Severe COVID-19 is characterized by immune dysregulation and an elevated risk of thromboembolic events, including pulmonary embolism (PE). Identifying dynamic immunological and coagulation markers that predict adverse outcomes in hospitalized patients remains clinically important. This prospective study aimed to evaluate associations between serial measurements of cytokines, coagulation parameters, cardiac injury biomarkers, and in-hospital all-cause mortality in patients with severe COVID-19 who were evaluated for suspected PE.
This was a prospective observational cohort study that enrolled 47 hospitalized adult patients with severe COVID-19 and clinical suspicion of pulmonary embolism. All patients underwent diagnostic workup for PE, and clinical data were collected at enrollment and on hospital day 5. The overall in-hospital mortality in the cohort was reported as 27.7% (13 of 47 patients).
Serum profiling included a panel of 15 cytokines measured at enrollment and on hospital day 5. Standard coagulation parameters and cardiac injury biomarkers were also assessed at the same time points. Confirmation of pulmonary embolism was performed by computed tomography pulmonary angiography (CTPA).
The investigators used univariable and multivariable binary logistic regression to identify predictors of in-hospital all-cause mortality. Model discrimination was evaluated using the area under the receiver operating characteristic curve (AUC). Internal validation of model performance and stability was performed by bootstrap resampling. Multiple-testing correction across the cytokine panel was addressed using the Benjamini-Hochberg false discovery rate procedure.
Pulmonary embolism was diagnosed by CTPA in 19 of 47 patients, representing 40.4% of the cohort. Thirteen patients (27.7%) died during hospitalization. These event rates frame the context for the biomarker analyses relating cytokine and coagulation dynamics to clinical outcomes.
On univariable analyses, higher standardized levels of day-5 interleukin-9 (IL-9) and IL-15 were associated with in-hospital mortality. However, when correcting for multiple comparisons across the 15-cytokine panel using the Benjamini-Hochberg procedure, neither IL-9 nor IL-15 retained statistical significance (reported q-values: IL-9 q=0.180; IL-15 q=0.210). These corrected q-values indicate that the observed univariable associations should be interpreted cautiously in the context of multiple testing.
In an age-adjusted multivariable model, day-5 IL-9 remained associated with mortality (adjusted odds ratio per 1-standard-deviation increase 3.331; 95% CI 1.289–8.608; P=0.013). The apparent discrimination of this age-adjusted model was moderate, with an AUC of 0.774 (95% CI 0.563–0.984; P=0.011). Despite these signals, internal validation via bootstrap resampling revealed marked coefficient instability, indicating that model estimates were not robust within this sample and may not generalize.
The authors report that elevated day-5 IL-9 levels may reflect an evolving risk state in patients with severe COVID-19. They stress that the association between day-5 IL-9 and in-hospital mortality was sensitive to multiple-testing correction and demonstrated instability on internal validation. Consequently, the study's findings are presented as exploratory rather than definitive.
The investigators note limitations inherent to the study design and sample size. The loss of statistical significance after false discovery rate correction across the cytokine panel, together with the marked coefficient instability observed on bootstrap validation, limits the strength of causal or prognostic inferences. The authors emphasize that their results require confirmation in larger cohorts with external validation to determine whether day-5 IL-9 or other immunologic or coagulation markers can reliably predict adverse outcomes in severe COVID-19.
In patients hospitalized with severe COVID-19 evaluated for suspected PE, serial cytokine profiling identified a signal for day-5 IL-9 related to in-hospital mortality in age-adjusted modeling, but the signal did not survive correction for multiple testing and showed model instability. These results highlight the potential value of dynamic biomarker assessment but also illustrate the methodological challenges (multiple comparisons, small sample effects, internal validation instability) that must be addressed before clinical application. External validation in larger, independent cohorts is necessary to determine whether IL-9 or other cytokines should be integrated into prognostic algorithms for severe COVID-19 accompanied by thromboembolic risk.