Published online Sep 20, 2026. doi: 10.5662/wjm.117916
Revised: January 21, 2026
Accepted: February 24, 2026
Published online: September 20, 2026
Processing time: 203 Days and 21.2 Hours
Artificial intelligence (AI) is transforming onco-anaesthesia by shifting practice from reactive physiological management toward predictive and precision-based care. This review outlines current AI applications across the perioperative cancer pathway. Preoperatively, machine learning and deep learning models enhance risk stratification through automated frailty assessment, electronic health record phenotyping, and prediction of cancer-specific outcomes. Intraoperatively, AI-enabled technologies such as closed-loop anaesthesia delivery systems, predictive haemodynamic monitoring, and automated depth-of-anaesthesia control optimize drug dosing, reduce physiological stress, and may help preserve perioperative immune function, with potential implications for long-term oncologic outcomes. Postoperatively, AI-driven integration of multimodal data-including genomics, radiomics, wearable biosignals, and high-resolution physiological waveforms-facilitates early detection of complications such as delirium, persistent pain, acute kidney injury, and anastomotic leakage. The review also examines the role of AI in evaluating the “onco-anaesthesia hypothesis” by clarifying links between anaesthetic techniques, inflammation, and cancer recurrence. Despite these advances, significant challenges persist, including data heterogeneity, limited generalisability, algorithmic opacity, regulatory uncertainty, and ethical concerns related to equity and clinical implementation. Future progress will depend on explainable AI, federated learning, real-time clinical decision-support systems, and validation through large, prospective studies to fully realise AI’s potential in personalised onco-anaesthetic care.
Core Tip: Artificial intelligence (AI) is redefining onco-anaesthesia by enabling the construction of “digital twins” for cancer patients, allowing clinicians to simulate and optimize anesthetic care before the first incision. This review highlights how AI algorithms, ranging from computer vision to natural language processing, are being deployed to predict adverse events like hypotension and delirium with unprecedented accuracy. By modulating the neuroendocrine stress response and facilitating opioid-sparing strategies, AI-augmented anesthesia may improve long-term cancer survival. However, the successful integration of these tools requires rigorous validation, ethical oversight, and a commitment to explainable AI to ensure clinical trust and patient safety.
- Citation: Sirohiya P, Maurya P, Gupta N, Ratre BK, Vig S, Puri S, Kumar B, Gupta R, Bhopale S, Pandit A. Artificial intelligence in onco-anaesthesia: Current applications, challenges, and future directions. World J Methodol 2026; 16(3): 117916
- URL: https://www.wjgnet.com/2222-0682/full/v16/i3/117916.htm
- DOI: https://dx.doi.org/10.5662/wjm.117916
Surgical resection of solid tumors remains the cornerstone of curative cancer therapy, yet the perioperative period represents a biologically vulnerable window during which anesthetic management may profoundly influence long-term oncological outcomes, including metastasis and recurrence. While the traditional focus of onco-anaesthesia has centered on providing insensibility and maintaining physiological stability, contemporary practice is increasingly shaped by mounting evidence linking perioperative stress modulation, inflammation, and immune preservation to the fundamental biology of cancer progression.
The mechanistic underpinnings of this relationship have been extensively studied, revealing that inhaled anesthetics such as isoflurane, sevoflurane, and halothane decrease cytotoxicity in natural killer (NK) cells when compared to propofol, which does not impair NK cell cytotoxicity[1]. Furthermore, sevoflurane specifically triggers apoptosis and increases hypoxia-inducible factor-1α expression in T lymphocytes through the PI3K/Akt pathway in a dose-dependent manner[1]. Beyond their effects on immune cells, inhalation anesthetics also upregulate pro-tumorigenic factors including vascular endothelial growth factor A, matrix metalloproteinase 11, transforming growth factor β, and C-X-C motif chemokine receptor 2 in ovarian cancer cells[1].
The picture is further complicated by evidence that high concentrations of sevoflurane may paradoxically inhibit tumor invasion by downregulating the p-38 MAPK pathway and suppressing MMP-2/-9 activity, though these effects occur at concentrations that may exceed clinical relevance, highlighting the complexity of translating laboratory findings to clinical practice[2].
Against this intricate biological backdrop, artificial intelligence (AI)-encompassing machine learning, deep learning, and natural language processing[3]-offers a transformative shift toward precision perioperative oncology. Unlike conventional statistical tools, AI systems exploit high-dimensional “big data”, including continuous physiological waveforms, complex electronic health records (EHR), radiomics, and genomic or treatment-related biomarkers, to uncover non-linear patterns and generate individualized, context-specific insights[4].
We conducted a comprehensive literature search using PubMed, Scopus, and Web of Science databases from January 2000 to December 2024. The search strategy employed a combination of keywords including “Artificial Intelligence”, “Machine Learning”, “Deep Learning”, and “Onco-anaesthesia” or “Surgical Oncology”. We utilized VOSviewer (version 1.6.19) to analyze publication trends and co-occurrence networks. Inclusion criteria focused on peer-reviewed articles discussing AI applications in the perioperative care of cancer patients, while editorials and non-English publications were excluded.
The growth of AI research in anesthesiology has been remarkable, with bibliometric analyses revealing rapid expansion beginning in 2019 at a 75% annual growth rate, with 80.81% of all publications totaling 1355 appearing between 2019 and 2023, ultimately reaching 326 articles by 2023[5]. This paradigm shift arrives at a particularly urgent moment as clinicians increasingly manage older, frailer patients undergoing ever more complex oncological surgeries, often after exposure to immunotherapies and targeted agents that fundamentally alter physiological reserves and pharmacological responses[6,7].
By integrating multimodal data streams, AI enables dynamic, real-time decision support across the entire perioperative continuum. This review therefore synthesizes current and emerging AI applications in onco-anaesthesia, evaluates their potential to clarify mechanistic links between anesthesia and cancer progression, and addresses the ethical, regulatory, and infrastructural barriers to implementation, while outlining key future directions such as federated learning, explainable AI, and real-time adaptive decision-support systems that may define the next era of precision onco-anaesthetic care[8].
Figure 1 provides a conceptual framework illustrating the AI applications across the perioperative continuum that will be examined in this review.
The preoperative phase constitutes the foundational pillar of perioperative safety, and in oncological surgery, accurate risk stratification determines not only the feasibility of resection but also the allocation of critical care resources and the implementation of prehabilitation strategies. Traditional risk indices, such as the American Society of Anesthesiologists physical status classification, are often subjective and lack the granularity necessary to predict specific oncological complications. AI methodologies are systematically dismantling these limitations by enabling objective, automated, and individualized risk profiling that transcends the capabilities of conventional assessment tools.
Frailty represents a syndrome of physiological decline characterized by reduced resilience to stressors, and it stands as a potent predictor of postoperative mortality, prolonged hospital stay, and discharge to institutional care in the cancer population[9]. Conventional frailty assessments, such as the fried frailty phenotype or the comprehensive geriatric assessment, are resource-intensive and often impractical in high-volume surgical clinics where time and personnel constraints limit thorough evaluation.
