More Draws, Better Decisions: Blood Biomarker Variability in Longitudinal Trial Design
A lumbar puncture schedule caps a trial at a handful of CSF timepoints per participant. A blood draw schedule removes that cap, and most trial teams book the extra timepoints as free statistical power. They are not free. Blood biomarker variability in neurodegenerative disease decides whether each additional draw sharpens a patient’s trajectory or adds noise to it, and that outcome is fixed before the first sample is collected: in the analyte, the handling protocol, the schedule, and where in the body the signal originated.
Key Takeaways
- Repeat blood sampling improves slope estimation only when blood biomarker variability within each person is small relative to the change a trial needs to detect.
- Published reference change values for plasma CNS markers run from roughly 14% to over 40%, which sets a floor on the individual-level effect a pharmacodynamic readout can resolve.
- Over a multi-year trial, kidney function and sample handling drift alongside disease, and mixed-origin plasma measurements cannot separate the two.
- Sampling schedules should be built backward from the decision each timepoint informs, with baseline replicates, locked handling, and change thresholds written into the protocol.
How Does Blood Biomarker Variability Affect Neurodegenerative Disease Trials?
Blood biomarker variability in neurodegenerative disease sets the smallest change a trial can detect in an individual participant. Repeat sampling tightens trajectory estimates only when within-person variation is small relative to the disease or drug effect, and only when that variation reflects brain biology rather than kidney function, body composition, or sample handling.
That definition splits the variance in any serial plasma measurement into three parts. Analytical variation comes from the assay. Within-subject biological variation is the week-to-week scatter a healthy or stable person shows around their own set point. The third part is the one trial designers want: change driven by pathology or by drug. Longitudinal designs succeed when the third part is large relative to the first two, and they fail quietly when it is not. The rest of this article works through what that means for schedule design, and why the provenance of the measured analyte becomes a design variable once sampling density is no longer limited by access.
More Timepoints, Fewer Patients: What a Lumbar Puncture Alternative Buys Statistically
The case for blood as a lumbar puncture alternative in neurology research usually starts with tolerability, and the data support it. In the Parkinson’s Progression Markers Initiative, 3,479 participants were scheduled for lumbar punctures annually for five years and every two years after that. Of those who attempted at least one procedure, 29.5% experienced an adverse event, and while baseline compliance was 90%, compliance in the Parkinson’s cohort fell by 39.4% between baseline and year five [1]. A protocol that depends on serial CSF loses data exactly where slope estimates need it most: late timepoints.
The statistical argument is less often made explicit. Most progression and treatment-effect analyses in neurodegeneration fit linear mixed-effects models, which estimate a slope for each participant and a mean slope for each arm. The precision of an individual slope depends on how many measurements define it and how much residual scatter surrounds each one. Add timepoints and the individual slope estimate tightens. Tighter individual slopes can, in turn, reduce between-participant variance in the fitted trajectories, which is what drives sample size.
How large can that effect be? In a 9-month study of at-home monitoring in ALS, investigators estimated that detecting a 30% slowing in functional decline would require 274 participants per arm with monthly assessment, but 73 per arm with weekly assessment [2]. A subsequent analysis of a large pooled ALS trial database found only modest sample-size gains from the same increase in frequency [3]. The original authors’ reply identified the reason for the gap: their self-administered measure carried more within-person noise than the same scale collected by trained raters, so frequent sampling had more noise to average out [4].
That exchange is the design lesson in miniature. Sampling density pays off in proportion to within-subject blood biomarker variability. The noisier the individual measurement, the more extra draws help, and the more a single draw misleads. Density does not change what the noise is made of, and it cannot turn a signal that tracks the wrong biology into one that tracks the right one. The wider case for denser sampling as an endpoint mechanism is covered in our analysis of how blood-based CNS biomarkers improve trial endpoints.
Figure 1. How sampling design changes the information a trial can extract
| Design feature | Serial CSF schedule | Serial blood schedule |
|---|---|---|
| Practical timepoints per participant per year | 1–2 | Limited by visit schedule, not procedure burden |
| Late-timepoint completeness | Declines with each procedure | Largely preserved |
| Early post-dose sampling | Rarely feasible | Feasible at days to weeks |
| Dominant source of individual slope error | Too few points | Within-subject and peripheral variation |
| Brain specificity of the analyte | High by compartment | Depends on analyte and sample preparation |
Measuring Blood Biomarker Variability Before Measuring Drug Effect
Clinical chemistry has a formal vocabulary for the question every longitudinal protocol has to answer. The within-subject coefficient of variation (CVI) describes how much one person’s values scatter around their own set point over weeks. The between-subject coefficient (CVG) describes spread across a population. Combined with analytical variation, CVI yields the reference change value (RCV): the percentage change between two serial results that must be exceeded before it can be attributed to something other than normal fluctuation.
