key: cord-0793518-lgwg262x
authors: Adeoye, John; Akinshipo, Abdulwarith; Thomson, Peter; Su, Yu-Xiong
title: Artificial intelligence-based prediction for cancer-related outcomes in Africa: Status and potential refinements
date: 2022-04-23
journal: Journal of global health
DOI: 10.7189/jogh.12.03017
sha: d980dbe647d90c304d2a2d5564c17ac4411c07c6
doc_id: 793518
cord_uid: lgwg262x

nan

G lobally, cancer ranks among the most common causes of death, especially among people under 70 years of age [1] . With this burden rapidly increasing globally, utilizing prediction tools to assist decision-making and encourage individualized treatment planning is gradually becoming paramount in cancer diagnosis and management. Notably, many tools constructed on the backend of artificial intelligence (AI) algorithms have been shown to improve the predictive accuracy and clinical impact of risk prediction compared to clinical scenarios not utilizing these models [2] . However, the actualisation of the potential of health care AI has mostly been assessed in high-income and resource-driven centres. The impact and efficiency of oncological AI-based prediction tools are expected to be better realised when applied in low-resource and rural settings fraught with a paucity of experienced clinicians and specialists.

Africa is considered the most vulnerable continent in the world with regards to the availability of resources for optimal health care provision and comprises about 85.2% of countries that belong to the low-income/ lower-middle-income economies [3] . Also, many individuals have to rely on underfunded public health care systems, even for lethal diseases like cancer, which may be further constrained by access to secondary or tertiary health centers and financial limitations [4] . Therefore, AI prediction tools would prove beneficial in triaging patients before they present to health institutions, as well as for treatment selection of patients with precancer and cancer in such settings. Furthermore, these platforms could potentially be employed alongside other digitized systems such as telemedicine for remote consultations and Internet of Medical Things (IoMT) to improve referral strategies [5, 6] .

We examine the implementation status of AI platforms for cancer outcomes in Africa and provide pointers that could maximize their utilization.

Artificial intelligence-based prediction for cancer-related outcomes in Africa: Status and potential refinements Efforts should be directed towards external validation and impact assessment of oncological AI tools in Africa.. AI -artificial intelligence, AUC -Area under the receiver operating characteristic curve, CRC -colorectal carcinoma *Phase I -Data collection and processing, Phase II -Model construction, Phase III -Model validation, Phase IV -Software application development, Phase V -Impact and efficiency analysis, Phase VI -Model implementation in daily oncology practices [21] .

The uptake of technology-driven systems available in developed regions is slower in Africa than in other parts of the world, as these platforms are often generated from integrative variables that may not be available for external validation and clinical application in many African settings. For oncology, this may be further challenged by the variation in molecular or etiological subtypes, as well as infrastructural and ethnocultural peculiarities of cancer management in the African region [7, 8] . A preliminary search in five electronic databases to map tools developed from January 2000 to January 2022 using patient cohorts in Africa found 12 prediction tools for different malignancies and applications (Appendix S1 in the Online Supplementary Document). Their characteristics are presented in Table 1 and Figure 1 .

Generally, the development of AI tools for assisted decision-making among patients with cancer in Africa can be considered a recent endeavour, with many models being reported from 2018 ( Table 1) . With more oncological AI platforms being proposed in the post-COVID-19 era, it may be that the refocusing of experienced health personnel towards pandemic control did motivate the search for automated tools aimed at facilitating the diagnosis, treatment selection, and prognostication of cancer patients [22] [23] [24] . Also, these tools may have been proposed with the aim of increasing the efficiency of specialists, easing the case burden caused by a worsening health provider-to-patient ratio [25] , and limiting person-to-person contact to clinical encounters of utmost importance [26] .

Relative to the different subregions in Africa, the pattern of AI penetrance mirrors the ongoing economic and infrastructural development in the continent [4, 27] . As such, populations in areas with larger economies and fair technological expertise like South Africa, Egypt, Nigeria, and Kenya are largely the focus of the development and validation of the AI models for cancer outcomes (Figure 1, Panel A) . This identifies a niche for intensified efforts in AI development and/or deployment among approximately 90% of the countries in Africa (especially within Central Africa), many of which are either low-income or lower-middle-income nations.

