Association between prehospital arterial hypercapnia and mortality in acute heart failure: a retrospective cohort study

Acute Heart Failure (AHF) is a common and life-threatening medical condition which requires urgent evaluation and management and is associated with a high burden of morbidity and mortality [1, 2]. In Switzerland, AHF led to the death or to the readmission of 8120 patients in 2005 [3]. The in-hospital mortality varies from 4 to 11%, and readmission within the next 3 months is frequent, with rates varying between 9 and 30% [4‐10].

The treatment of AHF includes vasodilators medication, as well as diuretics and oxygen [1]. Positive end-expiratory pressure, with or without pressure support, has also been shown to be beneficial in case of severe AHF [11‐13]. In the prehospital setting, NIV has been shown to reduce intubation rate in AHF [14]. It also seems to alleviate symptoms in the emergency room (ER) [15].

Some factors negatively affect the prognosis of AHF patients including low systolic arterial blood pressure [16], ischemic or parainfectious etiologies [17], renal dysfunction, and polymorbidity [9]. Other negative factors have been described like glucose variation during hospitalization [18], and multiples scores have been created to estimate the AHF-associated prognosis by taking these elements into account [19, 20].

Hypercapnia is common in AHF, with rates varying between 22 and 30% in the in-hospital and emergency setting [21, 22], and reaching up to 58% in the prehospital setting [23]. In this latter setting, hypercapnia has been shown to be associated with a higher rate of intensive care and high-dependent unit admission [23]. This might be linked to the increased need for non-invasive ventilation (NIV) and endotracheal intubation (ETI) described in AHF patients who are hypercapnic at hospital admission [21].

An association between hypercapnia and in-hospital mortality cannot be ruled out, but there is currently little data to support it. The goal of this study was to determine whether prehospital hypercapnia was associated with mortality in AHF.

The SMUR provided care to a total of 19,953 patients during the study period, 797 of whom met the inclusion criteria. After excluding 572 patients (Fig. 1) 225 patients (28.2%) were included in the analysis. The main exclusion criteria was a lack of ABG (448 patients). The rationale for no ABG being performed was unknown in most cases (291/448, 65%). Documented reasons for absence of ABG were a failed attempt by physician or collection of a venous sample (78), technical failure of the analyzer (31), medical decision (45), and patient refusal [3].

Baseline patient characteristics are described in Table 1. Overall, patients had a mean age of 83.9 years (with 27.6% of them being 90 years or older), and 58.7% of them had prehospital hypercapnia. Patients without hypercapnia had a median PaCO₂ of 5.1 kPa (IQR: 4.7–5.6), compared to 7.4 kPa (IQR: 6.6–8.9) for patients with hypercapnia. Most baseline characteristics were otherwise similar between groups. Female patients, which represented 63.6% of the population, were more likely to be hypercapnic (69.7% versus 54.8%, p = 0.023). Hypercapnic patients had higher diastolic blood pressures (103.6 mmHg versus 93.7 mmHg, p = 0.036) and were more likely to have abnormal GCS scores (62.1% had a normal GCS versus 81.7%, p = 0.002). They were also more likely to be treated by NIV in the field (93.9% versus 74.2%, p < 0.001).

The overall in-hospital mortality was 12.9% and was significantly higher in patients with prehospital hypercapnia than in those without (17.4% versus 6.5%, p = 0.016), with an unadjusted OR of 3.06 (95%CI 1.19–7.85) (Table 2). After adjustment for history of chronic renal failure, chronic obstructive pulmonary disease and hypertension, the adjusted OR was 3.18 (95%CI 1.22–8.26) (Table 3). Figure 2 presents the association between in-hospital mortality and prehospital PaCO₂, and shows a dose-dependent association. Finally, although in-hospital mortality was slightly higher in male patients than in female patients (17.1% versus 10.5%, p = 0.156), the association between prehospital hypercapnia and the outcomes did not differ by sex (p = 0.810).

There was also a significant difference in 7-day mortality (13.6% vs 5.5%, p = 0.044), with an unadjusted OR of 2.68 (95%CI 1.69–4.24). The median ER length of stay was 6.0 h and was longer for patients without hypercapnia than for patients with hypercapnia (7.1 versus 5.6, p = 0.018) (Table 2).

This cohort study shows that prehospital hypercapnia is associated with an increase in in-hospital and 7-day mortality in patients with AHF. A “dose-dependent” association between the value of PaCO₂ and mortality was also observed.

Even though association does not equal causation, these results are strengthened by physiologic plausibility, a dose-dependent association, and similar results reported in in-hospital studies [21]. Indeed, a prospective cohort study of patients with acute respiratory distress syndrome reported a similar association between hypercapnia and in-hospital mortality [25]. In addition, animal models also support this link, as hypercapnia was found to be associated with a reduced probability of pulmonary wound repair in a rat model of ARDS [26].

