Therapeutic Plasma Exchange and Alzheimer's Disease: What the Research Shows

For most of the last two decades, Alzheimer's disease has been a condition medicine could describe far better than it could modify. That changed only recently, and only partially. Two anti-amyloid monoclonal antibodies — lecanemab and donanemab — now carry FDA approval after demonstrating statistically significant slowing of clinical decline. Both are meaningful advances. Both are also restricted to patients in the earliest symptomatic stages, require serial MRI surveillance, and carry a class-specific risk called amyloid-related imaging abnormalities (ARIA), which can involve brain swelling or microhemorrhage.van Dyck 2023Sims 2023

That leaves a substantial population without evidence-backed options — in particular, patients who have progressed past the mild stage. It is against that backdrop that therapeutic plasma exchange has drawn sustained research attention. TPE is a well-established apheresis procedure in which a defined volume of a patient's plasma is separated out and replaced with a substitute fluid, most commonly 5% human albumin. It has been used clinically for decades in neurologic, hematologic, and autoimmune disease, so its procedural safety profile is unusually well characterized for an investigational approach in this field.

The central point of interest is this: TPE with albumin replacement is one of the very few non-pharmacologic interventions in Alzheimer's disease with randomized, sham-controlled trial evidence of slowed cognitive and functional decline — and, unlike the approved antibodies, the strongest signal appeared in patients with moderate disease. This review covers what that evidence establishes, what it does not, and how it sits alongside the approved drugs.

Amyloid-beta (Aβ) is not confined to the brain. It circulates in blood, and approximately 90% of the amyloid-beta in plasma is bound to albumin. Albumin is therefore the principal peripheral reservoir for the same peptide that aggregates into plaque in the central nervous system.

The peripheral sink hypothesis holds that amyloid-beta exists in a dynamic equilibrium across the blood–brain barrier. If circulating, albumin-bound amyloid is removed from plasma and replaced with fresh therapeutic albumin that has unoccupied binding capacity, the equilibrium may shift — encouraging net movement of amyloid out of the brain compartment and into the periphery, where it can be cleared. Plasma exchange is a direct way to act on that equilibrium, because it removes the carrier protein and its bound cargo together.

There is a second, less-discussed dimension. Albumin in Alzheimer's patients is not simply present or absent; it is functionally altered. Circulating albumin accumulates oxidative and glycation modifications with age and disease, which impair its antioxidant capacity, its binding and transport functions, and its role in immune modulation. Exchange replaces structurally compromised endogenous albumin with intact therapeutic albumin, which is a distinct mechanism from amyloid removal alone and may account for effects that amyloid binding does not explain.

The second rationale is broader. Plasma carries pro-inflammatory cytokines, chemokines, autoantibodies, and other signaling factors implicated in neurodegeneration. Removing a volume of plasma removes a proportional share of those circulating mediators.

This is not purely theoretical in the AMBAR dataset. A dedicated analysis of 142 AMBAR participants found that plasma exchange with albumin replacement significantly reduced interferon-gamma, eotaxin, MIP-1α, and ICAM-1 in serum, and eotaxin-3 and MIP-1β in cerebrospinal fluid, at various points during treatment. The authors characterized the overall pattern of change as favoring reduced inflammation in treated patients in both compartments. Notably, rising serum MIP-1α correlated with cognitive decline in the placebo group but not in treated patients.Gonzalo 2024

AMBAR was a phase 2b/3 randomized, blinded, sham-controlled multicenter trial sponsored by Grifols. It enrolled 496 patients across 41 centers — 19 in Spain and 22 in the United States — of whom 347 were randomized. Eligible patients had mild-to-moderate Alzheimer's disease, defined by a Mini-Mental State Examination (MMSE) score of 18–26.Boada 2019

Patients were randomized 1:1:1:1 across four arms — three active and one sham control. The active arms differed in their replacement scheme: low-dose albumin alone, low-dose albumin plus intravenous immunoglobulin (IVIG), and high-dose albumin plus IVIG. The treatment structure was consistent across active arms and is the pattern most subsequent protocols have followed:

• Induction: six weekly conventional (high-volume) plasma exchanges, replacing approximately 2,500–3,000 mL of plasma per session
• Maintenance: twelve monthly low-volume plasma exchanges, on the order of 650–880 mL per session
• Total treatment period: 14 months

The co-primary endpoints were change from baseline to end of treatment on the ADAS-Cog (Alzheimer's Disease Assessment Scale–Cognitive Subscale, a cognitive measure) and the ADCS-ADL (Alzheimer's Disease Cooperative Study–Activities of Daily Living, a functional measure). Pre-specified secondary work included cognitive, functional and behavioral assessments, plasma and CSF biomarkers, and structural and functional brain imaging.