Recent advancements in computer vision, a subset of AI, have enabled the automation of frailty assessment in ways that were previously unimaginable. Algorithms analyzing preoperative video data of patient gait, sit-to-stand mechanics, and facial expressions have demonstrated high concordance with formal geriatric assessments. By utilizing pose-estimation libraries such as OpenPose, these systems quantify kinematic features-including gait velocity, stride length variability, and joint angular velocity-within seconds, providing an objective "frailty biomarker" that can be obtained without specialized geriatric expertise.
Beyond visual analysis, machine learning models are leveraging the vast troves of data within the EHR to construct “digital phenotypes” of frailty that capture multidimensional aspects of patient vulnerability[10]. Studies utilizing algorithms such as XGBoost and random forest (RF) have successfully mined structured data including laboratory values and diagnosis codes alongside unstructured data extracted from clinical notes via natural language processing to predict frailty-related outcomes. One notable study using the “Hospital Frailty Risk Score” derived from ICD-10 codes demonstrated that machine learning models could predict 90-day mortality and readmission rates in elderly cancer patients with an area under receiver operating characteristic curve (AUROC) of 0.840 for XGBoost, substantially outperforming traditional regression models[11].
A multicenter analysis of 937 elderly patients undergoing gastrointestinal cancer surgery compared multiple machine learning algorithms including RF, gradient boosting, support vector machine, and logistic regression against the Gastrointestinal Surgery Frailty Index, with RF demonstrating superior performance achieving an area under the curve (AUC) of 0.822, outperforming the GiS-FI score for mortality prediction[12].
The implications of these tools are transformative for clinical practice. By identifying patients with occult frailty who might otherwise be cleared for surgery based on superficial assessment, anesthesiologists can initiate targeted prehabilitation programs focusing on nutritional optimization, exercise conditioning, and anemia correction, thereby modifying the patient's risk profile before they enter the operating room.
Cancer surgery carries specific risks ranging from anastomotic leakage in gastrointestinal surgery to acute kidney injury and sepsis, and the ability to predict these specific complications allows for truly informed consent and risk-adapted perioperative management strategies tailored to individual patient vulnerabilities.
In major abdominal oncologic surgery, anastomotic leakage represents one of the most dreaded complications, and a recent study utilizing machine learning trained on perioperative data from colorectal cancer patients demonstrated superior predictive performance for anastomotic leakage compared to experienced surgeons using standard clinical criteria[13]. A systematic review and meta-analysis of machine learning applications in colorectal cancer surgery indicated that predictive models achieved a pooled AUC of 0.84, suggesting good overall discriminatory ability[13]. The neural network integrated complex variables including intraoperative hypotension duration, blood loss, and nutritional indices to generate a personalized risk score that surpassed clinical intuition.
Similarly, in liver resection for hepatocellular carcinoma, machine learning models incorporating preoperative computed tomography (CT) radiomics and clinical data have outperformed the established Child-Pugh and model for end-stage liver disease scores in predicting post-hepatectomy liver failure, with studies comparing RF, support vector machine, gradient boosting, and neural networks using decision curve analysis to rank variable importance and optimize model selection[14].
For patients undergoing lung cancer resection, predicting postoperative pulmonary complications is critical for perioperative planning and resource allocation. Machine learning models analyzing preoperative pulmonary function tests, diffusing capacity, and comorbidities have achieved high discrimination in predicting these complications. One study, “predicthor”, utilizing an AI-powered predictive risk model, identified that specific combinations of comorbidities and tumor location were strong predictors of prolonged air leak and pneumonia, allowing for the prophylactic placement of specialized chest drainage systems or admission to high-dependency units when appropriate[15].
For early recurrence prediction in non-small cell lung cancer, a stacking ensemble model combining RF, XGBoost, and support vector machine with a logistic regression meta-learner achieved an AUC of 0.81 in the testing cohort and 0.80 in external validation, with key predictors identified via boruta algorithm and SHapley Additive exPlanations (SHAP) analysis including pathological T stage, maximal tumor diameter, CA125, and CYFRA21-1[16]. The MoLPre model for predicting metastasis in cT1 solid lung cancer using RF as the primary classifier achieved an AUC of 0.92 with 95% confidence interval (CI) of 0.88 to 0.94 in internal validation and 0.90 in external validation, with key predictors including age, largest nodule dimension, CEA, NSE, and CYFRA21-1[17].
A critical barrier to the clinical adoption of these models has been the “black box” nature of AI, where the rationale behind a prediction remains opaque to the clinician who must act upon it. To address this fundamental challenge, recent studies have employed Explainable AI techniques such as SHAP. A study developing a machine learning model for postoperative complications classified as Clavien-Dindo grade 3 or higher in cancer inpatients achieved AUROC of 0.77 in training and 0.73 in holdout testing[6]. The authors reported AUPRC values of 0.56 in training and 0.52 in holdout testing, which are more appropriate metrics than AUROC for imbalanced datasets where complications occur in only 28% of patients. SHAP values identified specific features driving risk predictions, including total hospital length of stay greater than 13 days in the past year, lymphocyte count below 10.6%, hemoglobin below 9.1 g/dL, heart rate above 103 beats per minute, and greater than 52% abnormal lab tests in the past 90 days. This transparency empowers clinicians to trust the AI’s output and explains the risk to patients in tangible terms, facilitating meaningful shared decision-making.
A comprehensive comparison of the various AI algorithms employed in onco-anaesthesia is presented in Table 1.
| Algorithm type | Primary mechanism | Key clinical application | Advantages | Limitations | Performance metrics |
| Random forest | Ensemble learning using multiple decision trees | Preoperative risk stratification; Metastasis prediction (MoLPre) | Handles non-linear data; robust to outliers; explainable | Can overfit; less effective for real-time waveforms | AUROC 0.73-0.77[6], AUC 0.92 (internal validation), 0.90 (external validation)[17] |
| Gradient boosting | Iterative ensemble correcting prior errors | Mortality prediction; delirium prediction | High performance; handles missing data | Computationally intensive | AUC 0.85[31]; AUC 0.840[11] |
| Stacking ensemble | Meta-learning combining RF, XGBoost, SVM | Early recurrence (NSCLC) | Improved generalization | Complex implementation | AUC 0.81 (test), 0.80 (external validation)[16] |
| CNN | Deep learning for grid-like data (images) | Ultrasound guidance; radiomics | State-of-the-art image recognition | Low interpretability; needs large datasets | 95%-100% accuracy[24] |
| RNN/LSTM | Deep learning for sequential data | Hemodynamic prediction (HPI); Nociception | Excellent for waveform analysis | High computational cost | AUROC 0.94, Accuracy 0.88[22] |
| BiLSTM with Attention | Bidirectional sequence processing | Depth of anesthesia (EEG) | Superior temporal extraction | Computationally intensive | 88.7% accuracy[3] |
| Gaussian process regression | Bayesian non-parametric regression | Remifentanil PK prediction | Uncertainty estimates | Intensive for large datasets | R² = 0.96[3] |
| Reinforcement learning | Trial-and-error optimization | Closed-loop anesthesia | Adapts to patient variability | Safety concerns during learning | Correlation 0.88[3] |
| NLP | Human language processing | EHR phenotyping | Unlocks unstructured text | Struggles with medical jargon | Variable[6] |
The cancer patient is often on a complex pharmacological regimen including chemotherapeutics, immunomodulators, and analgesics, and AI systems are increasingly being deployed to navigate this pharmacological complexity with greater precision than human cognition alone can achieve. Pharmacogenomic analysis, facilitated by AI, can identify patients with genetic variants affecting the metabolism of anesthetic drugs and opioids. Key genetic factors include CYP450 enzyme polymorphisms, particularly CYP2D6 and CYP3A4, pseudocholinesterase variants affecting succinylcholine metabolism, and receptor polymorphisms influencing drug response[18].
