The Population, Intervention, Comparison, and Outcome (PICO) framework was used to structure the search strategy. This is summarised in Table 1. This method ensured a comprehensive search strategy to gather relevant publications related to predictive modelling in cervical cancer risk assessment, covering key aspects such as population characteristics, intervention methods, comparison with traditional approaches, and outcome measures.

Science mapping, as defined by Donthu et al. (2021) examines the relationships between research constituents, focusing on intellectual interactions and structural connections among them. This analysis employed techniques such as citation analysis, thematic analysis, bibliographic coupling, co-word analysis, and co-authorship analysis (Donthu et al., 2021; Öztürk et al., 2024). When combined with network analysis, these techniques were instrumental in presenting the bibliometric structure and the intellectual structure of the research field.

The top 20 highly cited research articles presented in Table 4 explored various methods for improving cervical cancer screening, diagnosis, and risk prediction. Machine learning algorithms were a common theme, with studies applying them to analyse clinical data, Pap smear images, and self-collected samples. These approaches have achieved high accuracy in some cases, exceeding the performance of traditional methods.

Recent advancements in cervical cancer research are revolutionising screening, diagnosis, and risk prediction methods. Machine learning algorithms are at the forefront, with studies such as Ijaz et al. (2020) utilising them to analyse clinical data for early cancer detection. Deep learning, a specialised machine learning technique, is also proving valuable. Hussain et al. (2020) demonstrate its effectiveness in Pap smear image analysis and classification using convolutional neural networks.

Beyond machine learning, the table highlights other promising techniques. Yaman and Tuncer (2022) introduced a novel deep feature extraction method for accurate cancer detection, while Ali et al. (2021) showcased the efficacy of machine learning models in analysing various clinical datasets to identify early-stage cervical cancer. Additionally, research on HPV testing, a major risk factor, is ongoing, with studies exploring its potential to improve screening programs (Zhao et al., 2021). Furthermore, Figure 2 shows the most globally cited authors. Overall, these advancements offer significant hope for improved cervical cancer outcomes and enhanced patient care.

The top 20 highly cited publications in cervical cancer research revealed a multi-pronged approach using recent advancements. Machine learning and deep learning techniques were prominent, demonstrating effectiveness in analysing various data sources—clinical data, Pap smear images, and HPV test results. Algorithms such as Random Forest and convolutional neural networks achieved high accuracy in classification and detection tasks.

Furthermore, researchers developed methods for efficient feature selection and model optimisation (e.g., Boruta analysis) to refine these tools and enhance model performance. Early detection remains a critical focus, with novel models such as the Cervical Cancer Prediction Model utilising risk factors and clinical data for early identification. This focus aimed to enhance patient outcomes through timely intervention.

Enhanced screening strategies are another key theme. Integration of advanced technologies such as digital colposcopy and image analysis were explored to improve accuracy and efficiency. In addition, the role of HPV testing in identifying high-risk individuals was investigated for improved screening protocols. Finally, the importance of clinical validation and translation of research findings into practical applications was emphasised. Extensive validation studies ensure the generalisability of proposed methodologies, thus, bridging the gap between research and improved patient care.

Thematic content analysis revealed a multidisciplinary approach in cervical cancer research. Machine learning, deep learning, clinical validation, and translation contributed significantly to innovation in cervical cancer detection, diagnosis, and management.

The reviewed studies on machine learning in cervical cancer research reveal several gaps. Many studies lack external validation on independent datasets, limiting the robustness and generalisability of their findings. Most studies do not address model interpretability, crucial for understanding prediction mechanisms. Socioeconomic factors, such as income and access to healthcare, are largely ignored despite their significant impact on cervical cancer risk and outcomes. Furthermore, information on dataset size and diversity is often missing, hindering the assessment of model performance and applicability across different populations. Addressing these gaps is essential for advancing the practical application of machine learning in cervical cancer care.

The visual representation in Figure 3 depicts a co-citation network with authors as nodes connected by lines indicating their co-citations, with the central node labelled “Zhang” prominently displayed. Surrounding this central node are numerous red lines connecting to other author nodes, suggesting that Zhang is highly cited with these authors. Other author nodes, labelled with various names or abbreviations, are scattered around the central node, with connections represented by red lines of varying opacities.

The denser lines indicate stronger co-citation relationships, resembling a starburst pattern emanating from the central “Zhang” node. This network structure implies that “Zhang” holds significant influence in the field, with the connections revealing collaborative relationships and influential connections among authors. Co-citation networks offer valuable insights into scholarly collaboration, research trends, and influential figures, aiding researchers in understanding the dynamics of scientific collaboration and impact.

The analysis of abstracts identified emerging themes within cervical cancer risk prediction modelling, represented by codes extracted from the text data. Table 5 summarises the frequency counts of these codes, showcasing the prominence of keywords such as “cervical,” “cancer,” “model,” “patient,” and “accuracy.” Additionally, codes such as “pap,” “study,” and “machine” were also prevalent, indicating themes related to diagnostic methods, research studies, and machine learning techniques, respectively. These codes provided valuable insights into the key concepts and areas of focus within the abstracts, facilitating a deeper understanding of the research landscape in cervical cancer risk prediction.

