Deep cervical lymphatic–venous anastomosis attenuates cognitive dysfunction and biomarker abnormalities in severe Alzheimer's disease: A prospective single‐arm study

Abstract

INTRODUCTION

This study evaluated the efficacy of deep cervical lymphatic–venous anastomosis (dcLVA) for severe Alzheimer's disease (AD).

METHODS

A total of 139 severe AD patients undergoing dcLVA were enrolled, and changes in cognitive function and biomarkers were evaluated by comparing preoperative measures with postoperative outcomes over a 6‐month follow‐up period.

RESULTS

Patients undergoing dcLVA showed modestly elevated Mini‐Mental State Examination (MMSE) scores from 48 hour postoperatively to 6‐month follow‐up. At 6‐month, observable changes across functional status and neuropsychiatric symptoms were documented, as demonstrated by reductions in scores on the Activities of Daily Living (ADL) scale, Neuropsychiatric Inventory (NPI), Neuropsychiatric Inventory‐Caregiver Distress (NPI‐D), and Sleep Disorders Inventory (SDI). Biomarker analyses revealed decreased cerebrospinal fluid (CSF) β‐amyloid (Aβ) 42, Aβ40, and p‐Tau levels postoperatively, along with increased plasma concentrations of these markers. No deaths or serious procedure‐related adverse events occurred during 6‐month follow‐up.

DISCUSSION

These findings suggest that dcLVA may confer multifaceted clinical benefits in severe AD patients, including attenuation of certain neuropsychiatric dysfunctions and a potential slowing of symptom progression.

Highlights

• Deep cervical lymphatic‐venous anastomosis (dcLVA) was evaluated in 139 patients with severe Alzheimer's disease (AD).

• The dcLVA was associated with modest postoperative improvements in global cognition, functional status and neuropsychiatric symptoms over 6 months.

• Cerebrospinal fluid (CSF) β‐amyloid (Aβ) and tau levels decreased, while corresponding plasma biomarkers increased after surgery.

• No deaths or serious procedure‐related adverse events occurred during the 6‐month follow‐up.

1. BACKGROUND

Alzheimer's disease (AD) is a progressive neurodegenerative disorder primarily characterized by gradual declines in cognitive and memory impairment. ¹ , ² Globally, approximately 50 million people are currently living with dementia—of which AD accounts for 60%–80%—and this prevalence is projected to surpass 150 million by 2050. ³ , ⁴ The neuropathological hallmarks of AD in brain include the accumulation of β‐amyloid (Aβ) plaque deposition, hyperphosphorylated tau protein forming neurofibrillary tangles, and widespread neuronal and synaptic loss. ¹ , ⁵ Despite extensive efforts to develop therapies targeting Aβ and Tau, current treatment options only provide limited symptomatic benefits. ⁶ These limitations underscore the urgent need to explore novel therapeutic targets and innovative intervention strategies for AD.

Recent discoveries have highlighted the importance of the central nervous system (CNS) lymphatic changes pathways in AD pathogenesis. In 2015, Louveau et al. identified functional meningeal lymphatic system in mammals, raising the possibility that impaired brain lymphatic drainage contributes to metabolic waste accumulation in the aging brain. ⁷ Mechanistically, Aβ and tau proteins in interstitial fluid (ISF) and cerebrospinal fluid (CSF) are cleared through meningeal lymphatic vessels at the skull base and then transported to deep cervical lymph nodes before ultimately draining into the venous circulation. ⁸ This deep cervical lymphatic network thus serves as a critical “bridge” linking brain clearance with systemic circulation. Accumulating evidence has further indicated that dysfunction of meningeal lymphatic drainage exacerbates cerebral Aβ and Tau deposition and disrupts synaptic homeostasis, implicating lymphatic impairment as a key driver to AD progression. ⁹ , ¹⁰ , ¹¹ Preclinical studies have demonstrated that enhancing cervical lymphatic outflow reduces pathological protein burden, improves cognition in AD‐model mice, ¹² , ¹³ , ¹⁴ whereas obstructing lymphatic drainage exacerbates AD‐like pathology in mice. ¹⁵ , ¹⁶ Most recently, near‐infrared imaging in nonhuman primates confirmed real‐time CSF drainage from the meninges into deep cervical lymph nodes, offering critical large‐animal in vivo support for lymphatic‐mediated clearance relevant to AD. ¹⁷ Together, these findings establish a strong rationale for targeting the brain–lymphatic–venous drainage axis as a novel therapeutic strategy for AD.

Deep cervical lymphatic‐venous anastomosis (dcLVA) is a novel surgical intervention designed to augment cerebral lymphatic drainage by creating direct shunts between deep cervical lymphatic vessels and adjacent veins. ¹⁸ By bypassing obstructed or functionally inefficient lymphatic pathways, dcLVA can rapidly reduce intracerebral lymphatic pressure and facilitate the efflux of accumulated metabolic proteins, including Aβ and Tau, from the brain parenchyma. ¹⁹ Emerging clinical observations have provided preliminary support for the translational potential of lymphatic‐targeted surgery in AD. ²⁰ , ²¹ In an initial cohort reported by Xie et al., behavioral and cognitive measures showed notable improvement over a 9‐month postoperative follow‐up in 50 treated patients. ²¹ Similarly, clinicians at the Shanghai Mental Health Center observed postoperative gains—including increased Mini‐Mental State Examination (MMSE) scores, reduced Clinical Dementia Rating Scale‐Sum of Boxes (CDR‐SB) scores, and better sleep and alertness—in a 70‐year‐old AD patient treated with cervical lymphatic shunting. ²⁰ A recent study involving 26 patients also found that approximately 60% of caregivers reported symptomatic improvement at 1 month after dcLVA. ¹⁸ By the end of 2024, the hospitals and media sources has reported an actual number likely exceeding 1000 dcLVA cases. ²² However, prior studies have been constrained by notable limitations, including small sample sizes and short‐term follow‐up periods. As a result, key questions regarding the long‐term efficacy and safety of dcLVA remain unanswered, underscoring the need for more rigorous, systematically designed clinical investigations.

RESEARCH IN CONTEXT

1. Systematic review: We have synthesized existing literature demonstrating that impaired clearance of neurotoxic proteins—particularly amyloid‐beta (Aβ) and Tau—plays a pivotal role in Alzheimer's disease (AD) progression. However, therapies directly targeting cerebral lymphatic clearance pathways remain largely unexplored in clinical practice, with a notable absence of efficacy trials. To address this critical gap, we present findings from a single‐arm, prospective pre–post design evaluating deep cervical lymphatic–venous anastomosis (dcLVA) in patients with severe AD.

2. Interpretation: Our results provide compelling evidence for the feasibility and potential superiority of dcLVA as a surgical intervention. dcLVA produced observable cognitive improvement, with elevated Mini‐Mental State Examination (MMSE) scores over 6‐month follow‐up. Visible functional, neuropsychiatric, and sleep quality gains were evidenced by reduced Activities of Daily Living (ADL) Scale, Neuropsychiatric Inventory (NPI), Neuropsychiatric Inventory‐Caregiver Distress (NPI‐D), and Sleep Disorders Inventory (SDI) scores at 6 months. Decreased cerebrospinal fluid (CSF) Aβ42, Aβ40, and p‐Tau at 48 h alongside a concurrent rise in their plasma levels—a trend that persisted at the 6‐month follow‐up. Notably, plasma p‐Tau217 was significantly reduced 6 months postoperatively. Collectively, these findings underscore the potential of dcLVA as a feasible intervention to slow cognitive deterioration and delay disease progression in severe AD patients.

3. Future directions: This single‐arm, prospective pre–post design lays crucial methodological groundwork for designing large‐scale efficacy, effectiveness, and pragmatic trials in Alzheimer's disease and related dementias (ADRD). Future work requires longer follow‐up with denser biomarker sampling, complemented by multimodal imaging (TSPO, Aβ/tau positron emission tomography [PET], and magnetic resonance imaging [MRI] volumetrics/arterial spin labeling [ASL]) and exosomal mitochondrial assays to prospectively test these mechanisms.

In the present study, a single‐arm, prospective pre–post design was employed to systematically assess the effects of dcLVA surgery on cognitive function and AD‐related biomarker changes in patients with severe AD. Specifically, we examined whether postoperative cognitive changes are accompanied by alterations in Aβ and tau levels. This study delineated how surgically enhanced meningolymphatic drainage may influence AD pathology, offering preliminary guidance for patient selection and strategies to sustain treatment effects. Overall, our findings support dcLVA as a potential therapeutic option for AD and advance the understanding of lymphatic modulation in neurodegenerative diseases.

