Diagnostic accuracy
Blood GFAP and NfL predict dementia years ahead
Metadata + PaperFren explanation · cc by · source: Europe PMC
In 48,542 UK Biobank participants followed ~13 years, elevated plasma GFAP and NfL preceded dementia up to 15 years and improved prediction beyond CAIDE/DRS risk scores (AUC up to ~0.89).
How this study connects
Role on claims
Each row is a claim on a concept or method page where this paper supports, challenges, or qualifies the statement. Roles are hand-checked — not a model guess.
Blood GFAP and NfL associate with higher later dementia risk years before diagnosis.
In UK Biobank adults with baseline plasma measures, higher GFAP and NfL associated with roughly doubled hazards of incident all-cause dementia over ~13 years, with elevations detectable up to about 15 years before diagnosis and incremental prediction beyond demographic risk scores.
Evidence for the claim as stated.
In people with Alzheimer’s, tau and microglial PET can flag who declines faster.
In longitudinal AD imaging, baseline tau PET burden and microglial activation predicted subsequent cognitive decline; when markers were modelled together, posterior cortical tau and inflammation carried prediction while atrophy added little incremental signal.
Scope note — different population — volunteer biobank prognosis, not clinic AD decline
Limits the claim's scope: a different population, assay, or outcome.
Higher CSF tau can track early object-memory and hippocampal activity changes.
In cognitively unimpaired older adults, higher CSF tau correlated with worse object mnemonic discrimination and greater right hippocampal task fMRI activity—evidence that a fluid marker can track early functional change before a dementia diagnosis.
Scope note — different scale — biobank plasma prognosis, not small fMRI sample
Limits the claim's scope: a different population, assay, or outcome.
A clinic plasma panel plus genetics can discriminate clinical AD from older controls without being autopsy-confirmed diagnosis.
In the AD Cardiff Cohort (1,439 clinical AD cases, 508 older controls), plasma Aβ40/42, P-tau181, NfL, and GFAP plus APOE-ε4 and polygenic risk reached AUC 0.81. Aβ-related peptides were lower in cases; P-tau181, NfL, and GFAP were higher. Case-only GWAS linked the Aβ42/Aβ40 ratio to WWOX and COPG2. Authors note ~25% of clinical AD lack AD pathology at autopsy.
Scope note — biobank plasma GFAP/NfL forecast later dementia in volunteers — not same-day clinic discrimination
Limits the claim's scope: a different population, assay, or outcome.
Plasma p-tau217 can match CSF p-tau217 for PET amyloid/tau status in a memory clinic.
In 114 Geneva memory-clinic patients, plasma p-tau217 identified amyloid/tau PET positivity with mean AUC 0.96, outperforming plasma p-tau181/231 (AUC 0.76/0.79) and similar to CSF p-tau217 (AUC 0.95). CSF n was only 36.
Scope note — UK Biobank GFAP/NfL incident-dementia prognosis — not same-day PET discrimination
Limits the claim's scope: a different population, assay, or outcome.
Taiwan claims link gabapentin/pregabalin use to higher later dementia rates, especially under age 50.
Among 206,802 matched Taiwanese adults, gabapentin or pregabalin use associated with dementia incidence 980 vs 605 per 100,000 person-years (aHR 1.45), aHR 3.16 under age 50, and a cumulative-dose gradient.
Scope note — plasma injury-marker prognosis in volunteers — not a drug-exposure claims cohort
Limits the claim's scope: a different population, assay, or outcome.
Plasma GFAP/NfL forecast registry dementia years later in a volunteer cohort; tau/inflammation PET predict short-horizon decline in people already diagnosed with AD. Both can be valid—the limit is not treating prognosis and clinic decline as one assay.
Evidence for the claim as stated.
Clinic plasma+genetics discrimination, biobank GFAP/NfL prognosis, cortical tau PET, and postmortem retinal tau are related Alzheimer’s biology at different compartments, time horizons, and confirmation standards. They do not yield one interchangeable “accuracy number.”
Evidence for the claim as stated.
Phone pulse waveforms can detect atrial fibrillation accurately versus ECG in clinic samples.
Smartphone photoplethysmography detected atrial fibrillation with sensitivity about 95.6% and specificity about 96.6% versus ECG in an analysed clinic sample (adequate signal required; pacing excluded).
Scope note — different question — long-horizon prognosis, not point-of-care arrhythmia detection
Limits the claim's scope: a different population, assay, or outcome.
Diabetes risk scores trade accuracy against how many people you must screen locally.
Diabetes risk scores in an Amsterdam sample showed AUCs about 0.74–0.80 with numbers-needed-to-screen of 3–7 depending on the outcome definition—accuracy metrics tied to local prevalence and cut-offs.
