Evaluating AI Models and Predictors for COVID-19 Infection Dependent on Data from Patients with Cancer or Not: A Systematic Review
Takdon Kim1 and Heeyoung Lee2,3*
1Clinical Trials Center, Chungnam National University Hospital, Daejeon 35015, Republic of Korea
2College of Pharmacy, Inje University, Gimhae 50834, Republic of Korea
3Inje Institute of Pharmaceutical Sciences and Research, Inje University, Gimhae 50834, Republic of Korea
2College of Pharmacy, Inje University, Gimhae 50834, Republic of Korea
3Inje Institute of Pharmaceutical Sciences and Research, Inje University, Gimhae 50834, Republic of Korea
Correspondence to: Heeyoung Lee, College of Pharmacy, Inje University, Gimhae 50834, Republic of Korea
Tel: +82-55-320-3328, Fax: +82-55-320-3328, E-mail: phylee1@inje.ac.kr
Tel: +82-55-320-3328, Fax: +82-55-320-3328, E-mail: phylee1@inje.ac.kr
Received May 3, 2024; Revised June 13, 2024; Accepted June 14, 2024.
This is an Open Access journal distributed under the terms of the Creative Commons Attribution Non-Commercial License (http://creativecommons.org/licenses/by-nc/4.0) which permits unrestricted non-commercial use, distribution, and reproduction in any medium, provided the original work is properly cited.
- Abstract
- Background: As preexisting comorbidities are risk factors for Coronavirus Disease 19 (COVID-19), improved tools are needed for screening or diagnosing COVID-19 in clinical practice. Difficulties of including vulnerable patient data may create data imbalance and hinder the provision of well-performing prediction tools, such as artificial intelligence (AI) models. Thus, we systematically reviewed studies on AI prognosis prediction in patients infected with COVID-19 and existing comorbidities, including cancer, to investigate model performance and predictors dependent on patient data. PubMed and Cochrane Library databases were searched. This study included research meeting the criteria of using AI to predict outcomes in COVID-19 patients, whether they had cancer or not. Preprints, abstracts, reviews, and animal studies were excluded from the analysis. Majority of non-cancer studies (54.55 percent) showed an area under the curve (AUC) of >0.90 for AI models, whereas 30.77 percent of cancer studies showed the same result. For predicting mortality (3.85 percent), severity (8.33 percent), and hospitalization (14.29 percent), only cancer studies showed AUC values between 0.50 and 0.69. The distribution of comorbidity data varied more in non-cancer studies than in cancer studies but age was indicated as the primary predictor in all studies. Non-cancer studies with more balanced datasets of comorbidities showed higher AUC values than cancer studies. Based on the current findings, dataset balancing is essential for improving AI performance in predicting COVID-19 in patients with comorbidities, especially considering age.
- Keywords : Artificial intelligence models, cancer, comorbidity, coronavirus disease-19, non-cancer
- September 2024, 34 (3)
Full Text(PDF) Free- Social Network Service
-
- Services
- Cited By Articles
-
CrossRef (0)