2. METHODS

2.1. Study design

This is a prospective, single‐arm trial designed to evaluate the efficacy and safety of dcLVA in patients with severe AD. A priori sample size estimation was performed using G^*Power software (v. 3.1.9.2). Based on a repeated‐measures analysis of variance (ANOVA) model with the following parameters: significance level (α) of 0.05 (two‐tailed), statistical power (1–β) of 80%, effect size (Cohen's f) of 0.1, 5 measurement points, and an assumed correlation of 0.5 among repeated measures—the calculation indicated that a total of 121 patients would be required to detect a 20% improvement in MMSE. To ensure an adequate number of valid datasets, we ultimately enrolled 142 patients with severe AD (CDR  =  3) from October 2024 to March 2025 at the Third Affiliated Hospital of Zunyi Medical University (The First People's Hospital of Zunyi), which currently performs the largest number of dcLVA procedures in in a public hospital setting. Patient (or proxy) fully understood the study's purpose, procedures, and potential risks, and voluntarily agreed to participate with written informed consent. Findings are interpreted within this patient severity context. This study complied with the Declaration of Helsinki and local regulations. The study protocol was approved by the Ethics Committee of The First People's Hospital of Zunyi (Approval number: 2024‐1‐727), and the trial was prospectively registered at the Chinese Clinical Trial Registry (ChiCTR2400094603) prior to participant enrollment. Given decisional impairment in severe AD, we implemented a tiered consent model: surrogate/guardian authorization following standardized explanation, written materials, and a structured comprehension check; a cooling‐off period (≥ 24 hours) before final consent; and independent witnessing/documentation. Participants and caregivers were thoroughly counseled on risks, benefits, and alternatives, and participants could withdraw consent at any time without penalty.

2.2. Inclusion and exclusion criteria

Eligibility was determined based on strict inclusion and exclusion criteria. Inclusion criteria: (1) diagnosis of severe AD CDR score of 2 or 3); (2) age > 50 years; (3) disease duration > 6 months; (4) presence of behavioral and psychological symptoms of dementia; (5) no history of other psychiatric illness or epilepsy, or severe dysfunction of other organ systems.

Exclusion criteria: (1) patients with non‐AD cognitive impairment (e.g., mild cognitive impairment or other types of dementia); (2) cognitive dysfunction primarily due to major depression, Parkinson's disease, stroke, or other disorders; (3) comorbid malignancy, significant cardiovascular or cerebrovascular disease, bleeding disorders, or severe hepatic/renal insufficiency; (4) receipt of any investigational or AD‐related treatment in the month prior to enrollment; (5) use of cognition‐ or neuropsychiatric‐altering medications (e.g., donepezil, GV‐971 (sodium oligomannate), or olanzapine) in the 3 months before enrollment; or (6) inability to tolerate cervical lymphatic‐venous anastomosis surgery.

2.3. Quality assurance and oversight

Surgeons were credentialed with documented microsurgical proficiency. Preoperative targets were mapped by high‐resolution ultrasound. Anastomosis patency was verified intraoperatively using indocyanine green (ICG) fluorescence. Peri‐operative management followed standardized pathways (delirium/sleep prevention, multimodal analgesia, and individualized anticoagulant/antiplatelet discontinuation with enhanced 24‐ to 48‐hour monitoring). An independent Data and Safety Monitoring Board (DSMB) reviewed safety data at predefined intervals and retained stopping authority.

2.4. Surgical procedure

All patients underwent bilateral dcLVA under general endotracheal anesthesia. Intraoperative neurophysiological monitoring of the cervical plexus, accessory nerve, and phrenic nerve was employed throughout the procedure. Patients were positioned supine with the head rotated approximately 60° to the contralateral side. After standard surgical skin preparation and draping, a ∼5 cm longitudinal incision was made along the posterior border of the left sternocleidomastoid muscle. Local infiltration with ropivacaine provided subcutaneous anesthesia. Layered dissection was performed to expose the external jugular vein and its tributaries, and deep cervical lymphatic vessels were identified and isolated. Several lymphatic vessels were carefully freed, and some regional lymph nodes were partially resected. ICG tracer was injected to visualize lymphatic drainage. Venotomy incisions were made in the external jugular vein and selected tributaries, and end‐to‐side anastomoses were then performed between deep cervical lymphatic vessel branches and the venous openings under an operating microscope. After completing the anastomoses, the ICG and GLOW800 augmented reality (AR) fluorescence imaging confirmed the flow of the lymphatics into the venous circulation. The incision was then closed in layers. Representative intraoperative images of the dcLVA procedure were shown in Figure 1.

FIGURE 1.

Intraoperative images of the dcLVA procedure. (A) Surgical incision surface positioning. (B) Schematic diagram of neck anatomical structures as seen on ultrasound. (C) Intraoperative photograph of the left neck incision (∼5 cm along sternocleidomastoid posterior border). (D) Operative field showing isolated EJV. (E) Operative field showing transverse nerve of neck, EJV, and SCM. (F) Dissection of the carotid sheath. (G) Isolation of a lymphatic tissue (cluster of lymphatic vessels and node tissue). (H) Intraoperative GLOW800 AR fluorescence image. (I) Intraoperative ICG fluorescence imaging showing lymphatic vessels filling with dye within the isolated lymphatic valve. (J) Microsurgical end‐to‐side anastomosis of a deep cervical lymphatic vessel to the EJV. (K) Intraoperative GLOW800 AR fluorescence image after anastomosis. (L) ICG fluorescence angiography after anastomoses, confirming patency of the lymphatic–venous shunts (ICG flowing freely from lymphatics into venous circulation). AR, augmented reality; dcLVA, deep cervical lymphatic‐venous anastomosis; EJV, external jugular vein; ICG, indocyanine green; SCM, sternocleidomastoid.

2.5. Postoperative care

Postoperative care followed an enhanced recovery protocol. Early ambulation was encouraged as soon as feasible to reduce the risk of deep vein thrombosis, pressure ulcers, and other complications. All pre‐existing medications and/or cognitive rehabilitation therapies were continued postoperatively. Patients were closely monitored for any surgical or neurological complications, and standard postoperative management was provided. Anticoagulant drugs (rivaroxaban 5 mg once a day) were initiated 1 week after the operation and used continuously for 3 months. After the incision recovers, lymphatic detoxification massage was performed to promote lymphatic drainage in the neck.

2.6. Perioperative management and safety monitoring

We applied standardized measures to reduce delirium and sleep disturbance: preoperative risk stratification; minimization of deliriogenic/anticholinergic sedatives; orientation cues and light–dark hygiene; early mobilization; and multimodal analgesia. Anticoagulant/antiplatelet therapy was managed with individualized discontinuation plans and postoperative 24‐ to 48‐hour wound/drain surveillance. Safety events were prospectively logged with predefined criteria and reviewed by the clinical team.

2.7. Data collection

A dedicated data manager oversaw collection and management of all study data. Neuropsychological assessments were conducted by trained examiners blinded to surgical status at baseline and at 48 hours and 1, 3, and 6 months postoperatively, using standardized instruments including global cognition—MMSE and Montreal Cognitive Assessment (MoCA); dementia staging— CDR; functional status—Activities of Daily Living (ADL) scale; neuropsychiatric symptoms—Neuropsychiatric Inventory (NPI, 12 items), including the Neuropsychiatric Inventory–Caregiver Distress (NPI‐D); Sleep Disorders Inventory (SDI). Baseline demographic characteristics, clinical features, and perioperative information were recorded and summarized in Table 1. CSF biomarkers were detected at baseline and 48 hours, and plasma AD biomarkers (Aβ40, Aβ42, Aβ42/40, and p‐Tau217) were measured at baseline, 48 hours, and 6 months. Follow‐up adherence and reasons for attrition were documented (Figure 2), including loss of contact, caregiver withdrawal, financial burden, and transportation difficulty; no participant discontinued due to surgical complications or clinical worsening. Neuropsychological assessment, AD biomarkers, and serum cytokines measurement were performed preoperatively and postoperatively using paired pre‐ and postoperative data with prespecified control for multiple testing. Proteomic analysis of CSF and plasma was conducted to investigate the potential molecular changes associated with dcLVA.

TABLE 1.

Baseline demographic and preoperative characteristics of the patients (n = 139).