For example, in breast cancer patients taking tamoxifen, which is a prodrug requiring bioactivation by CYP2D6, AI tools can flag potential interactions with antidepressants or analgesics that inhibit this enzyme, ensuring that cancer therapy efficacy is not compromised by perioperative supportive care. Furthermore, AI algorithms can predict individual susceptibility to chemotherapy-induced toxicities, such as anthracycline-induced cardiomyopathy, guiding the anesthesiologist to choose cardio-stable induction agents and avoid fluid overload[19]. Ongoing randomized controlled trials (RCT) are investigating the effect of different intraoperative blood pressure regulation levels on postoperative myocardial injury in breast cancer patients after neoadjuvant chemotherapy, comparing mean arterial pressure targets of 80% to 100% vs 100% to 120% of baseline[19].
The intraoperative phase involves the continuous surveillance and adjustment of physiological parameters to maintain homeostasis amidst surgical perturbation, and AI applications in this domain are moving beyond simple monitoring toward predictive analytics and autonomous control.
Intraoperative hypotension is a common and preventable cause of postoperative morbidity that has been increasingly recognized as a critical target for intervention. Even short durations of hypotension with mean arterial pressure below 65 mmHg are associated with myocardial injury, acute kidney injury, and potentially compromised perfusion to the tumor bed, which may promote hypoxic signaling and metastasis. A retrospective study in 500 esophagectomy patients found that prolonged hypotension with mean blood pressure below 65 mmHg showed significant association with anastomotic leakage with an adjusted odds ratio of 1.02 per 10-minute interval and 95%CI: 1.01-1.04[20].
Traditional monitoring systems alarm only after a threshold has already been breached. In contrast, the Hypotension Prediction Index, an AI-driven algorithm, analyzes high-fidelity arterial pressure waveforms to predict the likelihood of hypotension occurring in the next 5 minutes to 15 minutes[21]. The algorithm, trained on millions of waveform segments, extracts features related to cardiac contractility, afterload, and preload.
As detailed in Table 1, gradient boosting machine and hybrid CNN-RNN models using multimodal waveform data including arterial blood pressure, electrocardiography (ECG), photoplethysmography, and capnography achieved an AUROC of 0.94 with 95%CI: 0.9-0.95 and accuracy of 0.88 for intraoperative hypotension prediction at a 5-minute lead time[22]. Validation studies have consistently shown high discriminatory performance for predicting hypotensive events, though external validity remains to be verified given potential selection bias from single tertiary centerdata[22].
Importantly, researchers have identified six endotypes of hypotension-myocardial depression, bradycardia, vasodilation with or without cardiac index increase, hypovolemia, and mixed type-which may inform targeted therapeutic interventions tailored to the underlying mechanism[21]. For the onco-anaesthetist, this tool proves invaluable, as preemptively treating impending instability through fluid bolus or vasopressor titration allows the clinician to maintain optimal organ perfusion, theoretically reducing the release of stress hormones and inflammatory cytokines that favortumor survival.
Maintaining the optimal depth of anesthesia represents a delicate equilibrium with significant consequences at either extreme. Excessive depth is linked to postoperative delirium and increased mortality, while light anesthesia carries the risk of awareness and severe stress response. Closed-loop anesthesia delivery systems represent the application of robotics and control theory to anesthesia, using real-time feedback from depth of anesthesia monitors that process electroencephalogram (EEG) signals to automatically adjust the infusion rates of propofol and opioids[3].
AI algorithms, particularly reinforcement learning agents, have substantially improved the performance of these systems by learning each individual patient's pharmacodynamic response. Specific advances include gaussian process regression for remifentanil pharmacokinetic prediction achieving an R² of 0.96[3]. In complex cancer surgeries where stimulus intensity varies wildly-for instance, during dissection vs closure phases-these closed-loop systems have demonstrated the ability to maintain the bispectral index within the target range more consistently than manual control, with reduced total drug consumption and faster emergence times[21].
The implications for oncology are significant given that volatile anesthetics and opioids have been implicated in immune suppression. Specifically, propofol enhances cytotoxic T lymphocyte activity, reduces pro-inflammatory factors, and inhibits hypoxia-inducible factor 1-alpha (HIF-1α) translation, potentially offering immunoprotectivebenefits[1]. By optimizing and often reducing the total anesthetic load through precise titration, closed-loop systems may help preserve the patient's immune competence during this critical window.
Pain assessment in the unconscious patient has traditionally relied on autonomic surrogates like heart rate and blood pressure, which are non-specific and confounded by numerous other factors. AI is enabling the development of a composite “Nociception Index” by integrating multimodal data streams including EEG, photoplethysmography, heart rate variability, pupilometry, and skin conductance[3].
Deep learning models, such as Convolutional Neural Networks and Long Short-Term Memory networks, can analyze the temporal relationships between these signals to distinguish true nociception from other causes of autonomic arousal. A multimodal deep learning approach integrating EEG, photoplethysmography, and ECG for intraoperative nociception monitoring has demonstrated correlations with established nociception level indices, utilizing specific signal processing and normalization techniques to optimize signal quality[23]. Accurate nociception monitoring allows for the precise titration of analgesics, particularly opioids, and in the context of “opioid-free” or “opioid-sparing” onco-anaesthesia, these AI tools provide the confidence to minimize opioid use without risking the deleterious effects of untreated surgical stress.
Regional anesthesia, encompassing epidurals, paravertebral blocks, and fascial plane blocks, constitutes a key component of enhanced recovery protocols in cancer surgery. These techniques provide superior analgesia, reduce opioid requirements, and blunt the neuroendocrine stress response that may promote tumor progression. However, the efficacy of regional techniques depends critically on the accurate identification of anatomical structures via ultrasound, which presents a steep learning curve even for experienced practitioners.
AI-based image analysis software is now available to assist in real-time interpretation of ultrasound images, democratizing access to these advanced techniques. These systems use deep learning, specifically U-net architectures, to segment anatomical structures, overlaying color-coded highlights on the screen to identify nerves, arteries, and fascial planes with remarkable accuracy. Commercial systems such as ScanNav Anatomy and Nerveblox have demonstrated 95% to 100% accuracy for anatomical structure identification, while AI-assisted B-line quantification has achieved 94.5% accuracy for pulmonary pathology detection, and the FDA-cleared Kosmos algorithm for ejection fraction calculation represents another validated application[24]. Studies have shown that AI assistance significantly improves the success rate of blocks performed by less experienced practitioners and reduces the time to procedure completion[24].
A central tenet of modern onco-anaesthesia is the hypothesis that the perioperative period constitutes a “perfect storm” for cancer progression. According to this framework, surgical stress, volatile anesthetics, and opioids suppress cell-mediated immunity, particularly NK cell function, potentially allowing residual micrometastases to survive and proliferate during this vulnerable window[1]. AI is playing a crucial role in investigating and validating this hypothesis through sophisticated analytical approaches that transcend traditional statistical methods.
