BGPT: Paper Review: A multicentre validation study of the diagnostic value of plasma neurofilament light

Explore by Goal

Quick Explanation Copied

Plasma NfL validation (multicentre, 2 cohorts + external ADNI) — what stands up, what doesn’t

This Nature Communications study reports that plasma neurofilament light (NfL) is broadly elevated across multiple neurodegenerative disorders and is particularly useful for (i) identifying atypical parkinsonian disorders vs PD and (ii) ruling in neurodegeneration with low false positives when using age-related cut-offs derived from amyloid-negative cognitively unimpaired controls.

Long Explanation

Paper Review: A multicentre validation study of the diagnostic value of plasma neurofilament light

Nature Communications • DOI: 10.1038/s41467-021-23620-z

Last update: 2026-04-30 Study type: diagnostic biomarker validation

Known vs inferred (what the paper directly shows vs what remains uncertain)

Known from the paper: plasma NfL is significantly higher in multiple neurodegenerative disorders and, in ROC analyses, can discriminate atypical parkinsonian disorders from PD with high specificity in both cohorts, with AUCs reported in the manuscript text.
Known from the paper: reference cut-offs using amyloid-negative cognitively unimpaired controls and age-related stratification are used to manage false positives; the paper reports low false-positive rates for several non-dementia groups with a 99% CI-based cutoff.
Inferred but not fully resolved: plasma NfL “indicates neurodegeneration” at the individual level; however, without neuropathological confirmation, clinical misclassification and overlapping pathologies could shift apparent diagnostic accuracy.
Uncertain: generalizability across assay platforms, pre-analytical pipelines, and different case-mix (e.g., community vs specialist clinics) remains dependent on external replication with standardized calibration materials—an issue the paper partially addresses via harmonization but still notes platform dependence.

Visuals (from reported metrics in the manuscript text)

Because the prompt provides only figure captions/selected numeric statements (not full raw tables), these plots use only the explicitly stated AUC/effect sizes and cutoff values shown in the supplied paper text.

Evidence basis for visuals: AUC and cutoff values come from the supplied manuscript text.

Structured critique

1) Design & cohort construction

Strength: Two independent multicentre cohorts with a relatively large total sample (2269) and explicit cross-cohort testing of cut-offs and harmonization via QC correction.
Concern: Primary diagnostic categories are clinically assigned; the paper does not provide neuropathology confirmation and acknowledges the risk of diagnostic uncertainty.
Concern: Some disorder categories have smaller n (e.g., certain atypical parkinsonian disorders), which can inflate variance in AUC estimates and widen confidence intervals even if point estimates look strong. The manuscript mentions underpowering in categories/comparisons.

2) Biomarker measurement & comparability

Strength: Plasma NfL is measured using Simoa (single-molecule array) with reported LLOQ and assay QC metrics in each cohort; the paper also evaluates correlation between assays via QC.
Concern: Even if two assays correlate, absolute cut-offs can shift; the authors note absolute concentrations differ between assays and that platform-dependent cutoffs would be needed.

3) Statistical strategy (ROC, effect sizes, and cut-off definitions)

Strength: ROC AUC is used to summarize discriminative performance; sensitivity/specificity and 95% CIs are reported via supplement tables (not provided in this prompt, but the manuscript text references them).
Key interpretability choice: The paper introduces multiple cut-off schemes: 99% CI, 95% CI, 90% CI (plus mean+2SD and Gaussian mixture modeling), derived in KCL and tested in Lund.
Potential blind spot: When using cut-offs, you must validate in a setting with similar prevalence and case-mix; AUC and “false positive rates” do not automatically translate into positive/negative predictive values in a new population without accounting for disease prevalence. The paper emphasizes false positives but does not (in the provided text) present PPV/NPV by assumed clinical prevalence.

Where the paper’s claims are strongest (and why)

Atypical parkinsonian syndromes vs PD: Multiple reported AUCs and very high specificities suggest that NfL may capture severity of axonal injury that differs between PD and atypical parkinsonism in these cohorts.
Age-aware “ruling in” with controlled false positives: The paper’s age-related cut-off approach aims to convert NfL (a broadly non-specific neuroaxonal injury marker) into a pragmatic clinical decision tool. This is conceptually consistent with known age-dependent baseline NfL distributions discussed in the manuscript background.
Concordance across cohorts: The manuscript describes replication of findings between KCL and Lund even when cut-offs derived in one cohort are tested in the other.

Blind spots / counterpoints (what could mislead a reader)

Disease specificity is limited by biology: NfL reflects axonal injury, which can occur across diverse neurodegenerative and non-degenerative contexts (e.g., vascular, inflammatory, traumatic). The paper acknowledges overlapping distributions across disorders, so “positive NfL” should be interpreted as “neuroaxonal injury/neurodegeneration is likely,” not a unique diagnosis.
Clinical misclassification risk: Without neuropathology confirmation, ROC/AUC performance might be inflated or deflated depending on diagnostic accuracy in the reference labels (especially non-AD dementias). The paper explicitly states this limitation.
Selection and spectrum bias: Memory clinics and neurology specialty cohorts can over-represent more clear-cut cases or particular severity strata; AUC is sensitive to the spectrum of disease severity and overlap with controls. This is a general diagnostic-study issue; while the paper is multicentre, it does not fully eliminate this. The paper discusses that some included diagnoses are likely at advanced stages, limiting conclusions about early disease.
Assay/protocol dependence remains: Even within Simoa approaches, the study uses harmonization correction factors. In other labs with different pre-analytical handling, absolute values may shift. This could undermine cut-off transferability.

Competing interests: what we can verify from the text provided

The manuscript acknowledgement/disclosure section indicates support and/or consulting/advisory relationships with multiple industry and foundation organizations for several authors, including pharmaceutical companies (e.g., Roche, Sanofi Genzyme, Biogen, Novartis) and assay/biomarker ecosystem stakeholders.

Conclusion (with skepticism)

Best-supported use-case: Using age-aware plasma NfL cut-offs to indicate neurodegeneration and help distinguish atypical parkinsonian disorders from PD, with reported strong discriminative metrics and low false positives in amyloid-negative CU-derived cut-off schemes.
Not solved: NfL is not an AD-pathology-specific marker, and the paper indicates limited ability to differentiate among certain cognitive impairment subtypes (e.g., low AUCs for AD dementia vs other cognitive impairment disorders in the provided narrative).
What would most likely disprove/reshape the conclusion: external replication showing markedly worse calibration/threshold transferability across labs and case-mix, and/or neuropathology-confirmed re-labeling that reduces reported diagnostic accuracy. The paper itself flags lack of neuropathology and platform dependence as key limitations.

Author reviews (bespoke links on BGPT)

Click to open BGPT’s author-focused reviews for the named authors available in the provided paper text.

Feedback:

Updated: April 30, 2026