The integrated reference co-citation network was shown in Figure 2. It contained 1,558 nodes, 7,598 linkes, and 150 clusters. The modularity was 0.8531, which means that the structure of the network is clear and well-organized. The silhouette score of 0.9629 was very high, suggesting the clustering result was highly reliable. Each tree ring of a node represents the citation history of an article. Its size, color and thickness imply the cited times, year of publication and citations in a single time slice, respectively. What stand out in this figure were the references with high centrality, such as “van Dijk et al. (2007),” “Uhlig et al. (2016),” “Zheng et al. (2013),” “Liu et al. (2009),” and “Abildstrom et al. (2004)”. They were key hubs connecting different groups of nodes, which can be regarded as structurally essential references in this network. The color bar, ranging from cold colors to warm colors, indicated a specified period starting from 2001 to 2021. For example, lines or areas in purple generated in 2012, were earlier than those in red or orange.

The top 30 references with strong citation bursts can be grouped by the beginning time of burst (Figure 3). Instead of introduced all the 30 references, the reference with the strongest citation burst of PNDs in each group will be discussed as the most representative article. Among citation bursts starting in 2001 and 2002, the strongest bursts are associated with a 2000 paper by Diegeler et al. and a 2001 paper by Newman et al. Coronary artery bypass grafting (CABG) is a well-established surgical procedure for heart disease and can be performed with or without cardiopulmonary bypass (CPB), namely on- or off-pump CABG. Diegeler et al. (2000) investigated the impact of CPB on neuropsychological function of patients who underwent on-pump and off-pump CABG using transcranial Doppler ultrasound (TCD) and S-100 analytical method. Around the same year, some researchers focused on the role of neurobiochemical markers of brain damage and the change of inflammatory response during cardiac surgery. In 2001, Newman et al. (2001) undertook a prospective study to confirm the significant effect of perioperative cognitive decline after CABG on long-term cognitive function 5 years later. Since then, many randomized controlled trials (RCTs) have compared cognitive outcomes in different postoperative stages after on- vs. off-pump CABG.

In order to avoid possible confounding of those adverse events in surgeries caused by failed non-technical operations, the article published by Yule et al. (2006) developed a behavior rating system specifically for surgeons' non-technical skills (e.g., awareness, communication, and teamwork), aiming to allow surgeons to structure observations and improve safety and efficiency of surgical performances further.

Cibelli, Fidalgo, Terrando, Ma, Monaco and Feldmann (2010) published in 2010 suggested that inflammation plays a prominent role in the pathogenesis of postoperative cognitive dysfunction (POCD) using a mouse model of orthopedic surgery. Citation bursts starting in 2016 and 2018 are, respectively, led by Meybohm et al. (2015) and Yumuk et al. (2015) article published in 2015. The former was a prospective, double-blind, multicenter, randomized, controlled trial aiming to prove whether remote ischemic preconditioning (RIPC) can improve clinical outcomes in patients undergoing elective cardiac surgery, while the latter provided many explicit guidelines for obesity management in adults. In 2018, Evered et al. (2018) revised the nomenclature for PNDs to enhance consistency of communication and reporting in this field. It has the strongest citation burst starting in 2019, and it is still going strong.

These articles, identified based on citation bursts, indicated that researchers primarily focused on the following issues in past 20 years: the assessment, comparison and definition of neurocognitive outcomes in all stages during cardiac surgery, the predictive value of markers and tools in PNDs such as trauma-related or inflammation-related substances, the prevention and treatment of PNDs such as performing RIPC or managing high-risk population groups. In the different period, there are different focuses on this topic.

Each cluster is displayed as a horizontal timeline from left to right. The three most cited references in every year are listed below each timeline. Figure 4 lists the first 18 large clusters and demonstrates that Cluster #0 is the largest cluster and it is active in recent years. The field's focus has been gradually expanded from Cluster #5 postoperative delirium (POD) and Cluster #4 POCD to the Cluster #0 PNDs. Interestingly, remote ischemic preconditioning and obese individual are another two active clusters now. This is probably a hint that we can try to investigate the level of specific markers in different conditions or population.

