing to AD dementia at various plasma NFL levels and MMSE scores, as estimated by the logistic regression model (C). Tips on how to use this type of model in a clinical setting (D). AUC, area under the curve; MCI, mild cognitive impairment; MMSE, Mini-Mental State Examination; NFL, neurofilament light chain.

Predicting progression to AD dementia within 5 years using change in pNFL and MMSE

Thirty percent of participants had a pNFL and MMSE measurement taken within 6–12 months of their baseline assessment allowing for slope (rate of change) calculation. We tested the predictive performance of baseline pNFL, MMSE, and their slopes (Table 2 and Fig. 4A). pNFL, MMSE, and MMSE slope were significant predictors of progression to AD dementia in the multivariable logistic regression model. Every additional 1 ng/L of pNFL above 40 ng/L was associated with 6%increased odds of MCI5p, every additional 1 MMSE-point above 24 reduced odds by 49%, and every 1 MMSE point/year gain reduced odds by 44%. Change in pNFL was not a significant predictor of risk. The AUC of the model containing pNFL, MMSE, and MMSE slope (hereafter full longitudinal model) was 0.84 (0.77–0.9), with a sensitivity of 0.77 (0.58–0.88), specificity of 0.84 (0.72–0.97) and overall correct classification of 0.80 (0.73–0.86) (Fig. 4). DeLong comparison of AUCs revealed that this full longitudinal model performed significantly better than the static model that did not contain slopes (MMSE and NFL at baseline only) (Z = 2.54; p < 0.01). At a specificity of 0.84 (optimal Youden specificity), the full longitudinal model helped identify 1 more MCI5p for every 5 tests compared to the static model (Supplementary Figure 4B). In contrast, the MMSE only longitudinal model (baseline MMSE and MMSE slope) identified 1 more MCI5p for every 8 assessments compared to the static model. Additionally, the full longitudinal model also performed better than the MMSE only longitudinal model (Z = 2.23; p < 0.05) where 1 more MCI5p was identified for every 13 tests (Supplementary Figure 4C). There were no significant bivariate interactions

Fig. 4

Combining plasma NFL, MMSE score, and MMSE slope improves classification accuracy relative to the cross-sectional approach. Receiver operating characteristic (ROC) curve shows performance of different models at differentiating between MCI individuals who remain stable from those who progress within 5 years (A). Sensitivities, specificities, and accuracies as a function of the logistic regression probability threshold (cut-off) used to separate stable MCI from those progressing to AD dementia (B). Clinicians wishing to maximize test sensitivity (at the cost of specificity) will opt for a low MCI progression probability threshold (as calculated in panel D), whereas clinicians wishing to maximize test specificity (at the cost of sensitivity) will opt for a high MCI progression probability threshold. High MCI progression probability scores may warrant accelerated follow-up, and greater consideration should be given to the likelihood of earlier progression to AD. Probability of progressing to AD dementia at various plasma NFL levels, MMSE scores and MMSE slopes, as estimated by the logistic regression model (C). MMSE scores and slopes were categorized as good, moderate, or bad based on the lower, middle, and upper tertiles of their distributions. Tips on how to use this type of model in a clinical setting (D). AUC, area under the curve; MCI, mild cognitive impairment; MMSE, Mini-Mental State Examination; NFL, neurofilament light chain.

DISCUSSION

This study revealed that higher pNFL is associated with a significantly increased risk of progressing to AD dementia (Fig. 1), and that pNFL is elevated at least 2 years before MCI progresses to AD dementia (Fig. 2B). Combining baseline pNFL and MMSE is an effective approach to estimate the risk of MCI progression within 5 years (Fig. 3A). Moreover, our ability to predict progression was further improved by taking into account the change in MMSE over 6 to 12 months from the original assessment (Fig. 4A). The results suggest that pNFL is therefore a useful biomarker in the prediction of disease progression in pre-clinical AD.