For most CNS blood markers, formal estimates of blood biomarker variability were unavailable until recently. A European biological variation study sampled healthy adults weekly for 10 weeks and measured amyloid-β peptides, three phosphorylated tau species, GFAP, and NfL in plasma [5]. CVI ranged from about 3% for the Aβ42/Aβ40 ratio to about 16% for p-tau181, with the remaining markers between 6% and 10%. Most RCVs fell between 20% and 30% for decreases and between 25% and 40% for increases. The same study observed unexpectedly high single values in healthy individuals, a direct warning against enrollment or response decisions built on one draw.
Healthy volunteers are not trial participants, so a memory clinic study repeated the exercise in patients: three plasma samples collected over 36 days from 27 people with Alzheimer’s disease and 20 without [6]. RCVs were lowest for the amyloid-β markers, at roughly a 14% decrease or 17% increase, and highest for p-tau181, at roughly a 30% decrease or 42% increase. Diagnostic group made no striking difference to RCV, which suggests these thresholds describe the measurement, not the disease stage.
Figure 2. Published short-term biological variation for plasma CNS biomarkers.
| Population | Sampling design | Within-subject CV | Reference change value | Source |
|---|---|---|---|---|
| Healthy adults | Weekly for 10 weeks | ~3% (Aβ42/Aβ40) to ~16% (p-tau181); others 6–10% | Mostly 20–30% decrease, 25–40% increase | [5] |
| Memory clinic patients (AD and non-AD) | Three draws within 36 days | Not summarized here | ~14% decrease / ~17% increase (Aβ) to ~30% decrease / ~42% increase (p-tau181) | [6] |
Three design consequences follow. First, a pharmacodynamic effect smaller than the RCV cannot be called in an individual participant from two draws; it can only be detected at group level, and only with enough participants and timepoints to average the scatter away. Second, baseline is the most important timepoint in a longitudinal protocol and the one most often collected once. A single baseline carries the full within-subject variance into every change score computed against it. Third, markers with low CVI are not automatically better monitoring tools. The Aβ ratio is the most stable marker in these data, but it also changes least between patients and controls [5]. What matters is the ratio of expected change to RCV, not stability in isolation.
Peripheral Tissue Confounds: What Drifts Over a Multi-Year Trial Besides the Disease
Biological variation studies capture blood biomarker variability over weeks. Disease-modifying trials in neurodegeneration run for years, and over that window a second category of variance appears: slow, directional change in the physiology that clears, dilutes, or contributes to the plasma pool. This is blood biomarker noise from peripheral tissue confounds, and it does not average out with more draws, because it is not random. It trends.
Kidney function is the clearest example. In the Gothenburg H70 Birth Cohort, 744 septuagenarians had plasma NfL, kidney function, and brain MRI measured, and 313 also provided CSF [7]. Plasma NfL was associated with both prevalent and incident chronic kidney disease. CSF markers of neurodegeneration and structural MRI measures did not differ by kidney status. The same analyte, measured in two compartments, told two different stories, and only the plasma measurement moved with a non-neurological variable.
Now place that finding inside a four-year trial in an older population. Estimated glomerular filtration rate declines with age, and it declines faster in some participants than others. A plasma NfL slope in that trial is a composite of neuroaxonal injury and renal clearance, and the placebo arm will show apparent progression that the brain did not produce. Blood biomarker variability in neurodegenerative disease trials, at this timescale, is not only scatter. It is drift with a physiological cause outside the CNS.
Sample handling adds a third, operational source. A Global Biomarker Standardization Consortium study tested collection tube type, hemolysis, centrifugation settings, processing and storage delays, tube transfers, and freeze-thaw cycles [8]. Every assessed marker varied by more than 10% depending on collection tube alone. Amyloid-β peptides declined by more than 10% under storage and centrifugation delays, faster at room temperature than at 2–8°C, while NfL and GFAP rose by more than 10% under room-temperature or −20°C storage. Phosphorylated tau species were comparatively stable. In a multi-site, multi-year study, tube lots change, courier times vary, and site staff turn over, so handling effects can enter the data as apparent longitudinal change.
