Preoperative APOE and Alzheimer’s disease polygenic risk profiling for perioperative neurocognitive disorders

Abstract

Perioperative neurocognitive disorders (PND) include postoperative delirium within 7 days after surgery, delayed neurocognitive recovery up to 30 days, and postoperative neurocognitive disorder up to 12 months. These outcomes are related, but they are not the same. They arise from the interaction of baseline brain vulnerability and perioperative stress, including inflammation, vascular instability, blood–brain barrier injury, metabolic strain, and reduced neural reserve. Preoperative genetic profiling is useful because it can estimate latent susceptibility before surgery. Among current signals, APOE is the strongest and most biologically relevant locus. At the same time, Alzheimer’s disease polygenic risk scores (AD-PRS) can capture non-APOE common-variant burden across lipid transport, endosomal trafficking, innate immune signaling, complement activity, microglial regulation, mitochondrial stress, and neurovascular integrity. Recent perioperative cohort studies have begun to test preoperative APOE-based and polygenic neurocognitive risk in surgical patients. Large delirium genetics studies also show a strong signal at the APOE locus and support overlap between delirium risk and Alzheimer’s disease-related common-variant architecture. These findings support an APOE-aware framework in which APOE genotype is modeled separately from non-APOE AD-PRS. In clinical use, this genomic layer should be combined with baseline cognition, frailty, vascular comorbidity, surgery-related risk, and circulating biomarkers such as neurofilament light chain. This review summarizes the loci, molecular pathways, and translational model designs that can move preoperative genomic profiling from association to perioperative risk stratification.

1. Introduction

Older adults now account for a large and growing share of the surgical population, and perioperative neurocognitive disorders (PND) have become an important clinical problem in this group (Evered et al., 2018a; Ren et al., 2025; Paternò et al., 2026). These disorders are associated with longer hospital stay, greater care burden, loss of independence, and worse longer-term cognitive outcomes (Evered et al., 2018a; Ren et al., 2025; Paternò et al., 2026). The problem is especially important in aging surgical populations because many patients already have reduced cognitive reserve, frailty, vascular comorbidity, sleep disturbance, or unrecognized neurodegenerative vulnerability before surgery (Evered et al., 2018a; Ren et al., 2025; Paternò et al., 2026). In this setting, surgery and anesthesia do not act on a uniformly healthy brain. They act on a brain with variable baseline resilience.

In this review, we use the 2018 perioperative nomenclature throughout. We use PND as the umbrella term, and we reserve postoperative delirium, delayed neurocognitive recovery, and postoperative neurocognitive disorder for their specific clinical contexts. We avoid the legacy term postoperative cognitive dysfunction except when referring to older literature. These phenotypes are related, but they are not identical (Evered et al., 2018a; Ren et al., 2025; Paternò et al., 2026; Leung et al., 2007; Vasunilashorn et al., 2015; Vasunilashorn et al., 2020).

Postoperative delirium is an acute syndrome marked by inattention, fluctuating awareness, and rapid changes in mental status. Delayed neurocognitive recovery and postoperative neurocognitive disorder are more often characterized by impairment in memory, executive function, attention, or overall cognitive performance (Leung et al., 2007; Vasunilashorn et al., 2015; Vasunilashorn et al., 2020). When available, this estimate may be further improved by considering baseline phenotype, early neuropsychiatric features, and biomarker evidence of active neurodegenerative stress.

Perioperative neurocognitive decline should also be viewed as a dynamic transition rather than a single static postoperative event. In older adults, baseline clinical phenotypes and early neuropsychiatric symptoms may reflect a brain that is already moving along a neurodegenerative trajectory before surgery. Recent cerebrospinal fluid proteomics studies in the mild cognitive impairment–Alzheimer’s disease continuum support this concept and suggest that genetic susceptibility is likely to interact with baseline phenotype and biomarker-defined biology, rather than act in isolation. This principle is relevant to perioperative screening, where surgical stress may unmask latent neurocognitive vulnerability (Han et al., 2026; Vromen et al., 2022).

In this setting, current perioperative risk assessment already identifies several important predictors, including age, frailty, baseline cognitive impairment, cerebrovascular disease, systemic inflammation, and the magnitude of surgical stress (Evered et al., 2018a; Ren et al., 2025; Paternò et al., 2026; Riaz et al., 2021; Kertai et al., 2021; Clark et al., 2022; Tang et al., 2024). However, these factors do not fully explain why patients with apparently similar clinical profiles can show very different postoperative neurocognitive outcomes. This gap is especially relevant in older adults, in whom subtle preclinical brain vulnerability may not be captured by routine perioperative assessment alone. In this setting, polygenic risk stratification is best understood as a way to estimate latent susceptibility before the perioperative insult occurs. It is not a stand-alone diagnostic test. Its value lies in its ability to add genomic information to established clinical variables and to refine preoperative estimates of neurocognitive vulnerability (Riaz et al., 2021; Kertai et al., 2021; Clark et al., 2022; Tang et al., 2024).

Among current genetic signals, APOE remains the strongest and most interpretable locus linked to perioperative neurocognitive risk (Thedim et al., 2025; Raptis et al., 2026; Yang et al., 2025). Still, a single-locus model is unlikely to capture the full range of biologic susceptibility. A broader approach uses Alzheimer’s disease-informed polygenic risk scores together with dementia-related loci to estimate background neurodegenerative and neuroinflammatory liability. This approach is relevant because perioperative neurocognitive decline may reflect the unmasking of pre-existing vulnerability rather than a wholly new disease process (Paternò et al., 2026; Thedim et al., 2025; Thedim et al., 2026). Recent perioperative biobank studies and large-scale delirium genetics support the potential relevance of APOE-aware and Alzheimer’s disease-informed genomic profiling in perioperative neurocognitive risk, although the current evidence remains stronger for association than for direct improvement over clinical-only prediction models (Thedim et al., 2025; Raptis et al., 2026; Thedim et al., 2026) (Figure 1).

FIGURE 1.

Clinical framework of perioperative neurocognitive disorders and the preoperative window for genomic risk stratification. This schematic shows the perioperative timeline across the preoperative, intraoperative, early postoperative (0–7 days), intermediate recovery (up to 30 days), and long-term postoperative (30 days-12 months) phases. It also places the main clinical phenotypes within perioperative neurocognitive disorders (PND) along this timeline. Pre-existing cognitive impairment is shown as an important baseline condition before surgery. Postoperative delirium, delayed neurocognitive recovery, and postoperative neurocognitive disorder are placed according to their usual time windows after surgery. Below the timeline, major perioperative stressors are shown, including surgical tissue injury, systemic inflammation, hemodynamic fluctuations, sleep disruption, pain, infection, metabolic stress, and medication exposure. These factors act together and increase brain vulnerability across the perioperative course. The highlighted preoperative risk stratification window shows the opportunity to combine baseline cognition, frailty, vascular comorbidity, APOE genotype, Alzheimer’s disease polygenic risk score (AD-PRS), and circulating or imaging biomarkers before surgery to identify patients at higher risk. These factors shape acute brain vulnerability, postoperative delirium, delayed neurocognitive recovery, postoperative neurocognitive disorder, and different postoperative recovery patterns. The figure provides the clinical basis for preoperative genomic risk assessment in patients at risk for PND.

In this review, we examine how APOE and Alzheimer’s disease-related polygenic burden may inform perioperative neurocognitive risk, how these signals may interact with baseline phenotype and biomarker-defined vulnerability, how they map onto molecular pathways relevant to surgery and recovery, and how a clinically useful perioperative genomic framework might be designed.

2. APOE as a core locus in perioperative neurocognitive risk

APOE is located on chromosome 19 and encodes a 299-amino-acid apolipoprotein that is central to lipid and cholesterol transport in the brain (Husain et al., 2021; Budny et al., 2025; Al-Ghraiybah et al., 2026). The three main human isoforms, APOE2, APOE3, and APOE4, are defined by amino acid differences at residues 112 and 158. APOE2 carries cysteine at both sites, APOE3 carries cysteine at position 112 and arginine at position 158, and APOE4 carries arginine at both sites. These small structural differences change the shape, stability, and lipid-binding behavior of the protein. They also change how APOE interacts with cell-surface receptors and lipoprotein particles (Husain et al., 2021; Troutwine et al., 2022; Islam et al., 2025). In practical terms, the isoforms do not function in the same way in neural tissue, and APOE4 is the form most strongly linked to late-onset alzheimer’s disease risk (Budny et al., 2025; Al-Ghraiybah et al., 2026).