Variable
Sex, n (%)
Male	57 (41.0)
Female	82 (59.0)
Age, mean (SD), years	70.2 (8.7)
< 60, n (%)	18 (12.9)
61‐69, n (%)	45 (32.9)
70‐79, n (%)	56 (40.3)
≥ 80, n (%)	20 (14.4)
Height, mean (SD), cm	160.0 (8.6)
Weight, mean (SD), kg	57.1 (10.3)
BMI, mean (SD), kg/m2	22.2 (2.9)
18.5–23.9, n (%)	86 (61.9)
< 18.5, n (%)	16 (11.5)
≥ 24.0, n (%)	37 (26.6)
Education level, n (%)
Illiterate	19 (13.7)
Primary	29 (20.9)
Secondary or above	81 (58.3)
Unknown	10 (7.2)
Comorbidities and past medical history, n (%)
Hypertension	31 (22.3)
Diabetes	15 (10.8)
Hyperlipidemia	1 (0.7)
Coronary artery disease	2 (1.4)
Allergy	1 (0.7)
History of surgery	24 (17.3)
History of tumor	6 (4.3)
Disease duration, mean (SD), years	4.1 (2.4)
MRI MTA score, mean (IQR)	3 (1.0)
0	12 (8.6)
1	8 (5.8)
2	35 (25.2)
3	48 (34.5)
4	15 (10.8)
Baseline MMSE score, mean (IQR)	2.0 (7.0)
Baseline MoCA score, mean (IQR)	0.0 (1.0)
Baseline CDR score, 3, n (%)	139 (100.0)
Baseline ADL score, mean (SD)	61.1 (14.8)
Baseline NPI score, mean (SD)	35.8 (21.2)
Baseline NPI‐D, mean (SD)	15.4 (8.7)

Abbreviations: ADL, activities of daily living; BMI, body mass index; CDR, clinical dementia rating; IQR, range interquartile; MMSE, Mini‐Mental Status Examination; MoCA, Montreal Cognitive Assessment; MRI, magnetic resonance imaging; MTA, medial temporal lobe atrophy; NPI, neuropsychiatric inventory; NPI‐D, Neuropsychiatric Inventory—Caregiver Distress; SD, standard deviation.

FIGURE 2.

Study flowchart.

2.8. Study Outcomes

2.8.1. Primary and secondary outcomes

The primary outcomes included cognitive function (MMSE scores [baseline, 48 hours and 1, 3, and 6 months]; ADL, NPI, NPI‐D [baseline and 6 months]) and paired AD biomarkers (CSF concentrations of Aβ40, Aβ42, total Tau [t‐Tau], and p‐Tau181 [baseline, 48 hours]; plasma changes of Aβ40, Aβ42, and p‐Tau217 [baseline, 48 hours, 6 months]). Secondary outcomes included the factors influencing dcLVA, SDI, proteomic analysis of CSF and plasma.

2.8.2. Exploratory analyses

We longitudinally assessed serum cytokines (interleukin [IL]‐6 and glial fibrillary acidic protein [GFAP]) in AD patients at baseline, 48 hours, and 6 months using enzyme‐linked immunosorbent assays (ELISA). These analyses were prespecified for exploratory purposes.

2.8.3. Safety outcomes

Safety was evaluated in terms of the frequency of adverse events, alterations in vital signs, and postoperative recovery.

2.8.4. Neuropsychological test

Trained neuropsychologists (blinded to group allocation) administered the MMSE preoperatively, then at 48 hours and 1, 3, and 6 months postoperatively (Figure S1). ADL, NPI, and NPI‐D scores were involved at baseline and 6 months. Besides, SDI were evaluated 1, 3, and 6 months postoperatively. Neuropsychological evaluations were administered by the same neurologist, neurosurgeon, or trained examiner in a standardized environment to ensure consistency. For the purposes of analysis, an effective neuropsychological improvement was defined as any increase in MMSE, but decease in ADL, NPI, NPI‐D, and SDI scores relative to preoperative score.

2.9. CSF and plasma collection

CSF was obtained at baseline and 48 hours postprocedure under standardized conditions to minimize preanalytical variability. Long‐term CSF collection was not performed for ethical reasons. Lumbar puncture was performed between 08:00 and 10:00 a.m.with participants in the lateral decubitus position using 22‐gauge Sprotte atraumatic needles. The first 2 mL of CSF was discarded, and the subsequent 3 mL was collected into prechilled polypropylene tubes. Samples were centrifuged at 2,000 × g for 10 minutes at 4°C within 15 minutes of collection. The supernatant was aliquoted in 500 µL portions and flash‐frozen within 30 minutes. All aliquots were stored at −80°C with no freeze–thaw cycles before batch analysis.

Venous blood was drawn at baseline, 48 hours, and 6 months after treatment. Samples were collected into ethylenediaminetetraacetic acid (EDTA) tubes and centrifuged at 1600 × g for 15 minutes at 4°C within 30 minutes of phlebotomy. Plasma was separated, aliquoted, and stored at −80°C with no freeze–thaw cycles prior to batch analysis. Informed consent was obtained from all patients. Laboratory staff performing biomarker assays were blinded to clinical data. All collections followed standardized protocols to minimize preanalytical variability. Paired observations were employed for analysis with prespecified multiple‐testing control.

2.10. ELISA measurement on AD biomarkers

ELISA were employed to detect changes of AD biomarkers in CSF collected at baseline (preoperatively) and 48 hours after the procedure, and plasma harvested at baseline, 48 hours, and 6 months postoperatively. To minimize interassay variability, all samples from the same patient (both pre‐ and postoperative) were analyzed within the same batch. Concentrations of Aβ40, Aβ42, t‐Tau, and p‐Tau181 in CSF, plasma levels of Aβ40, Aβ42, and p‐Tau217, serum changes of IL‐6 and GFAP were measured by ELISA Kits. Detailed information regarding these kits (including manufacturers, catalog numbers, and sensitivity) is provided in Table S1. All assays were performed according to the manufacturer's instructions, including sample dilution ratios, incubation times, and detection wavelength settings. Calibration curves were generated for each batch using standards supplied with the kits, and only curves with R ² ≥ 0.99 were accepted. All samples were run in duplicate, and plates were read using a microplate reader after colorimetric development. Intra‐ and inter‐assay coefficients of variation (CV%) met the predefined quality‐control criteria. Paired observations were analyzed with prespecified multiple‐testing control.

2.11. Proteomic analysis of CSF and plasma

To capture short‐latency molecular changes after dcLVA, we profiled CSF and plasma at 48 hours postoperation to align with early biomarker sampling. To relate proteomics to AD biology, we analyzed associations between proteomic features/modules and immunoassay‐based AD biomarkers. Proteomic sample preparation and mass spectrometry analysis were conducted by OE Biotech (Shanghai, China) using a data‐independent acquisition (DIA) approach.

Protein extraction and digestion: Total protein concentration of each CSF and plasma sample was determined by the bicinchoninic acid (BCA) assay. Equal amounts of protein from each sample were adjusted to a uniform volume and concentration with lysis buffer. Proteins were reduced with dithiothreitol (5 mM final concentration, incubated at 55°C for 30 minutes) and alkylated with iodoacetamide (10 mM, 15 minutes at room temperature in the dark). Proteins were then precipitated by adding six volumes of cold acetone and incubating at −20°C for > 4 hours. Pellets were collected by centrifugation (8,000×g, 10 minutes, 4°C) and air‐dried briefly. Pellets were re‐dissolved in 100 µL of 50 mM ammonium bicarbonate, and proteins were digested with sequencing‐grade trypsin (Trypsin‐TPCK, 1:50 enzyme‐to‐substrate ratio) at 37°C overnight. The resulting peptides were lyophilized and stored at −80°C. Peptide mixtures were desalted using SOLA SPE 96‐well plates (Thermo Fisher). Each well was activated with 200 µL methanol (three times) and equilibrated with 200 µL 0.1% formic acid in water (three times). Peptide samples (50–500 µL of digest per well, diluted in 0.1% formic acid) were loaded under vacuum at ∼1 mL/min flow. Wells were washed with 0.1% formic acid (200 µL, three times). Peptides were eluted with 150 µL of 50% acetonitrile/0.1% formic acid (three times, total ∼450 µL) and then dried under vacuum.

