Development of a machine learning-based nomogram for predicting chronic postsurgical pain after uniportal video-assisted thoracoscopic surgery

Introduction

Lung cancer has emerged as one of the most prevalent malignancies globally and in China, necessitating video-assisted thoracic surgery (VATS) for an increasing number of patients (1,2). In recent years, uniportal video-assisted thoracoscopic surgery (UVATS) has gained prominence due to its advantages over conventional multi-port VATS (MP-VATS), including reduced surgical trauma, lower hospitalization costs, and decreased postoperative pain (3-5). However, chronic postsurgical pain (CPSP) remains a common complication following UVATS, with the International Association for the Study of Pain (IASP) defining it as persistent or intermittent pain lasting beyond the normal tissue healing period, typically over 3 months (6). Notably, the reported incidence of CPSP after UVATS reaches 25.3%, significantly impairing patients’ quality of life (7,8).

With the growing emphasis on enhanced recovery after surgery (ERAS) protocols, CPSP prevention has become a critical focus. Early identification of high-risk patients is thus clinically imperative (9,10). Current research predominantly compares complications between MP-VATS and UVATS or examines MP-VATS outcomes, leaving a gap in UVATS-specific CPSP prediction models.

Machine learning, a sophisticated subset of artificial intelligence, possesses unique capabilities for analyzing high-dimensional clinical datasets, deciphering complex nonlinear relationships, and identifying optimal predictive factors for surgical outcomes. Its application in predicting both acute and CPSP has gained significant traction across various surgical disciplines (11-13). In the present study, we employ cutting-edge machine learning algorithms to systematically identify the most clinically significant predictors of CPSP from comprehensive perioperative data, subsequently developing and validating a robust predictive model. This approach not only establishes an evidence-based framework for risk stratification but also provides actionable insights for implementing targeted preventive measures and personalized pain management protocols in clinical practice. The following sections detail our methodology and present the key findings of this investigation. We present this article in accordance with the TRIPOD reporting checklist (available at https://jtd.amegroups.com/article/view/10.21037/jtd-2025-1215/rc).

Methods

Patients

This was a retrospective study involving patients who underwent UVATS at Nanjing Brain Hospital’s Department of Thoracic Surgery between January 2023 and January 2024. Additionally, patients between 1 June 2024 and 30 September 2024 were included in the external validation set (Figure 1). Inclusion criteria: (I) aged ≥18 years; (II) underwent UVATS lobectomy or sublobar resection. Exclusion criteria: (I) pre-existing chronic pain conditions; (II) history of re-do thoracic procedures; (III) chronic use of psychotropic or analgesic medications preoperatively; (IV) incomplete records. The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The study was approved by the Institutional Review Board of Nanjing Brain Hospital (No. 2023-KY077-01) and individual consent for this retrospective analysis was waived. CPSP was defined according to current diagnostic standards as surgically-related pain persisting ≥3 months postoperatively, localized to the surgical site or corresponding dermatomes, excluding pain from infection, tumor recurrence, or pre-existing conditions (14). All patients are scheduled for an outpatient follow-up 3-month after discharge. Pain was systematically assessed using the numerical rating scale (NRS), with detailed evaluation of pain characteristics, triggers, and prior history to confirm CPSP diagnosis according to these criteria.

Figure 1 Flowchart of this study. LASSO, least absolute shrinkage and selection operator; RCS, restricted cubic spline; RF, random forest; XGBoost, extreme gradient boosting.

Feature selection

In developing our prediction model, we reviewed literature, consulted experts, and leveraged clinical experience to select potential predictors. Ultimately yielding 22 clinically-relevant variables for evaluation. The selected feature set comprehensively captured: (I) demographic and comorbidity profiles, including age, sex, body mass index (BMI), hypertension, diabetes mellitus, chronic obstructive pulmonary disease (COPD) status, smoking history, alcohol use patterns, and educational attainment; (II) treatment-specific parameters, encompassing tumor invasiveness, utilization of preoperative CT-guided hookwire localization, serial C-reactive protein (CRP) measurements (preoperative and postoperative day 1), selected intercostal approach, operative duration, intraoperative blood loss, drainage tube configuration, duration of thoracic drainage, and postoperative analgesic strategies (both intravenous and oral regimens); and (III) pain trajectory documentation through standardized NRS assessments at postoperative day 1 and discharge.

