tetano
Editor, Senior Moderator
PeerJ
. 2026 Jul 9:14:e21417.
doi: 10.7717/peerj.21417. eCollection 2026.
Development and validation of a model for early prediction of severe/critical COVID-19 in elderly patients
Jiaxuan Li[SUP] #[/SUP][SUP] 1 2 [/SUP], Ruining Li[SUP] #[/SUP][SUP] 1 2 [/SUP], Chang Hong[SUP] 1 2 [/SUP], Richeng Mao[SUP] 3 [/SUP], Lushan Xiao[SUP] 2 [/SUP], Ziyong Zhang[SUP] 1 2 [/SUP], Min Ding[SUP] 2 [/SUP], Xuejing Zou[SUP] 2 [/SUP], Li Liu[SUP] 1 2 [/SUP]
Affiliations
Background: The mortality rate of severe/critical coronavirus disease (COVID-19) is high in the elderly, and early prediction of its prognosis can facilitate timely treatment and reduce mortality. This study aims to identify early predictors of severe COVID-19 in elderly and construct a validated risk prediction model.
Methods: This retrospective study included 722 elderly COVID-19 patients (those aged ≥60) who attended Nanfang Hospital of Southern Medical University between July 2022 and November 2023. They were categorized as mild/moderate or severe/critical according to the extent of their condition during hospitalization. Predictive models were constructed using logistic regression analysis and visualized using nomograms. Receiver operating characteristic (ROC) curves were used to assess the model's accuracy and predictive value. An external validation cohort containing 1,249 elderly COVID-19 patients who were admitted to Huashan Hospital of Fudan University between March and May 2022 was also collected.
Results: In multivariable logistic regression analysis, respiratory rate, comorbid diabetes, C-reactive protein (CRP), lymphocyte percentage, and D-dimer were independently associated with severe and critical COVID-19. Based on these findings, the final severity prediction model was constructed using three laboratory markers: CRP, lymphocyte percentage, and D-dimer. This model achieved an area under the curve (AUC) of 0.753 (0.713-0.794). For mortality prediction, CRP and D-dimer emerged as the significant independent predictors; the model showed an AUC of 0.722 (0.653-0.791) in the internal validation cohort and 0.877 (0.833-0.921) in the external validation cohort.
Conclusions: The predictive model incorporating features selected via logistic regression accurately predicts prognosis of severe COVID-19 in elderly, facilitating the implementation of early clinical interventions.
Keywords: COVID-19; Logistic regression; Personalized medicine; Predictive model; Risk stratification.
. 2026 Jul 9:14:e21417.
doi: 10.7717/peerj.21417. eCollection 2026.
Development and validation of a model for early prediction of severe/critical COVID-19 in elderly patients
Jiaxuan Li[SUP] #[/SUP][SUP] 1 2 [/SUP], Ruining Li[SUP] #[/SUP][SUP] 1 2 [/SUP], Chang Hong[SUP] 1 2 [/SUP], Richeng Mao[SUP] 3 [/SUP], Lushan Xiao[SUP] 2 [/SUP], Ziyong Zhang[SUP] 1 2 [/SUP], Min Ding[SUP] 2 [/SUP], Xuejing Zou[SUP] 2 [/SUP], Li Liu[SUP] 1 2 [/SUP]
Affiliations
- PMID: 42438710
- PMCID: PMC13356827
- DOI: 10.7717/peerj.21417
Background: The mortality rate of severe/critical coronavirus disease (COVID-19) is high in the elderly, and early prediction of its prognosis can facilitate timely treatment and reduce mortality. This study aims to identify early predictors of severe COVID-19 in elderly and construct a validated risk prediction model.
Methods: This retrospective study included 722 elderly COVID-19 patients (those aged ≥60) who attended Nanfang Hospital of Southern Medical University between July 2022 and November 2023. They were categorized as mild/moderate or severe/critical according to the extent of their condition during hospitalization. Predictive models were constructed using logistic regression analysis and visualized using nomograms. Receiver operating characteristic (ROC) curves were used to assess the model's accuracy and predictive value. An external validation cohort containing 1,249 elderly COVID-19 patients who were admitted to Huashan Hospital of Fudan University between March and May 2022 was also collected.
Results: In multivariable logistic regression analysis, respiratory rate, comorbid diabetes, C-reactive protein (CRP), lymphocyte percentage, and D-dimer were independently associated with severe and critical COVID-19. Based on these findings, the final severity prediction model was constructed using three laboratory markers: CRP, lymphocyte percentage, and D-dimer. This model achieved an area under the curve (AUC) of 0.753 (0.713-0.794). For mortality prediction, CRP and D-dimer emerged as the significant independent predictors; the model showed an AUC of 0.722 (0.653-0.791) in the internal validation cohort and 0.877 (0.833-0.921) in the external validation cohort.
Conclusions: The predictive model incorporating features selected via logistic regression accurately predicts prognosis of severe COVID-19 in elderly, facilitating the implementation of early clinical interventions.
Keywords: COVID-19; Logistic regression; Personalized medicine; Predictive model; Risk stratification.