TF-IDF vectorisation and LDA produced topics characterised by prominent keywords, delineating prevalent themes within the abstracts. These topics encompassed a spectrum of concepts, with Topic 0 focusing on cancer and cervical, Topic 1 emphasising patient-related terms, Topic 2 highlighting medical images, Topic 3 focusing on patient-centred models, and Topic 4 showcasing features pertinent to predictive modelling. The weightings in Tables 6, 7 represent the probability of a word belonging to a specific topic, with higher weights indicating a stronger association.

Additionally, Gensim and LDA topic modelling revealed five key themes within cervical cancer risk prediction research. Topics 0 and 1 featured core concepts such as “cancer,” “cervical,” “model,” and “accuracy,” emphasising model development and machine learning. Topic 2 suggested the use of medical images for detection, while Topic 3 focused on patient-centred models with interpretable results. Topic 4 pointed to research on feature engineering for improved model performance, highlighting a strong focus on model development and translating research into practical clinical applications for improved patient outcomes.

This convergence of findings from both methodologies emphasises the multifaceted nature of cervical cancer risk prediction research. The clusters derived from both approaches included:

The integration of findings from both the Braun and Clarke thematic analysis and the NLP approach underscored the multifaceted nature of cervical cancer risk prediction research. The thematic and NLP analyses provided specific, actionable insights into cervical cancer risk prediction research. Thematic analysis revealed emerging themes such as diagnostic methods, machine learning techniques, and patient-focused models, underscored by the frequent occurrence of keywords like “cervical,” “cancer,” “model,” and “accuracy.”

NLP techniques like TF-IDF vectorisation and LDA highlighted key topics, including the development of predictive models, the use of medical images, and patient-centred approaches, as seen in the prevalence of terms across topics. This integrated methodology found research clusters such as machine learning for diagnosis, HPV testing, prognosis and risk stratification, and recurrence analysis. These findings emphasise the multidisciplinary nature of cervical cancer risk prediction, showing the combination of qualitative and quantitative insights to enhance understanding and guide future research directions.

The provided Figure 4 illustrates the thematic understanding of cervical cancer research and machine learning, aligning with the identified themes using BiblioShiny. A quadrant plot visualises the thematic landscape of cervical cancer research and machine learning. Two axes, developmental degree (vertical) and relevance degree (horizontal), divide the plot into quadrants. Niche themes, like “pap smear images” and “machine learning algorithms,” occupy a specific space, reflecting their focused nature. Conversely, central themes like “cervical cancer prediction” reside in the core area due to their high relevance.

Emerging or declining themes, such as “challenging process due to detecting cervical cancer,” are positioned accordingly. Established themes like “confidence interval (CI)” form the foundational “basic themes” quadrant. Specific terms like “support vector machine,” “human papillomavirus (HPV),” or “squamous intraepithelial lesions” further illustrate the thematic breakdown within each quadrant. The study gained insights into the relative importance and developmental stages of various research areas, ultimately aiding in identifying promising avenues and knowledge gaps within the field of cervical cancer research and machine learning by analysing this thematic distribution.

The study aligns with current trends, as commonly used models in cervical cancer risk prediction include support vector machine (SVM), random forest (RF), and multivariable logistic regression. This is further supported by research comparing machine learning algorithms for disease prediction, which identified SVM and RF as preferred algorithms (Uddin et al., 2019; Zhang et al., 2023; Yang et al., 2019).

The niche themes “detect cervical cancer” and “machine learning algorithms” signify specialised areas of focus within the broader research field. This means that there is a concentrated effort on using advanced computational techniques to improve the accuracy and efficiency of cervical cancer detection.

Researchers are increasingly integrating machine learning algorithms to develop innovative diagnostic tools, which can potentially enhance early detection, optimise treatment plans, and improve patient outcomes. This targeted research highlights the importance of interdisciplinary collaboration in advancing medical technologies and addressing specific challenges within cervical cancer care.

The visualisation in Figure 5 revealed that between 2013 and 2019, research activities primarily focused on refining pap smear imaging techniques, a fundamental aspect of cervical cancer screening. This sustained emphasis on pap smear imaging highlights the pivotal role in identifying precancerous lesions and early-stage malignancies. Concurrently, investigations into cervical cancer risk spanned several years (2013–2021), reflecting ongoing efforts to elucidate risk factors, epidemiology, and preventive measures. Studies explored genetic predispositions, viral associations (such as human papillomavirus), and lifestyle factors contributing to cervical cancer risk, contributing to a comprehensive understanding of the disease's aetiology and prevention strategies.

The years 2020–2021 witnessed a significant shift with the emergence of machine learning algorithms in cervical cancer research. Researchers adopted artificial intelligence to enhance diagnostic accuracy and predictive models, analysing complex datasets comprising imaging results and patient histories for personalised risk assessment. Subsequently (2021–2022), predictive modelling gained prominence, aiming to forecast individualised outcomes such as disease progression, treatment response, and recurrence. Integrating clinical data with machine learning algorithms facilitated more precise risk stratification, marking a notable advancement in cervical cancer risk prediction and management strategies.