In the central nervous system, astrocytes are the main source of APOE under normal conditions. Microglia also produce APOE, and this expression increases in disease-related and activated states (Al-Ghraiybah et al., 2026; Lin and Holtzman, 2024; Strickland et al., 2026). After synthesis, APOE is secreted as part of lipoprotein particles and binds members of the LDL receptor family, including LDLR and LRP1. Through these receptor systems, APOE helps control lipid redistribution between glial cells and neurons, membrane turnover, synaptic maintenance, and the clearance of cellular debris (Troutwine et al., 2022; Liu et al., 2025). This places APOE in a central position in brain homeostasis. It is not only a marker of neurodegenerative risk. It is part of the machinery that supports membrane repair and metabolic balance in aging neural tissue.

The relevance of APOE to perioperative neurocognitive disorders comes from the fact that the same biologic systems regulated by APOE are stressed during surgery and in the postoperative period. One major mechanism is amyloid-β handling. APOE binds amyloid-β and influences its aggregation, deposition, and clearance (Al-Ghraiybah et al., 2026; Xia et al., 2024). Another major mechanism is immune regulation. APOE shapes microglial activation and inflammatory signaling, in part through receptor-linked pathways that affect downstream kinase activity and cytokine response (Lin and Holtzman, 2024; Strickland et al., 2026). APOE also affects synaptic lipid supply and membrane composition, which can alter how well synapses maintain transmission during acute systemic stress (Islam et al., 2025; Liu et al., 2025). In older adults, these effects can be especially important because baseline synaptic reserve is already reduced, and the brain may be less able to compensate for inflammation, sleep disruption, metabolic instability, and medication exposure after surgery (Thedim et al., 2026).

A second major mechanism is blood–brain barrier integrity. Experimental studies show that APOE4 is linked to a more permeable barrier phenotype than APOE3 (Kirchner et al., 2023). In APOE4 models, blood–brain barrier dysfunction has been associated with increased MMP9, degradation of tight junction proteins such as ZO-1, occludin, and claudin-5, and reduced astrocytic end-foot coverage of cerebral vessels (Kirchner et al., 2023; Chen et al., 2024). APOE4 is also linked to weaker control of the pericyte CYPA-MMP9 pathway, which promotes vascular injury over time (Kirchner et al., 2023). These vascular effects are directly relevant to perioperative cognition because surgery can increase systemic inflammation, endothelial stress, and hemodynamic instability (Thedim et al., 2026). In a brain with reduced vascular resilience, these perioperative insults may more easily lead to acute attention deficits, fluctuating cognition, and slower cognitive recovery.

3. Current perioperative genetic evidence

The strongest current common-variant signal linked to delirium comes from a large multi-ancestry genome-wide association study and meta-analysis (Raptis et al., 2026). That study included 1,059,130 individuals, including 11,931 delirium cases, from the United Kingdom Biobank, FinnGen, All of Us Research Program, and Michigan Genomics Initiative cohorts. The lead signal was rs429358 in APOE, with an odds ratio of 1.60 (95% CI 1.55–1.65; P = 9.7 × 10⁻¹⁷⁷). The linked rs7412 signal was also significant (OR 0.84, 95% CI 0.79–0.88; P = 1.8 × 10⁻¹¹). These two coding variants are especially important because they define the classic APOE ε2, ε3, and ε4 haplotypes (Raptis et al., 2026; Belloy et al., 2019). This means that the delirium association maps directly to the core APOE coding structure rather than to a nearby noncoding marker that only tracks with the locus through linkage disequilibrium (Raptis et al., 2026). In practical terms, this provides more direct genetic evidence for APOE involvement in delirium (Sepulveda et al., 2021). It also places the main delirium signal within the same APOE framework that is already central to late-life neurodegenerative risk (Raptis et al., 2026; Lake et al., 2023; Corder et al., 1993; Kunkle et al., 2019; Jansen et al., 2019).

The same study also showed that the APOE signal was not explained only by pre-existing dementia. After adjustment for dementia status, rs429358 remained significantly associated with delirium (OR 1.20, 95% CI 1.12–1.28; P = 3.7 × 10⁻¹⁵). The signal also remained significant in dementia-free cohorts (OR 1.27, 95% CI 1.20–1.34; P = 4.7 × 10⁻¹⁸) and in Alzheimer’s disease-free cohorts (OR 1.45, 95% CI 1.38–1.52; P = 3.9 × 10⁻⁵⁶) (Raptis et al., 2026). These findings suggest that APOE may reflect a broader form of acute brain vulnerability under stress, and not only background dementia liability. In perioperative terms, the APOE signal in delirium may capture reduced brain resilience under stress rather than only background dementia risk (Thedim et al., 2025; Raptis et al., 2026; Corder et al., 1993).

The same analysis also helped separate the main effect within the chromosome 19 region. When the model was conditioned on APOE ε4 haplotype count, the strong APOE-region association largely disappeared. This supports the view that APOE ε4 is the main independent driver of the signal in that region. After this conditioning step, additional loci became easier to detect. These included an intronic variant in ADAM32 (rs531178459) and, in sensitivity analyses, signals near SEC14L1 (Raptis et al., 2026). The investigators also used knowledge-based prioritization methods and highlighted several genes within the broader chromosome 19 interval, including APOE, TOMM40, PVRL2, and BCAM (Raptis et al., 2026). This is useful because it shows that the local signal is not biologically simple. Even though APOE is the dominant association, it sits in a dense genomic neighborhood that includes genes linked to mitochondrial function, membrane biology, vascular interactions, and cellular transport (Belloy et al., 2019; Roses et al., 2010). This local structure matters when the region is interpreted in translational models, since some nearby genes may influence perioperative brain stress responses even if they do not carry the strongest independent association.

The same delirium genetics work also supported a shared genetic background between delirium and Alzheimer’s disease (Raptis et al., 2026; Deiner and Silverstein, 2009). The investigators used summary statistics from large Alzheimer’s disease genome-wide studies and found that AD-linked information could improve discovery in delirium analyses (Kunkle et al., 2019; Jansen et al., 2019). This supports the idea that part of delirium risk is connected to the same common-variant architecture that shapes late-life neurodegenerative vulnerability (Deiner and Silverstein, 2009; Davis et al., 2012). Within this shared framework, several non-APOE genes were highlighted as important targets for future work, including MS4A4A, CR1, and TOMM40 (Roses et al., 2010; Lambert et al., 2013). These genes are notable because they are linked to biologic systems that are highly relevant to perioperative neurocognitive injury, such as microglial signaling, complement-related immune activity, and mitochondrial stress (Paternò et al., 2026; Yang et al., 2025; Hollingworth et al., 2011; Crehan et al., 2012; Shi et al., 2017). This makes them useful candidates for pathway-based models, especially if future studies move from single-locus association toward Alzheimer’s disease-informed polygenic risk scores or pathway-restricted scores built around immune, vascular, and membrane-trafficking networks (Thedim et al., 2025; Kunkle et al., 2019; Jansen et al., 2019; Deiner and Silverstein, 2009).

The perioperative literature is smaller and less consistent. In an early nested cohort study of 190 patients aged 65 years or older undergoing major noncardiac surgery, 15.3% developed delirium on postoperative days 1 and 2, and one APOE-ε4 allele was associated with a higher adjusted risk of early postoperative delirium (OR 3.64, 95% CI 1.51–8.77) (Leung et al., 2007). In contrast, a later study of 557 nondemented patients aged 70 years or older undergoing major noncardiac surgery found no association between APOE-ε4 carrier status and postoperative delirium incidence (RR 1.0, 95% CI 0.7–1.5), severity, or duration (Vasunilashorn et al., 2015). A further analysis in 553 noncardiac surgical patients showed that APOE-ε4 did not act as a clear main-effect predictor, but it modified the association between postoperative day 2 CRP and delirium. Among APOE-ε4 carriers, high CRP was associated with increased delirium incidence (RR 3.0, 95% CI 1.4–6.7), whereas no significant association was observed in non-carriers (Vasunilashorn et al., 2020). Taken together, these perioperative cohort studies suggest that APOE is biologically relevant, but the observed effect depends on the cohort, endpoint, and analytic framework used.

More recent perioperative biobank work has moved the field closer to direct preoperative genetic risk profiling in surgical populations (Clark et al., 2022). In the Mass General Brigham Biobank study of 33,526 surgical patients aged 40–89 years without previous Alzheimer’s disease, 86% of participants were of European ancestry. Among patients of European ancestry, APOE-ε4 carriage was associated with higher odds of delirium (OR 1.32, 95% CI 1.19–1.47), mild cognitive impairment (OR 1.70, 95% CI 1.49–1.94), and Alzheimer’s disease (OR 3.42, 95% CI 2.72–4.29), while the Alzheimer’s disease polygenic risk score was associated with higher odds of Alzheimer’s disease (OR 2.25, 95% CI 1.64–3.09) (Clark et al., 2022). These findings support the feasibility of perioperative genomic profiling in real-world surgical cohorts, but they mainly support association rather than incremental prediction beyond clinical variables.

