Previously in Part 1
In AMBAR Part 1, we explored the study design and primary clinical results. The trial demonstrated 52% less functional decline (ADCS-ADL, p=0.03) and 66% less cognitive decline (ADAS-Cog, p=0.06) over 14 months, with even stronger effects on global measures — 71% less decline on CDR-Sum of Boxes (p=0.002) and near-stabilization on clinical impression (p<0.0001). Moderate AD patients showed 61% less decline across all measures, while mild AD patients actually improved on global functional assessments. Benefits emerged gradually over 6–9 months and continued diverging through month 14. The clinical effects were clear and consistent. But as we'll see in this post, the mechanism of how plasma exchange works is far more complex — and potentially more interesting — than originally hypothesized.
The Biomarker Story: A Puzzle
One of the most intriguing — and perplexing — aspects of the AMBAR study is what happened with cerebrospinal fluid (CSF) biomarkers.
The study measured four key biomarkers at three timepoints:
- A-beta-40 and A-beta-42: The amyloid proteins that form plaques in Alzheimer's
- Total tau (T-tau): A protein that becomes elevated when neurons are damaged
- Phosphorylated tau (P-tau): The form of tau that aggregates into tangles
If plasma exchange worked purely by the "peripheral sink" mechanism — pulling amyloid out of the brain — we'd expect to see clear changes in CSF amyloid levels.
What They Found
Across all patients combined, amyloid markers showed no dramatic shifts — A-beta-40 showed no significant changes in either group, while A-beta-42 remained stable in the treatment group but decreased in the placebo group (borderline significant, p=0.06). Tau and P-tau showed no clear pattern at the population level.
The most meaningful signal emerged in moderate AD patients. A-beta-42 remained stable in the treatment group (+1.2 pg/mL at month 2, -7.3 pg/mL at month 14) but decreased significantly in placebo (-47.3 pg/mL at month 2, -59.8 pg/mL at month 14), with p=0.04 at month 2 and p=0.05 at month 14. T-tau increased substantially in placebo (+193.5 pg/mL) but remained stable in treatment (+35.5 pg/mL), p=0.002 at month 2. P-tau followed a similar pattern, increasing in placebo (+13.4 pg/mL) but stable in treatment (+5.7 pg/mL), p=0.04 at month 2.
In mild AD patients, the biomarker patterns were, in the authors' own words, "inconclusive or even counterintuitive" — no clear consistent effects emerged. This asymmetry between mild and moderate patients remains one of the study's unresolved puzzles.
What Does This Mean?
The biomarker results are puzzling but potentially revealing. Here's what we can infer:
1. It's not a simple amyloid-removal story
If plasma exchange worked solely by pulling amyloid out of the brain, we'd expect to see CSF A-beta-42 levels decrease in treated patients. Instead, they remained relatively stable. By contrast, placebo patients showed the expected progressive decline in CSF A-beta-42, as ongoing amyloid deposition traps soluble A-beta-42 into insoluble plaques, reducing its levels in the CSF.
The stabilization of CSF A-beta-42 in treated patients, despite ongoing removal of plasma-bound amyloid, suggests plasma exchange may restore a healthier dynamic equilibrium, preventing further net deposition in plaques rather than primarily dissolving existing ones. This is consistent with maintaining or enhancing physiological clearance mechanisms.
2. Tau dynamics are affected
The fact that tau proteins (markers of neuronal damage and tangle formation) increased in the placebo group but remained stable in treated patients with moderate AD suggests that plasma exchange may reduce ongoing neurodegeneration.
This is particularly interesting because plasma exchange doesn't directly target tau — tau isn't significantly present in blood plasma. Yet tau pathology was affected. This implies an indirect mechanism, perhaps through reducing inflammation or oxidative stress that drives tau pathology.
3. Multiple mechanisms are likely involved
The disconnect between strong clinical effects and unclear biomarker patterns suggests plasma exchange is working through multiple pathways, not just one simple mechanism.
The authors acknowledge this: "Our CSF biomarker results overall suggest that more than one mechanism may have been involved in the PE approach, perhaps including changes in oxidation status and inflammatory mediators, or could even be procedure-related."
"The disconnect between strong clinical effects and unclear biomarker patterns suggests plasma exchange is working through multiple pathways — not just one simple mechanism."
Independent Validation: The Argentina Study
One of the most common criticisms of AMBAR was that it was a single study sponsored by Grifols, the manufacturer of therapeutic albumin. Fair point — replication in independent populations is critical for confirming any medical finding.
In 2025, that replication arrived.
Real-World Evidence from Buenos Aires
A team in Argentina published results from a real-world cohort study comparing 32 patients who received plasma exchange to 194 historical controls from the same memory clinic. The study was published in the Journal of Alzheimer's Disease in January 2025.
Key differences from AMBAR:
- Smaller sample size (32 vs 322)
- Real-world setting, not a controlled trial
- Historical controls (patients treated 2008-2018) rather than concurrent randomization
- Modified protocol: 6 weekly intensive sessions (average 88.2% plasma volume exchange) followed by 10-12 monthly maintenance sessions (average 49.8% plasma volume exchange)
- All procedures done via peripheral access (no central lines)
- Independent investigators and funding source
The Results
The cognitive results were striking across every domain tested. On the MMSE, the treatment group declined 2.2 points less than controls over approximately 11 months — a 45% reduction in decline (p<0.001), nearly identical to AMBAR's effect size. Memory preservation was even more pronounced: 88% less decline in immediate recall (p<0.001), 74% less decline in delayed recall (p=0.04), and 69% less decline in recognition memory (p<0.001) on the RAVLT. Language and executive function followed the same pattern — 37% less decline in semantic fluency (p<0.001), 35% less decline in naming (p=0.03), and 49% less decline in phonemic fluency (p<0.001).