In these timelines, there were three clusters labeled with specific markers, which is Cluster #7, #8, and #20. Although they were neither the largest clusters nor the most active fields, they were closely associated with our purpose to explore the potential markers of PNDs. It was much important for us to understand how these markers work and why they were studied in this field.

S100b and Neuron-specific enolase (NSE) were both first described in 1965 (Moore and McGregor, 1965), while kynurenic acid was initially found as a constituent of canine urine in 1853 (Liebig, 1853). In 2001, Donato (2001) published a paper in which he demonstrated that S100b is a member of a multigenic family of Ca2+-modulated proteins of the EF-hand type. It not only acts as an intracellular regulator to modulate cell proliferation, migration and differentiation, but can be seen as an extracellular signal playing a neurotrophic or neurotoxic role under different conditions (Donato et al., 2009). Additionally, Kleindienst et al. (2007) analyzed the value of S100b as a marker of brain damage and possible treatment option, and they concluded that the positive effect of S100b on cognition and neurogenesis can improved the functional prognosis of patients with brain injury, which probably had a therapeutic potential. But it should be noted that serum S100b levels can't reflect the corresponding cerebral S100b release.

NSE, a glycolytic enzyme found in the neuronal cytoplasm and involved in the energy metabolic process, has been widely studied in many nervous system diseases. Neuronal damage or death can increases its levels in cerebrospinal fluid (CSF) or serum (Marangos and Schmechel, 1987; Rasmussen et al., 1999). Herrmann et al. (2000) reported that postoperative serum concentrations and kinetics of S100b and NSE after cardiac surgery not only reflected early different neuropsychological outcomes but also might mirror underlying damage to neuronal and glial brain tissue. As a metabolite of tryptophan produced in the kynurenine pathway, kynurenic acid (KYNA) possesses neuroactive function, which may influence important neurophysiological and neuropathological processes. Forrest et al. (2011) reported that KYNA was a significant correlate of performance in the Stroop C and Trail A and B tests, and predicted cognitive function in patients following cardiac surgery. However, due to their poor specificity, inadequate sensitivity and inaccurate assay methods (Smith et al., 2019), the correlations between these biomarkers and postoperative cognitive function have been controversial.

Nine major clusters were listed in the Table 4. The titles of citing articles were extracted to label the corresponding clusters, which are generally considered as the research fronts of these clusters. All labels chosen by the log-likelihood ratio test method (LLR) were used for analysis. Of those nine clusters, Cluster #7 and #8 have been mentioned in Section Timeline View, while other clusters such as Cluster #1, #2, #3, and #6 were related to operation ways and neurodevelopmental problems. Hence, We selectively interpreted the Cluster #0 PNDs, #4 POCD, #5 POD in detail since they were directly related to neurocognitive disorders after cardiac surgery, the topic we're interested in. Besides, in each cluster, we emphatically discussed those references with high citations and high topic relevance.

As Table 5 shows, the top 10 cited references in Cluster #0 were mostly published in the past 5 years. Inouye et al. (2014), the most cited reference in this cluster, summarized the epidemiology, etiology, diagnosis, evaluation and management of delirium in elderly people. It offered many potential pathophysiologic contributors to delirium, including neurotransmitters (e.g., acetylcholine, melatonin, and serotonin), inflammation (e.g., IFN α/β, IL-6/8/10, and TNF-α), physiological stressors (e.g., cortisol and neopterin), electrolyte disorders (e.g., sodium, calcium, and magnesium), metabolic derangements (e.g., lactate, glucose, and IGF-1) and so on. More specifically, Hughes et al. (2020), the citing article with coverage of 34%, provided evidence-based consensus statements and practice recommendations regarding POD identification and prevention, based on an established iterative Delphi process and Grading of Recommendations Assessment, Development and Evaluation (GRADE) Criteria in the Sixth Perioperative Quality Initiative (POQI-6) consensus conference.

Evered et al. (2018), the second most cited reference, recommended using PNDs to describe cognitive impairment identified in the perioperative period, which includes pre-existing mild or major neurocognitive disorders (NCD) before operation, POD after surgery, delayed neurocognitive recovery within postoperative 30 days, and mild or major POCD from postoperative 30 days to 12 months. Both of cited references are major milestones in Cluster #0. Other citing articles mainly focused on the assessment, prevention and comprehensive management of PNDs, especially the part of neuroinflammation-related medicators and pathways.