A 50%increase in pNFL above the population mean (40 ng/L) was associated with a 49%increase in the risk of progressing from MCI to AD dementia. This result suggests that axonal degeneration measured by pNFL is strongly associated with clinical deterioration in individuals with MCI. This is consistent with diffusion tensor imaging studies which found that reduced white matter (WM) integrity, particularly in large commissural tracts, was associated with diminished cognitive performance in MCI [37, 38]. While most WM defects in these studies arose secondary to grey matter (GM) atrophy in areas typically associated with AD (Wallerian degeneration), these studies also reported evidence of primary axonal degeneration not due to GM loss [37]. Thus, pNFL likely reflects both pathophysiological processes, and future studies should assess the relative contributions of each to pNFL since they may relate to different clinical stages of prodromal AD or different pathophysiological mechanisms [39].

The evidence of elevated pNFL preceding progression to AD dementia by 2 years in MCI5p individuals (Fig. 2A) is consistent with a recent study investigating pNFL in familial AD where pNFL was elevated 6.8 years before AD dementia onset [40]. This result provides converging validity supporting the use of pNFL to predict disease progression in MCI. The fact that WM degeneration in familial AD can be detected almost 5 years earlier than in sporadic AD may be explained by their younger age and higher brain reserve [40]. Indeed, even low relative neurodegeneration occurring years before AD dementia may result in large absolute pNFL elevations. This highlights the importance of considering the age and pre-existing brain health status of individuals when interpreting their pNFL.

When used in isolation, MMSE and pNFL had similar ability to predict MCI5p. Combining the two markers resulted in significant sensitivity, specificity, and overall accuracy gains (Fig. 3A). Combining pNFL and MMSE resulted in 9%more MCI5p detection (Supplementary Figure 4A) and 15%fewer MCI5p misdiagnoses (Supplementary Figure 5A) compared to models using only MMSE or pNFL. This is consistent with other predictive studies combining dementia screeners with AD biomarkers [13 , 42], and also suggests that pNFL and MMSE contain complementary and non-overlapping information regarding the risk of clinical deterioration to AD dementia. MMSE may act as a marker of cognitive reserve and past neurobiological damage, while pNFL reflects current axonal degeneration [22]. In keeping with the cognitive reserve theory [30], individuals with mild MCI may be able to sustain higher levels of axonal degeneration compared to individuals with more severe cognitive impairment. A clear understanding of an individual’s future functional state can be gained by evaluating their levels of axonal degeneration using pNFL, especially if there is prior evidence of significant cognitive and behavioral impairment.

The ability of baseline pNFL and MMSE to predict MCI5p was significantly improved by including the rate of change in MMSE in the 6–12 months following baseline assessment (Fig. 4A). Using MMSE change in addition to baseline pNFL and MMSE resulted in 20%more MCI5p detection (Supplementary Figure 4B) and 10%fewer MCI5p misdiagnoses (Supplementary Figure 5B) compared to using only baseline pNFL and MMSE. This is consistent with prior research which found that the rate of change in the Functional Assessment Questionnaire was a strong predictor of progression from MCI to AD dementia within 2 years [41] and that gradual cognitive decline (Fig. 2B) was a key indicator differentiating individuals who develop AD dementia from those who remain dementia-free [43].

In contrast, the rate of pNFL change in the 6–12 months following baseline assessment did not help improve MCI5p prediction. This is surprising because the rate of pNFL change was previously found to be a good predictor of disease progression in pre-symptomatic (CN) and MCI individuals carrying a familial AD mutation [40], and rates of pNFL change were found to be negatively correlated with rates of change in hippocampal volume, cortical thickness, cerebral metabolism and MMSE [24] (i.e., the faster pNFL rises, the faster these neurobiological and cognitive markers fall). It is possible that acceleration in axonal degeneration occurs earlier than 5 years prior to AD dementia before plateauing and persisting at high levels, or that the 6–12 months follow-up was insufficient to capture long term trends in pNFL. There may also be transient periods of accelerated neurodegeneration earlier in life which slowly erode cognitive reserves and take individuals ever so slightly closer to AD dementia [10]. To clarify these questions the longitudinal behavior of pNFL should be investigated in larger prospective studies and at earlier timepoints along the MCI-AD continuum.