DIA mass spectrometry: Prior to liquid chromatography–mass spectrometry (LC‐MS) analysis, a retention time calibration peptide (iRT standard) was spiked into each sample (volume ratio iRT: sample = 1:20). Equal aliquots of peptide digests were separated by nanoliquid chromatography (EASY‐nLC 1200 system) using a 30‐min gradient (mobile phase A: 0.1% formic acid in water; mobile phase B: 0.1% formic acid in acetonitrile). The gradient was: 5% B to 22% B over 0–20 minutes, 22% to 37% B from 20–24 minutes, 37% to 80% B from 24–27 minutes, then 80% B to 30 minutes. Eluted peptides were injected into a timsTOF Pro mass spectrometer (Bruker) for DIA analysis. MS settings included a capillary voltage of 1.4 kV, dry gas temperature 180°C, dry gas flow 3.0 L/min. The MS1 scan range was m/z 100–1700, with an ion mobility range 0.7–1.3 Vs/cm² and collision energy range 20–59 eV for fragmentation.

Protein identification and quantification: DIA raw data were processed with Spectronaut Pulsar (v.18.4, Biognosys). Protein and peptide identification used a false discovery rate (FDR) cutoff of 1% at both precursor and protein levels. Carbamidomethylation of cysteine was set as a fixed modification; oxidation of methionine and N‐terminal acetylation were variable modifications. Up to two missed tryptic cleavages were allowed. Peptide signals were quantitatively analyzed in Spectronaut. Proteins with a p‐value < 0.05 and fold‐change (FC) > 1.2 (or < 0.833) between pre‐ and post‐surgery were considered differentially expressed. Significantly up‐regulated proteins were defined by p < 0.05 and FC > 1.2, while significantly down‐regulated proteins had p < 0.05 and FC < 0.833 (Figure S2).

Bioinformatics analysis: We performed gene ontology (GO) enrichment analysis on the differentially expressed proteins (DEPs) to categorize the affected biological processes (BP), cellular components (CC), and molecular functions (MF). Pathway analysis was conducted using the Kyoto Encyclopedia of Genes and Genomes (KEGG) database to identify which biological pathways were significantly associated with the differential proteins. Additionally, gene set enrichment analysis (GSEA) was applied (with reference to KEGG pathways) to compare pre‐ vs post‐operative proteomic profiles.

2.12. Use of external control from a multicenter lecanemab cohort

To contextualize the biomarker changes observed in our single‐arm surgical cohort, we incorporated external comparative data from a recently published multicenter, prospective cohort study evaluating the real‐world efficacy, safety, and plasma biomarker trajectories of lecanemab in Chinese patients with AD. ²³ From this study, we extracted the subgroup of moderate‐to‐severe AD patients (n = 16) who fulfilled the following criteria: clinical diagnosis of AD, advanced disease severity, and availability of plasma biomarker data at baseline, 2.5, and 7 months after treatment initiation. Plasma concentrations of Aβ40, Aβ42, and p‐Tau217 at these time points were collected from the published summary statistics. In the comparative framework, the lecanemab cohort served as the Drug group, whereas our dcLVA cohort served as the Surgery group. Because the external dataset provided only median values and interquartile ranges, individual‐level statistics were not available for model‐based adjustment. Therefore, we conducted descriptive graphical comparison to evaluate whether the directionality and magnitude of biomarker trajectories in the Surgery group diverged from or paralleled those observed in the Drug group over a comparable follow‐up period. This approach is methodologically aligned with prior external‐control analyses in AD therapeutic studies, where summary‐level data were used to benchmark treatment‐related changes in the absence of internal controls. ²⁴ Consistent with these principles, our findings should be interpreted as trend‐level, hypothesis‐generating evidence, rather than as definitive between‐group statistical comparisons.

2.13. Statistical analysis

All statistical analyses were performed using SPSS Statistics version 29.0 (IBM Corp., Armonk, NY) and R software (v.3.5.0). Data are presented as mean ± standard deviation (SD) for approximately normally distributed variables, or as median (interquartile range, IQR) for non‐normally distributed variables. Continuous variables were tested for normality (Shapiro–Wilk test). Categorical variables are summarized as count (percentage). Reporting of cognitive outcomes. We report absolute mean changes (Δ) vs baseline with two‐sided 95% confidence intervals (CIs) and exact p values at each timepoint; To control for multiple comparisons, we applied the Benjamini–Hochberg FDR (BH‐FDR) method within predefined families of hypotheses (with adjusted p < 0.05 was considered statistically significant difference), covering analyses of CSF and plasma biomarkers, biomarker–MMSE correlations, proteomic differential expression, and functional enrichment (GO/KEGG/GSEA). Proteomic identification was controlled at 1% FDR. Longitudinal cognitive outcomes were analyzed using linear mixed‐effects models, with posthoc comparisons against baseline adjusted by BH‐FDR. We do not report “percent of patients improved” to avoid misinterpretation. For transparency, timepoint‐specific n (paired observations) are shown on the plots and in figure legends. To identify factors associated with neuropsychological improvement after surgery, we performed univariate logistic regression analysis. And a correlation analysis was made to investigate the association between changes in the concentration of multiple AD‐related biomarkers in plasma and CSF and changes in MMSE scores at 48 hours postoperatively. In addition to analyzing changes in multiple biomarkers simultaneously, we employed a longitudinal analysis strategy in R to assess overall changes pre‐ vs post‐surgery for the five key biomarkers (Aβ40, Aβ42, t‐Tau, p‐Tau181 in CSF, and Aβ40, Aβ42, p‐Tau217 in plasma). Missing values were addressed using Expectation‐Maximization (EM) imputation and sequential mean substitution. For biomarkers meeting the assumptions of normality (assessed via Shapiro–Wilk tests) and sphericity (Mauchly's test), repeated‐measures ANOVA (RM ANOVA) was used.

3. RESULTS

3.1. Patients’ baseline and perioperative characteristics

A total of 139 patients with severe AD were enrolled in this study (CDR = 3; 59% female; mean age 70.2 ± 8.7 years; 87.1% ≥ 60 years). At baseline, patients exhibited profound cognitive impairment (median MMSE 2.0 (7.0); median MoCA 0.0 (1.0)), alongside significant neuropsychiatric symptoms (mean NPI 35.8 ± 21.2) (Figure 2, Table 1). Functional dependence was markedly impaired (ADL: 61.1 ± 14.8), accompanied by considerable depressive symptoms (NPI‐D: 15.4 ± 8.7; Table 1). Baseline laboratory evaluations, including liver and renal function tests, as well as coagulation profiles, were within normal ranges for all patients (Table S2). All patients successfully underwent bilateral dcLVA under general anesthesia, with 71.2% (n = 99) of cases performed by a single surgeon. Mean operative duration was 212.1 ± 47.5 minutes and anesthesia duration was 286.3 ± 60.3 minutes. Anesthesia techniques were equally distributed (total intravenous anesthesia (TIVA): 50.4%; combined: 49.6%). Intraoperative fluid balance averaged 807.1 ± 379.5 mL (crystalloid: 887.7 ± 304.5 mL; colloid: 428.0 ± 223.3 mL), with minimal blood loss (22.6 ± 12.0 mL). Patients regained consciousness promptly (12.0 ± 4.5 minutes) and had brief Post‐Anesthesia Care Unit (PACU) stays (43.0 ± 8.0 minutes). No intraoperative hemodynamic instability occurred. Only one case (0.72%) required reoperation for postoperative bleeding/hematoma, but no mortality or incision infection occurred. Postoperative nausea/vomiting affected 7 patients (5.0%), 44 patients (31.7%) developed transient delirium, which resolved with supportive care. Most patients (59.7%) experienced self‐limited sleep disturbances within 48 hours. Mean postoperative stay was 2.9 ± 0.8 days, with a total of hospitalization was 10.0 ± 3.6 days (Table 2).

TABLE 2.

Perioperative surgical and anesthetic information (n = 139).

Variable
Intraoperative variables
Lead surgeon (X.H. Fu), n (%)	99 (71.2)
Anesthesia, n (%)
Total intravenous anesthesia (TIVA)	70 (50.4)
Intravenous‐inhalation combined anesthesia	69 (49.6)
Anesthesia duration, mean (SD), min	286.3 (60.3)
Surgery duration, mean (SD), min	212.1 (47.5)
Awakening time, mean (SD), min	12.0 (4.5)
PACU stay time, mean (SD), min	43.0 (8.0)
Intraoperative crystalloid, mean (SD), mL	887.7 (304.5)
Intraoperative colloid, mean (SD), mL	428.0 (223.3)
Blood loss, mean (SD), mL	22.6 (12.0)
Urine output, mean (SD), mL	486.0 (214.6)
Intraoperative fluid balance, mean (SD), mL	807.1 (379.5)
Intraop sufentanil dose, mean (SD), µg	20.9 (2.9)
Intraop remifentanil dose, mean (SD), µg	1.6 (0.6)
Hemodynamic fluctuations count, Md (IQR)	0.0 (2.0)
Postoperative variables, n (%)
Postop bleeding requiring intervention	1 (0.007)
Re‐operation for complication	1 (0.007)
Surgical site infection	0 (0)
Postoperative nausea and vomiting	7 (5.0)
Postoperative delirium	44 (31.7)
Sleep disorder 48 h after the operation	83 (59.7)
Postoperative hospital stay, mean (SD), days	2.9 (0.8)
Total hospital stay, mean (SD), days	10.0 (3.6)

Abbreviations: IQR, range interquartile; Md, median; PACU, postanesthesia care unit; SD, standard deviation.