Pooled across the three active arms and compared with sham at 14 months, plasma-exchange–treated patients declined 52% less on the ADCS-ADL (P = .03) and 66% less on the ADAS-Cog (P = .06).Boada 2020

That second p-value deserves to be stated plainly rather than buried: the cognitive co-primary endpoint did not reach the conventional threshold of statistical significance in the pooled analysis. The functional endpoint did. A trial with two co-primary endpoints that meets one and narrowly misses the other is a mixed result, and describing the 66% figure without that context would misrepresent the study.

The subgroup analysis is where the picture sharpens. Patients were pre-stratified by baseline severity, and the two subgroups behaved very differently:

• Moderate AD (MMSE 18–21): 61% less decline on both co-primary endpoints — ADCS-ADL (P = .002) and ADAS-Cog (P = .05). This is the strongest and most statistically robust signal in the trial, and it is the source of the "up to 61% slower decline" figure cited elsewhere in this research index.
• Mild AD (MMSE 22–26): no significant differences between treated and sham patients.

The direction of that gradient is the opposite of what the anti-amyloid antibodies show, and it is the single most clinically interesting feature of the AMBAR dataset. It is also a subgroup finding from a trial whose pooled cognitive endpoint missed significance, which is exactly the situation in which subgroup results warrant caution rather than confidence.

The neuroimaging analyses were published separately in the European Journal of Nuclear Medicine and Molecular Imaging and covered 198 patients with analyzable serial MRI and 213 with complete FDG-PET data from baseline to month 14.Cuberas-Borrós 2022

On structural MRI, the high-dose albumin plus IVIG arm showed significant volume loss in 3 of 15 assessed subcortical structures, and was the only arm in which the right hippocampus showed no statistically significant reduction. On FDG-PET, the same arm showed the least metabolic decline: −2.9% versus −4.3% in placebo across all patients, and −4.2% versus −6.6% in the moderate-AD subgroup. SPM analysis found smaller metabolic defect areas in the posterior cingulate, precuneus, and parieto-temporal regions characteristic of Alzheimer's pathology.

The honest caveat is that the structural results were not dose-ordered. Placebo showed significant volume loss in 4 of 15 structures — fewer than the low-dose albumin arm (9 of 15) and the low-dose albumin plus IVIG arm (8 of 15). Only the highest-dose arm outperformed placebo. A clean dose–response relationship would have been considerably more persuasive than what was actually observed, and this pattern is more consistent with a genuine effect confined to the highest-intensity regimen than with a uniform class effect across all active arms.

Pre-specified biomarker endpoints included CSF and plasma Aβ40, Aβ42, total tau and phosphorylated tau. In the moderate-AD subgroup, the placebo group showed increases in total tau and phosphorylated tau relative to treated patients — directionally consistent with the clinical findings in the same subgroup. The inflammatory-mediator analysis discussed in the mechanism section above provides the most granular biological data from the trial, with significant reductions in several serum and CSF inflammatory markers in treated patients.Gonzalo 2024

Taken together, the sub-studies show biological changes moving in the expected direction across three independent modalities — structural, metabolic, and inflammatory. That coherence strengthens the case that the clinical signal reflects something real. It does not, on its own, establish clinical benefit.

Published in the Journal of Alzheimer's Disease in 2025, Taragano and colleagues reported outcomes from an Argentine cohort of 32 patients with mild-to-moderate Alzheimer's disease (mean age 72.1 years, 42.4% female, baseline MMSE 15–26), compared against 194 historical controls drawn from a 2008–2018 registry and matched on inclusion criteria.Taragano 2025

The protocol closely mirrored AMBAR's shape: six weekly intensive sessions followed by at least ten monthly maintenance sessions, with albumin replacement. Mean exchange volumes were 88.2% of estimated plasma volume during the intensive phase and 49.8% during maintenance. Critically for real-world applicability, the entire program was delivered on an outpatient basis using peripheral venous access — no central lines, no hospital admission.

The memory findings are the most striking — an 88% reduction in decline on immediate recall is a large effect by any standard in this field. The executive-function measures were more equivocal: phonemic verbal fluency and Trail Making Test-B both showed sizeable numerical differences that did not reach statistical significance in a cohort this small.