The fact that pNFL is raised in several dementia and non-dementia neurological disorders makes it a relatively unspecific marker for AD [22, 25]. Additionally, neurodegeneration alone is not sufficiently specific to predict MCI progression to AD dementia. There are several neurodegenerative MCI phenotypes that are not primarily associated with AD pathology, as well as MCI individuals with underlying AD pathology who do not present with accelerated neurodegeneration [44]. Much like liver function biomarkers or cardiac troponins, pNFL should be interpreted in the wider clinical context and is no substitute for more specific testing to rule out non-AD causes of WM degeneration. AD is fundamentally characterized by the presence of Aβ and tau pathology, and blood biomarkers for these proteins could be used alongside pNFL to improve test specificity [45] (unfortunately, ADNI did not contain these plasma markers for our sample).

Strengths and limitations

This study has a number of limitations but also several strengths. While not directly comparable due to different sample characteristics, our models perform similarly to other published approaches which used neuroimaging, CSF and non-pNFL plasma-based biomarkers [13 , 46–49]. Our model has the benefit of being relatively simple and cheap, and to our knowledge is the first to use pNFL for prediction of progression from MCI to AD dementia. This article helps inform clinicians how pNFL can be used to obtain and interpret an MCI to AD dementia progression risk score. This study is limited by a relatively small sample size, particularly where longitudinal analyses are concerned. This study did not consider the Aβ, tau, and APOE ɛ4 status of individuals and as such must be interpreted with caution.

Future directions

Future studies should attempt to replicate our findings in a different population and investigate the prognostic usefulness of pNFL in subjective cognitive decline, vascular dementia, frontotemporal dementia and dementia with Lewy bodies. Our approach should be built on by including biomarkers such as plasma Aβ/tau which are more specific to AD. Studies would benefit from following individuals over longer periods with regular pNFL and MMSE testing, allowing careful investigation of the interaction between pNFL, MMSE and rates of change in these markers. Studies are needed to assess the effect of common comorbidities such as diabetes, kidney, and liver dysfunction on pNFL—as this may result in unreliable predictions.

CONCLUSIONS

pNFL is increased in MCI at least 2 years before clinical diagnosis of AD dementia and is associated with an increased risk of clinical deterioration. pNFL may be a useful biomarker for predicting the risk of MCI progression to AD dementia within 5 years but should be used in conjunction with a dementia screener, or other cognitive measures, and interpreted carefully due to its lack of specificity to AD and its inability to account for past neurobiological damage. External validation will be an important next step in further characterizing the psychometric properties of the proposed approach.

Footnotes

ACKNOWLEDGMENTS

Data collection and sharing for this work was funded by the Alzheimer’s Disease Neuroimaging Initiative (ADNI; National Institutes of Health Grant U01 AG024904). ADNI is funded by the National Institute on Aging, the National Institute of Biomedical Imaging and Bioengineering, and through generous contributions from the following: Alzheimer’s Association; Alzheimer Drug Discovery Foundation; BioClinica, Inc.; Biogen Idec, Inc.; Bristol-Myers Squibb Company; Eisai, Inc.; Elan Pharmaceuticals, Inc.; Eli Lilly and Company; F. Hoffmann-La Roche Ltd and its affiliated company Genentech, Inc.; GE Healthcare; Innogenetics, N.V.; IXICO Ltd.; Janssen Alzheimer Immunotherapy Research & Development, LLC.; Johnson & Johnson Pharmaceutical Research & Development LLC.; Medpace, Inc.; Merck & Co., Inc.; Meso Scale Diagnostics, LLC.; NeuroRx Research; Novartis Pharmaceuticals Corporation; Pfizer, Inc.; Piramal Imaging; Servier; Synarc, Inc.; and Takeda Pharmaceutical Company. The Canadian Institutes of Health Research is providing funds to support ADNI clinical sites in Canada. Private sector contributions are facilitated by the Foundation for the National Institutes of Health (). The grantee organization is the Northern California Institute for Research and Education, and the study is coordinated by the Alzheimer’s Disease Cooperative Study at the University of California, San Diego. ADNI data are disseminated by the Laboratory for Neuro Imaging at the University of Southern California.