3.2. Primary outcome: efficacy and safety

3.2.1. Cognitive function and AD biomarkers

A detailed summary of participant retention and follow‐up completion is presented. Among the 139 participants who underwent the operation, 126 (90.6%) completed the 1‐month follow‐up, 108 (77.7%) completed the 3‐month follow‐up, and 95 (68.3%) completed the 6‐month assessment (Figure 2, Figure 3A). A total of 44 participants (31.7%) were lost to follow‐up. As summarized, the main reasons included loss of contact, caregiver withdrawal due to prolonged follow‐up, and logistical or financial difficulties. Notably, no discontinuations were attributed to surgery‐related adverse events or clinical deterioration. In addition, the longitudinal changes in major outcome measures were illustrated at each timepoints, including MMSE scores, CSF/plasma concentrations of Aβ and p‐Tau, and serum IL‐6/GFAP (Table S3).

FIGURE 3.

Neuropsychological test and AD biomarkers measurements pre‐ and post‐dcLVA. (A) Research process diagram (supported by BioRender). (B) MMSE scores at 48 hours (n = 139), 1 month ( n = 126), 3 months (n = 108), 6 months post‐dcLVA (n = 95). (C) NPI scores at preop and 6 months post‐dcLVA (n = 95). (D) ADL scores at preop and 6 months post‐dcLVA (n = 95). (E) NPI‐D scores at preop and 6 months post‐dcLVA (n = 95). (F) SDI scores at preop and 1, 3, and 6 months post‐dcLVA. (G) The concentration of Aβ42 in CSF 48 hours post‐dcLVA (n = 29). (H) The concentration of Aβ40 in CSF 48 hours post‐dcLVA (n = 29). (I) The ratio of Aβ42/Aβ40 in CSF 48 hours post‐dcLVA (n = 29). (J) The concentration of t‐Tau in CSF 48 hours post‐dcLVA (n = 29). (K) The concentration of p‐Tau181 in CSF at 48 hours post‐dcLVA (n = 29). (L) The concentration of Aβ40 in plasma (n = 74). (M) The concentration of Aβ42 in plasma at 48 hours (n = 72), 6 months (n = 22) post‐dcLVA. (N) The ratio of Aβ42/ Aβ40 in plasma at 48 hours (n = 72) and 6 months (n = 22) postd‐cLVA. (O) The concentration of p‐Tau217 in plasma at 48 hours (n = 72) and 6 months (n = 22) post‐dcLVA. Statistics: The Friedman test was applied to the remaining variables due to non‐normal distribution. Regarding missing data, the EM algorithm was used for imputation of the Aβ42/40 ratio, and series mean imputation was applied to other variables; BH‐FDR applied within CSF and plasma biomarker families (adjusted p‐value were presented in the Figure, with p < 0.05 showed statistically significant). Aβ, β‐amyloid; AD, Alzheimer's disease; BH‐FDR, Benjamini–Hochberg false discovery rate; CSF, cerebrospinal fluid; EM, Expectation‐Maximization; MMSE, Mini‐Mental State Examination; NPI, Neuropsychiatric Inventory; NPI‐D, Neuropsychiatric Inventory–Caregiver Distress; SDI, Sleep Disorders Inventory.

MMSE scores increased significantly in the majority of patients post‐dcLVA (84.2%, n = 117), at 48 hours (69, 49.6%), 1 month (63, 49.6%), 3 months (53, 48.6%), and 6 months (42, 44.2%), with elevation in 41.3% (n = 45) of patients at 3 months and 29.5% (n = 28) at 6 months (Table S4, Table S5). The mean MMSE score change were 1.6 at 48 hours (95% CI: 1.079 to 2.158; p < 0.001), 1.25 at 1 month (95% CI: 0.603 to 1.888; p < 0.001), 1.17 at 3 months (95% CI: 0.343 to 2.01; p = 0.004), 1.28 at 6 months (95% CI: 0.311 to 2.344; p = 0.016) after surgery (Figure 3B, Figure S1, Table S3). In parallel, we observed significant reductions in ADL (−59.5; 95% CI: −62.541 to −56.436; p < 0.001), NPI (−9.9; 95% CI: −15.721 to −4.205; p < 0.001), NPI‐DI (−3.4; 95% CI: −6.313 to −0.471; p < 0.001]) scores at 6 months postoperatively (Figure 3C‐E, Table S3). Moreover, SDI scores get improved at 1 month (11.6; 95% CI: 9.0 to 14.2; p < 0.001), 3 months (12.7; 9.5 to 15.9; p < 0.001) and 6 months (11.8; 8.2 to 15.5; p < 0.001) postoperatively, consistent with reductions in NPI and NPI‐D (Figure 3F). Postsurgery follow‐up videos further documented observable changes across multiple functional domains (Supplementary videos). Patients showed increased spontaneous speech, faster verbal responses, improved efficiency in daily tasks (e.g., dressing, light housework), enhanced social engagement with clearer recognition of family members, longer spoken phrases, longer sleep duration, and more stable mood with greater willingness to express needs.

In terms of biomarker changes in CSF, significant decreases were observed in Aβ42 (−176.185; 95% CI: −225.999 to −126.371; p < 0.001), Aβ40 (−7706.746; 95% CI: −11287.154 to −4126.336; p < 0.001), and p‐Tau181 (−37.046; 95% CI: −61.557 to −12.535; p = 0.004) at 48 hours after operation (Figure 3G‐K, Table S3). Conversely, plasma analysis demonstrated an inverse pattern, with significant increases in Aβ40 (50.261; 95% CI: 9.433 to 91.090; p = 0.013) and p‐Tau217 (3.734; 95% CI: 2.901 to 4.566; p < 0.001) at 48 hours after surgery. Plasma Aβ40, Aβ42, Aβ42/Aβ40 ratio at 6 months post‐dcLVA were significantly elevated compared to preoperative levels (52.160; 95% CI: 5.760 to 98.560; p < 0.001), (12.750; 95% CI: 9.590 to 15.910; p < 0.001) (Figure 3L‐N); while p‐Tau217 levels at 6 months significantly decreased below baseline (37.957; 95% CI: 35.928 to 39.985; p < 0.001) (Figure 3O). Besides, IL‐6 and GFAP levels were transiently elevated at 48 hours postoperatively, but returned to baseline by 6 months (Figure S3). These shifts suggested dcLVA facilitated brain‐to‐blood transfer of pathogenic proteins and reduced inflammatory response.

Regarding safety, postoperative delirium occurred in 44 patients (31.7%) and sleep disturbance in 83 patients (59.7%) at 48 hours (Table 2). Delirium episodes were transient, typically resolving spontaneously within 3–7 days without lasting effects, and were managed effectively. ²¹ , ²² One patient (0.72%) underwent reoperation for postoperative bleeding/hematoma attributable to the improper withdrawal of anticoagulants before surgery. Events were addressed per protocol, and timepoint‐specific n and management details were provided in Table 2. Importantly, no deaths or serious procedure‐related adverse events occurred during 6‐month follow‐up. Our results indicated that dcLVA could be performed safely in AD patients with an acceptable perioperative risk profile, and that postoperative assessments revealed short‐term changes in neuropsychological and functional measures.