The real-world cohort provides the most directly applicable dataset. Across 514 procedures, 81.5% were entirely free of adverse events. The remaining 18.5% involved mild-to-moderate events, no severe events occurred, and the most frequent problems were related to venipuncture rather than to the exchange itself.Taragano 2025 AMBAR similarly reported a low overall adverse-event rate, with events occurring mainly during the high-volume induction phase rather than low-volume maintenance.Boada 2020

The events clinicians should anticipate are the familiar ones for apheresis:

• Hypotension during or shortly after a session, related to volume shifts — the most common event across TPE populations, and usually asymptomatic
• Citrate-related effects — perioral or peripheral tingling, numbness, or muscle cramping from the anticoagulant, reflecting transient reductions in ionized calcium
• Fatigue on the day of treatment and occasionally the day after
• Vascular access complications — bruising, infiltration, difficult venipuncture; the dominant category in the outpatient real-world cohort
• Hypogammaglobulinemia with repeated sessions, as immunoglobulins are removed along with other plasma proteins; this is cumulative and monitorable

Age is a live question in this population, since Alzheimer's patients skew older. A 2026 case-control study in the Journal of Clinical Apheresis examined this directly, comparing 249 TPE sessions in patients aged 75 and over with 257 sessions in patients under 75. Adverse event rates were statistically indistinguishable — 43.4% versus 40.6% of sessions (P = .592) — despite significantly higher rates of hypertension and cardiac disease in the older group. Asymptomatic hypocalcemia was more frequent in older patients, pointing toward closer calcium monitoring as a reasonable precaution.Coirier 2026

Standard practice before and during a treatment course includes baseline and periodic laboratory assessment (complete blood count, comprehensive metabolic panel, coagulation studies, immunoglobulin levels for extended courses), cardiovascular screening, assessment of venous access adequacy, and session-day vital sign monitoring. None of this requires imaging surveillance — a practical distinction from the anti-amyloid antibodies discussed below.

• Not FDA-approved for Alzheimer's disease. Any use for this indication is off-label. Regulatory approval requires a standard of evidence this literature has not met.
• One randomized trial, with a mixed primary result. AMBAR met one co-primary endpoint and missed the other. The field's convention — and the correct one — is that a single trial with a split primary outcome is a reason for a confirmatory trial, not a conclusion.
• The mild-subgroup ambiguity. Mild-AD patients showed no significant benefit. Whether this reflects a genuine biological threshold, a ceiling effect on the instruments used, or insufficient power in the subgroup is unresolved.
• Optimal protocol is unknown. AMBAR's four arms differed in albumin dose and IVIG, and the imaging sub-studies favored the highest-intensity arm. Exchange volume, albumin concentration, IVIG inclusion, session frequency, and maintenance duration have not been systematically optimized.
• Durability is unknown. Both AMBAR and the real-world cohort measured outcomes during ongoing maintenance. No published dataset establishes what happens after treatment stops.
• No head-to-head comparison with lecanemab, donanemab, or any other disease-modifying agent has been conducted, and none is currently registered.
• No long-term extension data. The antibody programs have published three- and four-year extension data suggesting benefits accumulate with continued treatment. TPE has no comparable long-term dataset — a real gap, and one that favors the antibodies on evidentiary maturity.

Research has continued. Grifols maintains an ongoing AMBAR clinical program in collaboration with Ace Alzheimer Center Barcelona, extending the original work.Grifols

Three contrasts are worth drawing out in plain terms.

The moderate-stage gap. Both approved antibodies are restricted to mild cognitive impairment and mild dementia. A patient who has progressed beyond that window is ineligible for either. Yet moderate-stage patients were precisely the group that responded most strongly in AMBAR — the only subgroup meeting significance on both co-primary endpoints. Whatever else is uncertain, this is the population with the fewest evidence-backed options, and TPE is one of very few interventions with randomized data pointing toward benefit in it.

No ARIA. ARIA is the defining safety liability of the anti-amyloid class, and it drives the entire MRI surveillance apparatus, the APOE4 genotyping, and the titration schedules built to manage it. It is a consequence of antibody-mediated amyloid clearance from cerebral vasculature. Plasma exchange has no such mechanism — it acts on the blood compartment and does not engage cerebral amyloid immunologically. This is a genuine structural difference, not a favorable trial result that might not replicate.

The antibodies hold the regulatory and evidentiary high ground — and this should be conceded in the same breath as the two points above. Lecanemab and donanemab each have large phase 3 trials that met their primary endpoints with high statistical significance, FDA approval, Medicare coverage, published long-term extension data, and demonstrated target engagement on amyloid PET. TPE has one trial that split its co-primary endpoints, one supportive non-randomized cohort, and no approval for this indication. A physician weighing options for an eligible early-stage patient is weighing a proven intervention against an investigational one, and the evidentiary asymmetry is substantial.

That combination places TPE in an unusual evidentiary position. It is better supported than most interventions marketed for cognitive health, and considerably less well supported than the two approved anti-amyloid antibodies. It carries no ARIA risk and requires no imaging surveillance, and it is the only approach with randomized data suggesting benefit in patients who have progressed past the mild stage — the population the approved drugs cannot treat. It also lacks a confirmatory trial, long-term durability data, and regulatory approval for this indication.

For a clinician or family evaluating options, the practical questions are the ordinary ones: disease stage, comorbidity, vascular access, the burden of a sustained treatment schedule, and cost in the absence of insurance coverage. Those are candidacy questions, and they are answered individually rather than from a page.