Currently, there is no standard for drain tube removal. Our department’s policy requires: (I) daily output of <200 mL; (II) computed tomography (CT) or X-ray confirmation of no major residual effusion; and (III) lack of aberrant drainage. CRP levels were tested on the first postoperative morning using rate nephelometry, and the results were reported in mg/L.

For patients undergoing wedge resection, preoperative tumor localization was achieved through CT-guided hookwire needle placement. Postoperative pain management consisted of intravenous patient-controlled analgesia (PCA) delivering butorphanol (4 mg; doubled to 8 mg for patients >70 kg), flurbiprofen axetil (200 mg), and tropisetron (10 mg). Supplemental oral tramadol was administered if patients reported requiring additional analgesia.

Anesthetic and surgical technique

All patients received standardized anesthetic management consisting of: (I) induction with ropivacaine plus dexamethasone acetate, rocuronium for neuromuscular blockade, and maintenance with propofol, remifentanil, and dexmedetomidine; (II) ultrasound-guided serratus anterior and intercostal nerve blocks using ropivacaine-dexamethasone acetate mixture post-induction. All surgical procedures were performed by a chief surgeon with 25 years of experience, through a single 3-cm incision at either the 4th intercostal space (for upper/middle lobe lesions) or the 5th intercostal space (for lower lobe lesions). Drainage strategy was tailored to resection extent: sublobar resections received a single 20F silicone chest tube through the operative incision, while lobectomies had additional 16F pigtail catheter placement at the 7th intercostal space along the posterior axillary line.

Machine learning-based feature selection for binary classification

To identify key clinical variables significantly associated with the binary outcome, this study employed three machine learning-based feature selection methods: random forest (RF), least absolute shrinkage and selection operator (LASSO) regression, and extreme gradient boosting (XGBoost). All models were evaluated using 10-fold cross-validation. The RF model, trained with 500 decision trees, ranked variable importance through mean decrease in Gini impurity, and the top 10 most influential features were retained. For LASSO regression, non-informative variables were compressed to zero coefficients via binomial penalization, with the optimal regularization parameter (λ) selected using the minimum criterion (λ.min) and the one-standard-error rule (λ.1se); variables retaining non-zero coefficients under λ.1se were considered significant predictors. In XGBoost, SHapley Additive exPlanations (SHAP) values were computed to interpret feature contributions, and the top 10 variables ranked by gain were selected. Finally, consensus features identified by all three methods were visualized via a Venn diagram. Predictive performance of each model was assessed using receiver operating characteristic (ROC) curves, with optimal cutoff thresholds determined by Youden’s index.

Nomogram construction

Based on the consensus variables identified by the three machine learning methods, a logistic regression model was constructed to develop a nomogram. This nomogram assigns weighted scores to each variable according to its regression coefficients, serving as a visual tool for individualized risk prediction. Calibration curves were plotted to evaluate the agreement between predicted probabilities and observed outcomes. Calibration curves were supplemented with the Hosmer-Lemeshow test to quantify goodness-of-fit (P>0.05 indicating adequate calibration). Additionally, 1,000 bootstrap resamples were applied to assess the model’s calibration performance, mitigating overfitting and ensuring robustness.

Statistical analysis

Statistical analyses were performed using RStudio (version 4.2.1). Normally distributed continuous variables were expressed as mean ± standard deviation, while non-normally distributed data were presented as median (interquartile range). Categorical variables were summarized as counts and percentages (%). Machine learning analyses were performed using the following R packages: randomForest for RF modeling, glmnet for LASSO regression, and xgboost for gradient boosting. Data visualization was generated using ggplot2. Model performance evaluation was conducted ROC curve analysis. Predictive model validation and nomogram construction were implemented using the rms package, including calibration curve plotting and bootstrap validation. Clinical utility assessment was performed through decision curve analysis (DCA) using the ggDCA package.

Results

Characteristics of the patients

A total of 501 patients were enrolled, meeting the minimum data requirement of 10 times the variables, including 207 males and 294 females, with a mean age of 55.7±11.6 years (Table 1). The cohort was randomly divided into a training set (n=352) and a validation set (n=149) at a 7:3 ratio. Within the training set, 119 patients developed CPSP, while 233 served as controls, yielding a CPSP incidence of 33.8% (Table 2).

Machine learning results

The LASSO model [area under the curve (AUC) =0.922, 95% confidence interval (CI): 0.880–0.964] selected 9 non-zero coefficients at λ.one standard error (1se), the RF model (AUC =0.919, 95% CI: 0.875–0.963) and XGBoost (AUC =0.896, 95% CI: 0.845–0.947) selected the top 10 variables in their importance rankings. Six key variables were consistently identified by all three methods, including age, postoperative NRS score, discharge NRS score, postoperative CRP, operative time and drainage tube duration (Figure 2). The raincloud plots (Figure 3) illustrate the distributions of six variables.

Figure 2 Machine learning results. (A-D) The optimal parameter (λ) selection in the LASSO model employed 10-fold cross-validation using a minimum criteria approach. Nine variables were selected with the one-standard-error rule (λ.1se=0.0311). (E-G) The RF model was trained with 500 decision trees to ensure stability, and 10-fold cross-validation was applied to optimize hyperparameters and prevent overfitting. MDA quantifies how much each variable contributes to prediction accuracy when permuted, and the top 10 most important variables were selected based on it. (H-J) XGBoost selected the top 10 most important features for the prediction. SHAP beeswarm plot demonstrating feature importance and the effects of the predictions. (K) A Venn diagram comparing variables selected by three different methods: RF (green), LASSO (yellow), and XGBoost (red) shows the overlap of selected variables. AUC, area under curve; BMI, body mass index; CI, confidence interval; COPD, chronic obstructive pulmonary disease; CRP, C-reactive protein; LASSO, least absolute shrinkage and selection operator; NRS, numerical rating scale; OOB, out-of-bag; ROC, receiver operating characteristic; RF, random forest; SHAP, SHapley Additive exPlanations; XGBoost, extreme gradient boosting.

Figure 3 Distribution of key predictors between CPSP and non-CPSP groups in the training set. CPSP, chronic postsurgical pain; CRP, c-reactive protein; NRS, numerical rating scale.

Restricted cubic splines (RCS) analysis

To evaluate potential nonlinear effects, we modeled all six consensus variables using RCS with three knots placed at the 10th, 50th, and 90th percentiles. The significance of nonlinearity was assessed via likelihood ratio tests comparing linear and spline models. The RCS analysis identified a statistically significant nonlinear association for operative time, exhibiting a U-shaped relationship with the outcome. The risk nadir occurred at 92 minutes (Figure 4). Other variables maintained linear relationships. We developed a multivariable logistic regression model using the rms package to incorporate this nonlinearity while preserving clinical utility. Operative time was modeled with RCS, while linear terms were retained for other predictors.

Figure 4 RCS analysis. All examined variables demonstrated statistically significant associations with the outcome (P<0.05). Notably, operative time exhibited a significant nonlinear component (χ²=15.19, df=1, P=0.0001), with peak risk observed at 92 minutes (95% CI: 84–110). The remaining variables maintained linear relationships with the clinical outcome. CI, confidence interval; CRP, c-reactive protein; NRS, numerical rating scale; RCS, restricted cubic spline.

Nomogram construction

Based on the Machine Learning and RCS analysis, a nomogram was established to predict CPSP (Figure 5). Each risk factor was assigned a score. The total score was calculated by adding the individual scores and locating the sum on the total-point scale axis. Given the U-shaped relationships identified for Operative Time, the contributions to the nomogram exhibit directional reversals around their respective inflection points. For operation times less than 92 minutes, there is a positive correlation with CPSP; however, when the operation time exceeds 92 minutes, this correlation becomes inverse.

Figure 5 Construction and validation of the nomogram. (A) The nomogram is used by summing all points identified on the scale for each variable. (B) To facilitate clinical application, we developed a dynamic nomogram (https://sylia95.shinyapps.io/dynnomapp-1/). Users can input individual variable scores, and the web-based tool will automatically calculate the patient’s predicted risk of CPSP. NRS, numerical rating scale; CPSP, chronic postsurgical pain; CRP, c-reactive protein.

Nomogram validation

The nomogram demonstrated a robust predictive performance in both the training and validation datasets, which are presented in Figure 6. The AUC for the training set was 0.893 (95% CI: 0.856–0.929) (Figure 6A), and the calibration curves also showed good agreement (Figure 6B,6C). Additionally, DCA for the training set (Figure 6D) demonstrated that a model-based decision provided a more significant net benefit, compared to no treatment or an all-treatment approach, for predicted probability thresholds ranging from 0 to 100%.

Figure 6 Nomogram validation. (A-D) ROC curve, calibration curve, Boostrap curve, and DCA curve for the training set. (E-H) ROC curve, calibration curve, Boostrap curve, and DCA curve for the validation set. (I-L) ROC curve, calibration curve, Boostrap curve, and DCA curve for the external validation set. AUC, area under curve; CI, confidence interval; DCA, decision curve analysis; ROC, receiver operating characteristic.

For the internal validation set, the AUC was 0.861 (95% CI: 0.802–0.920) (Figure 6E). The calibration curves (Figure 6F,6G) and DCA (Figure 6H) similarly confirmed the model’s good performance and clinical utility.

External validation was performed using a dataset of 114 patients between 1 June 2024 and 30 September 2024. The nomogram maintained a good predictive accuracy with an AUC of 0.881 (95% CI: 0.804–0.958) (Figure 6I). The calibration curve also performed well (Figure 6J,6K). Furthermore, the DCA (Figure 6L) indicated that patients achieved a high net benefit across all prediction probability thresholds.

Discussion

CPSP, defined as persistent pain lasting beyond 3 months postoperatively after excluding other causes, is a common complication following surgical procedures. This condition significantly impairs patients’ quality of life, often leading to anxiety, sleep disturbances, and even opioid misuse (15,16). Among patients undergoing VATS, CPSP is particularly prevalent, with reported incidence rates ranging from 20% to 60% (17,18). In recent years, UVATS has been increasingly adopted in multiple centers. Compared with MP-VATS, it shows a lower incidence of CPSP and has been increasingly adopted in clinical practice (8). However, the risk remains clinically significant, highlighting the need to identify risk factors and establish predictive models to improve postoperative outcomes (8). Currently, no validated predictive model exists specifically for CPSP after UVATS. Since 2014, the Department of Thoracic Surgery at Nanjing Brain Hospital Affiliated to Nanjing Medical University has routinely performed UVATS. In this study, we analyzed clinical data and 3-month follow-up outcomes from 501 patients to develop a predictive model for CPSP following UVATS. Our objective is to assist clinicians in the early identification of high-risk patients, enabling timely interventions to enhance postoperative recovery.

The exact etiology of CPSP remains incompletely understood, though current evidence suggests multifactorial involvement, including psychological factors, postoperative analgesic usage, age, sex, surgical approach, and inadequate acute pain control (19,20). Our findings demonstrate an inverse correlation between age and CPSP incidence, consistent with existing literature (21,22). Younger patients exhibit significantly higher susceptibility to developing chronic pain, a phenomenon potentially attributable to several neurobiological and psychosocial mechanisms. First, age-related differences in pain experience may play a crucial role. Older patients often have prior exposure to pain from comorbid conditions, which might confer psychological resilience against postoperative pain-related anxiety and distress. In contrast, younger patients’ limited pain history could exacerbate their vulnerability to maladaptive emotional responses following surgery (23). Neurobiological factors further explain this association. Younger individuals possess greater neural plasticity, potentially amplifying nociceptive signal transmission and central sensitization. This heightened sensitivity may facilitate the transition from acute to chronic pain states. Conversely, age-related neurological degeneration may attenuate pain perception in elderly patients (24).

