%0 Journal Article %1 kennion_machine_2022 %A Kennion, Oliver %A Maitland, Stuart %A Brady, Richard %D 2022 %J Health Sciences Review %K Artificial Big Cancer Colorectal Machine cancer data, intelligence, learning, prediction, %P 100041 %R 10.1016/j.hsr.2022.100041 %T Machine learning as a new horizon for colorectal cancer risk prediction? A systematic review %U https://www.sciencedirect.com/science/article/pii/S2772632022000307 %V 4 %X Background Machine learning algorithms have demonstrated high performance in the risk stratification of patients for colorectal cancer. Despite the promise, there has not yet been a proportionate clinical impact from this technology. We provide a comprehensive overview of machine learning and big data analysis in the field of colorectal cancer risk prediction and stratification. Methods Original articles in the English language between January 2011-December 2021 from databases Ovid (Embase and MedLine), PubMed and Google Scholar. Subject key terms including machine learning, artificial intelligence, neural networks, colorectal cancer, and diagnosis/risk. Primary outcome measures included area under the curve and the odds ratio. Results Of the 14 studies included in the final analysis, the number of patients varied from 17,095 to 2,550,119. All articles used for model synthesis contained a minimum of 70,000 patients. The area under the curve varied between 0.738 and 0.896. Commonly used methods included random forests, neural networks and logistic regression, but no specific machine learning method was found to be superior for colorectal cancer prediction. Comparison with STROBE guidelines showed consistent strength and clarity in reporting. Conclusion These studies demonstrate significant and effective risk stratification and prediction of colorectal cancer. Current barriers include lack of external verification, heterogeneity amongst machine learning algorithms and difficult cost/benefit analysis. Clinical integration is fundamental for further external validation which will improve understanding and trust in such algorithms. However, it is vital that this technology is assessed using the same rigorous cost/benefit analysis as for other medical technologies to help valuate the place of such technology in the modern treatment of colorectal cancer.

@article{kennion_machine_2022, abstract = {Background Machine learning algorithms have demonstrated high performance in the risk stratification of patients for colorectal cancer. Despite the promise, there has not yet been a proportionate clinical impact from this technology. We provide a comprehensive overview of machine learning and big data analysis in the field of colorectal cancer risk prediction and stratification. Methods Original articles in the English language between January 2011-December 2021 from databases Ovid (Embase and MedLine), PubMed and Google Scholar. Subject key terms including machine learning, artificial intelligence, neural networks, colorectal cancer, and diagnosis/risk. Primary outcome measures included area under the curve and the odds ratio. Results Of the 14 studies included in the final analysis, the number of patients varied from 17,095 to 2,550,119. All articles used for model synthesis contained a minimum of 70,000 patients. The area under the curve varied between 0.738 and 0.896. Commonly used methods included random forests, neural networks and logistic regression, but no specific machine learning method was found to be superior for colorectal cancer prediction. Comparison with STROBE guidelines showed consistent strength and clarity in reporting. Conclusion These studies demonstrate significant and effective risk stratification and prediction of colorectal cancer. Current barriers include lack of external verification, heterogeneity amongst machine learning algorithms and difficult cost/benefit analysis. Clinical integration is fundamental for further external validation which will improve understanding and trust in such algorithms. However, it is vital that this technology is assessed using the same rigorous cost/benefit analysis as for other medical technologies to help valuate the place of such technology in the modern treatment of colorectal cancer.}, added-at = {2023-02-08T12:07:40.000+0100}, author = {Kennion, Oliver and Maitland, Stuart and Brady, Richard}, biburl = {https://www.bibsonomy.org/bibtex/2162575d64a0ed25b480a6c9d63e513d4/stumaitland}, doi = {10.1016/j.hsr.2022.100041}, file = {Full Text:/Users/stuartbman/Zotero/storage/GU8YJHUK/Kennion et al. - 2022 - Machine learning as a new horizon for colorectal c.pdf:application/pdf;ScienceDirect Snapshot:/Users/stuartbman/Zotero/storage/2SV8NG97/S2772632022000307.html:text/html}, interhash = {15019e945398c545c4f05fc15d377875}, intrahash = {162575d64a0ed25b480a6c9d63e513d4}, issn = {2772-6320}, journal = {Health Sciences Review}, keywords = {Artificial Big Cancer Colorectal Machine cancer data, intelligence, learning, prediction,}, language = {en}, month = sep, pages = 100041, shorttitle = {Machine learning as a new horizon for colorectal cancer risk prediction?}, timestamp = {2023-02-08T12:08:14.000+0100}, title = {Machine learning as a new horizon for colorectal cancer risk prediction? {A} systematic review}, url = {https://www.sciencedirect.com/science/article/pii/S2772632022000307}, urldate = {2022-07-24}, volume = 4, year = 2022 }