Comparing machine learning screening approaches using clinical data and cytokine profiles for COVID-19 in resource-limited and resource-abundant settings - PubMed (original) (raw)
Comparative Study
. 2024 Jun 28;14(1):14892.
doi: 10.1038/s41598-024-63707-3.
Aamer Ikram 2, Luke T Dang 3, Adnan Bashir 4, Tanzeel Zohra 2, Amna Ali 2, Hamza Tanvir 2, Mohammad Mudassar 2, Resmi Ravindran 3, Nasim Akhtar 5, Rana I Sikandar 5, Mohammed Umer 6, Naeem Akhter 6, Rafi Butt 7, Brandon D Fennell 8, Imran H Khan 9
Affiliations
- PMID: 38937503
- PMCID: PMC11211475
- DOI: 10.1038/s41598-024-63707-3
Comparative Study
Comparing machine learning screening approaches using clinical data and cytokine profiles for COVID-19 in resource-limited and resource-abundant settings
Hooman H Rashidi et al. Sci Rep. 2024.
Abstract
Accurate screening of COVID-19 infection status for symptomatic patients is a critical public health task. Although molecular and antigen tests now exist for COVID-19, in resource-limited settings, screening tests are often not available. Furthermore, during the early stages of the pandemic tests were not available in any capacity. We utilized an automated machine learning (ML) approach to train and evaluate thousands of models on a clinical dataset consisting of commonly available clinical and laboratory data, along with cytokine profiles for patients (n = 150). These models were then further tested for generalizability on an out-of-sample secondary dataset (n = 120). We were able to develop a ML model for rapid and reliable screening of patients as COVID-19 positive or negative using three approaches: commonly available clinical and laboratory data, a cytokine profile, and a combination of the common data and cytokine profile. Of the tens of thousands of models automatically tested for the three approaches, all three approaches demonstrated > 92% sensitivity and > 88 specificity while our highest performing model achieved 95.6% sensitivity and 98.1% specificity. These models represent a potential effective deployable solution for COVID-19 status classification for symptomatic patients in resource-limited settings and provide proof-of-concept for rapid development of screening tools for novel emerging infectious diseases.
Keywords: Binary classifier; COVID-19; CTACK (CCL27); Chemokines; Cytokines; Eotaxin (CCL11); FGF basic (FGF2); G-CSF (CSF3); GM-CSF (CSF2); GRO-α (CXCL1); HGF; IFN-α2 (IFNA2); IFN-γ (IFNG); IL-12 (p40) (IL12B); IL-12 (p70) (IL12A); IL-18 (IL18); IL-1ra (IL1RN); IL-1α (IL1A); IL-1β (IL1B); IL-2Rα (IL2RA); IL10; IL13; IL15; IL16; IL17A; IL2; IL3; IL4; IL5; IL6; IL7; IL8 (CXCL8); IL9; IP-10 (CXCL10); LIF; M-CSF (CSF1); MCP-1 (MCAF) (CCL2); MCP-3 (CCL7); MIF; MIG (CXCL9); MIP-1α (CCL3); MIP-1β (CCL4); Machine learning; PDGF-BB (PDGFB); RANTES (CCL5); SCF (KITLG); SCGF-β (CLEC11A); SDF-1α (CXCL12); TNF-α; TNF-β (LTA); TRAIL (TNFSF10); VEGF (VEGFA); β-NGF (NGF).
© 2024. The Author(s).
Conflict of interest statement
The authors declare the following competing interests: the Automated Machine Learning platform MILO (Machine Intelligence Learning Optimizer) in which Hooman Rashidi is an inventor is an Intellectual Property of the University of California. Dr. Rashidi also serves as a board member for the University of California start up (MILO-ML Inc) that has acquired this technology. Brandon Fennell also serves as a shareholder and board member of MILO-ML Inc. Besides the aforementioned authors, no other authors have any competing interests.
Figures
Figure 1
ROC AUC and precision recall curves for approach 3.
Figure 2
Pearson standard correlation coefficients for clinical features.
Figure 3
Pearson standard correlation coefficients for cytokine/chemokine features.
References
Publication types
MeSH terms
Substances
LinkOut - more resources
Full Text Sources
Medical
Research Materials
Miscellaneous