Cervical cancer survival prediction by machine learning algorithms: a systematic review

{"Background": "Cervical cancer is the fourth most common cancer of the female reproductive system and the seventh most common cancer worldwide. According to the WHO 2020 report, there were approximately 604,127 diagnosed cases and 341,831 deaths globally. The average 5-year overall survival rate for cervical cancer is 66% per the American Society of Clinical Oncology. Survival prediction traditionally relies on classical statistical methods such as Cox regression, but machine learning algorithms offer advantages in handling non-linear relationships and learning hidden patterns in data without requiring implicit assumptions. This systematic review aimed to evaluate published studies on machine learning algorithms for predicting survival in patients with cervical cancer, considering overall survival (OS), disease-free survival (DFS), and progression-free survival (PFS).", "Methods": "A systematic review was conducted following PRISMA guidelines. An electronic search of PubMed, Scopus, and Web of Science was performed on October 1, 2022, using a search strategy combining terms for cervical cancer ('Uterine Cervical Neoplasms' OR 'Cervical Cancer'), survival ('Survival'), and artificial intelligence ('Machine Learning' OR 'Deep Learning' OR 'Artificial Intelligence'). Inclusion criteria were original full-text English articles using machine learning algorithms as predictive models for cervical cancer survival. Excluded were books, review articles, meta-analyses, case reports, posters, case studies, and articles insufficiently focused on ML implementation for cervical cancer survival. The initial search returned 229 articles; after removing 45 review articles and 85 duplicates, 99 articles remained. Following title/abstract screening (70 excluded) and full-text review (16 excluded based on method, results, or study design), 13 articles were included. Data extracted included authors, publication year, dataset details, survival type, evaluation criteria, machine learning models, and algorithm execution methods.", "Key Results": "Of the 13 included articles, most were published from 2018 onwards, with the most recent from 2022. Eight studies were conducted in Asia, four in Europe, and one in the United States. Eight articles addressed OS, six addressed DFS, and three addressed PFS. The most common machine learning models were random forest (6 articles, 46%), logistic regression (4 articles, 30%), support vector machines (3 articles, 23%), ensemble and hybrid learning (3 articles, 23%), and deep learning (3 articles, 23%). Dataset sizes ranged from 85 to 14,946 patients. Ten articles used hospital/clinic datasets, while three used public databases (SEER, TCGA, GEO). Eleven articles used internal validation, and only two used external validation. The AUC range for OS was 0.40 to 0.99, for DFS was 0.56 to 0.88, and for PFS was 0.67 to 0.81. C-index values ranged from 0.39 to 0.94 for OS, 0.41 to 0.89 for DFS, and 0.69 to 0.79 for PFS. Accuracy ranged from 0.61 to 0.89 for OS, 0.84 to 0.92 for DFS, and 0.73 to 0.84 for PFS. Sensitivity ranged from 0.75 to 0.97 for OS and 0.20 to 0.93 for DFS. F1-score ranged from 0.22 to 0.92 for DFS. Fifteen variables with an effective role in predicting cervical cancer survival were identified, including FIGO stage (most frequently cited), histology, treatment type, tumor size, lymph node metastasis, age, and race/ethnicity. Among articles with multiple models, ensemble/hybrid models (3 articles), random forest (3 articles), logistic regression (1 article), and deep learning (1 article) had the best performance.", "Clinical Implications": "This review demonstrates that machine learning algorithms can effectively predict cervical cancer survival, with random forest and ensemble methods showing particular promise. The combination of heterogeneous multidimensional data (clinical, imaging, molecular) with machine learning techniques may significantly improve prediction accuracy. However, several barriers prevent clinical adoption: only 2 of 13 studies used external validation, limiting generalizability; imbalanced datasets remain a serious challenge affecting model reliability; and the lack of interpretability and explainability in ML algorithms prevents official use in healthcare. The authors note that classical statistical models like Cox are increasingly incompatible with complex medical data, and machine learning offers advantages in pattern recognition and feature selection. For clinical practice, the identified important variables (FIGO stage, histology, treatment type, tumor characteristics) align with known prognostic factors, suggesting ML models are capturing clinically meaningful patterns. The authors conclude that while ML is more accurate than statistical methods for survival prediction, further studies are needed before these algorithms can become a standard tool in cervical cancer management."}