Evidence for prediction gain remains limited. In the delirium genetic and proteomic study by Raptis et al., the proteomic analysis included 32,652 United Kingdom Biobank participants, including 541 incident delirium cases. Adding 18 selected proteins to a basic model that included age, sex, and BMI increased the AUC from 0.764 to 0.791, but this increase was not significant (P = 0.09). When APOE-ε4 status was added together with the proteins, the AUC increased from 0.764 to 0.794, and this improvement was significant (P = 0.049) (Raptis et al., 2026). This result suggests modest but measurable improvement beyond a simple non-genetic model. However, comparable model-comparison analyses remain limited in perioperative cohorts with clearly defined postoperative endpoints (Table 1).

TABLE 1.

Key studies on APOE and polygenic risk in perioperative neurocognitive disorders.

Reference	Study design/ Population	Sample size	Endpoint(s)	Main genetic variable	Main quantitative finding	Increment beyond clinical-only model
Leung et al. (2007)	Nested cohort; patients ≥65 years undergoing major noncardiac surgery	190	Early postoperative delirium on postoperative days 1–2	APOE-ε4 carrier status	Adjusted OR 3.64, 95% CI 1.51–8.77	Not reported
Vasunilashorn et al. (2015)	Prospective cohort; nondemented patients ≥70 years undergoing major noncardiac surgery	557	Delirium incidence, severity, duration	APOE-ε2/ε4 carrier status	ε4 not associated with delirium incidence; RR 1.0, 95% CI 0.7–1.5	Not reported
Vasunilashorn et al. (2020)	Prospective cohort; noncardiac surgical patients ≥70 years	553	Delirium incidence, severity, duration	APOE-ε4 × postoperative CRP	In ε4 carriers, high CRP associated with delirium incidence; RR 3.0, 95% CI 1.4–6.7	Not reported
Thedim et al. (2025)	Surgical biobank study; patients aged 40–89 years without prior AD	33,526	Delirium, MCI, AD	APOE-ε4; AD-PRS	Delirium OR 1.32, 95% CI 1.19–1.47; MCI OR 1.70, 95% CI 1.49–1.94; AD OR 3.42, 95% CI 2.72–4.29; AD-PRS and AD OR 2.25, 95% CI 1.64–3.09	Not clearly established in perioperative model comparison
Raptis et al. (2026)	Multi-ancestry GWAS/meta-analysis using United Kingdom biobank, FinnGen, all of us, and Michigan genomics Initiative	1,059,130 total participants; 11,931 delirium cases	Delirium genetic association	APOE rs429358 (lead variant)	rs429358 was the lead delirium signal: OR 1.60, 95% CI 1.55–1.65; after dementia adjustment, the association remained significant: OR 1.20, 95% CI 1.12–1.28	Not applicable/not reported as a clinical prediction model comparison
Raptis et al. (2026)	United Kingdom biobank European-ancestry proteomic cohort; PWAS plus prediction modeling for incident delirium	32,652 participants; 541 incident delirium cases	Incident delirium prediction (up to 16 years of follow-up)	APOE-ε4 status+18 stability-selected plasma proteins	109 of 2,919 plasma proteins were associated with incident delirium; prediction models were built using age, sex, BMI, 18 stability-selected proteins, and APOE-ε4	Adding proteins + APOE-ε4 to the basic model (age, sex, BMI) improved AUC from 0.764 to 0.794 (P = 0.049)

4. Alzheimer’s disease polygenic risk beyond APOE

A perioperative neurocognitive model that includes only APOE can capture an important part of baseline genetic vulnerability, but it does not represent the full common-variant structure linked to late-life brain susceptibility (Thedim et al., 2025; Skoog et al., 2021). Alzheimer’s disease genetics now includes many validated common-risk loci outside the APOE region (Andrews et al., 2023; Belloy et al., 2023). These loci do not act through a single pathway. They cluster across several biologic systems that are also relevant to perioperative brain stress. These systems include lipid transport, innate immune signaling, complement activity, endosomal and lysosomal trafficking, membrane protein recycling, amyloid precursor protein handling, cytoskeletal stability, synaptic maintenance, and mitochondrial stress responses (Andrews et al., 2023; Chandler et al., 2025; Xu et al., 2024). Because of this, an Alzheimer’s disease polygenic risk score, or AD-PRS, is useful in perioperative research as a way to capture distributed baseline neurobiological risk that cannot be explained by APOE alone (Najar et al., 2023; Hou et al., 2024).

From a computational perspective, AD-PRS construction involves several steps that can materially affect score performance. Investigators must first select the discovery GWAS, because the size of the training dataset, case definition, ancestry composition, and variant-level quality control of the source study all influence the effect estimates used for score construction (Clark et al., 2022; Jiang et al., 2024; Privé et al., 2021; Ge et al., 2019). They must then harmonize summary statistics with the target dataset, including genome build, SNP identifiers, effect alleles, and strand orientation, and remove variants with poor imputation quality, very low frequency, duplication, or ambiguous alignment when needed (Clark et al., 2022; Ge et al., 2019). After this preprocessing step, SNP weights are assigned by the PRS algorithm. Early Alzheimer’s disease studies often used clumping-and-thresholding approaches, which prune correlated variants and retain SNPs according to predefined association P-value thresholds (Clark et al., 2022). This method is simple and transparent, but it may discard part of the polygenic signal. By contrast, newer linkage disequilibrium-aware methods, such as LDpred2 and PRS-CS, model correlation between nearby variants more directly and apply shrinkage to effect sizes across the genome (Jiang et al., 2024; Privé et al., 2021). These methods often yield more stable scores and can improve predictive performance, although their results still depend on the quality of the input GWAS and the analytic setting used (Jiang et al., 2024; Privé et al., 2021; Ge et al., 2019).

Several additional design choices are also important. Score performance depends on the linkage disequilibrium reference panel, the parameter-tuning strategy, and the ancestry match between the discovery dataset, the LD reference dataset, and the target cohort (Clark et al., 2022; Privé et al., 2021; Ge et al., 2019). In Alzheimer’s disease, handling of the APOE region is especially important because this locus has a very large effect and can dominate the overall score if it is left inside a genome-wide PRS without special treatment (Clark et al., 2022). For this reason, many studies separate APOE from the broader polygenic component and calculate a non-APOE AD-PRS after removing the extended APOE region (Clark et al., 2022; Xu et al., 2024; Sofer et al., 2023). In practical terms, perioperative studies should report the discovery GWAS, variant filtering and harmonization procedures, PRS algorithm, LD reference panel, tuning method, APOE handling strategy, and validation framework. They should also report whether the score improves discrimination, calibration, and risk stratification beyond clinical-only models (Clark et al., 2022; Ge et al., 2019). These details are important because two AD-PRS built from the same Alzheimer’s disease GWAS can perform differently if the computational pipeline differs.

In practical terms, APOE should still be modeled separately from the broader polygenic background. The APOE signal is stronger than most other common variants in Alzheimer’s disease genetics, and it can dominate a global score if it is not handled carefully (Belloy et al., 2023; Sofer et al., 2023; Fan et al., 2020). A perioperative model is therefore easier to interpret when it is built in two layers. The first layer is an APOE-aware component, which treats APOE genotype, especially ε4 carriage, as a distinct variable (Leung et al., 2007; Thedim et al., 2025). The second layer is a non-APOE AD-PRS, which is constructed from many common variants across validated Alzheimer’s disease loci outside the APOE region (Skoog et al., 2021; Xu et al., 2024; Najar et al., 2023; Sofer et al., 2023; Sampatakakis et al., 2024). This structure makes the model clearer in both biologic and statistical terms. It allows the clinician or researcher to distinguish between a patient whose risk is mainly driven by APOE and a patient whose risk is shaped by broader common-variant burden across multiple pathways. It also reduces the chance that the very large APOE effect will mask weaker but still meaningful signals from other loci (Xu et al., 2024; Sofer et al., 2023).