Our study further demonstrates a strong association between acute postoperative pain and the development of chronic pain, with higher NRS scores on postoperative day 1 and at discharge being positively correlated with CPSP risk. This relationship can be attributed to multiple mechanisms (25,26). First, neuroplasticity plays a crucial role by altering neuronal and synaptic function and morphology, thereby contributing to pain chronification (27). Second, peripheral and central sensitization mechanisms are critically involved. Persistent surgical trauma renders spinal cord neurons hypersensitive to nociceptive signals, a hyperexcitable state driven by enhanced synaptic efficacy and increased neuronal excitability (28). The observed relationship between operative duration and CPSP in our study provides indirect support for these mechanisms. We found that 92 minutes as a critical inflection point. Below this threshold, each additional minute of surgery increased CPSP risk. Strikingly, beyond 92 minutes, prolonged operations exhibited a protective effect against CPSP development. This biphasic pattern may be attributed to stress-induced analgesia—a neurophysiological phenomenon wherein sufficiently intense or prolonged surgical stress triggers activation of the endogenous analgesic system as an adaptive response to potential threats (29). This complex mechanism involves both opioid and non-opioid systems, though its exact pathways remain incompletely understood and warrant further investigation.

The association between prolonged drainage tube duration and CPSP development may be attributed not only to trauma-induced central sensitization as previously discussed, but also to neuropathic pain mechanisms. During tube insertion, fixation, and indwelling periods, the drainage tube may directly compress, traction, or injure intercostal nerves, potentially leading to neuropathic pain. Notably, intercostal nerve damage represents one of the most common etiological factors for postoperative CPSP following thoracic procedures. This dual-pathway mechanism (central sensitization and neuropathic pain) may synergistically contribute to pain chronicization, particularly in cases with extended drainage tube duration (30,31).

In recent years, the association between inflammatory factors and pain has become a research hotspot (32,33). In this study, we included CRP levels on the first postoperative day as a predictive factor, and the results demonstrated a significant positive correlation with CPSP. Current research suggests that inflammatory markers, such as CRP are closely linked to both acute and chronic pain following surgery, with some scholars proposing CRP as a potential blood biomarker for pain assessment (34-36). Postoperative tissue damage typically triggers the release of inflammatory mediators, which can directly or indirectly sensitize nociceptive nerve endings, thereby amplifying pain signaling. Elevated CRP levels may also reflect the degree of systemic inflammation, which could further modulate central pain processing pathways, leading to heightened pain perception (35).

There are limitations in this study. First, the predictive performance could potentially be improved by incorporating additional clinically relevant factors. Second, this study was conducted as a single-center investigation, inherently limited by the demographic characteristics of patients and clinical practice patterns specific to our center. Therefore, in future studies, we could establish larger-scale patient databases incorporating additional predictive factors, enabling comprehensive analysis of diverse postoperative symptoms to better implement ERAS principles.

Conclusions

In summary, our findings identify six key predictors of CPSP following uniportal VATS: younger age, prolonged chest tube duration, elevated postoperative CRP, and higher NRS scores (both postoperative day 1 and at discharge). The developed nomogram enables accurate CPSP risk stratification, allowing early identification of high-risk patients for targeted interventions. Clinical applications include: (I) early chest tube removal in younger patients when clinically appropriate; (II) preventive and multimodal analgesia for patients with elevated CRP or pain scores (37); and (III) aggressive acute pain management to prevent pain chronification. These stratified approaches may significantly improve postoperative quality of life by mitigating CPSP development.

Acknowledgments

None.

Footnote

Reporting Checklist: The authors have completed the TRIPOD reporting checklist. Available at https://jtd.amegroups.com/article/view/10.21037/jtd-2025-1215/rc

Data Sharing Statement: Available at https://jtd.amegroups.com/article/view/10.21037/jtd-2025-1215/dss

Peer Review File: Available at https://jtd.amegroups.com/article/view/10.21037/jtd-2025-1215/prf

Funding: This work was supported by Nanjing Medical Technology Development Fund (No. YKK23137) and Nanjing Medical Technology Development Fund (No. YKK23148).

Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://jtd.amegroups.com/article/view/10.21037/jtd-2025-1215/coif). The authors have no conflicts of interest to declare.

Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The study was approved by the Institutional Review Board of Nanjing Brain Hospital (No. 2023-KY077-01) and individual consent for this retrospective analysis was waived.

Open Access Statement: This is an Open Access article distributed in accordance with the Creative Commons Attribution-NonCommercial-NoDerivs 4.0 International License (CC BY-NC-ND 4.0), which permits the non-commercial replication and distribution of the article with the strict proviso that no changes or edits are made and the original work is properly cited (including links to both the formal publication through the relevant DOI and the license). See: https://creativecommons.org/licenses/by-nc-nd/4.0/.