Moreover, oncological AI-based models in Africa were being considered for outcome determination among patients with breast, cervical, and colorectal carcinomas (Figure 1, Panel B) . This fairly coincides with the incidence of malignant neoplasms in the continent, with breast, cervical, prostate, liver, and colorectal cancers being the five most common subtypes [1] . However, actionable AI patient-based platforms have not been proposed to facilitate the screening, diagnosis, and management of hepatocellular and prostatic carcinoma in the region. While one model was found for prostatic carcinoma [17] , it was fashioned to improve incidence reporting in cancer registries from free-text pathology reports, which may not have a direct consequence on the diagnosis or management of the malignancy. Hence, Africa will benefit from research seeking to propose novel platforms and/or external validation of existing AI-based platforms, especially for liver and prostate cancer, as well as other, previously unconsidered malignancies [28] [29] [30] [31] [32] . Specifically, malignancies such as oesophageal cancer and Kaposi's sarcoma, with higher age-standardized incidences in Africa than in other parts of the world, would markedly benefit from AI-based prediction to improve the standard and reach of diagnosis and management [1] .

The technical aspects of the development and validation of cancer-related AI models in Africa is generally satisfactory ( Table 1) . This may be credited to the availability of knowledgeable individuals with sufficient skills to perform optimal data pre-processing, model selection, tuning, validation, and software deployment. However, many novel AI platforms that are being developed by African institutions affiliates may be employing 

The authors completed the ICMJE Unified Competing Interest Form (available upon request from the corresponding author) and declare no conflicts of interest.

Online Supplementary Document open-source databases comprised of patients from other populations [33] [34] [35] [36] [37] . This alludes to the unavailability of clinical, imaging, pathologic, and molecular data from cancer patients, especially in electronic formats, within many parts of the regions. The data paucity challenge may have informed the modelling of many one-time point outcomes such as tumour diagnosis and profiling instead of prognosis, which requires time-to-event monitoring that is likely unavailable (Figure 1, Panel C) . While this problem may be inseparable from the overall technological advancements and e-health penetrance in the continent [38] , it is expected that an increase in the knowledge and potential impacts of digital health among African clinicians involved in cancer management, as well as health information technology stakeholders and policymakers, would encourage digitization of salient patient records and treatment follow-up data for model construction. Essentially, these knowledge provision avenues and platforms may also come in the form of practical experiences while participating in multinational collaborative work for external validation of prediction models. This strategy could potentially lay the foundation for the integration of these AI-based models with other digital health systems, especially telehealth [18] . In addition, the digital integrative framework of AI tools and telehealth can be adopted to mitigate the lack of direct access to experienced specialists and tertiary facilities, while delivering comparable health care standards in rural Africa. Training platforms may also emphasize the role of AI in assisting clinicians in their duties rather than replacing them, as the need to secure their jobs may discourage interest in digital health tools and their implementation [39] .

The performances of the set of models proposed for use in African oncological settings so far were satisfactory to excellent upon validation [9] [10] [11] [12] 14, 18] (Table 1) . However, to fully actualize the potential of AI-based prediction for cancer outcomes in Africa, inquiries into the impact and efficiency of the models in comparison to the current standards of management are paramount. Since the training and validation of AI prediction platforms have only been described recently within the region, the future should see intensified efforts into elucidating their impact when implemented in practice. While this requires the use of prospective rigorous study approaches like the randomized controlled trial (RCT) design, findings of these pivotal research before full-scale implementation may take a few years. In the interim, Africa may benefit from studies seeking to perform an external validation of any AI-based prediction tools for cancer-related outcomes, especially when their clinical effectiveness is being investigated using an RCT. However, this will require an initial feasibility check to determine whether the input parameters of the model may be obtained within the region.

Applying AI in developing credible tools for the prediction of cancer outcomes in Africa is a relatively recent endeavour. Model construction in the continent reflected the degree of technological advancement among the nations, as well as the availability of electronic health records and digitized clinical investigations. None of the models developed locally and internationally have been assessed for their impact and efficiency in automated assistive clinical decision-making and risk stratification. Hence, this should represent the phase of AI implementation to be addressed soon, especially in numerous remote areas of the low-and lower-middle-income areas. Likewise, for these settings, it is expected that the future will see the integration of AI-based prediction and other digital health technologies to mitigate the challenges of reduced manpower, substandard facilities, and lack of access to care currently experienced in cancer diagnosis and management in Africa.