Scope note — different task — prognostic plasma proteins, not cross-sectional diabetes scoring
Limits the claim's scope: a different population, assay, or outcome.
Higher blood GFAP and NfL associate with later dementia risk — prognosis, not same-day diagnosis.
In UK Biobank, higher baseline plasma GFAP and NfL were associated with roughly doubled hazards of incident dementia over ~13 years, with incremental discrimination when added to demographic risk scores—prognostic biomarker performance, not a bedside diagnostic rule-in test.
Evidence for the claim as stated.
Other tools (sickle cell, pain subtype, frailty trajectories) answer their own accuracy questions.
Other validated tools in this set include a point-of-care sickle-cell device detecting HbS/HbC at low percentages suitable for neonates, StEP pinprick signs highly sensitive/specific for neuropathic vs non-neuropathic pain in specialist clinics, and rapidly rising electronic frailty index trajectories associated with doubled 12-month mortality versus stable frailty.
Scope note — different prognostic biomarker — plasma proteins for dementia
Limits the claim's scope: a different population, assay, or outcome.
Point-of-care classification accuracy (AF, blood culture ID, sickle cell, pain subtype) is not the same claim as low-prevalence screening PPV or long-horizon prognostic discrimination. Mixing sensitivities with hazards or PPVs invents a false head-to-head.
Evidence for the claim as stated.
Higher plasma NfL associates with higher later dementia risk in UK Biobank adults.
In UK Biobank adults with baseline plasma measures, higher NfL (and GFAP) associated with roughly doubled hazards of incident all-cause dementia over ~13 years, with protein elevations detectable up to about 15 years before diagnosis and useful incremental prediction alongside demographic risk scores.
Evidence for the claim as stated.
In SOD1-ALS care, treatment can lower serum NfL as injury markers fall.
In German early-access SOD1-ALS care, tofersen treatment was accompanied by significant falls in serum NfL and CSF pNfH, with ALSFRS-R progression slower than patients’ pre-treatment slopes in the analyzed subset.
Scope note — different question — long-horizon prognosis, not treatment monitoring
Limits the claim's scope: a different population, assay, or outcome.
Higher CSF tau can track early memory and hippocampal activity changes before dementia.
In cognitively unimpaired older adults, higher CSF tau correlated with worse object mnemonic discrimination and greater right hippocampal task fMRI activity—evidence that a fluid neurodegeneration marker can track early functional brain changes before a dementia diagnosis.
Scope note — different scale and analyte — biobank plasma NfL/GFAP, not CSF tau fMRI
Limits the claim's scope: a different population, assay, or outcome.
UK Biobank treats baseline plasma NfL as a long-horizon prognostic marker for dementia in a general volunteer cohort. The tofersen series treats serial NfL/pNfH as a pharmacodynamic injury marker in SOD1-ALS. Both can be valid for their questions—the limit is not picking a winner, but not swapping prognostic and treatment-monitoring readings.
Evidence for the claim as stated.
Plasma GFAP/NfL at biobank scale forecast registry dementia years later; CSF tau in a small imaging sample tracks hippocampal hyperactivity and object memory. They share the idea that fluid markers can flag neurodegeneration early, while answering different assay, sample, and outcome questions.
Evidence for the claim as stated.
Open questions
Tensions this paper is part of
From concept pages' “where studies disagree.” Disagreement means the same question; scope means different assays, populations, or outcomes.
Plasma GFAP/NfL forecast registry dementia years later in a volunteer cohort; tau/inflammation PET predict short-horizon decline in people already diagnosed with AD. Both can be valid—the limit is not treating prognosis and clinic decline as one assay.
Clinic plasma+genetics discrimination, biobank GFAP/NfL prognosis, cortical tau PET, and postmortem retinal tau are related Alzheimer’s biology at different compartments, time horizons, and confirmation standards. They do not yield one interchangeable “accuracy number.”
Point-of-care classification accuracy (AF, blood culture ID, sickle cell, pain subtype) is not the same claim as low-prevalence screening PPV or long-horizon prognostic discrimination. Mixing sensitivities with hazards or PPVs invents a false head-to-head.
UK Biobank treats baseline plasma NfL as a long-horizon prognostic marker for dementia in a general volunteer cohort. The tofersen series treats serial NfL/pNfH as a pharmacodynamic injury marker in SOD1-ALS. Both can be valid for their questions—the limit is not picking a winner, but not swapping prognostic and treatment-monitoring readings.
Plasma GFAP/NfL at biobank scale forecast registry dementia years later; CSF tau in a small imaging sample tracks hippocampal hyperactivity and object memory. They share the idea that fluid markers can flag neurodegeneration early, while answering different assay, sample, and outcome questions.