The safety profile was equally reassuring. 81.5% of 514 procedures were uneventful, there were no severe adverse events, no ARIA, and all procedures were completed via peripheral venous access — no central lines required. This last point is significant: AMBAR had shown higher adverse event rates with central venous catheters. The Argentina study demonstrated that excellent results are achievable with peripheral access alone, making the procedure less invasive and more broadly implementable.
Why This Matters
The Argentina study is critical for several reasons.
1. Independent replication: Different investigators, different country, different patient population — yet similar results. This dramatically strengthens confidence in the findings.
2. Real-world feasibility: This wasn't a highly controlled trial with strict protocols. This was actual clinical practice in a memory clinic. The fact that it worked reinforces that the treatment is practical and implementable.
3. Peripheral access confirmation: AMBAR showed higher adverse event rates with central venous catheters. The Argentina study demonstrated that excellent results can be achieved using only peripheral access, making the procedure less invasive and safer.
4. Extended cognitive battery: The Argentina study used different cognitive tests than AMBAR (RAVLT instead of ADAS-Cog). The fact that memory showed 74-88% preservation across different test paradigms suggests the effect is robust and not test-specific.
5. Effect size consistency: A 45% reduction in MMSE decline in the Argentina study is remarkably close to AMBAR's 52% effect on ADCS-ADL and the overall pattern of 50-70% effect sizes across measures.
When two independent studies, using slightly different protocols, in different populations, with different assessment tools, show similar magnitude of benefit, that's compelling evidence.
Putting the Results in Context: How Does AMBAR Compare?
Now that we understand what AMBAR (and the Argentina study) showed, let's put these results in context by comparing them to other Alzheimer's treatments.
Comparison to FDA-Approved Therapies
The following efficacy comparison draws on data presented at the ASFA 2024 conference in a webinar by Ace Alzheimer Center Barcelona titled "Alzheimer's disease, current treatments and the role of Plasma exchange" (February 7th, 2024), alongside published trial results for lecanemab and donanemab.
| Treatment | Patients | ADCS-ADL | ADAS-Cog | CDR-Sb | ARIA Risk |
|---|---|---|---|---|---|
| Aducamumab (ENGAGE) | 1,647 | 18% | 11% | -2% | 40% |
| Aducamumab (EMERGE 10mg/kg) | 1,638 | 40% | 27% | 22% | 40% |
| Lecanemab | 1,795 | 37% | 26% | 27% | 12.6% / 14% |
| Donanemab | 1,182 | 40% | 32% | 37% | 24% / 31.4% |
| AMBAR | 347 | 52% | 66% | 71% | 0.6% |
Several observations jump out:
1. AMBAR showed the largest effect sizes across all three primary outcome measures
This is particularly striking for the CDR-Sum of Boxes, where AMBAR's 71% effect size is nearly double that of the best-performing monoclonal antibody (donanemab at 37%).
2. Sample size context
AMBAR was considerably smaller (322 patients) than the antibody trials (1,182-1,795 patients). Smaller studies generally have more variable results, so the fact that AMBAR showed larger effects despite smaller size suggests a robust treatment effect.
3. Safety advantage
AMBAR's ARIA risk was essentially zero (0.6%, likely representing incidental findings unrelated to treatment), compared with 12.6%-40% for monoclonal antibodies.
ARIA (Amyloid-Related Imaging Abnormalities) comes in two forms:
- ARIA-E (edema): Brain swelling that can cause headaches, confusion, and in severe cases, seizures
- ARIA-H (hemorrhage): Brain microbleeds or larger hemorrhages
These are serious safety concerns that require MRI monitoring before and during treatment, potential dose interruption or discontinuation, management of symptoms, and increased risk in APOE4 carriers.
With plasma exchange, these risks are essentially absent.
4. Mechanism difference
All the monoclonal antibodies target amyloid specifically (shown in the blue circle in the Ace Alzheimer Center webinar diagram) and require amyloid-positive status on PET scan or CSF testing. AMBAR worked in both amyloid-positive and amyloid-negative patients.
As illustrated in the ASFA 2024 presentation, Alzheimer's disease involves multiple pathological processes working together — amyloid accumulation, tau tangles, vascular dysfunction, neuro-inflammation, infection vulnerability, genetic factors, and aging/senescence (including oxidative stress and glycation). Although monoclonal antibodies target only the amyloid pathway, AMBAR may affect multiple pathways simultaneously (particularly aging/senescence and neuroinflammation, as indicated by the red circle in the webinar diagram).
This could explain why it works in amyloid-negative patients, why the biomarker patterns were "counterintuitive," why the effect sizes were larger across multiple outcome measures, and why global functional measures showed such strong effects.
If Alzheimer's is a multi-factorial disease (which mounting evidence suggests it is), then a multi-factorial treatment approach may be more effective than targeting a single pathway.
Clinical Significance of Effect Sizes
Even small differences in these scales are meaningful. For context:
- A 3-point difference on CDR-Sb represents about 6 months of disease progression
- AMBAR's effect on CDR-Sb was more than twice as large as the best-performing mAb (71% vs 37%)
- This translates to potentially years of preserved function
For a family caring for someone with Alzheimer's, the difference between 52% decline and full decline is enormous. It's the difference between someone who can still manage their medications with reminders vs. needing complete medication management, prepare simple meals vs. being unable to use the stove safely, handle money for small purchases vs. being unable to understand currency, and recognize family members consistently vs. frequent confusion about who people are.
These aren't abstract numbers on a test — 52% less decline means years of preserved independence, more time recognizing loved ones, and maintaining dignity and quality of life.
"52% less decline means years of preserved independence — more time recognizing loved ones, and maintaining dignity and quality of life."
Read the Complete Series: