Physiologic dead space is independently associated with mortality and discharge of mechanically ventilated patients with COVID-19 ARDS: a retrospective study

Study design

Retrospective, multicentre, and observational study. Data were collected from the MargheritaTre database²⁹, an electronic health record (EHR) developed by the Italian Group for the Evaluation of Interventions in Intensive Care (GiViTI) and the Mario Negri Institute for Pharmacological Research, with the objectives of supporting intensive care practitioners in everyday clinical activities and collecting high-quality data for research purposes. The study was conducted according to the Declaration of Helsinki. The experimental protocol received the approval from the Ethics Committee of the Coordinating Center, Comitato Etico Indipendente di Area Vasta Emilia Centro, CE-VAC (protocol code 17164), and each local ethics committee of the hospitals adopting the MargheritaTre software (Comitato Etico ATS Sardegna, Comitato Etico Regionale della Liguria, Comitato Etico Interaziendale A.S.O. SS. Antonio e Biagio e C. Arrigo di Alessandria, Comitato Etico dell’Insubria, Comitato Etico Brianza, Comitato Etico Regionale per la Sperimentazione Clinica della Regione Toscana Area Vasta Centro, Comitato Etico Regionale per la Sperimentazione Clinica della Regione Toscana Area Vasta Nord Ovest, Comitato Etico di Brescia, Comitato Etico Area Pavia-Policlinico San Matteo, Comitato Etico Fondazione IRCCS Istituto Nazionale dei Tumori, Comitato Etico per la Sperimentazione Clinica delle Province di Treviso e Belluno). The informed consent was collected in agreement with national regulations.

Study population

All adult patients (older than 17 years old) with COVID-19-associated ARDS admitted to ICUs from different Italian hospitals between February 21st, 2020 and March 31st, 2021 were enrolled in this study. Infection from acute respiratory syndrome coronavirus 2 (SARS-CoV-2) was confirmed by reverse transcriptase-polymerase chain reaction (RT-PCR) tests on naso-pharyngeal swabs or lower respiratory tract aspirates.

Data collection

Age, sex, and comorbidities upon admission to the ICU were recorded for each patient. In addition to baseline data, the following time series were extracted from the MargheritaTre database: systolic blood pressure (SBP, in mmHg), set or measured tidal volume (V\(_{\textrm{T}}\), in mL), positive end-expiratory pressure (PEEP, in cmH\(_{2}\)O), PaO\(_{2}\)/FiO\(_{2}\) ratio as the ratio of arterial oxygen partial pressure (PaO\(_{2}\), in mmHg) to fractional inspired oxygen (FiO\(_{2}\)), and physiologic dead space (DS). Specifically, physiologic dead space was estimated using the Enghoff’s modification of the Bohr’s Equation³⁰ as \(DS=(PaCO_2-EtCO_2)/PaCO_2\), where PaCO\(_{2}\) is the arterial partial pressure of carbon dioxide (in mmHg) and EtCO\(_{2}\) is the end-tidal partial pressure of carbon dioxide (in mmHg). For time series of physiological data (SBP, V\(_{\textrm{T}}\), PEEP, PaO\(_{2}\)/FiO\(_{2}\), EtCO\(_{2}\), PaCO\(_{2}\)) only sets of observations recorded in the same 1-h time window throughout the ICU stay (since admission to the last available data) were considered valid. Multiple observations recorded in the same hour window were averaged. Patients belonging to the same intensive care unit were clustered to account for possible differences across centers, which may depend on local hospital policies and resources (e.g. breathing circuit configurations, EtCO\(_{2}\) monitors, sedation practices, among others).

Objectives and outcomes

The main aim of the study was to investigate the association between physiologic dead space and ICU mortality or ICU discharge within 7 days since the last available measure. Figure 3 graphically depicts the study design.

Statistical analysis

For a single patient, admission to the ICU can result in two main competing outcomes: death or discharge from the unit. The competing risk modeling approach allows for correctly estimating the marginal probability of an event in presence of competing events. With this perspective, we tested a “Competing Risk Extended Cox Proportional Hazard Model for Time-Dependent Covariates”³¹ to determine the association between the temporal evolution of pulmonary dead space measures and both the patients’ risk of death or probability of discharge. A further description of the proposed modeling approach is provided in the Supplementary material.

Model characteristics

The proposed model contains the following features: PaO\(_{2}\)/FiO\(_{2}\) ratio, DS, PEEP, V\(_{\textrm{T}}\), SBP, age, sex, presence of diabetes, presence of chronic obstructive pulmonary disease (COPD).

Age and sex were included in the model as they proved to be significant predictors of death in patients with COVID-19-associated ARDS²⁶, together with diabetes and COPD³², which are highly prevalent conditions associated with pre-existing endothelial damage and pulmonary dysfunction, respectively. SBP was included to correct for the global hemodynamic status. The set or measured tidal volume is a major determinant of minute ventilation and CO\(_{2}\) elimination, while PEEP may influence the venous admixture and affect lung recruitment or overdistension²⁵. V\(_{\textrm{T}}\) and PEEP were used to correct for the effect of ongoing ventilatory support; however, they may also be proxies of the severity of disease, with higher PEEP or lower set V\(_{\textrm{T}}\) being used in more critical patients.

PaO\(_{2}\)/FiO\(_{2}\) and DS are widely-recognized predictors of outcome in ARDS^10,33 and were included to account for the severity of impairment in pulmonary gas exchange arising from intrapulmonary shunt, ventilation/perfusion mismatch and diffusion impairment.

The inclusion of DS was also justified by the observed significant improvement in the performance of the model, as well as by the significant role of DS in predicting both death and discharge, with a relative importance comparable to PaO\(_{2}\)/FiO\(_{2}\) (especially in patients experiencing ICU discharge). These observations strengthened the significant role of DS when correcting for the other confounding variables.

Observations were clustered according to the corresponding patient and ICU. The proportional hazards assumption was tested for all the variables included in the model with a significance level of 0.05.

The likelihood ratio test was used to evaluate whether the inclusion of DS led to a significant improvement in the model performances. The same model was also fitted after normalizing the variables with a Z-score transformation (i.e., subtracting the average value and diving by the standard deviation), in order to have an estimate of the feature importance from the obtained coefficients and hazard ratios, and to provide a more quantitative interpretation of the Hazard Ratios. Categorical variables are reported as frequencies (percentages) and continuous variables as means (with standard deviations, SDs) or medians (with interquartile ranges, IQRs) as appropriate. PostgreSQL 9.5.25, Python 3.8.10, and R 3.6.3 softwares were used to perform the analysis.