3.3. Secondary outcomes

3.3.1. Predictors of cognitive improvement after dcLVA

We conducted univariate analysis to identify factors associated with better cognitive outcomes after dcLVA. Several factors emerged as significant predictors of cognitive improvement on univariate screening, including female sex (odds ratio [OR] = 2.148), body weight (OR = 0.960) and body mass index (BMI) (OR = 0.867), baseline MMSE score (OR = 0.897), baseline ADL score (OR = 1.026), CT abnormality (OR = 2.917), pulmonary inflammatory changes on CT (OR = 7.075), surgery duration (OR = 1.008), and opioid doses (OR = 0.882) (Table 3). Further multivariable logistic regression including key candidate variables identified two independent predictors of improvement: higher preoperative BMI (OR = 0.786; 95% CI: 0.642 to 0.962; p = 0.020) and higher baseline MMSE score (OR = 0.809; 95% CI: 0.671 to 0.974; p = 0.025). Specifically, patients with higher BMI or less severe baseline cognitive impairment had greater odds of improvement (Table 4). We performed a correlation analysis to assess the relationship between changes in MMSE scores and changes in biomarkers. However, at 48 hours and 6 months postoperatively, no statistically significant correlations (q > 0.05) were found between the changes in the selected plasma and CSF biomarkers and the concurrent changes in MMSE scores (Tables S6 and S7). We further conducted stratified analyses by sex to explore the potential subgroup effects. Notably, within the female cohort, a significant association emerged between the reduction in plasma p‐tau217 at 6 months and changes in both MMSE and ADL scores (Tables S8, S9 ; all q < 0.05). This sex‐specific correlation was not observed in male participants. These findings suggest that postoperative biomarker–clinical relationships may vary across subgroups, warranting further validation in larger, adequately powered cohorts.

TABLE 3.

Univariate analysis of factors associated with cognitive improvement at 48 hours (n = 139).

Variable	p‐Value	OR (95% CI)
Sex: female	0.029*	2.148 (1.080–4.275)
Age, years	0.986	1.000 (0.962–1.039)
≥ 60 (vs. < 60)	0.475	0.691 (0.251–1.903)
≥ 65 (vs. < 65)	0.928	0.967 (0.460–2.030)
≥ 70 (vs. < 70)	0.618	0.843 (0.431–1.648)
≥ 75 (vs < 75)	0.745	1.123 (0.558–2.258
≥ 80 (vs < 80)	0.864	1.086 (0.419–2.814)
Height, cm	0.135	0.970 (0.932–1.009)
Weight, kg	0.017*	0.960 (0.928–0.993)
BMI	0.023*	0.867 (0.766–0.980)
BMI subgroup:vs normal	0.019*
< 18.5	0.185	2.265 (0.676–7.592)
≥ 24.0	0.028*	0.409 (0.184–0.909)
Education degree:vs illiterate	0.976
Primary	0.951	0.964 (0.303–3.071)
Secondary or above	0.971	1.018 (0.374–2.770)
Unknown	0.705	1.350 (0.286–6.379)
Disease duration, years	0.800	0.982 (0.850–1.134)
Hypertension	0.839	1.086 (0.487–2.423)
Diabetes	0.590	0.744 (0.254–2.179)
Hyperlipidemia	1.000	/
Coronary artery disease	0.999	/
Allergy	1.000	/
History of surgery	0.583	1.283 (0.527–3.126)
History of tumor	0.506	1.800 (0.319–10.165)
Baseline MMSE score	0.006*	0.897 (0.831–0.969)
Baseline MoCA score	0.137	0.898 (0.780–1.035)
Baseline ADL score	0.049*	1.026 (1.000–1.053)
Baseline NPI score	0.745	1.003 (0.986–1.020)
Baseline NPI‐D score	0.334	1.021 (0.979–1.065)
Baseline MRI MTA score: vs 0	0.575
1	0.206	3.333 (0.515–21.584)
2	0.217	2.375 (0.602–9.367)
3	0.101	3.053 (0.805–11.569)
4	0.303	2.286 (0.475–11.003)
Pulse rate: vs normal	0.839
< 60 bpm	0.553	0.663 (0.170–2.583)
> 100 bpm	0.999	/
WBC: vs normal	0.776
< 4*109/L	0.868	0.845 (0.116–6.181)
> 10*109/L	0.486	0.423 (0.037–4.775)
HCT < 35: vs ≥ 35%	0.557	0.748 (0.283–1.973)
Chest CT showed abnormalities: vs normal	0.005*	2.917 (1.374–6.190)
Combined with pulmonary bullae	0.464	2.354 (0.238–23.306)
Combined with emphysema	0.063	3.500 (0.933–13.128)
Combined with pulmonary inflammation	0.012*	7.075 (1.540–32.518)
Combined with partial pulmonary fibrosis	0.594	1.300 (0.495–3.415)
Combined with aortic sclerosis and/or coronary atherosclerosis	0.532	1.581 (0.376–6.642)
Combined with pulmonary nodules or calcification foci	0.852	0.923 (0.398–2.141)
Combined with atelectasis	0.999	/
ECG showed abnormalities: vs normal	0.231	1.525 (0.764–3.045)
T‐wave or ST‐T changes	0.199	1.843 (0.725–4.683)
Combined conduction block	0.909	1.075 (0.312–3.702)
Combined left ventricular high voltage	0.999	/
Combined with abnormal sinus rate	0.418	0.609 (0.184–2.023)
Anesthesia duration, min	0.245	1.003 (0.998–1.009)
Surgery duration, min	0.040*	1.008 (1.000–1.016)
Awakening time, min	0.241	0.954 (0.881–1.032)
PACU stay time, min	0.794	0.992 (0.937–1.051)
Blood loss, mL	0.705	1.004 (0.983–1.026)
Intraoperative crystalloid, mL	0.542	1.000 (0.999–1.001)
Intraoperative colloid, mL	0.835	1.000 (0.999–1.002)
Urine output, mL	0.571	1.000 (0.998–1.001)
Net fluid balance, mL	0.353	1.000 (1.000–1.001)
Intravenous‐inhalation combined anesthesia: vs total intravenous anesthesia	0.556	1.222 (0.627–2.382)
Intraop sufentanil dose, µg	0.042*	0.882 (0.781–0.996)
Intraop remifentanil dose, µg	0.911	1.036 (0.560–1.917)
Hemodynamic fluctuations count	0.458	1.065 (0.901–1.260)
Lead surgeon (X.H. Fu): vs other	0.390	1.382 (0.662–2.885)

Abbreviations: ADL, activities of daily living; BMI, body mass index; CDR, clinical dementia rating; CI, confidence interval; CT, computed tomography; ECG, electrocardiogram; HCT, hematocrit; MMSE, mini‐mental status examination; MoCA, Montreal Cognitive Assessment; MRI, magnetic resonance imaging; MTA, medial temporal lobe atrophy; NPI, Neuropsychiatric Inventory; OR, odds ratio; PACU, postanesthesia care unit; SDI, Sleep Disorders Inventory; WBC, white blood cell.

^* p< 0.05.

TABLE 4.

Multivariable logistic regression of factors associated with cognitive improvement at 48 hours.

Factors	p‐Value	OR (95% CI)
Female sex (vs. male)	0.672	1.304 (0.382–4.453)
Age, per year	0.197	0.956 (0.893–1.024)
BMI, per 1 kg/m2	0.020*	0.786 (0.642–0.962)
Baseline ADL, per 1 point	0.995	1.000 (0.947–1.056)
Baseline MMSE, per 1 point	0.025*	0.809 (0.671–0.974)
MRI MTA grade 1 (vs. grade 0)	0.088	14.797 (0.671–326.390)
MRI MTA grade 2 (vs. grade 0)	0.631	1.651 (0.213–12.786)
MRI MTA grade 3 (vs. grade 0)	0.858	1.191 (0.176–8.044)
MRI MTA grade 4 (vs. grade 0)	0.762	0.709 (0.076–6.587)
Chest CT with pneumonia	0.998	N/A (effect not estimable)
Surgery duration, per 1 min	0.079	1.010 (0.999–1.021)
Intraop sufentanil dose, per 1 µg	0.179	0.879 (0.728–1.061)

Abbreviations: ADL, activities of daily living; BMI, body mass index; CI, confidence interval; CT, computed tomography; min, minute; MMSE, Mini‐Mental Status Examination; MRI, magnetic resonance imaging; MTA, medial temporal lobe atrophy; OR, odds ratio.

^*Indicates p < 0.05 (statistically significant).