Determining the impact of anesthetic technique, such as total intravenous anesthesia vs volatile anesthesia or regional vs systemic approaches, on cancer recurrence requires massive sample sizes and extended follow-up periods, making RCTs difficult and expensive to conduct. Retrospective database analyses, while more feasible, are prone to confounding that can obscure true causal relationships. AI techniques, specifically causal inference machine learning, are being deployed to analyze large cancer registries with greater rigor than previously possible[18].
These algorithms can identify non-linear confounders and interaction effects that traditional propensity score matching might miss entirely. For instance, AI analyses have suggested that the protective effect of total intravenous anesthesia might be heterogeneous, benefiting patients with specific tumor biology or inflammatory profiles more than others who may show no benefit at all.
Importantly, the relationship between intraoperative opioid dosage and cancer recurrence appears to be stage-dependent rather than following a simple dose-response curve that would apply universally. A retrospective analysis of non-small cell lung cancer patients with median intraoperative fentanyl equivalents of 10.15 μg/kg found a trend toward significance for stage I disease with a P value of 0.053, with opioid consumption being a significant risk factor for overall survival in stage I patients with P value of 0.036, but no effect was observed for stage II to III patients[25]. A systematic review of opioids and breast cancer recurrence similarly revealed conflicting results across retrospective and prospective studies, examining impacts on the immune system, angiogenesis, and tumor growth modulation[26]. These findings suggest that the “threshold effect” may vary considerably by cancer stage and type, and the authors recommend that opioids should continue as a key component of balanced anesthesia until RCTs clarify these associations[25].
AI is also being applied to model the patient’s immune response to surgery with unprecedented precision. By integrating preoperative data encompassing genomics, inflammatory markers, and nutritional status, machine learning models can predict the degree of postoperative immunosuppression a specific patient is likely to experience[27]. A systematic review and meta-analysis of immunomodulatory effects of anesthetic techniques in lung cancer surgery has quantified effect sizes for different anesthetic modalities on immune markers, providing the foundation for evidence-based technique selection[27]. This capability allows for a “precision immuno-oncology” approach to anesthesia that tailors technique to individual patient vulnerability.
For example, a patient predicted to have severe suppression of NK cell function might be targeted for an aggressive opioid-sparing regimen utilizing regional anesthesia and anti-inflammatory agents such as non-steroidal anti-inflammatory drugs and lidocaine infusion. Furthermore, AI models are being developed to predict response to neoadjuvant immunotherapies including checkpoint inhibitors based on histological and genomic features, with national institutes of health scientists developing AI tools to predict how cancer patients will respond to immunotherapy[28]. This information helps anesthesiologists avoid drugs like high-dose dexamethasone that could theoretically antagonize these life-saving therapies in responders.
Advanced AI models are facilitating basic science research into the mechanisms of metastasis in ways that generate clinically relevant insights. Machine learning algorithms are used to analyze proteomic and genomic changes in tumor cells exposed to anesthetic agents, identifying signaling pathways such as HIF-1α and VEGF that are modulated by drugs like sevoflurane or propofol[2]. The MoLPre model and similar machine learning approaches are being used to predict metastasis risk based on clinical and imaging features, with an online prediction tool available at https://molpre.cqmu.edu.cn/ for clinical application[17]. These insights are generating new hypotheses for clinical trials and identifying biomarkers that could serve as early warning signs of recurrence.
The postoperative phase is critical for the return to function and the timely initiation of adjuvant therapies, and AI applications in this period focus on the early detection of complications and the management of long-term sequelae that significantly impact cancer survivorship.
Postoperative delirium is a devastating complication in the geriatric cancer population, with incidence ranging from 9% to 46% depending on cancer type and surgery. Specifically, incidence rates vary by cancer type with colorectal at 2.2%, gastric at 8%, hepatic at 8.6%, and pancreatic or biliary at 11.4%, while abdominal malignancies show overall rates of 22.91% in elderly patients[29]. Beyond its immediate impact, delirium is associated with increased mortality, long-term cognitive decline, and substantial patient distress. The pathophysiology is complex, involving neuroinflammation, neurotransmitter imbalance, oxidative stress, and network dysconnectivity[29]. AI models leveraging preoperative cognitive testing, intraoperative EEG signatures such as burst suppression duration, and medication data have achieved high accuracy in predicting this complication[30].
The “AI-delirium guard” model using XGBoost achieved best performance with an AUC of 0.85, outperforming RF’s AUC of 0.84, on test data from the National Surgical Quality Improvement Program database[31]. A web-based implementation tool is available at https://ai-delirium-guard.streamlit.app for clinical use. A notable study incorporated sleep architecture data collected via polysomnography and wearable sensors into a machine learning model, integrating clinical features with sleep variables to significantly improve the prediction of delirium compared to clinical variables alone[32]. Identifying high-risk patients in advance allows for the implementation of non-pharmacological prevention bundles focusing on reorientation and sleep hygiene alongside the avoidance of deliriogenic drugs, effectively shifting care from reactive treatment to proactive prevention.
Chronic postsurgical pain is a major survivorship issue that often leads to long-term opioid use with all its attendant risks. Predicting which patients will transition from acute to chronic pain remains challenging with traditional clinical assessment. AI models analyzing trajectories of acute pain scores, genomic markers such as COMT variants, and psychological factors including catastrophizing can predict the risk of chronic postsurgical pain and persistent opioid use with high precision. Machine learning models for chronic postsurgical pain in breast cancer patients have been developed using logistic regression, support vector machine, RF, and XGBoost, with key predictive features identified including higher pain ratings within 48 hours post-surgery, post-menopausal status, history of prior surgery, fentanyl with sevoflurane anesthesia, and axillary lymph node dissection[33].
AI applications in cancer pain assessment now encompass multiple modalities including camera-based systems analyzing facial expressions, contact-sensor approaches using physiological signals, audio analysis of voice patterns, and multimodal integration of these data streams[34]. These systems enable personalized treatment plans, predictive analytics, telehealth integration, and opioid optimization strategies.
Studies integrating preoperative EHR data with Fitbit wearable device data encompassing sleep duration and physical activity levels from 347 patients across eight surgical types demonstrated that a stacking ensemble model achieved 89% accuracy with AUC of 0.90 and 95%CI: 0.88-0.92 for chronic opioid use prediction, substantially outperforming logistic regression which achieved AUC of 0.74[18]. SHAP analysis identified very active minutes, lightly active minutes, and cancer history as critical predictive factors[35]. Using these models, clinicians can identify patients who require aggressive early intervention through ketamine infusions or gabapentinoids and extended follow-up by transitional pain services. This personalized approach is crucial for mitigating the risk of opioid addiction in cancer survivors.
Postoperative deterioration on the general ward often goes unnoticed until it reaches severity that demands emergent intervention. AI-based early warning systems analyze trends in vital signs and laboratory values to predict events like sepsis or respiratory failure hours before they manifest clinically[36]. In oncology patients who may be neutropenic or profoundly debilitated by their disease and treatments, these systems provide a critical safety net that can trigger early escalation of care. The integration of wearable sensors that continuously monitor heart rate, respiratory rate, and oxygen saturation feeds these algorithms with high-resolution data that far exceeds the temporal granularity of intermittent nursing assessments, further enhancing predictive power[35].