Hypercapnic patients were more likely to present with higher systolic and diastolic blood pressures than normocapnic patients. Theoretically, higher blood pressures could potentially contribute to an increased risk of pulmonary edema and thus a decrease of CO₂ clearance and oxygen uptake, a pathophysiologic process that has already been theorized to explain hypercapnia in AHF [27]. The resulting hypoxemia could therefore contribute to the increased mortality rate. Another hypothesis is that higher blood pressure are found in sicker patients given their higher sympathetic stress.

This study reports a higher proportion of hypercapnic patients compared to in-hospital studies, as well as a higher rate of hypoxemic patient [21, 22]. Even though some initially hypercapnic patients might have normalized their PaCO₂ and PaO₂ after prehospital treatment, a selection bias cannot be ruled out. Indeed, the SMUR units are only called to attend to the sickest patients with AHF, while patients with milder symptoms can be managed by paramedics only. The subset of patients examined in the present study is rather limited, and most patients admitted in the ED for heart failure are not breathless at rest [28]. Nevertheless, patients with AHF taken care of by prehospital teams deserve special consideration as they are usually critically ill and therefore more likely to require advanced therapies. This theory is supported by the overall mortality rate which was also slightly higher than the rates reported in larger cohort studies [5, 9].

Normocapnic patients spent more time in the ER than hypercapnic patients, an effect already observed when only considering patients who were eligible for admission in an intensive care or high-dependency unit [23]. The most likely explanation is that given the worse clinical status of these patients, clinicians can rapidly decide upon admission disposition and therefore need less time and data to make their admission decision. Thus, prehospital PaCO₂ could be used as a tool to improve patient disposition and reduce the time spent in the ER. Further prospective trials, which could compare ER LOS depending on the systematic analysis of prehospital ABG, could be designed to investigate the clinical applicability of this finding.

This study has several limitations. First, as already mentioned above, SMUR units are probably called to help manage the sickest of AHF patients, thereby inducing potential selection bias. Another potential selection bias is related to the local AHF treatment protocol, which strongly recommends performing ABG analysis whenever NIV is considered, but also acknowledges that physicians should not waste time before initiating treatment. Therefore, several AHF patients who would otherwise have met the inclusion criteria could not be analysed because an ABG was not performed. Consequently, even though the population should not be markedly different, the risk of bias still exists. Another limitation is the lack of information regarding the fraction of inspired oxygen when ABG was performed. This potential confounder could not be assessed in our study given its unreliable reporting in the prehospital medical file. We were also prevented from determining the effect a normalization (or even a decrease) of PaCO₂ might have on mortality as in-hospital ABG analysis was missing in 33% of all analysed cases. Other potential variables of interest, such as volume status, left ventricular ejection fraction, and troponin were not reported as they were inconsistently recorded and as the high proportion of missing data would have prevented their interpretation. In addition, although the study was adequately powered to show a difference in mortality, the sample size was too small to adjust for many covariates. Therefore, only the most clinically relevant covariates were chosen for adjustment, and residual confounding resulting in overestimated or underestimated results cannot be excluded. Even though usual, prehospital and in-hospital medication could have influenced the patients’ outcome, such data was not straightforward to retrieve and its reliability would have been limited. We therefore refrained from extracting and including this data, which would have been difficult to analyse given the limited sample size. Even though there is little reason to believe that different treatments would have been given according to the prehospital PaCO₂ values, a bias linked to differences in the choice of medications cannot be formally excluded. Likewise, the analyses regarding effect modification by sex should be interpreted with caution given the small sample size. Finally, the generalization of these results is limited by the fact that this was a monocentric study performed in a supervised and physician-staffed prehospital system. Due to the exclusion of respiratory distress aetiologies other than AHF, our findings cannot be transposed to other populations.

Nevertheless, this study also has several strengths. First, the use of automatically extracted data guarantees, at least in part, its validity. Second, the originality of the analysis, including the use of restricted cubic splines to model the association between PaCO₂ and mortality in addition to the more traditional approach using a usual cut-off, improves the presentation and interpretability of the results. Indeed, binarization or categorization of continuous variables assumes that there is a discontinuity in response, which is extremely unlikely.

This study should encourage further research on the topic, and the actual clinical utility of prehospital ABG analysis should be assessed by means of a randomised controlled trial. Moreover, a multi-centre study with a bigger sample size would allow to adjust for more covariates. Finally, should the added value of prehospital ABG be confirmed, specific interventions could be devised to increase the utilization of the ABG in a pre-hospital setting.

Prehospital hypercapnia is associated with an increase in in-hospital and 7-day mortality in patients with AHF. Prospective randomized studies should be performed before systematic prehospital analysis can be recommended in these patients.