In CiteSpace, key concepts can be algorithmically extracted from the titles of citing articles in Cluster #0 (Supplementary Figure 2). When it comes to potential diagnostic or prognostic indicators, some of them referred to certain soluble molecular substances in biological fluids such as serum, plasma or CSF, mostly associated with inflammation, brain injury, stress response, metabolism, neurotransmitter or key regulators in these processes, such as C-reactive protein (CRP), glial fibrillary acidic protein (GFAP), visinin-like/galectin-3 protein.

In Figure 4, some of more recent and highly cited members in Cluster #0 include a review of different experimental approaches and models that have been used in preclinical postoperative cognitive dysfunction research (Eckenhoff et al., 2020), a prospective biomarker cohort study of POD is associated with increased plasma neurofilament light (NfL) (Casey et al., 2020), and a mechanism research of BDNF/TrkB signaling disruption mediated by NMDAR/Ca2+/calpain might contribute to POCD in aging mice (Qiu et al., 2020).

Cited references and citing articles of Cluster #4 POCD were listed in Table 6. van Harten et al. (2012) suggested that more biochemical markers specific to neuronal damage should be identified, such as glial fibrillary-associated protein and brain- and heart-type fatty acid binding protein. The five citing articles are all reviews, which largely summarized many common studied markers associated with Alzheimer's disease pathophysiology (e.g., ApoE, Aβ1-42, and p-tau), inflammatory process (e.g., NF-kappaB, MMP9, and MCP-1), neuronal injury. Androsova et al. (2015) emphasized the importance of systems biology methods encompassing experimental measurements, imaging techniques and mathematical/computational modeling, so that we can not only understand the pathomechanisms of PNDs at different levels, but also integrate sparse markers into a complete knowledge network through some known molecular interactions and pathways. In this way, more potential markers and underlying pathways can be found easier than ever before with purposeful design.

The most cited members of Cluster #5 represent important mileposts in relation to POD (Table 7), such as Saczynski et al. (2012) article on the cognitive trajectory after POD during the first year, and Lin, Chen and Wang (2012) article on the risk factors of delirium after cardiac surgery. The citing article with the highest citation coverage of 21% in this cluster, Maldonado (2013), summarized seven complementary theories and their interrelation: the hypotheses of “neuroinflammatory,” “neuronal aging,” “oxidative stress,” “neuroendocrine,” “diurnal dysregulation,” and “network disconnectivity” can acted together to lead to a final pathway associated with neurotransmitter alterations, which generally were deficiencies in acetylcholine (Ach) and melatonin; excess of dopamine (DA), norepinephrine (NE), and/or glutamate (GLU); and decrease or increase in serotonin (5HT), histamine, and/or γ-aminobutyric acid (GABA).

The structural variation analysis can identify the potentially transformative articles published in recent years according to their contributions to the global or local structure of the knowledge domain of interest. To great extent, the method compensates for the limitation of citation-based analysis relies on citations that are greatly influenced by the publication time of articles.

Table 8 lists the first five articles in each of the last 3 years, which were ranked by the harmonic mean (H) of three different structural variation variables, namely ΔM, ΔCLw, and C_KL (Chen, 2012). The five highly ranked articles in 2021 focused on the pathophysiologic mechanisms of PNDs, such as anesthetic regimen, cerebral perfusion pressure management, cerebral oximetry monitoring and so on (Akhtar et al., 2021; Evered and Goldstein, 2021; Pisano et al., 2021; Snyder et al., 2021; Zugni et al., 2021). In 2019 and 2020, there were two valuable articles providing evidence-based recommendations on perioperative delirium care and prevention (White et al., 2019; Hughes et al., 2020). Also, several articles studied related neurovascular and immune mechanisms of PNDs and their possible markers (Luo et al., 2019; Yang and Terrando, 2019; Fong et al., 2020; Majewski et al., 2020; Wang et al., 2020).