3.3.2. Comparison of longitudinal changes in CSF biomarkers between the surgical cohort and the reference pharmacological cohort

To contextualize the biomarker changes observed in our single‐arm surgical cohort, we incorporated external comparative data from a recently published multicenter, prospective cohort study evaluating the real‐world efficacy, safety, and plasma biomarker trajectories of lecanemab in Chinese AD patients. ²³ The Drug group (baseline and 2.5 and 7 months post‐treatment) was used as the external reference and compared with the corresponding time points in the Surgery group (baseline and 3 and 6 months postoperation). In terms of global cognition, the Drug group exhibited an initial decline in MMSE scores at 2.5 months, followed by a slight rebound at 7 months, though values remained below baseline overall (Figure S4A, B). In contrast, the Surgery group showed a modest early improvement at 3 months, followed by a slight decrease at 6 months, with scores remaining slightly above baseline (Figure S4A, B). The between‐group difference appeared to narrow by approximately 6 months, suggesting that the short‐term cognitive trajectory after dcLVA may not be inferior to that observed with lecanemab. With respect to soluble Aβ species, both cohorts demonstrated increases following treatment, but the magnitude differed between groups. The Surgery group exhibited significantly larger increases in Aβ1‐42 (p < 0.001) and Aβ1‐40 (p = 0.001) relative to the Drug group at the corresponding follow‐up points (Figure S4C‐E). A plausible explanation for these larger postoperative increases is the restoration or enhancement of lymphatic drainage following dcLVA, which may facilitate the mobilization of brain‐derived Aβ into the peripheral circulation via the deep cervical lymphatic pathway. This mechanism is consistent with emerging evidence that dcLVA augments extracranial lymphatic outflow and may transiently elevate peripheral Aβ levels as central compartments unload their accumulated burden. ²⁵ For plasma p‐tau217, both groups demonstrated a reduction from baseline, and the extent of decline appeared broadly comparable based on available summary statistics (Figure S4F). This similarity indicates that dcLVA may elicit a degree of tau‐related biomarker modulation within a timeframe analogous to that observed with lecanemab. Nonetheless, given the summary‐level nature of the external data, these findings should be interpreted strictly as supportive contextual information rather than as evidence of comparative efficacy. Direct equivalence cannot be inferred due to differences in study design and available data granularity. Taken together, these descriptive, exploratory comparisons indicate that while dcLVA was associated with larger postoperative increases in soluble Aβ species—possibly reflecting enhanced lymphatic‐mediated drainage of Aβ from the brain—the reduction in p‐tau217 closely paralleled the trajectory reported in the pharmacological cohort. As only aggregated data were available for the Drug group, these findings should be viewed as contextual trend‐level evidence intended to support interpretation of our single‐arm study, rather than definitive comparative conclusions.

3.4. Proteomic analysis of CSF and plasma

DIA proteomics of paired CSF samples at 48 hours post‐dcLVA revealed 237 DEPs, with 229 significantly downregulated and 8 upregulated (p < 0.05, log2(FC) > 0.263; Figure 4A‐B). CSF DEPs were enriched in pathways related to mitochondrial energy metabolism (e.g., “Mitochondrial proton‐transporting ATP synthase complex”, “Mitochondrion”, “Mitochondrial matrix”, “Inner mitochondrial membrane organization”, “Proton transmembrane transporter activity”) and oxidative stress and neuroinflammation pathways (e.g., “Cellular oxidant detoxification”, “Haptoglobin–hemoglobin complex”, “Complement activation”, “lectin pathway”) (Figure S5A, B). KEGG and GSEA analyses indicated postoperative shifts in pathways annotated to Alzheimer's disease pathway and HIF‐1 signaling postoperatively (Figure 4C‐D). In plasma, 128 proteins showed significant postoperative differences post‐dcLVA, with 81 downregulated and 47 upregulated (p < 0.05, log2(FC) > 0.263; Figure 4A,G). GO analysis highlighted enrichment in protein homeostasis (e.g., “serine‐type endopeptidase activity”, “protease binding”) and epigenetic regulation (e.g., “histone/arginine methylation”) (Figure S5C,D). KEGG analysis further showed alterations in N‐glycan biosynthesis pathways (Figure S5C,D), while GSEA identified postoperative upregulation of serine‐type endopeptidase activity (Figure 4H). Besides, the proteomic outcomes further revealed that several AD‐related proteins—including PRKCBA, PRKCB, and PLCG2 in CSF, and APOE and FGA in plasma—exhibited postoperative changes (Figure 4E,F,I,J). These findings align with established roles of these proteins in AD pathways. ²⁶ , ²⁷ , ²⁸ Since proteomic shifts observed at 48 hours postdcLVA may reflect perioperative responses, these pathway findings are considered hypothesis‐generating and should be interpreted without claims of disease modification.

FIGURE 4.

The CSF and plasma proteomic findings. (A) The workflow for CSF proteomic sample collection and analysis (supported by BioRender) (preop n = 6, postop n = 6). (B) The volcano plot of CSF differential protein expression (red dots for proteins significantly up or down after surgery). (C) GSEA results for the HIF‐1 signaling pathway, with enrichment before versus after surgery. (D) GSEA results for the Alzheimer's disease pathway, with enrichment before versus after surgery. (E,F) Expression of Aβ42 or p‐Tau217 related proteins in pre‐ and postoperative CSF of AD patients. (G) The volcano plot of plasma differential protein expression (red dots for proteins significantly up or down after surgery). (H) GSEA results for the serine‐type endopeptidase activity, with enrichment before versus after surgery. (I,J) Expression of Aβ42 or p‐Tau217 related proteins in pre‐ and postoperative plasma of AD patients. Note: The 48‐hour window can include perioperative stress responses; differential proteins and enriched pathways are therefore hypothesis‐generating and are interpreted without claims of disease modification. Multiple testing was controlled using BH‐FDR; identification FDR was controlled at standard levels as detailed in the Methods section. Aβ, β‐amyloid; AD, Alzheimer's disease; BH‐FDR, Benjamini–Hochberg false discovery rate; CSF, cerebrospinal fluid; GSEA, gene set enrichment analysis.

4. DISCUSSION

This study investigated the effects of dcLVA on both cognitive function and biomarker dynamics in patients with severe AD. Our results demonstrated multifaceted therapeutic benefits of dcLVA in treated patients, including visible restoration of patient responsiveness and social interaction, enhancements in functional status and neuropsychiatric symptoms, and favorable biomarker alterations (decreased CSF Aβ42, Aβ40, and p‐Tau levels with concurrent increases in their plasma concentrations). These findings reinforced that dcLVA may enhance the efflux of pathological proteins (Aβ and tau) from the brain into the peripheral circulation, and surgically enhancing lymphatic drainage via dcLVA slowed cognitive deterioration and delayed severe AD progression.

In 2022, Lu et al. reported the first dcLVA case using a 3D exoscope, where an elderly cognitively impaired patient showed recovery of basic cognitive function by 9 months postoperatively. ²⁹ Similarly, Gan et al. observed short‐term cognitive improvement at 1‐month in four AD patients receiving dcLVA. ³⁰ However, these early studies lacked AD biomarker data, leaving unanswered whether surgery alters neurotoxic protein dynamics. A recent preprint prospective cohort study from Southwest Hospital detected no significant perioperative changes in CSF Aβ or tau, limited by a small sample size (n = 26) and short 1‐month follow‐up (https://doi.org/10.21203/rs.3.rs‐7584273/v1). In contrast, our study enrolled the largest cohort to date (n = 139) and extended neuropsychological and biomarker assessments to 6 months. This design enabled us to demonstrate not only measurable postoperative changes across multiple clinical domains but also complementary shifts in AD‐related biomarkers, providing a more comprehensive evaluation of the potential effects of dcLVA on disease‐related processes. Our findings align with emerging research implicating the meningeal lymphatic dysfunction in neurodegenerative disease. ³¹ , ³² In our cohort, complementary postoperative biochemical changes were exhibited: CSF concentrations of Aβ42, Aβ40, and p‐Tau decreased, while their plasma levels increased, with plasma p‐Tau217 additionally reduced at 6 months postoperatively. Given the restrictive nature of the blood–brain barrier (BBB) and blood‐CSF barrier on macromolecules, ³³ the divergence observed in our study therefore supports that dcLVA surgery facilitates the redistribution or efflux of these AD‐related pathological proteins from the CNS into the systemic circulation.