The provided Figure 6 offers insight into the evolving landscape of topics related to cervical cancer and machine learning from 2016 to 2021. Each topic is depicted by blue dots, with their sizes indicating the term frequency, denoted by a legend on the right. Notable topics include cervical cancer detection, machine learning models, Pap smear images, cervical cancer screening, machine learning algorithms, support vector machine, cervical cancer risk, cervical cancer patients, and human papillomavirus (HPV). The graph highlights the progressive nature of research and development in cervical cancer detection, machine learning, and associated domains, emphasising the ongoing efforts to advance diagnostic and predictive capabilities for improved healthcare outcomes.

The significance of the word cloud (see Figure 7) analysis lies in its ability to highlight key themes and trends within cervical cancer screening research. It reveals the central focus on cervical cancer screening, indicating the ongoing efforts to improve detection and diagnosis methods. The prominence of terms like “Pap Smear Images,” “Human Papillomavirus (HPV),” “Low-Grade Squamous Intraepithelial,” “High-Grade Squamous Intraepithelial,” and “Machine Learning Algorithms” suggests a shift towards incorporating advanced technology for more accurate and efficient screening processes.

This implies a potential transformation in clinical practise, with the adoption of innovative approaches to enhance early detection rates and improve patient outcomes. Terms like “Cervical Cancer Detection,” “Random Forest (RF),” and “Support Vector Machine” show the importance of early diagnosis and the specific techniques employed to achieve this goal. Overall, the word cloud analysis shows the evolving field of cervical cancer screening research, highlighting the integration of technology and the ongoing pursuit of improved screening methods for better patient care.

A key finding was the recurring theme of “diagnostics” (see Figure 8). This theme appeared as a central element in the visualisation, highlighting its widespread relevance across the analysed articles. The prominence of diagnostics suggested its importance as a major area of research focus.

The visualisation also revealed a diverse range of journals and research fields connected to diagnostics. Each surrounding node represented a specific journal or field of study, such as the Asian Journal of Information Technology, Healthcare Technology Letters, International Journal of Molecular Sciences, Abdominal Radiology, Multimedia Tools and Applications, Computational Mathematics International Journal, and Indian Journal of Science & Technology. This variety of nodes showcased the broad scope of research areas associated with diagnostics.

Red lines connect the central “diagnostics” theme to other nodes, indicating bibliographic coupling. This means there were close relationships between diagnostics research and various research domains. These connections highlighted the interdisciplinary nature of diagnostics research, suggesting its interconnectedness with multiple fields. The visualisation provided evidence for the role diagnostics plays in integrating different areas of research.

This approach went beyond traditional citation analysis by uncovering interconnected relationships between research topics and journals, offering valuable insights into the multidisciplinary nature of diagnostics research. Unlike traditional citation analysis, which primarily focuses on citation counts and direct references, Figure 8 provided a more comprehensive view of the complex interplay between different areas of scholarly inquiry.

The findings demonstrated the collaborative nature of interdisciplinary research efforts in driving advancements in cervical cancer diagnosis by highlighting the interconnectedness between diagnostics research and various research domains. This deeper understanding emphasises the importance of interdisciplinary collaboration in addressing complex healthcare challenges and highlights the pivotal role of diagnostics research in advancing diagnostic capabilities for cervical cancer.

The map in Figure 9 illustrates the distribution of scientific output across different countries regarding predictive modelling for cervical cancer risk. Countries are shaded in varying shades of blue to denote the volume of publications, with darker shades indicating higher production. Notably, India, United States, China, and Australia emerge as significant contributors to this field. Conversely, Africa appears mostly unshaded, indicating minimal research output, underscoring the disparity in scientific contributions between developed nations and African countries.

This observation emphasises the necessity for heightened investment and focus on scientific research concerning cervical cancer risk prediction in African nations. Closing this research gap not only aids in addressing the burden of cervical cancer within Africa but also holds promise for enhancing screening, prevention, and treatment strategies on a global scale. Redirecting resources and support towards scientific activities in Africa can pave the way for achieving greater equity in healthcare access and mitigating the global burden of cervical cancer.

The provided Figure 10 illustrates a succinct overview of the number of articles produced each year in the field of predictive modelling for cervical cancer risk. Spanning from ~2013 to 2023, the x-axis denotes the years, while the y-axis indicates the number of articles published. Notably, until around 2018, there was a steady production of fewer than five articles per year. However, post-2018, there is a notable surge in article production, reaching a peak in 2022 with over 15 articles. Subsequently, there is a sharp decline in 2022. This spike in 2021 indicates a surge in research activity, while the decline in 2022 may suggest a shift in research focus or external factors influencing publication trends.

This analysis of publication trends highlights the growing importance of predictive modelling in cervical cancer risk assessment, signifying its potential to improve preventative healthcare strategies. While the decline in 2023 publications requires further exploration, it highlights the dynamic nature of this research field. Continued monitoring of publication trends alongside a deeper understanding of the underlying reasons for these shifts can provide valuable insights for researchers and stakeholders invested in advancing this critical field.