The disparity in access to safe surgical cancer care between high-income countries and low- and middle-income countries represents a global health crisis of staggering proportions. A shortage of trained anesthesiologists in low- and middle-income countries (LMICs) contributes significantly to perioperative mortality and limits the capacity for curative cancer surgery[37]. AI offers innovative solutions to bridge this gap and extend the reach of specialty expertise.
AI-driven tele-anesthesia platforms allow specialist anesthesiologists in central hubs to support non-physician anesthesia providers in remote or resource-limited settings, effectively multiplying the impact of scarce expertise. These systems transmit vital signs in real-time while AI algorithms monitor the data for anomalies including disconnection, hypotension, and hypoxia, triggering alerts that enable timely intervention[3]. This “virtual supervisor” empowers local providers to manage more complex oncologic cases safely, expanding surgical capacity in regions where cancer care has historically been limited or unavailable[37]. AI integration with telemedicine platforms for rural healthcare delivery and resource-limited setting adaptations represents an emerging frontier with enormous potential impact[38].
While high-end AI systems require expensive infrastructure that may be unavailable in resource-limited settings, “frugal AI” applications are emerging that bring sophisticated capabilities to low-resource environments. Smartphone-based apps that use the microphone to analyze pulse oximeter tones or the camera to assess anemia through conjunctival pallor bring diagnostic capabilities to the bedside without the need for expensive hardware[38]. These tools are particularly relevant for preoperative screening in rural cancer camps where formal laboratory infrastructure is unavailable. Opportunistic use of existing imaging data, such as body composition measures from routine CT scans including skeletal muscle area and visceral or subcutaneous adipose tissue at the L3 level, can provide meaningful risk stratification without additional testing or expense.
The integration of AI into onco-anaesthesia, despite its promise, faces significant hurdles that must be addressed for successful clinical implementation.
While current AI models demonstrate impressive metrics (e.g., AUC > 0.90), substantial barriers to real-world deployment remain. First, most models are trained on retrospective data from single academic centers, limiting their generalizability to diverse populations. Second, the “human-in-the-loop” concept is often cited but rarely operationalized; we argue that for liability purposes, AI must function strictly as a decision-support tool, with the ultimate responsibility residing with the clinician[39].
From a regulatory standpoint, we align with the FDA’s “software as a medical device” framework, which advocates for continuous post-market surveillance of adaptive algorithms to ensure they do not “drift” or degrade over time[40]. Furthermore, to mitigate algorithmic bias, future models must be trained on datasets that explicitly include underrepresented populations, such as those from LMICs, to prevent the widening of global health disparities.
AI models are fundamentally data-dependent, and medical data is often fragmented, noisy, and unstructured in ways that complicate algorithmic learning. In oncology, the heterogeneity of tumor types and treatment protocols makes it particularly difficult to build “one-size-fits-all” models[3]. A model trained on colorectal cancer patients may not generalize to thoracic surgery populations with different physiological demands and complication profiles. Ensuring data quality and interoperability between different EHR systems is a prerequisite for scalable AI implementation. Key challenges include limited external validation of deep learning programs, data source heterogeneity across institutions and healthcare systems, and the need for standardized outcome definitions that enable meaningful comparison across studies[36,41].
Deep learning models are often described as “black boxes” because their decision-making process is opaque even to their developers. In medicine, trust is paramount, and clinicians are understandably reluctant to act on an AI prediction such as “high risk of mortality” if they cannot understand why the prediction was made or verify its plausibility against their clinical judgment. Explainable AI techniques such as SHAP values are essential to decode these models, identifying the specific features driving the output[39]. This transparency is also crucial for detecting algorithmic errors and bias that might otherwise go unnoticed[6].
A survey of 293 anesthesiologists revealed that while 69.6% expressed positive attitudes toward AI, liability concerns were paramount, with 87.3% citing “ambiguity of responsibility in cases of complications” as a major barrier, a figure significantly higher than the 47% reported in United States surveys[8]. Strong endorsement existed for non-invasive applications including preoperative assessments at 93.1% and medical education at 91.2%, while skepticism persisted for invasive procedures where the stakes of algorithmic error are highest[8].
AI models can perpetuate or even exacerbate existing healthcare disparities if they are trained on biased datasets that do not represent the full diversity of patient populations. If a dataset underrepresents certain racial or socioeconomic groups, the resulting model will be systematically less accurate for those populations[8]. Sources of bias in oncologic data include selection bias in who receives certain treatments, measurement bias in how outcomes are recorded, and class imbalance problems where rare but important outcomes are underrepresented. In onco-anaesthesia, where minority populations already face documented disparities in surgical outcomes, biased algorithms could lead to unequal risk stratification and resource allocation that compounds existing inequities. Guidelines for mitigating bias include diverse dataset curation, algorithmic auditing for differential performance across demographic groups, and strategies for inclusive and transferable AI development[42]. Rigorous auditing of datasets and models for bias is an ethical imperative that must be built into the development process from the outset.
The regulation of AI in medicine is in its infancy, with frameworks struggling to keep pace with rapidly evolving technology. Regulatory bodies like the FDA with its 510(k) pathway and the European Union with its medical device regulation are developing frameworks to certify “software as a medical device”, but substantial challenges remain regarding “adaptive” algorithms that learn and change post-deployment[40]. Post-market surveillance requirements for adaptive AI, transparency and accountability standards, and algorithmic bias detection in devices all require adaptive policy frameworks that have yet to be fully developed. Liability questions regarding who is responsible when an AI makes a mistake also need clear legal answers to encourage adoption without exposing clinicians to undue risk[39]. Patient privacy and information security concerns add additional layers of complexity that must be navigated carefully[36].
The future of AI in onco-anaesthesia lies in the convergence of multimodal data into a “digital twin” of the patient. This virtual model would integrate genomics, radiomics, physiological data, and EHR history to simulate the patient’s likely response to surgery and anesthesia before any intervention occurs[22]. The integration framework would enable specific use cases including simulation of anesthesia plans, optimization of drug choices and hemodynamic targets, and prediction of individual responses to various anesthetic techniques. Clinicians could effectively “rehearse” the anesthesia plan on the digital twin, optimizing every element of care to minimize risk for that specific patient.
Furthermore, the integration of Large Language Models and Generative AI will streamline administrative burdens, automate documentation, and provide conversational decision support at the point of care[43]. As shown in Table 1, natural language processing algorithms that process and understand human language are already being applied to EHR phenotyping, extracting comorbidities and frailty markers from unstructured clinical notes, and these capabilities will expand substantially with advances in large language models. LLM applications in oncology span clinician decision support, diagnosis and screening, treatment and management recommendations, patient-facing applications, and research facilitation. However, limitations including hallucinations where models generate plausible but incorrect information, generalization challenges across different clinical contexts, and ethical considerations must be addressed before widespread deployment[43]. As these technologies mature, onco-anaesthesia will evolve from an art based on intuition and experience to a precision science grounded in data and individualized prediction.