Authors’ disclosures available online ().

The supplementary material is available in the electronic version of this article: .

References

Prince

, Bryce

, Albanese

, Wimo

, Ribeiro

, Ferri

(2013) The global prevalence of dementia: A systematic review and metaanalysis. Alzheimers Dement 9, 63–75.e.

(2020) 2020 Alzheimer’s disease facts and figures. Alzheimers Dement 16, 391–460.

Gauthier

, Reisberg

, Zaudig

, Petersen

, Ritchie

, Broich

, Belleville

, Brodaty

, Bennett

, Chertkow

, Cummings

, de Leon

, Feldman

, Ganguli

, Hampel

, Scheltens

, Tierney

, Whitehouse

, Winblad

(2006) Mild cognitive impairment. Lancet 367, 1262–1270.

Busse

, Angermeyer

, Riedel-Heller

(2006) Progression of mild cognitive impairment to dementia: A challenge to current thinking. Br J Psychiatry 189, 399–404.

Mitchell

, Shiri-Feshki

(2008) Temporal trends in the long term risk of progression of mild cognitive impairment: A pooled analysis. J Neurol Neurosurg Psychiatry 79, 1386–1391.

Pandya

, Clem

, Silva

, Woon

(2016) Does mild cognitive impairment always lead to dementia? A review. J Neurol Sci 369, 57–62.

El-Hayek

, Wiley

, Khoury

, Daya

, Ballard

, Evans

, Karran

, Molinuevo

, Norton

, Atri

(2019) Tip of the iceberg: Assessing the global socioeconomic costs of Alzheimer’s disease and related dementias and strategic implications for stakeholders. J Alzheimers Dis 70, 323–341.

Belleville

, Fouquet

, Hudon

, Zomahoun

HTV

, Croteau

(2017) Neuropsychological measures that predict progression from mild cognitive impairment to Alzheimer’s type dementia in older adults: A systematic review and meta-analysis. Neuropsychol Rev 27, 328–353.

Price

, Morris

(1999) Tangles and plaques in nondemented aging and “preclinical” Alzheimer’s disease. Ann Neurol 45, 358–368.

10.

Jack

, Knopman

, Jagust

, Petersen

, Weiner

, Aisen

, Shaw

, Vemuri

, Wiste

, Weigand

, Lesnick

, Pankratz

, Donohue

, Trojanowski

(2013) Tracking pathophysiological processes in Alzheimer’s disease: An updated hypothetical model of dynamic biomarkers. Lancet Neurol 12, 207–216.

11.

Villemagne

, Burnham

, Bourgeat

, Brown

, Ellis

, Salvado

, Szoeke

, Macaulay

, Martins

, Maruff

, Ames

, Rowe

, Masters

(2013) Amyloid β deposition, neurodegeneration, and cognitive decline in sporadic Alzheimer’s disease: A prospective cohort study. Lancet Neurol 12, 357–367.

12.

Devanand

, Liu

, Tabert

, Pradhaban

, Cuasay

, Bell

, de Leon

, Doty

, Stern

, Pelton

(2008) Combining early markers strongly predicts conversion from mild cognitive impairment to Alzheimer’s disease. Biol Psychiatry 64, 871–879.

13.

Tabatabaei-Jafari

, Walsh

, Shaw

, Cherbuin

(2018) A simple and clinically relevant combination of neuroimaging and functional indexes for the identification of those at highest risk of Alzheimer’s disease. Neurobiol Aging 69, 102–110.