Plasma NfL Variability and Patient Stratification Over Time
Plasma NfL variability matters for patient stratification as much as for endpoints. Enrichment designs frequently use a baseline NfL threshold to select faster progressors, and in presymptomatic carriers of frontotemporal lobar degeneration mutations, higher baseline plasma NfL predicted short-term clinical progression in both a discovery and a validation cohort [9]. That prognostic value is real. It is also exposed to every source of variance described above. A participant whose kidney function declines during a long run-in may cross a stratification threshold without any change in the brain, and a participant sampled once at screening may be classified on a value that sits at the high end of their normal scatter.
The common thread is provenance. Within-subject scatter can be averaged with replicates, and handling effects can be controlled with protocol discipline. Peripheral drift cannot be controlled at the collection site, because it is a property of what is being measured. When the analyte in the tube is a mixture of brain-derived and peripheral contributions, no sampling schedule separates them.
One way through is to measure brain-origin cargo directly. Neuron-derived extracellular vesicles carry intracellular proteins from their parent cells, and a study of 704 patients showed that tau and TDP-43 measured in plasma extracellular vesicles distinguished tau and TDP-43 proteinopathies, including behavioral-variant frontotemporal dementia subgroups defined by genetics or neuropathology [10]. For longitudinal design, the relevant property is cell-of-origin restriction: immunoaffinity enrichment for neuronal origin narrows the contributing cell population before the assay runs, so the readout reports on a defined source rather than the soluble pool. Whether that restriction also insulates serial measurements from renal and other peripheral drift is a testable question, and it has not yet been characterized to the standard of the biological variation studies above. The ExoSORT™ platform was built around this constraint, isolating neuron-derived vesicles from plasma so that serial measurements are anchored to a defined cellular source rather than to the whole-body plasma pool.
Pharmacodynamic Monitoring in Neurodegeneration Trials: Building the Sampling Schedule
A sampling schedule should be built backward from the decision each timepoint informs, and from the blood biomarker variability expected at each one. Pharmacodynamic monitoring in neurodegeneration trials and disease progression monitoring ask different questions, and a single evenly spaced schedule rarely answers both well.
Pharmacodynamic questions ask whether the drug engaged its target and how fast. They need density early: a pre-dose baseline, then draws at intervals matched to the expected kinetics of the analyte after dosing, often at days to weeks rather than months. The objective is to characterize onset and magnitude before disease progression contributes meaningful change. Here, RCVs determine whether an individual-level response is interpretable at all, and whether the pharmacodynamic claim must rest on group-level differences.
Progression questions ask whether the drug changed the disease trajectory. They need duration more than density: evenly spaced timepoints across the longest feasible window, with enough points per participant to fit a slope that is not dominated by a single outlying value. At this timescale, peripheral drift becomes the dominant threat, and the covariates that drive it must be captured at every visit.
Several design elements apply to both. Baseline should be a replicate, not a single draw: two or three samples collected within the screening window reduce the within-subject variance carried into every subsequent change score. Handling should be locked in the protocol, including tube type, time to centrifugation, storage temperature, and freeze-thaw limits, because the consortium data show each can shift values by more than 10% [8]. Samples from the same participant should be analyzed in the same analytical run where the design allows it, and reagent lot changes across a multi-year study need a pre-specified bridging plan.
Regulatory expectations point the same way. FDA’s January 2025 guidance on bioanalytical method validation for biomarkers endorses a fit-for-purpose approach, with the extent of validation set by the intended use of the measurement [11]. For pharmacodynamic monitoring in neurodegeneration trials, fit-for-purpose means demonstrating that the assay and the handling protocol can resolve changes smaller than the effect the trial is powered to detect. That demonstration is far easier to build before first patient in than to reconstruct from banked samples after unblinding.
Figure 3. Longitudinal blood sampling design checklist
| Design decision | What to specify | Variance it controls |
|---|---|---|
| Decision mapping | Which timepoints serve pharmacodynamic questions and which serve progression | Wasted density; underpowered late timepoints |
| Baseline replicates | Two or three draws within the screening window | Within-subject variation carried into change scores |
| Response thresholds | Individual-level change definitions tied to published or study-derived RCVs | False responders and non-responders |
| Handling protocol | Tube type, processing time, storage temperature, freeze-thaw limits | Pre-analytical shifts above 10% |
| Covariate capture | Kidney function and body composition at every visit | Peripheral drift in plasma levels |
| Analytical plan | Same-run analysis per participant; reagent lot bridging | Assay drift across years |
| Analyte provenance | Whether the measured signal is restricted to a CNS source | Peripheral contributions no schedule can remove |
The checklist is platform-agnostic by design, with one exception. Its first six rows reduce noise around a measurement. The last row decides what the measurement is. Teams at NeuroDex work with sponsors on that last question during protocol design, when the choice of analyte and sample preparation can still shape the schedule rather than being retrofitted to it. For related context on how serial markers behave in motor neuron disease, see our overview of biomarkers in ALS diagnosis and disease monitoring, and for the proteinopathy background behind vesicle-based measurement, our review of TDP-43 in ALS and neurodegeneration.
