The Persistent Threat of Chronic Inflammation on the Mortality Among Cervical Cancer Survivors: A Mendelian Randomization and Machine Learning Analysis Using UK Biobank and Chinese Cohort Data

Abstract

Jing Wang,1,2,* Zhichao Chen,3,* Mingfei Guan,4 Zebiao Ma,4 Lin Peng,5 Jiongyu Chen,5 Pier Luigi Fiori,2 Ciriaco Carru,2 Giampiero Capobianco,6 Donatella Coradduzza,2 Li Zhou4 1Department of Obstetrics and Gynecology, Second Affiliated Hospital of Shantou University Medical College, Shantou, People’s Republic of China; 2Department of Biomedical Sciences, University of Sassari, Sassari, Italy; 3Department of Cardiology, Second Affiliated Hospital of Shantou University Medical College, Shantou, People’s Republic of China; 4Department of Gynecologic Oncology, Cancer Hospital of Shantou University Medical College, Shantou, People’s Republic of China; 5Department of Central Laboratory, Cancer Hospital of Shantou University Medical College, Shantou, People’s Republic of China; 6Gynecologic and Obstetric Clinic, Department of Medicine, Surgery and Pharmacy, University of Sassari, Sassari, Italy*These authors contributed equally to this workCorrespondence: Li Zhou, Department of Gynecologic Oncology, Cancer Hospital of Shantou University Medical College, Shantou, People’s Republic of China, Email [email protected]: The association between inflammatory dysregulation and cervical carcinogenesis and progression has not yet been fully elucidated. We aimed to comprehensively evaluate the genetic association between inflammation and cervical cancer, and construct an accurate prognosis model based on circulating inflammatory parameters and indexes with machine learning (ML) algorithms.Patients and Methods: We tested the genome-wide association of circulating inflammatory molecules (CIMs) (91 circulating inflammatory cytokines and 10 inflammatory cells) and summary data retrieved from the UK biobank (cases = 1659 and controls =381,902) with two-sample Mendelian randomization (MR) and colocalization analyses. Nine ML and logistic regression (LR) integrated prognosis models were developed for 1042 subjects with cervical cancer (random allocation into training and validation cohorts at 6:4 ratio).Results: Three potential causative CIMs for cervical cancer were identified via a two-sample MR. However, neither reverse MR, nor Bayesian colocalization analyses supported shared causal variation. After feature selection with 3 algorithms (LASSO regression, Boruta and Support vector machines), the gradient boosting machine (GBM) model outperformed other models by achieving an area under the curve (AUC) of 0.930 and a Brier score of 0.027 in 1-year overall survival (OS) prediction. Similarly, the GBM model delivered the best overall performance in 5-year OS prediction with an AUC of 0.893 and a Brier score of 0.089. Following the Shapley Additive explanations (SHAP), the lymphocyte monocyte ratio, neutrophil count, platelet count, and platelet lymphocyte ratio were associated with 1-year OS, while the systemic immune-inflammation index, platelet neutrophil ratio, and monocyte count were significantly related to 5-year OS.Conclusion: No substantial causal associations were observed between CIMs and cervical cancer. The cohort study findings reveal the persistent impact of inflammation on cervical cancer prognosis, highlighting the crucial role of chronic inflammation when investigating the biomarkers of cervical cancer progression and developing pharmacological interventions. The GBM model consistently achieved satisfactory performance in cervical cancer prognosis prediction with demographics and CIMs, meriting further validation and potential clinical implementation.Keywords: cervical cancer, Mendelian randomization, colocalization analysis, machine learning, inflammation, overall survival

Keywords

• Cervical cancer
• Mendelian randomization
• colocalization analysis
• machine learning
• inflammation
• overall survival