14.

Ottoy

, Niemantsverdriet

, Verhaeghe

, De Roeck

, Struyfs

, Somers

, wyffels

, Ceyssens

, Van Mossevelde

, Van den Bossche

, Van Broeckhoven

, Ribbens

, Bjerke

, Stroobants

, Engelborghs

, Staelens

(2019) Association of short-term cognitive decline and MCI-to-AD dementia conversion with CSF, MRI, amyloid- and 18F-FDG-PET imaging. Neuroimage Clin 22, 101771.

15.

Korolev

, Symonds

, Bozoki

, Alzheimer’s Disease Neuroimaging Initiative (2016) Predicting progression from mild cognitive impairment to Alzheimer’s dementia using clinical, MRI, and plasma biomarkers via probabilistic pattern classification. PLoS One 11, e0138866.

16.

Molinuevo

, Ayton

, Batrla

, Bednar

, Bittner

, Cummings

, Fagan

, Hampel

, Mielke

, Mikulskis

, O’Bryant

, Scheltens

, Sevigny

, Shaw

, Soares

, Tong

, Trojanowski

, Zetterberg

, Blennow

(2018) Current state of Alzheimer’s fluid biomarkers. Acta Neuropathol 136, 821–853.

17.

Blennow

, Hampel

, Weiner

, Zetterberg

(2010) Cerebrospinal fluid and plasma biomarkers in Alzheimer disease. Nat Rev Neurol 6, 131–144.

18.

Olsson

, Lautner

, Andreasson

, Öhrfelt

, Portelius

, Bjerke

, Hölttä

, Rosén

, Olsson

, Strobel

(2016) CSF and blood biomarkers for the diagnosis of Alzheimer’s disease: A systematic review and meta-analysis. Lancet Neurol 15, 673–684.

19.

Baker

, Lim

, Pietrzak

, Hassenstab

, Snyder

, Masters

, Maruff

(2017) Cognitive impairment and decline in cognitively normal older adults with high amyloid-β: A meta-analysis. Alzheimers Dement (Amst) 6, 108–121.

20.

Hansson

, Zetterberg

, Buchhave

, Londos

, Blennow

, Minthon

(2006) Association between CSF biomarkers and incipient Alzheimer’s disease in patients with mild cognitive impairment: A follow-up study. Lancet Neurol 5, 228–234.

21.

Brier

, Gordon

, Friedrichsen

, McCarthy

, Stern

, Christensen

, Owen

, Aldea

, Su

, Hassenstab

, Cairns

, Holtzman

, Fagan

, Morris

, Benzinger

TLS

, Ances

(2016) Tau and Aβ imaging, CSF measures, and cognition in Alzheimer’s disease. Sci Transl Med 8, 338ra66.

22.

Bridel

, van Wieringen

, Zetterberg

, Tijms

, Teunissen

, Alvarez-Cermeño

, Andreasson

, Axelsson

, Bäckström

, Bartos

(2019) Diagnostic value of cerebrospinal fluid neurofilament light protein in neurology: A systematic review and meta-analysis. JAMA Neurol 76, 1035–1048.

23.

Mattsson

, Andreasson

, Zetterberg

, Blennow

; Alzheimer’s Disease Neuroimaging Initiative (2017) Association of plasma neurofilament light with neurodegeneration in patients with Alzheimer disease. JAMA Neurol 74, 557–566.

24.

Mattsson

, Cullen

, Andreasson

, Zetterberg

, Blennow

(2019) Association between longitudinal plasma neurofilament light and neurodegeneration in patients with Alzheimer disease. JAMA Neurol 76, 791–799.

25.

Khalil

, Teunissen

, Otto

, Piehl

, Sormani

, Gattringer

, Barro

, Kappos

, Comabella

, Fazekas

, Petzold

, Blennow

, Zetterberg

, Kuhle

(2018) Neurofilaments as biomarkers in neurological disorders. Nat Rev Neurol 14, 577–589.

26.