This non-APOE burden is biologically relevant because the additional loci point to mechanisms that may influence how the brain responds to surgery and anesthesia. For example, variants in BIN1, PICALM, CD2AP, and SORL1 are linked to endocytosis, vesicle trafficking, and membrane recycling. These processes are important for synaptic function and receptor turnover (Andrews et al., 2023). Variants in TREM2, CD33, INPP5D, PLCG2, and MS4A-family genes are linked to microglial activation, phagocytic behavior, and inflammatory signaling (Andrews et al., 2023). These pathways are highly relevant to postoperative delirium, since delirium often develops in the setting of acute systemic inflammation and altered neuroimmune signaling (Leung et al., 2007; Li et al., 2025; Fournier et al., 2015). Variants in CR1 point to complement-related immune activity, and variants near TOMM40 point to mitochondrial transport and oxidative stress handling (Andrews et al., 2023; Chandler et al., 2025). Together, these loci support the view that non-APOE polygenic burden may reflect a broad state of reduced brain resilience, especially in older adults exposed to inflammatory, vascular, and metabolic stress during the perioperative period (Thedim et al., 2025; Supiyev et al., 2023; Thedim and Vacas, 2024).

There is already precedent for separating APOE from non-APOE polygenic risk in cognitive aging research. In older adults without baseline dementia, studies using small non-APOE polygenic panels have shown that higher non-APOE PRS is associated with a modest increase in dementia risk, while APOE ε4 produces a stronger and earlier effect (Skoog et al., 2021; Kauppi et al., 2020; Tomassen et al., 2022; Vasiljevic et al., 2023). This pattern is highly relevant to perioperative neurocognitive risk (Thedim et al., 2025; Najar et al., 2023). It suggests that APOE is likely to provide the clearest immediate signal of vulnerability, while the broader non-APOE score captures smaller, distributed effects that may become useful when combined with clinical variables such as age, frailty, baseline cognition, vascular burden, and surgery-related stress (Thedim et al., 2025; Hou et al., 2024; Supiyev et al., 2023; Loomis et al., 2024). In this setting, the value of AD-PRS is not that it replaces APOE. Its value is that it adds another layer of biologic information, especially when the goal is to estimate global neural vulnerability rather than to rely on one major locus alone (Thedim et al., 2025; Xu et al., 2024; Najar et al., 2023; Sampatakakis et al., 2024) (Figure 2).

FIGURE 2.

APOE-centered and non-APOE molecular networks relevant to perioperative neurocognitive vulnerability. This schematic shows an APOE-centered molecular framework that integrates major Alzheimer’s disease-related loci linked to perioperative neurocognitive vulnerability. APOE is placed at the center because it is involved in lipid transport, amyloid-β handling, microglial signaling, and blood–brain barrier integrity. Surrounding non-APOE risk genes are organized into major functional modules. The lipid transport and extracellular handling module includes CLU and ABCA7, which are linked to lipoprotein remodeling, extracellular chaperone activity, phagocytic lipid handling, and membrane repair support. The endocytosis, trafficking, and APP routing module includes BIN1, PICALM, CD2AP, and SORL1, which regulate endocytosis, vesicle recycling, endosomal sorting, amyloid precursor protein trafficking, and synaptic receptor turnover. The innate immunity and microglial signaling module includes TREM2, CD33, INPP5D, PLCG2, and the MS4A family, which influence microglial activation, immune signaling, and inflammatory state control. The complement and opsonic clearance module highlights CR1, which is linked to complement-related clearance, synaptic tagging, microglial recognition, and immune amplification. Mitochondrial stress and mitochondrial function are represented through TOMM40-associated and downstream bioenergetic pathways, with emphasis on mitochondrial protein import, oxidative stress, and cellular resilience. The neurovascular and blood–brain barrier integrity module includes EPHA1, CD2AP, and APOE ε4-related vascular effects, with emphasis on endothelial responses, barrier stability, astrocyte–vascular support, and permeability control. These pathways converge on reduced perioperative brain resilience and greater vulnerability to postoperative delirium, delayed neurocognitive recovery, and postoperative neurocognitive disorder. The figure provides an integrated molecular map linking APOE and non-APOE Alzheimer’s disease risk architecture to perioperative cognitive vulnerability.

5. Non-APOE loci and their functional molecular networks

5.1. Lipid transport and lipoprotein remodeling

Among non-APOE Alzheimer’s disease risk loci, CLU and ABCA7 are closely linked to extracellular lipid handling, membrane maintenance, and cellular clearance (Duchateau et al., 2024; Zhao et al., 2025; Sprenger et al., 2025; Laslo et al., 2024). CLU, which encodes clusterin, is a secreted glycoprotein with chaperone-like properties (Laslo et al., 2024; Yuste-Checa et al., 2025). It binds lipids and misfolded proteins in the extracellular space, interacts with lipoprotein particles, and helps regulate protein solubility under stress conditions (Zhao et al., 2025). In neural tissue, clusterin is produced mainly by glial cells and participates in lipid redistribution, apoptosis-related signaling, and immune response pathways (Zhao et al., 2025; Milinkeviciute and Green, 2023). It also affects amyloid-β assembly and solubility, which links it to extracellular protein homeostasis around synapses and vessel walls (Laslo et al., 2024; Yuste-Checa et al., 2025). When CLU-linked function is weaker, damaged proteins and lipid-rich debris may persist longer in the extracellular environment, which can impair synaptic stability and increase local stress signaling (Yuste-Checa et al., 2025; Milinkeviciute and Green, 2023).

ABCA7 encodes an ATP-binding cassette transporter that regulates phospholipid and cholesterol movement across cellular membranes (Duchateau et al., 2024; Sprenger et al., 2025; Santos-García et al., 2025). In the brain, ABCA7 is strongly linked to membrane lipid balance and to phagocytic activity, especially in microglia (Duchateau et al., 2024; Sprenger et al., 2025; Santos-García et al., 2025; Aikawa et al., 2021). It contributes to the engulfment and clearance of cellular debris and has been associated with amyloid-β uptake and removal (Santos-García et al., 2025). Reduced ABCA7 activity may weaken phagocytic clearance and disturb membrane composition, which in turn can affect receptor localization, vesicle fusion, and membrane repair (Duchateau et al., 2024; Sprenger et al., 2025). These effects are relevant in older adults because aging neurons and glial cells depend more heavily on efficient lipid turnover to maintain function under stress (Sprenger et al., 2025; He et al., 2025).

During the perioperative period, tissue injury, inflammation, oxidative stress, and metabolic fluctuation increase membrane turnover and generate oxidized lipids and protein debris. Under these conditions, impaired extracellular chaperone support or weaker lipid transport can reduce the capacity of the brain to stabilize damaged membranes and clear stress-related material (Santos-García et al., 2025; Aikawa et al., 2021). Common-variant burden in CLU and ABCA7 may therefore contribute to perioperative neurocognitive vulnerability by reducing lipid-linked repair capacity and by weakening debris handling during acute systemic stress (Duchateau et al., 2024; Santos-García et al., 2025; Aikawa et al., 2021).

5.2. Endocytosis, vesicle trafficking, and APP routing

A large part of non-APOE Alzheimer’s disease genetics converges on membrane trafficking, endocytosis, vesicle recycling, and intracellular cargo sorting (Kunkle et al., 2019; Jansen et al., 2019; Maninger et al., 2024; Szabo et al., 2022). Key genes in this network include BIN1, PICALM, CD2AP, and SORL1. These genes regulate how synaptic membranes are internalized and reshaped, how receptors are recycled, and how membrane-associated proteins move through endosomal and lysosomal pathways. Because synaptic function depends on constant membrane turnover and receptor repositioning, disruption in this network can impair neuronal signaling even before permanent structural injury appears.

BIN1 encodes bridging integrator 1, a membrane-binding protein involved in membrane curvature and vesicle formation (Vandal et al., 2025a; Mishra et al., 2022). In neurons, BIN1 is linked to endocytosis, cytoskeletal organization, calcium balance, and synaptic vesicle dynamics. It has also been connected to both amyloid-related and tau-related biology. Changes in BIN1-related function can alter membrane remodeling and intracellular trafficking, which may reduce the ability of neurons to maintain efficient synaptic communication during periods of physiologic stress.

PICALM, which encodes phosphatidylinositol-binding clathrin assembly protein, is a central component of clathrin-mediated endocytosis (Kunkle et al., 2019; Szabo et al., 2022; Mishra et al., 2022). It regulates receptor internalization, synaptic vesicle recycling, and movement of membrane proteins through trafficking pathways. It also influences autophagy-related transport and has been linked to amyloid precursor protein processing (Maninger et al., 2024). If PICALM-dependent trafficking is less efficient, neurons may clear damaged cargo more slowly and may recover less well after inflammatory activation or metabolic disruption.

CD2AP encodes an adaptor protein that links membrane trafficking to the actin cytoskeleton (Szabo et al., 2022; Dourlen et al., 2025; Zhang et al., 2024). It supports endocytosis, membrane stability, and vesicle movement. Because actin remodeling is essential for dendritic spine structure and receptor anchoring, CD2AP-related changes may alter the structural stability of synapses under acute stress. SORL1, which encodes sortilin-related receptor 1, regulates intracellular sorting of amyloid precursor protein and related cargo (Limone et al., 2022). Reduced SORL1 function shifts APP trafficking toward compartments that favor amyloidogenic processing, but its significance extends beyond amyloid generation. It also reflects the efficiency of intracellular cargo routing more broadly, including pathways needed for membrane protein turnover and receptor homeostasis (Kunkle et al., 2019; Limone et al., 2022).