History
When this study was placed
Dated entries from the concept change log — when this paper was added or removed as support, challenge, or qualifier on a claim.
Placed as a scope qualifier on Alzheimer’s and MCI markers
In the AD Cardiff Cohort (1,439 clinical AD cases, 508 older controls), plasma Aβ40/42, P-tau181, NfL, and GFAP plus APOE-ε4 and polygenic risk reached AUC 0.81. Aβ-related peptides were lower in cases; P-tau181, NfL, and GFAP were higher. Case-only GWAS linked the Aβ42/Aβ40 ratio to WWOX and COPG2. Authors note ~25% of clinical AD lack AD pathology at autopsy.
Placed as supporting evidence on Alzheimer’s and MCI markers
Clinic plasma+genetics discrimination, biobank GFAP/NfL prognosis, cortical tau PET, and postmortem retinal tau are related Alzheimer’s biology at different compartments, time horizons, and confirmation standards. They do not yield one interchangeable “accuracy number.”
Placed as a scope qualifier on Alzheimer’s and MCI markers
In 114 Geneva memory-clinic patients, plasma p-tau217 identified amyloid/tau PET positivity with mean AUC 0.96, outperforming plasma p-tau181/231 (AUC 0.76/0.79) and similar to CSF p-tau217 (AUC 0.95). CSF n was only 36.
Placed as a scope qualifier on Alzheimer’s and MCI markers
Among 206,802 matched Taiwanese adults, gabapentin or pregabalin use associated with dementia incidence 980 vs 605 per 100,000 person-years (aHR 1.45), aHR 3.16 under age 50, and a cumulative-dose gradient.
Placed as a scope qualifier on Diagnostic Accuracy
Smartphone photoplethysmography detected atrial fibrillation with sensitivity about 95.6% and specificity about 96.6% versus ECG in an analysed clinic sample (adequate signal required; pacing excluded).
Placed as a scope qualifier on Diagnostic Accuracy
Diabetes risk scores in an Amsterdam sample showed AUCs about 0.74–0.80 with numbers-needed-to-screen of 3–7 depending on the outcome definition—accuracy metrics tied to local prevalence and cut-offs.
Placed as supporting evidence on Diagnostic Accuracy
In UK Biobank, higher baseline plasma GFAP and NfL were associated with roughly doubled hazards of incident dementia over ~13 years, with incremental discrimination when added to demographic risk scores—prognostic biomarker performance, not a bedside diagnostic rule-in test.
Placed as a scope qualifier on Diagnostic Accuracy
Other validated tools in this set include a point-of-care sickle-cell device detecting HbS/HbC at low percentages suitable for neonates, StEP pinprick signs highly sensitive/specific for neuropathic vs non-neuropathic pain in specialist clinics, and rapidly rising electronic frailty index trajectories associated with doubled 12-month mortality versus stable frailty.
Placed as supporting evidence on Diagnostic Accuracy
Point-of-care classification accuracy (AF, blood culture ID, sickle cell, pain subtype) is not the same claim as low-prevalence screening PPV or long-horizon prognostic discrimination. Mixing sensitivities with hazards or PPVs invents a false head-to-head.
Study at a glance
- Design
- Cohort — UK Biobank prospective plasma GFAP/NfL with registry dementia follow-up
- N
- N=48542 · 1,312 incident all-cause dementia over ~13 years
- Population
- UK Biobank adults with baseline Olink plasma GFAP and NfL
- Outcome
- Incident all-cause dementia / ADRD HRs and prediction AUC for GFAP and NfL
Structured fields used in claim comparison tables when every cited study has a complete layer.
Key findings
Over 13.18 ± 2.42 years, 1,312 incident all-cause dementia cases occurred. GFAP >0.363 NPX linked to HR 2.25 for all-cause dementia (model 3); NfL >0.353 to HR 1.98. Proteins rose up to 15 years before diagnosis. DRSm plus GFAP/NfL reached AUC 0.867 (all-cause dementia) and 0.892 (ADRD). AD genetic risk correlated with both proteins.
Methodology
Prospective analysis of 48,542 UK Biobank adults with Olink plasma GFAP and NfL (NPX units) at baseline (2006–2010), followed through November 2022 for registry-based dementia. Authors used Cox models, cognition regressions, genetic correlations (AD-GRS, APOE*E4), and leave-one-region-out prediction validation.
Limitations
UK Biobank is healthier and mostly White; 97.7% had only baseline protein measures. GFAP/NfL are nonspecific—elevated in many neurological conditions. Registry dementia may miss mild cases.
Related papers in this topic
Same topic cluster — not a recommendation engine.