Although reduced p‐Tau217 may indicate a shift in tau‐related pathology, these findings may do not imply disease reversal. The modest increase in MMSE score at 6 months may still represent end‐stage dementia. Correlation analysis showed no statistically significant associations between MMSE and changes in plasma or CSF biomarker at 48 hours and 6 months postoperatively, underscoring the complexity of linking cognitive performance to biological signals, as well as the potential delay between biomarker alterations and measurable cognitive changes. Given that our cohort consisted of patients with severe AD, MMSE improvement is inherently limited, and performance is susceptible to a well‐recognized floor effect in advanced neurodegeneration. ³⁴ , ³⁵ This corresponds with a recent multicenter prospective cohort study of 68 Chinese AD patients treated with Lecanemab, in which MMSE scores remained stable in moderate‐to‐severe patients despite a significant decrease in plasma p‐tau217. ²³ Consistent results from OmegAD study likewise showed no significant correlations between key biomarkers (YKL‐40, NfL) and MMSE scores. ³⁶ Such findings support the notion that improvements in biomarkers may precede or exceed what global cognitive scales can detect in late‐stage AD. Although the MMSE increases were modest, patients exhibited broader array of clinically meaningful improvements in other measures, including reductions in plasma p‐tau217, serum IL‐6, and GFAP, and improvements in neuropsychiatric (NPI, NPI‐D) and functional (ADL) measures. Additionally, longitudinal video documentation of patients showed better alertness, engagement, and communication behaviors. These results, when taken together, provide compelling evidence that the dcLVA procedure may help mitigate aspects of clinical and biological deterioration in severe AD, although controlled studies are needed to confirm whether such effects translate into delayed disease progression.

To our knowledge, this study for the first time integrated CSF and plasma proteomic profiling to elucidate early molecular changes associated with a surgical intervention in AD. Given that aging is a dominant risk factor for AD and is closely tied to mitochondrial dysfunction, oxidative stress, and inflammation, ³⁷ , ³⁸ , ³⁹ , ⁴⁰ it is notable that postoperative CSF proteomic alterations were enriched in these aging‐related biological processes. Because CSF and plasma samples were collected 48 hours after dcLVA, these proteomic signatures predominantly reflect acute postoperative physiology. As such, enrichment results in our study should be interpreted as exploratory and hypothesis‐generating rather than as evidence of mechanistic effects or durable disease modification. Longer‐term profiling using additional time points, matched surgical controls, and external validation cohorts will be essential to determine whether any of these patterns have sustained relevance.

Chronic neuroinflammation is a key pathological feature of AD, characterized by gliosis, elevated proinflammatory cytokine levels, and synaptic loss. ⁴¹ , ⁴² , ⁴³ Since p‐Tau217 is a central mediator of CNS immune activation, the transient rise in plasma p‐Tau217 at 48 hours postoperatively may reflect perioperative stress responses, which typically resolve within days (returning to baseline levels quickly). ⁴⁴ , ⁴⁵ In contrast, the significant reduction of plasma p‐Tau217 below baseline at 6 months strongly suggests a longer‐term shift in tau‐related inflammatory activity. Additional reduced serum IL‐6 and GFAP further indicate attenuation of both systemic and astroglial inflammatory activity. Together with sustained postoperative reductions in NPI, NPI‐D, SDI, and ADL scores, and visible cognition gains documented in follow‐up videos, imply that the 6‐months p‐Tau217 decline may reflect a sustained biological effect related to enhanced lymphatic clearance following dcLVA. Nonetheless, causal inferences remain premature, and controlled mechanistic studies are needed to clarify the pathways underlying these observations.

The lack of a control arm is an acknowledged limitation of this study. However, in patients with severe AD (participants in this study), it is ethically challenging to establish a control group, particularly given the absence of a directly comparable standard of care. Because of ethical and practical constraints, a parallel control or a sham surgery group was not feasible. Therefore, we implemented a single‐arm design with a self‐controlled comparison of pre‐ and post‐operative data from the same subjects, which helps mitigate potential biases and observe within‐subject changes over time. Single‐arm, open‐label studies are common in early‐phase AD research, particularly when placebo‐controlled trials are not yet feasible. ⁴⁶ , ⁴⁷ An open‐label phase 2 trial of ALZ‐801 employed a similar self‐comparative design to assess longitudinal biomarker and cognitive changes in the absence of a control arm, ²⁴ providing valuable evidence of target engagement and disease‐modifying potential. To further validate our approach, we further compare our results with a recently single‐arm and self‐controlled study on lecanemab in Chinese AD patients, which, like our study, assessed changes in plasma biomarkers. ²³ Consistent with our findings, the external lecanemab treatment group (matched control) showed similar trends in biomarker changes of moderate‐to‐severe AD patients, along with stable cognitive outcomes (MMSE scores) over the course of the study. Although direct comparisons are limited, these trends strengthen the validity of our findings. We acknowledge that the absence of a parallel control group limits causal inference, and larger and longer‐time controlled studies are essential. To this end, we have initiated a multicenter, prospective, and randomized controlled trial evaluating the efficacy and safety of dcLVA for AD, which has received full ethics approval, with patient recruitment planned to begin shortly.

Early adverse events included delirium and sleep disturbance at 48 hours, plus one reoperation, reflecting the invasiveness and anesthesia exposure in a severe AD cohort. Delirium and sleep disturbances are common in elderly patients with neurodegenerative conditions like AD, and their presence may be exacerbated by pre‐existing cognitive dysfunction. Studies have shown that the severity of preoperative cognitive impairment is a significant predictor of postoperative delirium risk in older adults. ⁴⁸ In our cohort, all patients had severe AD, increasing their risk for postoperative delirium and other neuropsychiatric symptoms. These events were transient and managed under protocol, with delirium resolving spontaneously within 3–7 days without lasting effects. ⁴⁹ By 6 months, sleep‐related measures and broader neuropsychiatric indices had improved compared to baseline. Overall, the balance between early risks and multidomain benefits appears favorable under optimized protocols; however, longer controlled studies are needed to define safety and durability in more diverse populations.

The use of dcLVA as a therapeutic strategy for AD has rapidly gained international scientific interest, driven by the expanding understanding of brain–peripheral lymphatic interactions. ¹⁹ Chinese researchers have pioneered the application of LVA for AD, a microsurgical technique that reconnects deep cervical lymphatic vessels with adjacent veins, to restore impaired brain metabolic waste clearance. ²¹ As of now, at least 15 clinical studies worldwide are actively investigating lymphatic‐based surgical interventions for AD (Table S10), including an international single‐arm, open‐label proof‐of‐concept trial launched in April 2025 by Prof. Vincent Tay's team at Changi General Hospital, Singapore (NCT06965062). Our study, among the earliest to report outcomes in a relatively large cohort, therefore provides valuable early clinical data that may help inform the design of future controlled trials.

University of Zurich has also reported two clinical cases of dcLVA for central lymphatic dysfunction, discussing future innovations including robotic‐assisted lymphatic microsurgery. ⁵⁰ In parallel, Singapore researchers provided the first real‐time evidence of CSF efflux from the meninges into deep cervical lymph nodes in nonhuman primates, offering critical large‐animal in vivo support for lymphatic‐mediated clearance in AD. ¹⁷ Additionally, the U.S. Food and Drug Administration (FDA) granted an Investigational Device Exemption (IDE) for a first‐in‐human robotic microsurgery trial in AD patients with confirmed lymphatic obstruction (https://www.mmimicro.com/ide‐approval‐for‐first‐robotic‐microsurgery‐alzheimers‐study/). Preliminary clinical reports suggest cognitive benefit for some patients, though treatment responses appear heterogeneous, highlighting the need to identify reliable preoperative predictors of treatment response. Recent investigations have identified MRI‐derived perivascular space (PVS) burden as a strong predictor of dcLVA efficacy in AD patients, with larger PVS volume linked to greater postoperative improvement. ⁵¹ These findings position PVS burden as a clinically actionable biomarker for patient selection in lymphatic‐targeted interventions.

In summary, dcLVA is technically feasible and holds the potential to improve cognitive function, functional status, and neuropsychiatric symptoms in patients with severe AD patients, probably through brain–peripheral lymphatic interactions. ¹⁹ Our study contributes timely and meaningful data to this rapidly evolving field.

CONFLICT OF INTEREST STATEMENT

The authors declare no conflicts of interest. Author disclosures are available in the supporting information.

CONSENT STATEMENT

This study was conducted in accordance with the Declaration of Helsinki and was approved by the hospital's Ethics Committee (Approval No. 2024‐1‐727). The trial was registered at ClinicalTrials.gov (Registration No. ChiCTR2400094603). Written informed consent was obtained from all patients or their legal representatives; additional consent was obtained for the use of patient images taken before, during, and after surgery.

Supporting information

Supporting information

Supporting information

Supporting information

Supporting information

Supporting information

Supporting information

Supporting information

Supporting information

Supporting information

Fu X, Zhang J, Xiao Q, et al. Deep cervical lymphatic–venous anastomosis attenuates cognitive dysfunction and biomarker abnormalities in severe Alzheimer's disease: A prospective single‐arm study. Alzheimer's Dement. 2026;22:e71150. 10.1002/alz.71150