Later perioperative neurocognitive phenotypes can emerge from reversible synaptic failure long before irreversible neurodegeneration is present (Jansen et al., 2019; Maninger et al., 2024; Mishra et al., 2022). Surgery, anesthesia, sleep disruption, inflammatory cytokines, and transient perfusion changes all place pressure on synaptic homeostasis. In that setting, efficient vesicle recycling and intracellular sorting become critical for preserving transmission. Common-variant burden across BIN1, PICALM, CD2AP, and SORL1 may reduce this adaptive reserve and make the aging brain less able to maintain stable synaptic signaling after perioperative stress (Evered et al., 2018a).

5.3. Innate immunity and microglial signaling

A large group of Alzheimer’s disease risk loci acts through innate immune signaling and microglial state regulation (McFarland and Chakrabarty, 2022; Miao et al., 2023; Jorfi et al., 2023). The most relevant genes in this set include TREM2, CD33, INPP5D, PLCG2, and members of the MS4A family (McFarland and Chakrabarty, 2022; Miao et al., 2023; Jorfi et al., 2023; Hou et al., 2022). These loci are important because microglia control inflammatory sensing, phagocytosis, synaptic pruning, and the response to cellular injury (McFarland and Chakrabarty, 2022; Miao et al., 2023; Wu et al., 2025; Shi et al., 2025). In postoperative delirium and related perioperative neurocognitive states, neuroimmune activation is often a major short-term mechanism, so variation in this network has direct translational relevance (Yao et al., 2024; Paunikar and Chakole, 2024).

TREM2 encodes triggering receptor expressed on myeloid cells 2, a receptor expressed mainly on microglia in the central nervous system (Hou et al., 2022; Zheng and Wang, 2025). It supports microglial survival, sensing of lipid-rich debris, phagocytic activity, and shifts into activated states (Shi et al., 2025; Yao et al., 2024; Zheng and Wang, 2025). Reduced TREM2-linked signaling can impair microglial clearance of damaged material and can alter how microglia respond to acute injury signals (Wu et al., 2025; Yao et al., 2024). This may change both the magnitude and the duration of neuroinflammatory activation after systemic stress (Wu et al., 2025; Yao et al., 2024).

CD33 encodes a sialic acid-binding receptor expressed on myeloid cells. In the brain, CD33 is associated with suppression of phagocytic activity and reduced uptake of amyloid-related material by microglia (McFarland and Chakrabarty, 2022; Miao et al., 2023; Jorfi et al., 2023; Akinluyi et al., 2026). Higher CD33-linked inhibitory tone may therefore limit debris clearance and prolong local inflammatory stress (Jorfi et al., 2023; Akinluyi et al., 2026). INPP5D, also known as SHIP1, is a phosphatase that modulates myeloid signaling pathways and can dampen downstream activation related to phagocytosis. Increased INPP5D-linked inhibitory signaling may further weaken clearance function (McFarland and Chakrabarty, 2022; Jorfi et al., 2023). PLCG2 encodes phospholipase C gamma 2, which acts downstream of immune receptors in myeloid cells and influences calcium-linked signaling, activation-state transitions, and broader transcriptional responses (Bedford et al., 2025; Li et al., 2022). MS4A-family genes appear to affect receptor processing and immune membrane signaling, and part of their relevance may involve regulation of TREM2-related biology and microglial activation balance (Miao et al., 2023; Akinluyi et al., 2026).

Surgery triggers systemic cytokine release, endothelial activation, acute phase signaling, and interaction between circulating immune cells and the neurovascular interface. In patients with pre-existing genetic bias toward altered microglial activation or weaker phagocytic control, the same perioperative stress may produce a stronger or more prolonged inflammatory response inside the brain. This creates a plausible route from Alzheimer’s disease-related innate immune loci to postoperative delirium, especially in older adults who already have limited neuroimmune reserve (Yao et al., 2024; Paunikar and Chakole, 2024).

5.4. Complement signaling and opsonic clearance

CR1 is one of the main complement-related loci linked to Alzheimer’s disease risk and remains highly relevant to perioperative neurocognitive injury (Daskoulidou et al., 2025; Daskoulidou et al., 2023; Veteleanu et al., 2023; Torvell et al., 2021). It encodes complement receptor 1, which binds complement-opsonized material and supports its capture and clearance (Daskoulidou et al., 2025; Daskoulidou et al., 2023). In the immune system, this receptor helps process tagged cellular debris and immune complexes. In the brain, complement signaling has a broader role because it also participates in synaptic tagging, microglial recognition of synaptic elements, and amplification of inflammatory signaling (Daskoulidou et al., 2023; Tenner and Petrisko, 2025; Dejanovic et al., 2022; Gomez-Arboledas et al., 2024; Ayyubova and Fazal, 2024; Batista et al., 2024).

Altered CR1 function has been associated with reduced complement-mediated handling of amyloid-associated material (Daskoulidou et al., 2025; Veteleanu et al., 2023). This suggests that CR1 variation may reduce extracellular immune clearance efficiency (Daskoulidou et al., 2025). In recent delirium-focused genetic and proteomic work, CR1 was specifically highlighted as a candidate locus of interest, and plasma CR1-related signal showed nominal association (Leung et al., 2007; Raptis et al., 2026). That observation is biologically plausible because complement activation is increasingly recognized as a contributor to both acute and chronic synaptic dysfunction (Tenner and Petrisko, 2025; Dejanovic et al., 2022; Gomez-Arboledas et al., 2024; Ayyubova and Fazal, 2024; Batista et al., 2024).

During the perioperative period, tissue injury activates systemic inflammatory cascades and may increase complement activity (Paunikar and Chakole, 2024). If complement signaling becomes excessive, poorly regulated, or less effectively resolved, synaptic elements may be tagged inappropriately, and microglial pruning-like responses may increase (Dejanovic et al., 2022; Gomez-Arboledas et al., 2024; Ayyubova and Fazal, 2024; Batista et al., 2024). Even without permanent neurodegeneration, this can destabilize synaptic transmission and contribute to inattention, confusion, and delayed cognitive recovery (Gomez-Arboledas et al., 2024). In this way, CR1-related common-variant burden can reflect susceptibility to immune-mediated synaptic disturbance under acute surgical stress (Raptis et al., 2026; Paunikar and Chakole, 2024; Veteleanu et al., 2023).

5.5. Mitochondrial transport and oxidative stress

TOMM40 is located adjacent to APOE on chromosome 19 and is often difficult to separate statistically because of linkage disequilibrium across the region (Kunkle et al., 2019; Jansen et al., 2019; Roses et al., 2010). For this reason, researchers should not place highly correlated TOMM40-and APOE-region variants into the same additive PRS without special handling, because this can count the same local signal more than once. A more robust strategy is to remove the extended APOE-region signal from the non-APOE PRS, model APOE genotype separately, and then test residual TOMM40 effects with conditional regression, haplotype-based analysis, or fine-mapping after adjustment for APOE ε2/ε3/ε4 status (Ware et al., 2020; Jun et al., 2012). Even so, TOMM40 has a biologically plausible role that is not limited to APOE tagging. It encodes a core component of the translocase of the outer mitochondrial membrane complex, which is required for import of nuclear-encoded proteins into mitochondria (Araiso et al., 2019; Pfanner et al., 2019; Chacinska et al., 2009). This process is essential for maintaining oxidative phosphorylation, calcium buffering, reactive oxygen species control, and mitochondrial stress responses (Araiso et al., 2019; Pfanner et al., 2019; Wang et al., 2020).

Neurons are highly dependent on mitochondrial stability because synaptic transmission, membrane repolarization, and intracellular ion homeostasis all require sustained ATP production (Devine and Kittler, 2018; Rangaraju et al., 2019). If mitochondrial protein import becomes less efficient, stress buffering capacity falls. This can make neurons more vulnerable to oxidative load, calcium dysregulation, and metabolic instability (Wang et al., 2020).

The perioperative setting places direct pressure on mitochondrial function. Anesthetic exposure, transient hypoxia, blood pressure fluctuation, inflammation, pain, and postoperative metabolic shifts all increase bioenergetic demand (Subramaniyan and Terrando, 2019; Vacas et al., 2014). In older brains, mitochondrial reserve is already reduced (Sun et al., 2016; Swerdlow, 2018). Genetic burden that affects TOMM40-linked pathways may therefore lower the threshold for cognitive decompensation under acute perioperative stress. In this context, TOMM40 may index vulnerability to the energetic and oxidative dimension of perioperative neurocognitive injury, not only to classic amyloid-linked disease processes.