What Comes Next for Clinical Trial Biomarkers Measured Over Time
The biological variation data now available for soluble plasma markers took a coordinated, multi-center effort to generate, and it remains limited to a small set of analytes and short sampling windows. No equivalent dataset on blood biomarker variability yet exists for brain-derived vesicle cargo at the standard those studies set: weekly sampling, healthy and patient cohorts, formal RCV estimates. That gap is the next piece of infrastructure the field needs, and whoever generates it will define how serial vesicle-based measurements are interpreted in trials.
Sampling density will keep increasing. Decentralized and home-based collection is already being piloted, and it moves the handling problem from a controlled site to a participant’s kitchen. That makes pre-analytical stability and analyte provenance more important, not less. Clinical trial biomarkers measured over time will be judged less by how sensitive the assay is and more by whether each timepoint can be trusted to report on the brain. Blood biomarker variability in neurodegenerative disease is where that judgment starts, and the teams that characterize it before their first draw will be the ones whose longitudinal data hold up at the decision point.
References
- Barbosa W, Schneider RB, Choi SH, et al. Safety and feasibility of serial lumbar punctures: long-term results from the Parkinson’s Progression Markers Initiative. Clin Park Relat Disord. 2025;13:100385. doi:10.1016/j.prdoa.2025.100385
- Rutkove SB, Narayanaswami P, Berisha V, et al. Improved ALS clinical trials through frequent at-home self-assessment: a proof of concept study. Ann Clin Transl Neurol. 2020;7(7):1148–1157. doi:10.1002/acn3.51096
- van Eijk RPA. Frequent self-assessments in ALS clinical trials: worthwhile or an unnecessary burden for patients? Ann Clin Transl Neurol. 2020;7(10):2074–2075. doi:10.1002/acn3.51178
- Rutkove SB, Narayanaswami P, Berisha V, et al. Reply to: Frequent self-assessments in ALS clinical trials: worthwhile or an unnecessary burden? Ann Clin Transl Neurol. 2020;7(10):2076–2077. doi:10.1002/acn3.51177
- Brum WS, Ashton NJ, Simrén J, et al. Biological variation estimates of Alzheimer’s disease plasma biomarkers in healthy individuals. Alzheimers Dement.2024;20:1284–1297. doi:10.1002/alz.13518
- Clemmensen FK, Gramkow MH, Simonsen AH, et al. Short-term variability of Alzheimer’s disease plasma biomarkers in a mixed memory clinic cohort. Alzheimers Res Ther. 2025;17(1):26. doi:10.1186/s13195-024-01658-7
- Dittrich A, Ashton NJ, Zetterberg H, et al. Association of chronic kidney disease with plasma NfL and other biomarkers of neurodegeneration: the H70 Birth Cohort Study in Gothenburg. Neurology. 2023;101:e277–e288. doi:10.1212/WNL.0000000000207419
- Verberk IMW, Gouda M, Antwi-Berko D, et al. Evidence-based standardized sample handling protocol for accurate blood-based Alzheimer’s disease biomarker measurement: results and consensus of the Global Biomarker Standardization Consortium. Alzheimers Dement. 2025;21(10):e70752. doi:10.1002/alz.70752
- Rojas JC, Wang P, Staffaroni AM, et al. Plasma neurofilament light for prediction of disease progression in familial frontotemporal lobar degeneration. Neurology.2021;96(18):e2296–e2312. doi:10.1212/WNL.0000000000011848
- Chatterjee M, Özdemir S, Fritz C, et al. Plasma extracellular vesicle tau and TDP-43 as diagnostic biomarkers in FTD and ALS. Nat Med. 2024;30(6):1771–1783. doi:10.1038/s41591-024-02937-4
- US Food and Drug Administration. Bioanalytical Method Validation for Biomarkers: Guidance for Industry. January 2025. https://www.fda.gov/regulatory-information/search-fda-guidance-documents/bioanalytical-method-validation-biomarkers
Leave a Reply