Mayeli

, Mirshahvalad

, Aghamollaii

, Tafakhori

, Abdolalizadeh

, Rahmani

, Initiative

ADN

(2019) Plasma neurofilament light chain levels are associated with cortical hypometabolism in Alzheimer disease signature regions. J Neuropathol Exp Neurol 78, 709–716.

27.

Mielke

, Syrjanen

, Blennow

, Zetterberg

, Vemuri

, Skoog

, Machulda

, Kremers

, Knopman

, Jack

, Petersen

, Kern

(2019) Plasma and CSF neurofilament light. Neurology 93, e252–e260.

28.

Nyberg

, Lundquist

, Nordin Adolfsson

, Andersson

, Zetterberg

, Blennow

, Adolfsson

(2020) Elevated plasma neurofilament light in aging reflects brain white-matter alterations but does not predict cognitive decline or Alzheimer’s disease.e. Alzheimers Dement (Amst) 12, 12050.

29.

Osborn

, Khan

, Kresge

, Bown

, Liu

, Moore

, Gifford

, Acosta

LMY

, Bell

, Hohman

, Blennow

, Zetterberg

, Jefferson

(2019) Cerebrospinal fluid and plasma neurofilament light relate to abnormal cognition. Alzheimers Dement (Amst) 11, 700–709.

30.

Stern

(2009) Cognitive reserve. Neuropsychologia 47, 2015–2028.

31.

McKhann

, Knopman

, Chertkow

, Hyman

, Jack

Jr , Kawas

, Klunk

, Koroshetz

, Manly

, Mayeux

, Mohs

, Morris

, Rossor

, Scheltens

, Carrillo

, Thies

, Weintraub

, Phelps

(2011) The diagnosis of dementia due to Alzheimer’s disease: Recommendations from the National Institute on Aging-Alzheimer’s Association workgroups on diagnostic guidelines for Alzheimer’s disease. Alzheimers Dement 7, 263–269.

32.

Petersen

, Aisen

, Beckett

, Donohue

, Gamst

, Harvey

, Jack

, Jagust

, Shaw

, Toga

, Trojanowski

, Weiner

(2010) Alzheimer’s Disease Neuroimaging Initiative (ADNI). Neurology 74, 201–209.

33.

Folstein

, Folstein

, McHugh

(1975) “Mini-mental state”: A practical method for grading the cognitive state of patients for the clinician. J Psychiatr Res 12, 189–198.

34.

Therneau

, Crowson

, Atkinson

(2017) Using time dependent covariates and time dependent coefficients in the cox model. Surviv Vignettes.

35.

Bates

, Mächler

, Bolker

, Walker

(2015) Fitting linear mixed-effects models using lme4. J Stat Softw 67, doi: 10.18637/jss.v067.i01.

36.

Kuhn

(2008) Building predictive models in R using the caret package. J Stat Softw 28, doi: 10.18637/jss.v028.i05.

37.

Bosch

, Arenaza-Urquijo

, Rami

, Sala-Llonch

, Junqué

, Solé-Padullés

, Peña-Gómez

, Bargalló

, Molinuevo

, Bartrés-Faz

(2012) Multiple DTI index analysis in normal aging, amnestic MCI and AD. Relationship with neuropsychological performance. Neurobiol Aging 33, 61–74.

38.

Zhang

, Schuff

, Camacho

, Chao

, Fletcher

, Yaffe

, Woolley

, Madison

, Rosen

, Miller

, Weiner

(2013) MRI markers for mild cognitive impairment: Comparisons between white matter integrity and gray matter volume measurements. PLoS One 8, e66367.

39.

Serra

, Cercignani

, Lenzi

, Perri

, Fadda

, Caltagirone

, Macaluso

, Bozzali

(2010) Grey and white matter changes at different stages of Alzheimer’s disease. J Alzheimers Dis 19, 147–159.

40.