5.6. Blood–brain barrier and neurovascular vulnerability

The neurovascular system is another major point where Alzheimer’s disease genetics intersects with perioperative neurocognitive risk. Relevant loci include EPHA1, CD2AP, and the vascular effects associated with APOE, especially APOE ε4(111, 112). These genes influence the integrity of the neurovascular unit, which includes endothelial cells, pericytes, astrocytic end-feet, basement membrane components, and local immune signals (Kirchner et al., 2023; Vandal et al., 2025b; Cochran et al., 2015).

EPHA1 encodes ephrin type-A receptor 1, a receptor tyrosine kinase involved in cell signaling, tissue organization, and membrane-related signaling. Functional studies have linked EPHA1 to immune pathways, membrane trafficking, and blood–brain barrier stability (Owens et al., 2024). Changes in EPHA1-related signaling may alter endothelial responses during inflammation and may affect how well the barrier resists systemic stress. CD2AP, beyond its role in vesicle trafficking, also contributes to cytoskeletal support and membrane organization, which makes it relevant to barrier architecture as well as intracellular transport (Vandal et al., 2025b; Cochran et al., 2015). APOE ε4 adds a stronger and better-characterized vascular signal. Experimental work has linked APOE4 to increased barrier permeability, altered astrocyte–vascular support, greater matrix-degrading activity, and reduced tight junction integrity (Kirchner et al., 2023; Reas et al., 2024).

Surgery stresses the blood–brain barrier through several linked mechanisms. Systemic inflammation activates endothelial cells. Hemodynamic instability changes perfusion pressure (Su et al., 2025). Hypoxia, infection, pain, and metabolic disturbance can further weaken barrier regulation. If the neurovascular unit is already less stable because of genetic burden, these perioperative stressors may push the barrier toward dysfunction more easily (Kirchner et al., 2023). Once permeability increases, peripheral cytokines, complement components, and immune cells can influence brain tissue more directly (Su et al., 2025). This can disturb neuronal signaling, amplify microglial activation, and increase the risk of postoperative delirium or delayed neurocognitive recovery (Kirchner et al., 2023; Owens et al., 2024; Reas et al., 2024).

6. Genotype-to-phenotype links and perioperative genomic model design

6.1. Biological pathways linking AD-related genetic burden to perioperative neurocognitive disorders

The connection between Alzheimer’s disease-related common-risk loci and perioperative neurocognitive disorders becomes clearer when the genetic signal is placed inside the biologic systems that are stressed during surgery (Kunkle et al., 2019; Bellenguez et al., 2022). A large part of this link runs through neuroimmune activation. APOE ε4, together with variation in TREM2, CD33, INPP5D, PLCG2, and MS4A-family genes, changes how microglia sense lipid debris, how strongly they respond to inflammatory input, and how efficiently they clear damaged synaptic material (Kunkle et al., 2019; Heneka et al., 2015). Surgical injury produces a strong systemic response that includes damage-associated molecular patterns, cytokines, complement activation, and endothelial signaling. These signals reach the neurovascular interface early after tissue trauma. In an older brain that already carries a higher inflammatory set point, microglia can enter a reactive state more quickly, release more inflammatory mediators, and remove synaptic debris less efficiently. This provides a biologically direct route to postoperative delirium, where inattention, fluctuating awareness, and acute network dysfunction develop over a short time window. Current reviews of PND increasingly place neuroinflammation at the center of this process, which is consistent with the way these loci are interpreted in Alzheimer’s disease genetics (Evered et al., 2018b; Hovens et al., 2014).

A large part of delayed neurocognitive recovery and postoperative neurocognitive disorder can also be understood through stress on synaptic membranes and intracellular trafficking. BIN1, PICALM, CD2AP, and SORL1 regulate membrane curvature, receptor internalization, vesicle recycling, endosomal sorting, and cargo routing. These functions are essential for maintaining synaptic transmission when neurons are exposed to sleep disruption, pain, sedative drugs, fluctuating neurotransmitter tone, transient hypoperfusion, and inflammatory signaling (Hovens et al., 2014). During the perioperative period, synapses need rapid receptor turnover and stable membrane recycling to preserve network activity. If endocytosis, vesicle reuse, or endosomal sorting is already less efficient because of distributed common-variant burden, then a short-lived physiologic disturbance can more easily produce longer-lasting network instability (Harold et al., 2009; Small et al., 2005). That mechanism fits especially well with delayed neurocognitive recovery and later postoperative neurocognitive disorder, where the patient remains cognitively impaired after the acute delirium phase has passed (Hovens et al., 2014). The current PND framework recognizes these later phenotypes as distinct from delirium, which makes this trafficking-based interpretation more useful than treating all postoperative cognitive outcomes as one endpoint (Evered et al., 2018b; Hovens et al., 2014).

Neurovascular stress adds another direct path from genotype to perioperative phenotype. APOE ε4 is linked to increased blood–brain barrier fragility, and EPHA1 and CD2AP are also connected to endothelial support and membrane architecture that can influence barrier stability. In this setting, clinically important dysfunction does not require a large infarct or visible structural lesion. Small increases in permeability can alter the ionic and inflammatory environment around neurons, expose brain tissue to circulating cytokines and complement proteins, and intensify local glial activation. Surgery places repeated pressure on the neurovascular unit through systemic inflammation, blood pressure fluctuation, hypoxia, infection, pain, and metabolic instability. In patients with weaker endothelial–pericyte–astrocyte coordination, these insults can produce subtle barrier leakage, impaired neuronal-glial coupling, and unstable network synchrony. This offers a strong biologic explanation for why some older adults show marked cognitive decompensation after operations that appear routine from a purely surgical point of view (Montagne et al., 2020).

Reduced mitochondrial buffering capacity is also a plausible part of the perioperative phenotype. TOMM40, APOE-linked lipid imbalance, and broader AD-related polygenic burden converge on weaker cellular resilience under metabolic stress. Neurons need sustained ATP production to preserve membrane gradients, calcium balance, synaptic transmission, and repair pathways. The perioperative period sharply increases energy demand through anesthetic exposure, inflammation, pain, transient hypoperfusion, and postoperative metabolic shifts. In that context, delirium can be viewed in part as a failure of network energy homeostasis under acute systemic stress. A higher AD-informed polygenic burden may therefore capture reduced resilience across lipid repair, oxidative control, and mitochondrial maintenance at the same time. That interpretation is more clinically useful than treating the PRS as a generic dementia score, because it places the signal inside the stress biology that is actually active during surgery and recovery (Area-Gomez et al., 2012) (Figure 3).

FIGURE 3.

Mechanistic bridge from AD-related genetic burden to perioperative neurocognitive phenotypes. This schematic shows a layered framework linking Alzheimer’s disease-related genetic burden to perioperative neurocognitive phenotypes. In the left column, APOE genotype, especially APOE ε4, and non-APOE Alzheimer’s disease polygenic burden are shown as converging genetic inputs that contribute to pre-existing neural vulnerability before surgery. In the second column, this baseline vulnerability is translated into four major biologic stress pathways. The first pathway is microglial priming and neuroinflammation. It includes TREM2, CD33, INPP5D, PLCG2, and it is linked to altered microglial activation, cytokine amplification, and impaired clearance of synaptic debris. The second pathway is synaptic membrane and trafficking fragility. It includes BIN1, PICALM, CD2AP, and SORL1, and it reflects disrupted endocytosis, impaired vesicle recycling, failed receptor turnover, and broader network instability. The third pathway is blood–brain barrier and neurovascular instability. It includes APOE ε4, EPHA1, and CD2AP, and it highlights tight junction disruption, barrier leakage, and altered neuronal–glial coupling. The fourth pathway is mitochondrial stress and reduced energy buffering. It includes TOMM40 and APOE-linked lipid imbalance, and it reflects ATP supply stress, oxidative load, and reduced metabolic resilience. In the third column, representative perioperative triggers are shown, including surgery, systemic inflammation, pain, hemodynamic fluctuations, hypoxia, sleep disruption, medication exposure, and metabolic stress. These insults interact with the vulnerable biologic pathways and worsen them. In the right column, the combined effects of these mechanisms are linked to postoperative delirium, delayed neurocognitive recovery, and postoperative neurocognitive disorder. These outcomes may occur in different patterns across patients, and recovery trajectories may vary after surgery. The figure shows how Alzheimer’s disease-related genetic architecture may be translated into clinically relevant perioperative cognitive phenotypes through biologic susceptibility pathways exposed by surgical stress.