AI is fundamentally reshaping the landscape of onco-anaesthesia in ways that promise to benefit patients at every stage of the perioperative journey. By augmenting human intelligence with computational power, AI enables a level of precision in risk assessment, hemodynamic management, and outcome prediction that was previously unattainable through clinical judgment alone. As demonstrated throughout this review and summarized in Tables 1 and 2, diverse AI methodologies-from ensemble learning approaches like RF and XGBoost to deep learning architectures including convolutional neural networks and long short-term memorys-are achieving clinically meaningful performance across applications ranging from frailty assessment and complication prediction to hemodynamic forecasting and postoperative delirium prevention. The potential to modulate the biological impact of surgery on cancer progression through AI-optimized anesthesia represents a frontier that could improve survival for millions of patients worldwide.
| Outcome domain | Specific target | AI methodology | Key findings/performance | Clinical implication |
| Hemodynamics | Intraoperative hypotension | Gradient boosting; CNN-RNN | AUROC 0.94; Accuracy 0.88; 6 endotypes[21,22] | Preemptive endotype-specific treatment |
| Hemodynamics | Anastomotic leakage (esophagectomy) | Threshold analysis | MAP < 65 mmHg: OR 1.02 per 10-minute[20] | Avoid prolonged hypotension |
| Complications | Anastomotic leakage (colorectal) | ANN | Superior sensitivity vs surgeons[13] | Early intervention/diversion |
| Complications | CD ≥ 3 complications | XGBoost with SHAP | AUROC 0.77 (train), 0.73 (test)[6] | Transparent risk stratification |
| Neurological | Postoperative delirium | XGBoost (AI-delirium guard) | AUC 0.85[31] | Targeted prevention; web tool1 |
| Pain | Chronic postsurgical pain | Stacking ensemble (EHR + Wearable) | AUC 0.90; 89% accuracy[35] | Precision analgesia intervention |
| Prognosis | Recurrence (NSCLC) | Cox regression | Stage I: P = 0.036; stage II-III: NS[25] | Stage-specific counseling |
| Prognosis | Metastasis (cT1 lung) | Random forest (MoLPre) | AUC 0.92 (internal validation), 0.90 (external validation)[17] | Online prediction tool2 |
| Prognosis | Early recurrence (NSCLC) | Stacking ensemble | AUC 0.81 (test), 0.80 (external validation)[16] | Adjuvant therapy decisions |
| Liver surgery | Post-Hepatectomy failure | Multimodal ML (radiomics + clinical) | Outperformed MELD/Child-Pugh[14] | Resection vs transplant selection |
However, realizing this potential requires a disciplined approach that acknowledges both the promise and the limitations of these technologies. We must address the challenges of data quality, demand explainability from our algorithms, and rigorously ensure equity in their application. The conflicting evidence regarding anesthetic effects on cancer outcomes-with stage-dependent opioid effects and concentration-dependent volatile anesthetic effects-highlights the need for large, prospective, multicenter trials guided by AI-generated hypotheses. As we move forward, the role of the onco-anaesthetist will evolve from being a solitary pilot navigating the operating room to the architect of a complex, data-driven perioperative system, with AI serving as the indispensable engine of precision care.
| 1. | Teng Y, Yin Y, Shi Y, Zhao J, Sun M, Zhao X. The impact of perioperative anesthesia management-induced immunosuppression on postoperative cancer recurrence and metastasis: a narrative review. Front Oncol. 2025;15:1558652. [RCA] [PubMed] [DOI] [Full Text] [Full Text (PDF)] [Cited by in Crossref: 2] [Cited by in RCA: 7] [Article Influence: 7.0] [Reference Citation Analysis (0)] |
| 2. | Choi H, Hwang W. Anesthetic Approaches and Their Impact on Cancer Recurrence and Metastasis: A Comprehensive Review. Cancers (Basel). 2024;16:4269. [RCA] [PubMed] [DOI] [Full Text] [Cited by in RCA: 17] [Reference Citation Analysis (0)] |
| 3. | Cao Y, Wang Y, Liu H, Wu L. Artificial intelligence revolutionizing anesthesia management: advances and prospects in intelligent anesthesia technology. Front Med (Lausanne). 2025;12:1571725. [RCA] [PubMed] [DOI] [Full Text] [Full Text (PDF)] [Cited by in Crossref: 3] [Cited by in RCA: 7] [Article Influence: 7.0] [Reference Citation Analysis (0)] |
| 4. | Huhulea EN, Huang L, Eng S, Sumawi B, Huang A, Aifuwa E, Hirani R, Tiwari RK, Etienne M. Artificial Intelligence Advancements in Oncology: A Review of Current Trends and Future Directions. Biomedicines. 2025;13:951. [RCA] [PubMed] [DOI] [Full Text] [Full Text (PDF)] [Cited by in RCA: 22] [Reference Citation Analysis (0)] |
| 5. | Ou Y, Hu X, Luo C, Li Y. Global trends in artificial intelligence research in anesthesia from 2000 to 2023: a bibliometric analysis. Perioper Med (Lond). 2025;14:47. [RCA] [PubMed] [DOI] [Full Text] [Cited by in RCA: 3] [Reference Citation Analysis (0)] |
| 6. | Hernandez MC, Chen C, Nguyen A, Choong K, Carlin C, Nelson RA, Rossi LA, Seth N, McNeese K, Yuh B, Eftekhari Z, Lai LL. Explainable Machine Learning Model to Preoperatively Predict Postoperative Complications in Inpatients With Cancer Undergoing Major Operations. JCO Clin Cancer Inform. 2024;8:e2300247. [RCA] [PubMed] [DOI] [Full Text] [Full Text (PDF)] [Cited by in Crossref: 1] [Cited by in RCA: 7] [Article Influence: 3.5] [Reference Citation Analysis (0)] |
| 7. | Negahi A, Khosravi-Mashizi M, Najdsepas H, Negahban H, Mousavi-Beni SA, Shahrokhi Damavand R, Naseri A, Jayervand F, Shahbazi A, Rahmani A, Neamatzadeh H. Precision Medicine and Artificial Intelligence in Next-Generation Cancer Surgery: A Comprehensive Analysis of Clinical Applications, Therapeutic Outcomes, and Implementation Strategies. Asian Pac J Cancer Prev. 2025;26:4299-4312. [RCA] [PubMed] [DOI] [Full Text] [Cited by in RCA: 1] [Reference Citation Analysis (0)] |
| 8. | Erel S, Kılıç AG. The use of artificial intelligence in anesthesiology: Attitudes and ethical concerns of anesthesiologists. Saudi J Anaesth. 2025;19:498-504. [RCA] [PubMed] [DOI] [Full Text] [Full Text (PDF)] [Cited by in Crossref: 3] [Cited by in RCA: 1] [Article Influence: 1.0] [Reference Citation Analysis (0)] |