Preische

, Schultz

, Apel

, Kuhle

, Kaeser

, Barro

, Gräber

, Kuder-Buletta

, LaFougere

, Laske

, Vöglein

, Levin

, Masters

, Martins

, Schofield

, Rossor

, Graff-Radford

, Salloway

, Ghetti

, Ringman

, Noble

, Chhatwal

, Goate

, Benzinger

TLS

, Morris

, Bateman

, Wang

, Fagan

, McDade

, Gordon

, Jucker

; Dominantly Inherited Alzheimer Network (2019) Serum neurofilament dynamics predicts neurodegeneration and clinical progression in presymptomatic Alzheimer’s disease. Nat Med 25, 277–283.

41.

Gomar

, Bobes-Bascaran

, Conejero-Goldberg

, Davies

, Goldberg

, Alzheimer’s Disease Neuroimaging Initiative (2011) Utility of combinations of biomarkers, cognitive markers, and risk factors to predict conversion from mild cognitive impairment to Alzheimer disease in patients in the Alzheimer’s Disease Neuroimaging Initiative. Arch Gen Psychiatry 68, 961–969.

42.

Tabatabaei-Jafari

, Shaw

, Walsh

, Cherbuin

(2020) Cognitive/functional measures predict Alzheimer’s disease, dependent on hippocampal volume. J Gerontol B Psychol Sci Soc Sci 75, 1393–1402.

43.

Wilson

, Leurgans

, Boyle

, Bennett

(2011) Cognitive decline in prodromal Alzheimer disease and mild cognitive impairment. Arch Neurol 68, 351–356.

44.

Burnham

, Bourgeat

, Doré

, Savage

, Brown

, Laws

, Maruff

, Salvado

, Ames

, Martins

, Masters

, Rowe

, Villemagne

(2016) Clinical and cognitive trajectories in cognitively healthy elderly individuals with suspected non-Alzheimer’s disease pathophysiology (SNAP) or Alzheimer’s disease pathology: A longitudinal study. Lancet Neurol 15, 1044–1053.

45.

Palmqvist

, Janelidze

, Quiroz

, Zetterberg

, Lopera

, Stomrud

, Su

, Chen

, Serrano

, Leuzy

, Mattsson-Carlgren

, Strandberg

, Smith

, Villegas

, Sepulveda-Falla

, Chai

, Proctor

, Beach

, Blennow

, Dage

, Reiman

, Hansson

(2020) Discriminative accuracy of plasma Phospho-tau217 for Alzheimer disease vs other neurodegenerative disorders. JAMA 324, 772–781.

46.

Hye

, Riddoch-Contreras

, Baird

, Ashton

, Bazenet

, Leung

, Westman

, Simmons

, Dobson

, Sattlecker

, Lupton

, Lunnon

, Keohane

, Ward

, Pike

, Zucht

, Pepin

, Zheng

, Tunnicliffe

, Richardson

, Gauthier

, Soininen

, Kłoszewska

, Mecocci

, Tsolaki

, Vellas

, Lovestone

(2014) Plasma proteins predict conversion to dementia from prodromal disease. Alzheimers Dement 10, 799–807.e2.

47.

Davatzikos

, Bhatt

, Shaw

, Batmanghelich

, Trojanowski

(2011) Prediction of MCI to AD conversion, via MRI, CSF biomarkers, and pattern classification. Neurobiol Aging 32, 2322.e19–2322.e27.

48.

Zheng

, Kong

, Cui

, Wei

, Zhang

, Wei

(2016) Conversion from MCI to AD in patients with the APOE ɛ4 genotype: Prediction by plasma HCY and serum BDNF. Neurosci Lett 626, 19–24.

49.

Ritchie

, Smailagic

, Noel-Storr

, Takwoingi

, Flicker

, Mason

, McShane

(2014) Plasma and cerebrospinal fluid amyloid beta for the diagnosis of Alzheimer’s disease dementia and other dementias in people with mild cognitive impairment (MCI). Cochrane Database Syst Rev 2014, CD008782.