Abstract
OBJECTIVE: Prediction models, which estimate disease or outcome probabilities, are widely used in cancer research. This study aims to identify hotspots and future directions of cancer-related prediction models using bibliometrics. METHODS: A comprehensive literature search was conducted in the Science Citation Index Expanded (SCIE) from the Web of Science Core Collection (WoSCC) up to November 15, 2024, focusing on cancer-related prediction models research. Co-occurrence analyses of countries, institutions, authors, journals, and keywords were conducted using VOSviewer 1.6.20. Additionally, keyword clustering, timeline visualization, and burst term analysis were performed with CiteSpace 6.3. RESULTS: A total of 1,661 records were retrieved from the SCIE. After deduplication and eligibility screening, 1,556 publications were included in the analysis. The bibliometric analysis revealed a consistent annual increase in cancer-related prediction model research, with China and the United States emerging as the leading contributors. The United States, England, and the Netherlands had the strongest collaborative networks. The most frequent keywords, excluding "prediction model" and "predictive model", included nomogram (frequency=192), survival (191), risk (121), prognosis (112), breast cancer (103), carcinoma (93), validation (87), surgery (85), diagnosis (83), chemotherapy (80), and machine learning (77). Besides, the timeline view analysis indicated that the "#7 machine learning" cluster was experiencing vigorous growth. CONCLUSION: Cancer-related prediction models are rapidly advancing, especially in prognostic models. Emerging modeling techniques, such as neural networks and deep learning algorithms, are likely to play a pivotal role in current and future cancer-related prediction model research. Systematic reviews of cancer-related predictive models, which could help clinicians select the optimal model for specific clinical conditions may emerge as potential research directions in this field.