We planned to include a total of 1071 patients (357 per group) to have a statistical power of 80% to demonstrate, with a two-sided test at 0.025 level (to consider the two comparisons by applying a Bonferroni correction), a difference of 10% in 28-day cumulative incidence of first episodes of VA-LRTI in favor of SARS-CoV-2 pneumonia group against Influenza pneumonia or no viral infection groups. The sample size was calculated using PASS 12 (Logrank Test Accounting for Competing Risks) [19] on the basis of an expected rate of 20% in the two non-SARS-CoV-2 groups and 10% in SARS-CoV-2 group and an expected competing event rate of 30% in non-SARS-CoV-2 groups and 50% in SARS-CoV-2 group [4].

Quantitative variables were expressed as medians (interquartile range) and categorical variables were expressed as numbers (percentage). Patient characteristics at ICU admission and during ICU stay were described according to study groups (SARS-CoV-2 pneumonia vs. Influenza pneumonia vs. no viral infection) without formal statistical comparisons.

The 28-day cumulative incidence of first episodes of VA-LRTI (VA-LRTI, VAT, VAP) were estimated using the Kalbfleisch and Prentice method [20], considering extubation within 28-day (dead or alive) as a competing event. For VAT and VAP incidence, occurrence of VAP and VAT was respectively treated as a competing event, in addition to extubation. Comparison of the incidence of first episodes of VA-LRTI between study groups was done using Gray’ test to consider competing events. Subhazard ratios (and their 95% confidence intervals (CIs)) associated with SARS-CoV-2 pneumonia, against each other group, were calculated using univariable Fine-and Gray models as effect sizes. Comparisons were further adjusted for pre-specified confounders known to be associated with VA-LRTI incidence (age, gender, SAPS II, MacCabe classification, Immunosuppression, recent hospitalization, recent antibiotic, and ARDS) [8] by using multivariable Fine-and Gray models. Regarding the causal relationship, we also assessed the association of VA-LRTI with SARS-CoV-2 pneumonia using univariable and multivariable cause-specific Cox’s proportional hazard models.

To avoid case-deletion in multivariate analyses due to presence of missing data in covariates, multivariable Fine-and Gray models, and cause-specific Cox’s, were performed after handling missing data on patient’s characteristics at ICU admission using multiple imputation procedure [21]. Imputation procedure was performed using regression switching approach (chained equations with m = 20 imputations obtained) under the missing at random assumption by considering outcomes (event status and log of event time) and all baseline characteristics (see ESM Table 1 for missing data pattern and see ESM Table 2 for baseline data in complete and incomplete cases) in imputed models. Predictive mean matching method, and logistic regression model (binary, ordinal or multinomial) were used for quantitative, and categorical variables; respectively. We used 20 imputations, allowing a maximal fraction of missing information (FMI)/m < 0.1 in all analyses (see ESM Table 3). Estimates obtained in the different imputed data sets were combined using Rubin’s rules [22].

Data were analyzed using the SAS software package, release 9.4 (SAS Institute, Cary, NC).

Note: The content above has been extracted from a research article, so it may not display correctly.

Please log in to submit your questions online.
Your question will be posted on the Bio-101 website. We will send your questions to the authors of this protocol and Bio-protocol community members who are experienced with this method. you will be informed using the email address associated with your Bio-protocol account.

We use cookies on this site to enhance your user experience. By using our website, you are agreeing to allow the storage of cookies on your computer.