tetano
Editor, Senior Moderator
J Inflamm Res
. 2024 Sep 2:17:5923-5942.
doi: 10.2147/JIR.S472099. eCollection 2024. A Multi-Machine Learning Consensus Model Based on Clinical Features Reveals That Interleukin-10 Derived from Monocytes Leads to a Poor Prognosis in Patients with Coronavirus Disease-2019
Jing Yu[SUP] 1 [/SUP], Yike Zhang[SUP] 1 [/SUP], Zhixiong Ye[SUP] 1 [/SUP], Kun Tang[SUP] 1 [/SUP], Yiming Ma[SUP] 1 [/SUP], Linlin Fu[SUP] 1 [/SUP], Tongtong Cui[SUP] 1 [/SUP], Hening Kang[SUP] 2 [/SUP], Yadong Yuan[SUP] 1 [/SUP], Wensen Pan[SUP] 1 [/SUP]
Affiliations
Background: Despite ongoing interventions, SARS-CoV-2 continues to cause significant global morbidity and mortality. Early diagnosis and intervention are crucial for effective clinical management. However, prognostic features based on transcriptional data have shown limited effectiveness, highlighting the need for more precise biomarkers to improve COVID-19 treatment outcomes.
Methods: We retrospectively analyzed 149 clinical features from 189 COVID-19 patients, identifying prognostic features via univariate Cox regression. The cohort was split into training and validation sets, and 77 prognostic models were developed using seven machine learning algorithms. Among these, the least absolute shrinkage and selection operator (Lasso) method was employed to refine the selection of prognostic variables by ten-fold cross-validation strategy, which were then integrated with random survival forests (RSF) to build a robust COVID-19-related prognostic model (CRM). Model accuracy was evaluated across training, validation, and entire cohorts. The diagnostic relevance of interleukin-10 (IL-10) was confirmed in bulk transcriptional data and validated at the single-cell level, where we also examined changes in cellular communication between mononuclear cells with differing IL-10 expression and other immune cells.
Results: Univariate Cox regression identified 43 prognostic features. Among the 77 machine learning models, the combination of Lasso and RSF produced the most robust CRM. This model consistently performed well across training, validation, and entire cohorts. IL-10 emerged as a key prognostic feature within the CRM, validated by single-cell transcriptional data. Transcriptome analysis confirmed the stable diagnostic value of IL-10, with mononuclear cells identified as the primary IL-10 source. Moreover, differential IL-10 expression in these cells was linked to altered cellular communication in the COVID-19 immune microenvironment.
Conclusion: The CRM provides accurate prognostic predictions for COVID-19 patients. Additionally, the study underscores the importance of early IL-10 level testing upon hospital admission, which could inform therapeutic strategies.
Keywords: IL-10; SARS-CoV-2; clinical decision making; machine learning; therapeutic guidance.
. 2024 Sep 2:17:5923-5942.
doi: 10.2147/JIR.S472099. eCollection 2024. A Multi-Machine Learning Consensus Model Based on Clinical Features Reveals That Interleukin-10 Derived from Monocytes Leads to a Poor Prognosis in Patients with Coronavirus Disease-2019
Jing Yu[SUP] 1 [/SUP], Yike Zhang[SUP] 1 [/SUP], Zhixiong Ye[SUP] 1 [/SUP], Kun Tang[SUP] 1 [/SUP], Yiming Ma[SUP] 1 [/SUP], Linlin Fu[SUP] 1 [/SUP], Tongtong Cui[SUP] 1 [/SUP], Hening Kang[SUP] 2 [/SUP], Yadong Yuan[SUP] 1 [/SUP], Wensen Pan[SUP] 1 [/SUP]
Affiliations
- PMID: 39247837
- PMCID: PMC11378990
- DOI: 10.2147/JIR.S472099
Background: Despite ongoing interventions, SARS-CoV-2 continues to cause significant global morbidity and mortality. Early diagnosis and intervention are crucial for effective clinical management. However, prognostic features based on transcriptional data have shown limited effectiveness, highlighting the need for more precise biomarkers to improve COVID-19 treatment outcomes.
Methods: We retrospectively analyzed 149 clinical features from 189 COVID-19 patients, identifying prognostic features via univariate Cox regression. The cohort was split into training and validation sets, and 77 prognostic models were developed using seven machine learning algorithms. Among these, the least absolute shrinkage and selection operator (Lasso) method was employed to refine the selection of prognostic variables by ten-fold cross-validation strategy, which were then integrated with random survival forests (RSF) to build a robust COVID-19-related prognostic model (CRM). Model accuracy was evaluated across training, validation, and entire cohorts. The diagnostic relevance of interleukin-10 (IL-10) was confirmed in bulk transcriptional data and validated at the single-cell level, where we also examined changes in cellular communication between mononuclear cells with differing IL-10 expression and other immune cells.
Results: Univariate Cox regression identified 43 prognostic features. Among the 77 machine learning models, the combination of Lasso and RSF produced the most robust CRM. This model consistently performed well across training, validation, and entire cohorts. IL-10 emerged as a key prognostic feature within the CRM, validated by single-cell transcriptional data. Transcriptome analysis confirmed the stable diagnostic value of IL-10, with mononuclear cells identified as the primary IL-10 source. Moreover, differential IL-10 expression in these cells was linked to altered cellular communication in the COVID-19 immune microenvironment.
Conclusion: The CRM provides accurate prognostic predictions for COVID-19 patients. Additionally, the study underscores the importance of early IL-10 level testing upon hospital admission, which could inform therapeutic strategies.
Keywords: IL-10; SARS-CoV-2; clinical decision making; machine learning; therapeutic guidance.