| 9. | Yang H, Chang J, He W, Wee CF, Yit JST, Feng M. Frailty Modeling Using Machine Learning Methodologies: A Systematic Review With Discussions on Outstanding Questions. IEEE J Biomed Health Inform. 2025;29:631-642. [RCA] [PubMed] [DOI] [Full Text] [Cited by in Crossref: 1] [Cited by in RCA: 5] [Article Influence: 5.0] [Reference Citation Analysis (0)] |
| 10. | Yin C, Yin M, Zhen J, Xue C, Zhu Z. Development of a Risk Prediction Model for Preoperative Pre-Frailty in Elderly Patients with Gastric Cancer. Clin Interv Aging. 2025;20:2763-2772. [RCA] [PubMed] [DOI] [Full Text] [Full Text (PDF)] [Cited by in RCA: 1] [Reference Citation Analysis (0)] |
| 11. | Lee SW, Lee EH, Choi IC. An ensemble machine learning approach to predict postoperative mortality in older patients undergoing emergency surgery. BMC Geriatr. 2023;23:262. [RCA] [PubMed] [DOI] [Full Text] [Cited by in RCA: 9] [Reference Citation Analysis (4)] |
| 12. | Frezza B, Nurchis MC, Capolupo GT, Carannante F, De Prizio M, Rondelli F, Alunni Fegatelli D, Gili A, Lepre L, Costa G. A Comparison of Machine Learning-Based Models and a Simple Clinical Bedside Tool to Predict Morbidity and Mortality After Gastrointestinal Cancer Surgery in the Elderly. Bioengineering (Basel). 2025;12:544. [RCA] [PubMed] [DOI] [Full Text] [Full Text (PDF)] [Cited by in RCA: 1] [Reference Citation Analysis (0)] |
| 13. | Alshammari A, Boabbas A, Nassar B, Shaikhah A. The Role of Artificial Intelligence in General Surgery: A Systematic Review and Meta-Analysis of Machine Learning Applications in Colorectal Cancer Treatment Outcomes. Cureus. 2025;17:e96919. [RCA] [PubMed] [DOI] [Full Text] [Full Text (PDF)] [Cited by in RCA: 1] [Reference Citation Analysis (0)] |
| 14. | Zeng S, Li L, Hu Y, Luo L, Fang Y. Machine learning approaches for the prediction of postoperative complication risk in liver resection patients. BMC Med Inform Decis Mak. 2021;21:371. [RCA] [PubMed] [DOI] [Full Text] [Full Text (PDF)] [Cited by in Crossref: 3] [Cited by in RCA: 20] [Article Influence: 4.0] [Reference Citation Analysis (0)] |
| 15. | Durand X, Hédou J, Bellan G, Thomas PA, Pages PB, D'Journo XB, Brouchet L, Rivera C, Falcoz PE, Gillibert A, Baste JM. Predicthor: AI-Powered Predictive Risk Model for 30-Day Mortality and 30-Day Complications in Patients Undergoing Thoracic Surgery for Lung Cancer. Ann Surg Open. 2025;6:e578. [RCA] [PubMed] [DOI] [Full Text] [Full Text (PDF)] [Cited by in Crossref: 1] [Cited by in RCA: 3] [Article Influence: 3.0] [Reference Citation Analysis (0)] |
| 16. | Lin GY, Chen RN, Wu S, Gao ZS, Guo XQ, Zheng XH, Chen SZ. Development and validation of a machine learning model to predict early recurrence after surgery in NSCLC patients. Sci Rep. 2025;15:41952. [RCA] [PubMed] [DOI] [Full Text] [Full Text (PDF)] [Cited by in RCA: 1] [Reference Citation Analysis (0)] |
| 17. | Lan J, Wang H, Huang J, Li W, Ao M, Zhang W, Mu J, Yang L, Ran L. MoLPre: A Machine Learning Model to Predict Metastasis of cT1 Solid Lung Cancer. Clin Transl Sci. 2025;18:e70186. [RCA] [PubMed] [DOI] [Full Text] [Full Text (PDF)] [Cited by in RCA: 2] [Reference Citation Analysis (0)] |
| 18. | Zeng S, Qing Q, Xu W, Yu S, Zheng M, Tan H, Peng J, Huang J. Personalized anesthesia and precision medicine: a comprehensive review of genetic factors, artificial intelligence, and patient-specific factors. Front Med (Lausanne). 2024;11:1365524. [RCA] [PubMed] [DOI] [Full Text] [Cited by in RCA: 17] [Reference Citation Analysis (0)] |
| 19. | Zhou Z, Liu H, Wang X, Sang X, Zhang Y, Liu Y, Zhang X. Effect of different intraoperative blood pressure regulation levels on postoperative myocardial injury in patients undergoing radical mastectomy for breast cancer after receiving neoadjuvant chemotherapy: a study protocol for a randomised controlled trial. BMJ Open. 2025;15:e088559. [RCA] [PubMed] [DOI] [Full Text] [Cited by in RCA: 1] [Reference Citation Analysis (0)] |
| 20. | Juri T, Suehiro K, Yasuhara H, Takai M, Yasuda S, Kimura A, Takahashi K, Fujimoto Y, Mori T. Association between intraoperative hypotension and adverse clinical outcomes after esophagectomy for esophageal cancer: retrospective observational study. JA Clin Rep. 2025;11:63. [RCA] [PubMed] [DOI] [Full Text] [Full Text (PDF)] [Cited by in RCA: 1] [Reference Citation Analysis (0)] |
| 21. | Davoud SC, Kovacheva VP. On the Horizon: Specific Applications of Automation and Artificial Intelligence in Anesthesiology. Curr Anesthesiol Rep. 2023;13:31-40. [RCA] [PubMed] [DOI] [Full Text] [Cited by in RCA: 6] [Reference Citation Analysis (0)] |
| 22. | Shim JG, Yoon W, Lee SJ, Chang SH, Jung SR, Chung JY. Machine Learning Methods for the Prediction of Intraoperative Hypotension with Biosignal Waveforms. Medicina (Kaunas). 2025;61:2039. [RCA] [PubMed] [DOI] [Full Text] [Full Text (PDF)] [Cited by in RCA: 1] [Reference Citation Analysis (0)] |
| 23. | Abdel Deen OMT, Fan SZ, Shieh JS. A Multimodal Deep Learning Approach to Intraoperative Nociception Monitoring: Integrating Electroencephalogram, Photoplethysmography, and Electrocardiogram. Sensors (Basel). 2025;25:1150. [RCA] [PubMed] [DOI] [Full Text] [Full Text (PDF)] [Cited by in RCA: 8] [Reference Citation Analysis (6)] |
| 24. | Jones A, Tang R, Dabo-Trubelja A, Yeoh CB, Richards L, Gottumukkala V. Optimising artificial intelligence ultrasound tools in anaesthesiology and perioperative medicine: The next frontier for advanced technology application. Indian J Anaesth. 2024;68:1016-1021. [RCA] [PubMed] [DOI] [Full Text] [Full Text (PDF)] [Cited by in RCA: 2] [Reference Citation Analysis (0)] |
| 25. | Cata JP, Keerty V, Keerty D, Feng L, Norman PH, Gottumukkala V, Mehran JR, Engle M. A retrospective analysis of the effect of intraoperative opioid dose on cancer recurrence after non-small cell lung cancer resection. Cancer Med. 2014;3:900-908. [RCA] [PubMed] [DOI] [Full Text] [Full Text (PDF)] [Cited by in Crossref: 83] [Cited by in RCA: 106] [Article Influence: 8.8] [Reference Citation Analysis (0)] |