These pathway-level mechanisms are also likely to converge at the level of large-scale brain networks. Recent bidirectional Mendelian randomization work supports this systems-level view by linking resting-state network integrity and dementia risk in both directions (Zhu et al., 2025a). Although these data are not perioperative, they support the idea that surgical stress may unmask latent vulnerability by pushing already fragile networks beyond their adaptive range (Zhu et al., 2025a). Recent Mendelian randomization analyses also suggest that smoking-related traits are associated with higher Alzheimer’s disease risk (Chen et al., 2025). This further supports the view that modifiable exposures can interact with polygenic background and reduce perioperative brain resilience (Chen et al., 2025).

6.2. APOE-aware genomic modeling strategy

A clinically useful perioperative genomic model should be built around this biology, but it should not rely on genetics alone. The clearest entry point is explicit APOE genotyping. APOE should remain a separate variable because its coding structure, effect size, and biologic meaning are distinct from the rest of the common-variant background. The same rs429358 and rs7412 framework that defines ε2, ε3, and ε4 in dementia genetics also anchors the strongest common-variant delirium signal (Raptis et al., 2026). A model that keeps APOE separate allows direct estimation of whether risk is largely ε4-driven, which is easier to interpret than a single blended score. A broader non-APOE Alzheimer’s disease polygenic risk score can then be added to capture distributed burden across lipid transport, innate immunity, complement activity, endosomal trafficking, and mitochondrial stress. Keeping this component outside the APOE region makes calibration easier to assess and reduces the chance that the large APOE effect will mask smaller but still meaningful signals from other loci. Recent surgical biobank work supports the biologic and translational relevance of this APOE-aware structure by showing that APOE-ε4 carriage and Alzheimer’s disease polygenic burden are both associated with neurocognitive outcomes in surgical populations (Thedim et al., 2025; Raptis et al., 2026). However, stronger evidence is still needed from perioperative studies that compare APOE-aware genomic models directly against clinical-only models.

6.3. Integration with clinical variables, biomarkers, and perioperative workflow

The genomic layer becomes more useful when it is placed on top of a strong perioperative clinical model (Evered et al., 2018a; Marcantonio, 2017; Inouye et al., 2014). Age, baseline cognitive testing when available, frailty, prior stroke or transient ischemic attack, vascular disease burden, sleep disorder burden, education or other proxies of cognitive reserve, surgery complexity, ICU exposure risk, and medication profiles that increase delirium risk should remain in the base model. This point matters because PND is not caused by one risk source. It reflects the interaction between baseline brain vulnerability and perioperative stress (Evered et al., 2018a; Marcantonio, 2017; Inouye et al., 2014). Genomic data become clinically meaningful when they refine this baseline estimate, not when they are treated as an isolated signal (Evered et al., 2018a; Cerejeira et al., 2014). The same logic extends to biomarker integration. Stable genetic susceptibility and current biologic stress are different kinds of information. Plasma or serum markers of neuroaxonal injury and glial activation, such as neurofilament light chain, tau-related measures, and inflammatory readouts, can add a current-state layer that the genome alone cannot provide.

Current evidence for model improvement remains early and uneven. The clearest quantitative example comes from the delirium genetic and proteomic study by Raptis et al., in which adding 18 selected proteins to a basic model with age, sex, and BMI increased the AUC from 0.764 to 0.791, but this change was not significant. When APOE-ε4 status was added together with the proteins, the AUC increased from 0.764 to 0.794, and this improvement was significant (P = 0.049) (Raptis et al., 2026). By contrast, most perioperative cohort studies have focused on association rather than formal comparison against clinical-only prediction models (Leung et al., 2007; Vasunilashorn et al., 2015; Vasunilashorn et al., 2020; Thedim et al., 2025). For this reason, genomic information should presently be viewed as an adjunct layer within multimodal perioperative risk assessment, rather than as a replacement for standard clinical evaluation.

An important limitation of this framework is ancestry. Most large Alzheimer’s disease genome-wide association studies that underpin current Alzheimer’s disease polygenic risk scores were built mainly in European ancestry cohorts. For example, the large Bellenguez study was based on the European Alzheimer & Dementia Biobank and United Kingdom Biobank datasets, and recent multi-ancestry work has also noted that Alzheimer’s disease genome-wide association studies remain predominantly European ancestry in composition (Kunkle et al., 2019; Jansen et al., 2019). The current perioperative biobank evidence is also not ancestry-balanced. In the Mass General Brigham surgical biobank study, 86% of participants were of European ancestry (Thedim et al., 2025). This matters because polygenic risk score performance is not stable across ancestries. A recent transferability study showed that a European-derived Alzheimer’s disease polygenic score performed poorly in African ancestry populations, and the association weakened further as African ancestry increased (Sun et al., 2024). In a recent multi-ancestry Alzheimer’s disease polygenic risk score analysis, the score performed best in Europeans and more modestly in Caribbean Hispanic, African American, and Native American groups, while no significant association was observed in the South Asian and East Asian groups in that dataset (Schork and Elman, 2023). For this reason, any perioperative framework that uses Alzheimer’s disease-informed polygenic risk scores should be presented as ancestry-aware. It should require validation, calibration, and threshold assessment in the target population before clinical translation.

An additional barrier to implementation is ethical and practical governance. Preoperative APOE or Alzheimer’s disease polygenic risk testing should not be treated as routine surgical laboratory work. Current Alzheimer’s disease genetic counseling guidelines state that counseling is an integral part of the testing protocol, and current polygenic risk score guidance also emphasizes the need for appropriate counseling and informed consent before testing (Sun et al., 2024; Schork and Elman, 2023). For this reason, any future perioperative workflow would need a defined pathway for test ordering, interpretation, and result disclosure, with access to genetics expertise rather than ad hoc disclosure in the preoperative clinic. Consent should explain that APOE ε4 is a susceptibility marker with implications beyond perioperative risk, that polygenic risk scores are probabilistic rather than diagnostic, and that results may also have implications for family members (Sun et al., 2024; Schork and Elman, 2023). Risk communication in the preoperative setting should therefore use plain language, make uncertainty explicit, and allow patients to decide whether they wish to receive this information. Existing APOE disclosure frameworks support standardized education, assessment of psychological readiness, and structured post-test discussion, but these models were developed mainly in research settings and are not yet established as routine perioperative care (Romic et al., 2026). For this reason, the proposed perioperative genomic workflow should be viewed as a future implementation framework rather than a current standard of care.

For implementation, the output of the model should be linked to clinical action rather than presented as a raw genetic probability (Evered et al., 2018a; Fong et al., 2015; Robinson et al., 2009). If validated in future studies, a higher-risk result could be used to identify patients who may benefit from more structured delirium-prevention pathways, closer medication review, stronger sleep protection, family reorientation strategies, and earlier postoperative cognitive monitoring (Fong et al., 2015; Robinson et al., 2009). A lower-risk result may still support routine monitoring, even if it leads to less intensive intervention. In the future, this step may also support mechanism-based target prioritization rather than only general prevention bundles. Recent genetic causality work identified five potential delirium drug targets, including C4BPA, A2M, GRIK4, C1R, and SUMF1, and these targets were linked mainly to immune-related biology (Zhu et al., 2025b). In parallel, recent experimental work showed that pharmacologic activation of TERT attenuated hippocampal inflammation, improved neurogenesis-related signaling, and reduced cognitive deficits in a cigarette-smoke model (Zheng et al., 2026). Although these data are not yet perioperative treatment standards, they suggest that future precision perioperative medicine may move toward genetically informed therapeutic prioritization (Zhu et al., 2025b; Zheng et al., 2026).

Operational translation also depends on perioperative management itself, not only on the patient’s genomic profile. A recent trial protocol comparing etomidate-based and propofol-based total intravenous anaesthesia for postoperative quality of recovery highlights that anaesthetic choice may influence early recovery, although outcome data are still pending. This suggests that future genomic profiling may eventually be interpreted together with anaesthetic strategy, but current evidence does not support genotype-based selection of a specific agent (Zhu et al., 2025c). A brief systems-level view is also relevant, because occupational stress in anaesthesiologists has been associated with worse psychological health, which supports the broader idea that perioperative safety depends on the care environment as well as on patient-level biologic risk (Xu et al., 2025). At present, however, this step remains conceptual. Without a defined downstream care pathway and prospective validation, a perioperative polygenic risk score remains mainly an association tool rather than a fully translational tool (Evered et al., 2018a; Fong et al., 2015; Robinson et al., 2009).