References

1. Bray F, Laversanne M, Sung H, et al. Global cancer statistics 2022: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries. CA Cancer J Clin 2024;74:229-63. [Crossref] [PubMed]
2. Oudkerk M, Liu S, Heuvelmans MA, et al. Lung cancer LDCT screening and mortality reduction - evidence, pitfalls and future perspectives. Nat Rev Clin Oncol 2021;18:135-51. [Crossref] [PubMed]
3. Mizukami Y, Takahashi Y, Adachi H. Single-Port vs Conventional Three-Port Video-Assisted Thoracoscopic Pulmonary Wedge Resection: Comparison of Postoperative Pain and Surgical Costs. Ann Thorac Cardiovasc Surg 2021;27:91-6. [Crossref] [PubMed]
4. Tosi D, Mazzucco A, Musso V, et al. Pulmonary Lobectomy for Early-Stage Lung Cancer with Uniportal versus Three-Portal Video-Assisted Thoracic Surgery: Results from a Single-Centre Randomized Clinical Trial. J Clin Med 2023;12:7167. [Crossref] [PubMed]
5. Wang L, Ge L, Song S, et al. Clinical applications of minimally invasive uniportal video-assisted thoracic surgery. J Cancer Res Clin Oncol 2023;149:10235-9. [Crossref] [PubMed]
6. Scholz J, Finnerup NB, Attal N, et al. The IASP classification of chronic pain for ICD-11: chronic neuropathic pain. Pain 2019;160:53-9. [Crossref] [PubMed]
7. Azanjac-Arsić A. Neuropathic pain: definition, classification and treatment. Galenika Med J 2024;3:26-30.
8. Jin J, Du X, Min S, et al. Comparison of Chronic Postsurgical Pain Between Single-Port and Multi-Port Video-Assisted Thoracoscopic Pulmonary Resection: A Prospective Study. Thorac Cardiovasc Surg 2022;70:430-8. [Crossref] [PubMed]
9. Molla YD, Alemu HT. The Role of Gabapentin in Enhanced Recovery After Surgery (ERAS) for Patients Undergoing Abdominal Procedures, A Systematic Review and Meta-Analysis. Health Sci Rep 2025;8:e70813. [Crossref] [PubMed]
10. Girgiss C, Oladele M, Alice Vijjeswarapu M. Anesthesia and pain management in ERAS for spine Surgery. Seminars in Spine Surgery 2025;37:101185.
11. Sajdeya R, Narouze S. Harnessing artificial intelligence for predicting and managing postoperative pain: a narrative literature review. Curr Opin Anaesthesiol 2024;37:604-15. [Crossref] [PubMed]
12. Hassan AM, Rajesh A, Asaad M, et al. Artificial Intelligence and Machine Learning in Prediction of Surgical Complications: Current State, Applications, and Implications. Am Surg 2023;89:25-30. [Crossref] [PubMed]
13. Lötsch J, Ultsch A, Mayer B, et al. Artificial intelligence and machine learning in pain research: a data scientometric analysis. Pain Rep 2022;7:e1044. [Crossref] [PubMed]
14. Treede RD, Rief W, Barke A, et al. Chronic pain as a symptom or a disease: the IASP Classification of Chronic Pain for the International Classification of Diseases (ICD-11). Pain 2019;160:19-27. [Crossref] [PubMed]
15. Yoon S, Hong WP, Joo H, et al. Long-term incidence of chronic postsurgical pain after thoracic surgery for lung cancer: a 10-year single-center retrospective study. Reg Anesth Pain Med 2020;45:331-6. [Crossref] [PubMed]
16. Gan TJ. Poorly controlled postoperative pain: prevalence, consequences, and prevention. J Pain Res 2017;10:2287-98. [Crossref] [PubMed]
17. Bayman EO, Parekh KR, Keech J, et al. A Prospective Study of Chronic Pain after Thoracic Surgery. Anesthesiology 2017;126:938-51. [Crossref] [PubMed]
18. Zhang L, Yuan Y, Zhang Y, et al. Construction of prediction model for chronic postsurgical pain after video-assisted thoracoscopic surgery. Basic Clin Med 2023;4:651.
19. Glare P, Aubrey KR, Myles PS. Transition from acute to chronic pain after surgery. Lancet 2019;393:1537-46. [Crossref] [PubMed]