| 26. | Thomas TE, Bowers K, Gomez D, Morgan O, Borowsky PA, Dutta R, Abu Y, Roy S, Rojas KE. The association between perioperative opioids and breast cancer recurrence: a narrative review of the literature. Transl Breast Cancer Res. 2023;4:12. [RCA] [PubMed] [DOI] [Full Text] [Full Text (PDF)] [Cited by in RCA: 5] [Reference Citation Analysis (0)] |
| 27. | Konstantis G, Katsadouros I, Tsaousi G, Grosomanidis V, Pourzitaki C. Immunomodulatory Effects of Anesthetic Techniques in Lung Cancer Surgery: A Systematic Review and Meta-Analysis. Medicina (Kaunas). 2025;61:1263. [RCA] [PubMed] [DOI] [Full Text] [Full Text (PDF)] [Cited by in Crossref: 2] [Cited by in RCA: 4] [Article Influence: 4.0] [Reference Citation Analysis (0)] |
| 28. | Hoang DT, Dinstag G, Shulman ED, Hermida LC, Ben-Zvi DS, Elis E, Caley K, Sammut SJ, Sinha S, Sinha N, Dampier CH, Stossel C, Patil T, Rajan A, Lassoued W, Strauss J, Bailey S, Allen C, Redman J, Beker T, Jiang P, Golan T, Wilkinson S, Sowalsky AG, Pine SR, Caldas C, Gulley JL, Aldape K, Aharonov R, Stone EA, Ruppin E. A deep-learning framework to predict cancer treatment response from histopathology images through imputed transcriptomics. Nat Cancer. 2024;5:1305-1317. [RCA] [PubMed] [DOI] [Full Text] [Full Text (PDF)] [Cited by in Crossref: 46] [Cited by in RCA: 82] [Article Influence: 41.0] [Reference Citation Analysis (0)] |
| 29. | Arun O, Arun F. Postoperative delirium: A tragedy for elderly cancer patients. World J Gastrointest Oncol. 2024;16:3765-3770. [RCA] [PubMed] [DOI] [Full Text] [Full Text (PDF)] [Cited by in RCA: 3] [Reference Citation Analysis (0)] |
| 30. | Li H, Zang Q, Li Q, Lin Y, Duan J, Huang J, Hu H, Zhang Y, Xia D, Zhou M. Development of a Machine Learning-Based Predictive Model for Postoperative Delirium in Older Adult Intensive Care Unit Patients: Retrospective Study. J Med Internet Res. 2025;27:e67258. [RCA] [PubMed] [DOI] [Full Text] [Full Text (PDF)] [Cited by in Crossref: 4] [Cited by in RCA: 7] [Article Influence: 7.0] [Reference Citation Analysis (0)] |
| 31. | Boppana SH, Tyagi D, Komati S, Boppana SL, Raj R, Mintz CD. AI-delirium guard: Predictive modeling of postoperative delirium in elderly surgical patients. PLoS One. 2025;20:e0322032. [RCA] [PubMed] [DOI] [Full Text] [Full Text (PDF)] [Cited by in Crossref: 3] [Cited by in RCA: 4] [Article Influence: 4.0] [Reference Citation Analysis (0)] |
| 32. | Ha WS, Choi BK, Yeom J, Song S, Cho S, Chu MK, Kim WJ, Heo K, Kim KM. Development and Validation of Machine Learning Models to Predict Postoperative Delirium Using Clinical Features and Polysomnography Variables. J Clin Med. 2024;13:5485. [RCA] [PubMed] [DOI] [Full Text] [Cited by in RCA: 3] [Reference Citation Analysis (0)] |
| 33. | Sun C, Li M, Lan L, Pei L, Zhang Y, Tan G, Zhang Z, Huang Y. Prediction models for chronic postsurgical pain in patients with breast cancer based on machine learning approaches. Front Oncol. 2023;13:1096468. [RCA] [PubMed] [DOI] [Full Text] [Cited by in RCA: 12] [Reference Citation Analysis (0)] |
| 34. | Ghane G, Karimi R, Chekeni AM, Darvishi M, Imani R, Vafaeinezhad FZ. Pain Management in Cancer Patients With Artificial Intelligence: Narrative Review. Scientifica (Cairo). 2025;2025:6888213. [RCA] [PubMed] [DOI] [Full Text] [Cited by in RCA: 3] [Reference Citation Analysis (0)] |
| 35. | Soley N, Speed TJ, Xie A, Taylor CO. Predicting Postoperative Pain and Opioid Use with Machine Learning Applied to Longitudinal Electronic Health Record and Wearable Data. Appl Clin Inform. 2024;15:569-582. [RCA] [PubMed] [DOI] [Full Text] [Full Text (PDF)] [Cited by in Crossref: 5] [Cited by in RCA: 5] [Article Influence: 2.5] [Reference Citation Analysis (0)] |
| 36. | Giri R, Firdhos SH, Vida TA. Artificial Intelligence in Anesthesia: Enhancing Precision, Safety, and Global Access Through Data-Driven Systems. J Clin Med. 2025;14:6900. [RCA] [PubMed] [DOI] [Full Text] [Full Text (PDF)] [Cited by in RCA: 8] [Reference Citation Analysis (0)] |
| 37. | Nkenguye W. Artificial intelligence in surgical care for low- and middle-income countries: Challenges, opportunities, and the path forward. Surg Pract Sci. 2025;22:100290. [RCA] [PubMed] [DOI] [Full Text] [Full Text (PDF)] [Cited by in RCA: 9] [Reference Citation Analysis (0)] |
| 38. | Obst M, Arensmeyer J, Bonsmann H, Kolbinger A, Kigenyi J, Oneka F, Owere B, Schmidt J, Feodorovici P, Wynands J. AI-Enhanced 3D Models in Global Virtual Reality Case Conferences for Surgical Care in a Low-Income Country: Exploratory Study. JMIR Form Res. 2025;9:e69300. [RCA] [PubMed] [DOI] [Full Text] [Full Text (PDF)] [Cited by in RCA: 4] [Reference Citation Analysis (0)] |
| 39. | Fangerau H. Artifical intelligence in surgery: ethical considerations in the light of social trends in the perception of health and medicine. EFORT Open Rev. 2024;9:323-328. [RCA] [PubMed] [DOI] [Full Text] [Cited by in RCA: 6] [Reference Citation Analysis (0)] |
| 40. | Santra S, Kukreja P, Saxena K, Gandhi S, Singh OV. Navigating regulatory and policy challenges for AI enabled combination devices. Front Med Technol. 2024;6:1473350. [RCA] [PubMed] [DOI] [Full Text] [Full Text (PDF)] [Cited by in RCA: 18] [Reference Citation Analysis (1)] |
| 41. | Adams MCB, Nelson AM, Narouze S. Daring discourse: artificial intelligence in pain medicine, opportunities and challenges. Reg Anesth Pain Med. 2023;48:439-442. [RCA] [PubMed] [DOI] [Full Text] [Cited by in Crossref: 5] [Cited by in RCA: 15] [Article Influence: 5.0] [Reference Citation Analysis (0)] |
| 42. | Tasci E, Zhuge Y, Camphausen K, Krauze AV. Bias and Class Imbalance in Oncologic Data-Towards Inclusive and Transferrable AI in Large Scale Oncology Data Sets. Cancers (Basel). 2022;14:2897. [RCA] [PubMed] [DOI] [Full Text] [Full Text (PDF)] [Cited by in Crossref: 86] [Cited by in RCA: 53] [Article Influence: 13.3] [Reference Citation Analysis (0)] |
| 43. | Chen D, Parsa R, Swanson K, Nunez JJ, Critch A, Bitterman DS, Liu FF, Raman S. Large language models in oncology: a review. BMJ Oncol. 2025;4:e000759. [RCA] [PubMed] [DOI] [Full Text] [Full Text (PDF)] [Cited by in RCA: 24] [Reference Citation Analysis (1)] |