This was a retrospective cohort study, based on data obtained from an electronic chart review, exploring the association between prehospital hypercapnia and mortality. The study protocol was approved by the institutional ethics committee of Geneva, Switzerland (project ID 2019–01559, amended 12.11.2019). Patient consent was waived by this committee.
The methods are similar to those used in a previous work , the aim of which was to determine the presence of an association between prehospital hypercapnia and admission in intensive care or high-dependency units. Thus, even though the population and outcomes are different, the methods used to collect the data are analogous.
This study was carried out using data from the prehospital medical mobile unit (called SMUR for Service Mobile d’Urgence et de Réanimation) of the Geneva University Hospitals in Switzerland. This covers a population of 500′000 persons. This physician-staffed prehospital system has already been described in prior studies [14, 23]. Briefly, a SMUR unit is called whenever a critical situation, such as severe AHF, is identified. These units can either be dispatched by the emergency medical call center or requested by the paramedics already on site. The paramedics and physicians who staff this unit are trained to provide prehospital NIV (defined as application of bilevel positive airway pressure) and perform arterial blood gas (ABG) analysis . Even though venous blood gas analysis could also be considered to assess capnia , the current prehospital treatment protocol requires arterial blood to be drawn if prehospital physicians decide to perform an analysis of blood gases. In our system, ABG analysis is strongly encouraged whenever NIV is initiated in the prehospital setting and physicians are only allowed one attempt at drawing arterial blood, in one minute at most.
Electronic charts of all patients for whom an intervention took place between July 1, 2016 to January 31, 2020 were screened for inclusion. Prehospital interventions performed after this latter date were excluded to avoid any potential bias that could have been introduced by the COVID-19 pandemic.
Only 126 different diagnoses can be coded in the SMUR’s prehospital medical charts, and only two relate to AHF: “acute pulmonary edema” (APE) and “heart failure” (HF). All patients 18 years or older in whom any of these two diagnoses was coded were screened for inclusion. Patients for whom a diagnosis of APE was coded were automatically included. All patients with a diagnosis of HF were then manually screened to include patients who did present with an acute component to their HF. The ESC definition of AHF i.e., “a rapid onset or worsening of HF symptoms threatening life” , was used to determine whether an acute component was present. Any uncertainty regarding diagnosis and inclusion of a patient was settled by consensus between MF and two of the other authors (CAF and LS). All patients for whom prehospital arterial ABG could not be obtained were excluded. The other exclusion criteria were cardiac arrest before or upon SMUR team’s arrival, presence of a mixed diagnosis (such as a concomitant acute chronic obstructive pulmonary disease exacerbation) described by the physician in charge, secondary transfer from another hospital or emergency care structure, and patients who were left on site without transportation to a hospital (and for whom information about mortality was therefore unavailable).
All electronically recorded relevant data were automatically extracted to a Microsoft Excel file (Microsoft Corporation, Redmond, Washington, USA). These data were then imported to an electronic case report form (CRF) created under REDCap (Vanderbilt University, Nashville, Tennessee, USA). All of the data which could not be automatically extracted were manually retrieved and entered into this CRF.
Exposure and outcomes
Hypercapnia (defined as a PaCO2 equal to or higher than 6.0 kPa or 45 mmHg) was the main predictor. The primary outcome was in-hospital mortality. Secondary outcomes were 7-day mortality and ER length of stay (LOS).
Week-end interventions were those performed on Saturdays and Sundays. Night interventions occurred between 7 PM and 7 AM. Length of intervention was the time elapsed between arrival on site and arrival at the hospital.
All statistical analyses were performed using Stata 16 (StataCorp LLC, College Station, Texas, USA). Values are presented as mean and standard deviation (SD) or median and interquartile range (IQR), based on visual inspection of the normality of the distribution. Patients with and without hypercapnia were compared using either the Student t-test, the Wilcoxon–Mann–Whitney test or the Chi2 test, as appropriate. The outcomes for each group were then tabulated. For the primary outcome, univariable logistic regression was performed to compute the crude odds ratio (OR) for the association between hypercapnia and in-hospital mortality. Exploratory multivariable analyses were then performed, adjusting for pre-specified potential confounders, and respecting a ratio of 7-to-10 events per degree of freedom. Confounders were selected based on the literature and physiological plausibility. The adjusted OR with its 95% confidence interval is reported. Finally, the crude association between PaCO2 and in-hospital mortality was modelled using restricted-cubic splines and this result was presented graphically. For secondary outcomes, the outcomes were also tabulated. Associations were then tested using a Chi  test for the 7-day mortality and the Wilcoxon-Mann-Whitney test for the ER LOS, as this variable was positively skewed. For all tests, a p-value lower than 5% was considered statistically significant. Exploratory analyses looking at effect modification by sex were performed using stratification and computing the Mantel–Haenszel statistic.
As the proportion of missing data was extremely low (< 1%) for all vital signs but diastolic blood pressure (29%), lack of data was deemed random and multiple imputation by chained equations was therefore performed, leading to the creation of 10 data sets. All baseline characteristics were used in the imputation model. When not specifically detailed in the medical charts, the Glasgow Coma Scale (GCS) score was assumed to be 15. For the secondary mortality outcome, patients who were discharged alive before 7 days were considered alive at day 7 if no other information regarding their vital status was available on their record.
Assuming an overall mortality of 15%, it was estimated that 192 patients would be sufficient to show a 3-fold increase in mortality in the hypercapnic group with a power of 80% and an alpha error of 5%.