20. Chen D, Yang H, Yang L, et al. Preoperative psychological symptoms and chronic postsurgical pain: analysis of the prospective China Surgery and Anaesthesia Cohort study. Br J Anaesth 2024;132:359-71. [Crossref] [PubMed]
21. Kim JH, Sohn JH, Lee JJ, et al. Age-Related Variations in Postoperative Pain Intensity across 10 Surgical Procedures: A Retrospective Study of Five Hospitals in South Korea. J Clin Med 2023;12:5912. [Crossref] [PubMed]
22. van Dijk JFM, Zaslansky R, van Boekel RLM, et al. Postoperative Pain and Age: A Retrospective Cohort Association Study. Anesthesiology 2021;135:1104-19. [Crossref] [PubMed]
23. Wilson JM, Colebaugh CA, Flowers KM, et al. Early postoperative psychological distress as a mediator of subsequent persistent postsurgical pain outcomes among younger breast cancer patients. Breast Cancer Res Treat 2022;196:363-70. [Crossref] [PubMed]
24. Andreoletti H, Dereu D, Combescure C, et al. A systematic review and meta-analysis of three risk factors for chronic postsurgical pain: age, sex and preoperative pain. Minerva Anestesiol 2022;88:827-41. [Crossref] [PubMed]
25. Clephas PRD, Hoeks SE, Singh PM, et al. Prognostic factors for chronic post-surgical pain after lung and pleural surgery: a systematic review with meta-analysis, meta-regression and trial sequential analysis. Anaesthesia 2023;78:1005-19. [Crossref] [PubMed]
26. Gül ŞK, Tepetam H, Gül HL. Duloxetine and pregabalin in neuropathic pain of lung cancer patients. Brain Behav 2020;10:e01527. [Crossref] [PubMed]
27. Song Q, E S, Zhang Z, et al. Neuroplasticity in the transition from acute to chronic pain. Neurotherapeutics 2024;21:e00464. [Crossref] [PubMed]
28. Tassou A, Richebe P, Rivat C. Mechanisms of chronic postsurgical pain. Reg Anesth Pain Med 2025;50:77-85. [Crossref] [PubMed]
29. Butler RK, Finn DP. Stress-induced analgesia. Prog Neurobiol 2009;88:184-202. [Crossref] [PubMed]
30. MacDougall P. Chronic post-thoracotomy pain. In: Principles and practice of anesthesia for thoracic surgery. Springer; 2019:1045-1057.
31. Lee JI, Kim GW, Park KY. Intercostal bundle-splitting thoracotomy reduces chronic post-thoracotomy pain. Thorac Cardiovasc Surg 2007;55:401-2. [Crossref] [PubMed]
32. Zheng Y, Wang H, Zhang H, et al. Associations Between Circulating Inflammatory Cytokines and Neuropathic Pain: A Two-Sample Mendelian Randomization Study. J Pain Res 2025;18:1525-44. [Crossref] [PubMed]
33. Christensen JO, Nilsen KB, Hopstock LA, et al. Shift work, low-grade inflammation, and chronic pain: a 7-year prospective study. Int Arch Occup Environ Health 2021;94:1013-22. [Crossref] [PubMed]
34. Dainese P, Mahieu H, De Mits S, et al. Associations between markers of inflammation and altered pain perception mechanisms in people with knee osteoarthritis: a systematic review. RMD Open 2023;9:e002945. [Crossref] [PubMed]
35. Geißler K, Brock R, Meißner W, et al. Association between preoperative serum C-reactive protein level and leukocyte count and postoperative pain after otolaryngological surgery. Eur Arch Otorhinolaryngol 2023;280:3877-84. [Crossref] [PubMed]
36. Tonelli Enrico V, Schneider M, Haas M, et al. The association of biomarkers with pain and function in acute and subacute low back pain: a secondary analysis of an RCT. BMC Musculoskelet Disord 2022;23:1059. [Crossref] [PubMed]
37. Chen YK, Boden KA, Schreiber KL. The role of regional anaesthesia and multimodal analgesia in the prevention of chronic postoperative pain: a narrative review. Anaesthesia 2021;76:8-17. [Crossref] [PubMed]