6.4. Score architecture for perioperative risk stratification

The most straightforward score architecture is an APOE-aware dual-score design (Thedim et al., 2025; Raptis et al., 2026). In this structure, APOE is modeled categorically, with ε2 treated as relatively protective, ε3 as reference, and ε4 as higher risk (Thedim et al., 2025; Xu et al., 2024). Alongside that, a continuous non-APOE AD-PRS is calculated after removing the APOE region (Xu et al., 2024; Sofer et al., 2023). This separation is also statistically important, because it reduces the chance that correlated APOE- and TOMM40-region variants will be counted twice in a simple additive score and makes the non-APOE component easier to interpret (Ware et al., 2020). This creates four clinically interpretable combinations: low APOE burden with low non-APOE burden, high APOE burden with low non-APOE burden, low APOE burden with high non-APOE burden, and high APOE burden with high non-APOE burden (Sun et al., 2024). These groups are easy to understand in practice, and they allow direct testing of whether APOE and broader common-variant burden contribute partly distinct information (Raptis et al., 2026; Sofer et al., 2023). That pattern is already consistent with dementia modeling and is a logical fit for perioperative stratification (Thedim et al., 2025; Raptis et al., 2026).

A more biologically explicit design uses pathway-restricted scores instead of one single global non-APOE score (Schork and Elman, 2023). In that approach, the common-variant burden is divided into function-based components (Schork and Elman, 2023; Romic et al., 2026). One score can represent lipid transport and extracellular handling through genes such as CLU and ABCA7(130). Another can represent innate immune and microglial signaling through TREM2, CD33, INPP5D, PLCG2, and MS4A loci (Romic et al., 2026). Another can represent endosomal and trafficking burden through BIN1, PICALM, CD2AP, and SORL1 (Schork and Elman, 2023). A vascular and blood-brain barrier component can emphasize EPHA1 and barrier-related loci, while a mitochondrial stress component can capture TOMM40-linked burden. This type of structure is attractive because it maps more directly onto perioperative subphenotypes. A stronger immune-pathway burden may align more closely with delirium driven by neuroinflammation (Romic et al., 2026). A stronger trafficking or lipid-related burden may fit patients with delayed neurocognitive recovery or later postoperative neurocognitive disorder after the acute postoperative phase (Schork and Elman, 2023). A stronger vascular burden may align with greater sensitivity to hemodynamic instability or barrier dysfunction. This kind of pathway-based design produces a PRS that is easier to interpret biologically and more consistent with mechanism-based perioperative translation (Andrews et al., 2023; Schork and Elman, 2023; Romic et al., 2026). This vascular domain should also be interpreted in light of longitudinal comorbidity patterns rather than as a static perioperative feature alone. Recent longitudinal evidence shows that new-onset hypertension is not linked to an immediate drop in cognition, but is linked to faster subsequent decline in global cognition, attention/calculation, and orientation in middle-aged and older adults (Zhu et al., 2025d). This pattern suggests that a single PRS value should be read together with the patient’s vascular trajectory and treatment history. In the same way, persistent depressive-symptom trajectories may also add clinical context to genomic risk. In a recent longitudinal ADNI study, persistently high depressive symptoms over 36 months after MCI diagnosis provided prognostic information for later conversion to Alzheimer’s disease (Ding et al., 2025). Although these data are not perioperative, they support the idea that longitudinal symptom trajectories may act as practical clinical proxies for underlying vulnerability during neurodegenerative progression. In this setting, pathway-based design is most useful when genetic burden is interpreted together with comorbidity history, symptom trajectory, and perioperative stress exposure (Figure 4).

FIGURE 4.

APOE-aware preoperative genomic risk stratification workflow for perioperative neurocognitive disorders. This schematic shows a conceptual five-step framework for APOE-aware preoperative genomic risk stratification in patients at risk for perioperative neurocognitive disorders (PND). Any future clinical use would require explicit consent, genetics-informed result interpretation, and structured communication of probabilistic risk. In Step 1, preoperative information is collected across three domains. The clinical domain includes age, frailty, baseline cognition, vascular disease, sleep disorder burden, surgery complexity, and medication exposure/risk. The genetic domain includes APOE genotype and non-APOE Alzheimer’s disease polygenic risk score (AD-PRS). The biomarker domain includes neurofilament light chain, tau-related markers, and inflammatory or glial markers. In Step 2, these data are combined in a central perioperative neurocognitive risk model. This model integrates clinical variables, APOE status, AD-PRS, and biomarker data to estimate individual perioperative cognitive vulnerability. In Step 3, the model output is organized using complementary score designs and interpreted in the context of longitudinal clinical history. These include an APOE-aware dual-score model, a pathway-restricted model, and clinical context from vascular comorbidity patterns and depressive-symptom trajectories. In Step 4, the model output is translated into low-, intermediate-, and high-risk tiers. In Step 5, the assigned risk tier could help guide perioperative management. In the future, this framework may also support mechanism-based therapeutic prioritization when targets supported by genetic or causal evidence become available. In addition, future implementation may need to consider modifiable perioperative factors such as anaesthetic strategy, rather than interpreting genomic risk in isolation. High-risk patients may receive a delirium prevention bundle, medication review, targeted preventive measures, sleep protection, family reorientation, early postoperative cognitive screening, and closer monitoring. Intermediate-risk patients may receive enhanced monitoring and selected preventive strategies. Low-risk patients may proceed with standard perioperative monitoring. The figure also highlights the possible use of electronic health record (EHR)-linked clinical decision support. This figure outlines a possible future route for combining genomics, biomarkers, and bedside clinical variables in perioperative cognitive risk assessment.

7. Conclusion

APOE remains the most established genetic marker linked to perioperative neurocognitive vulnerability, but it does not capture the full range of baseline risk (Thedim et al., 2025; Raptis et al., 2026; Kirchner et al., 2023). Current delirium genetics and early perioperative cohort studies support a broader model in which APOE is combined with non-APOE Alzheimer’s disease polygenic burden (Thedim et al., 2025; Raptis et al., 2026). In this framework, genetic susceptibility is distributed across several biologic systems that are directly relevant to perioperative brain stress, including lipid transport, endosomal trafficking, microglial activation, complement-related immune signaling, mitochondrial stress handling, and blood–brain barrier stability (Miao et al., 2023; Zheng and Wang, 2025). These pathways shape how the aging brain responds to inflammation, vascular stress, metabolic fluctuation, oxidative load, sleep disruption, and other common perioperative insults (Paunikar and Chakole, 2024; Su et al., 2025). The clinical importance of this model is that it links genetic variation to specific forms of reduced brain resilience rather than treating perioperative neurocognitive disorders as a nonspecific extension of dementia risk (Andrews et al., 2023).

A practical translational approach is to use preoperative APOE-aware, AD-informed polygenic profiling as one layer within a larger perioperative risk model (Thedim et al., 2025; Raptis et al., 2026). In this setting, genomic information is most useful when it is integrated with baseline cognitive status, frailty, vascular comorbidity, surgery-related stress, circulating biomarkers, and structured perioperative workflows. This allows genetic susceptibility to be interpreted as part of a broader estimate of neurocognitive reserve and perioperative vulnerability (Paunikar and Chakole, 2024; Su et al., 2025). The current evidence does not support the use of polygenic profiling as a stand-alone diagnostic tool for perioperative neurocognitive disorders (Thedim et al., 2025). Instead, the available data support its use as a biologically informed risk-stratification layer, with early evidence of modest incremental value in selected models, while larger perioperative studies are still needed to define its added predictive value over clinical-only assessment and to determine how it should guide routine perioperative decision-making (Thedim et al., 2025; Paunikar and Chakole, 2024; Su et al., 2025).

Funding Statement

The author(s) declared that financial support was received for this work and/or its publication. This work was supported by the National Natural Science Foundation of China (No. 82505722), National Natural Science Foundation of China (No.82301698) and the Fujian Provincial Natural Science Foundation (No. 2025J08106).

Author contributions

MT: Conceptualization, Formal Analysis, Methodology, Software, Validation, Visualization, Writing – original draft. JZ: Conceptualization, Formal Analysis, Methodology, Software, Validation, Visualization, Writing – original draft. HZ: Validation, Visualization, Writing – original draft. SY: Validation, Visualization, Writing – original draft. TgW: Validation, Visualization, Writing – original draft. MZ: Validation, Visualization, Writing – original draft. TuW: Conceptualization, Formal Analysis, Methodology, Validation, Visualization, Writing – review and editing. ZL: Resources, Supervision, Validation, Visualization, Writing – review and editing. YD: Resources, Validation, Visualization, Writing – review and editing. SS: Methodology, Resources, Supervision, Validation, Visualization, Writing – review and editing.

Conflict of interest

The author(s) declared that this work was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Generative AI statement

The author(s) declared that generative AI was not used in the creation of this manuscript.

Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.

Publisher’s note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.