Photo: Oncological AI in Africa. Source: From the authors' own collection, used with permission.

Global Cancer Statistics 2020: GLOBOCAN Estimates of Incidence and Mortality Worldwide for 36 Cancers in 185 Countries

Clinical impact and quality of randomized controlled trials involving interventions evaluating artificial intelligence prediction tools: a systematic review. npj

Measuring health science research and development in Africa: mapping the available data

The World by Income and Region

Can digital technologies improve health?

Digital Technologies and Data Science as Health Enablers: An Outline of Appealing Promises and Compelling Ethical, Legal, and Social Challenges

Cancer in Africa: the way forward

Inequities in breast cancer treatment in sub-Saharan Africa: findings from a prospective multi-country observational study

EpiPanGI Dx: A Cell-free DNA Methylation Fingerprint for the Early Detection of Gastrointestinal Cancers

Use of Machine Learning and Statistical Algorithms to Predict Hospital Length of Stay Following Colorectal Cancer Resection: A South African Pilot Study

Predicting Colorectal Cancer Recurrence and Patient Survival Using Supervised Machine Learning Approach: A South African Population-Based Study

Tumor Mutation Burden Prediction Model in Egyptian Breast Cancer patients based on Next Generation Sequencing

Identification of Malignancies from Free-Text Histopathology Reports Using a Multi-Model Supervised

Breast cancer risk prediction in African women using Random Forest Classifier

Rule-Based Information Extraction from Free-Text Pathology Reports Reveals Trends in South African Female Breast Cancer Molecular Subtypes and Ki67 Expression

Automatic contouring system for cervical cancer using convolutional neural networks

A Text Mining Approach in the Classification of Free-Text Cancer Pathology Reports from the South African National Health Laboratory Services. Information (Basel)

Point-of-Care Digital Cytology With Artificial Intelligence for Cervical Cancer Screening in a Resource-Limited Setting

Robust breast cancer prediction system based on rough set theory at National Cancer Institute of Egypt

Accurate Machine-Learning-Based classification of Leukemia from Blood Smear Images

Guidelines and quality criteria for artificial intelligence-based prediction models in healthcare: a scoping review

Re-organising primary health care to respond to the Coronavirus epidemic in Cape Town, South Africa. Afr J Prim Health Care Fam Med

Africa Succeeded Against COVID-19's First Wave, but the Second Wave Brings New Challenges

Allocation of scarce resources in Africa during COV-ID-19: Utility and justice for the bottom of the pyramid?

COVID-19 in Africa: Survey Analysis of Impact on Health-Care Workers

The Role of Artificial Intelligence in Fighting the COVID-19 Pandemic

Economic growth in Sub-Saharan Africa

Development and Validation of an Artificial Intelligence-Powered Platform for Prostate Cancer Grading and Quantification

An artificial intelligence algorithm for prostate cancer diagnosis in whole slide images of core needle biopsies: a blinded clinical validation and deployment study

Deep learning in prostate cancer diagnosis and Gleason grading in histopathology images: An extensive study

Machine learning based liver disease diagnosis: A systematic review

Deep Learning for Accurate Diagnosis of Liver Tumor Based on Magnetic Resonance Imaging and Clinical Data

An Enhanced Deep Learning Framework for Skin Lesions Segmentation. 11th International Conference on Computational Collective Intelligence (ICCCI)

Deep Learning Model for Skin Lesion Segmentation: Fully Convolutional Network. 16th International Conference on Image Analysis and Recognition (ICIAR)

FCN-Based DenseNet Framework for Automated Detection and Classification of Skin Lesions in Dermoscopy Images

A Comparison of Two Deep Learning Architectures to Automatically Define Patient-Specific Beam Apertures

A stacking ensemble deep learning approach to cancer type classification based on TCGA data

Factors affecting sustainable adoption of e-health technology in developing countries: an exploratory survey of Nigerian hospitals from the perspective of healthcare professionals

The integration of artificial intelligence in medical imaging practice: Perspectives of African radiographers

Correspondence to: Yu-Xiong Su Division of Oral and Maxillofacial Surgery Faculty of Dentistry University of Hong Kong 34 Hospital