The findings of this post hoc analysis of a large cohort of well-defined ARDS patients suggest estimations of impaired ventilation on the second day of ventilation to have an independent association with mortality at 30 days. The estimated additive effects of the predictors are small, statistically significant but with limited improvement and our study did not suggest that one is better than another. Instead, these data suggest that all methods of dead space estimation have similar prognostic value. The added predictive accuracy on top of known predictors of outcome was not observed on the first day of invasive mechanical ventilation. Impaired ventilation does not seem to be associated with the severity of ARDS using the Berlin classification but rather adds an additional dimension to an estimation of the severity of ARDS.
We found that a significant number of ARDS patients had high estimated dead space fractions in the early course of ARDS. These findings confirm those from previous studies that have demonstrated that impaired ventilation can occur within hours after onset of ARDS [8,9,10, 24]. Dead space is the portion of each tidal volume that does not contribute to CO2 clearance and represents a good global index of lung function efficiency. In ARDS, elevated VD/VT results from increased V/Q heterogeneity due to injury of capillaries by thrombotic and inflammatory factors with obstruction of pulmonary blood in the pulmonary circulation and perhaps some degree of overdistention by mechanical ventilation. Shunt and low cardiac output states are known determinants of high VD/VT. However, these are features of impaired perfusion rather than ventilation impairment per se .
We did not find a statically significant association between PaO2/FiO2 recorded at the first day of ARDS and the primary outcome. Previous works in ARDS showed that PaO2/FiO2 may vary due to changes in ventilatory parameters during the first 24 h of mechanical ventilation. These observations may explain why we did not find an association with mortality on the first day in our cohort [24,25,26].
Our results showed statistically significant but limited improvement in the prediction for mortality for each of the estimated dead space fraction, including the VR. An elevated physiological dead space fraction, calculated with the Bohr–Enghoff equation through volumetric capnography, has been shown to be a strong independent predictor of mortality in the early and intermediate phase of ARDS . The first reports describing the independent association between dead space and mortality in ARDS used a Vt 10 ml/kg of IBW, which is an important factor for increased dead space due to regional pulmonary overdistention or reduced cardiac output . Our findings continue to show an association in mortality among the predictors studied in our cohort with Vt 6 ml/kg IBW, which may suggest intrinsic phenomena on ARDS pathophysiology or maybe other factors different from tidal volume overdistention, such as PEEP.
Dead space fraction was not included in the Berlin Definition. The consensus panel argued that because dead space fraction is challenging to measure, they choose to evaluate minute ventilation standardized at a PaCO2 of 40 mmHg as a surrogate instead [3, 27]. Addition of this surrogate resulted in the selection of a smaller group of patients and did not improve mortality stratification. Given the prognostic value of the hereby evaluated surrogates of dead space ventilation, we propose that ARDS patients could be categorized into those with failure of oxygenation and/or those with failure of ventilation. Such categories may facilitate a better understanding between those patients who suffer most from shunt or dead space ventilation. This differentiation could lead to a better understanding of pathophysiological processes and lead to more consistent management strategies. Previous studies have highlighted that the degree of impaired ventilation also depends on the etiology of ARDS , e.g., pulmonary ARDS has a significantly higher VD/VT than those with non-pulmonary ARDS. We did not test if there was a difference in the ventilation impairment measurements between pulmonary and extrapulmonary ARDS in our cohort.
The VR has been proposed as an alternative surrogate for dead space fraction. VR, which includes predicted minute ventilation and predicted PaCO2, had been described to monitor ventilatory efficiency . Recently, it was found to be independently associated with mortality in ARDS patients , and we confirmed this finding. In the current study, VR was independently associated with mortality on day 2, but not on day 1 of ARDS. The advantage of VR is that it is easier to calculate than the estimations of dead space fraction with a similar, but maybe slightly lower predictive accuracy for the outcome.
Important limitations of this study exist. Estimated methods for dead space fraction use equations for energy expenditure or physiological variables [11, 15, 17,18,19]. The Harris-Benedict estimate predicted the measured dead space fraction with moderate accuracy but was associated with mortality . Recently, the ventilatory ratio has been found to positively correlate with dead space in ARDS . However, the VR has been only validated in mandatory modes of ventilation. It is worth to know that CO2 production (V̇CO2) changes dramatically once patients are off sedation and in spontaneous modes of ventilation . Both the estimated VD/VT and VR are determined by V̇CO2 which is an important factor in CO2 homeostasis in critically ill patients and impacts any measure of impaired ventilation [30, 31].
In this study, we did not routinely use controlled ventilation, but spontaneous effort was allowed and likely, due to limited use of sedation. This certainly has led to differences in estimations of dead space fraction as compared to a scenario where all patients would have been under controlled ventilation, and could explain the lack of independent association with mortality in our study, except for the VD/VT phys which not include V̇CO2 measurements in its formula. In the future, the evaluation of the estimated measurements of impaired ventilation should incorporate a consideration of altered metabolism as it will impact the variation of CO2 production in ICU patients.
It has been described in previous works that the PaO2/FiO2 values during ARDS may vary and that the values measured during the first 24 h of mechanical ventilation are not as predictive of outcomes [24, 25, 32]. This may be explained by the fact that PaO2/FiO2 due to some causes can be improved by changing the mechanical ventilation settings, or due to rapid resolution of ARDS, perhaps because the underlying cause has been addressed effectively or because significant fluid overload was treated or prevented. These changes are typically made in the first 24 h leading to rapid improvement in oxygenation in those patients with reversible causes. We believe that this not only applies to oxygenation but also globally all gas exchange derangements, including ventilation impairment. We suspect that impaired ventilation and shunt that can be reversed effectively in the first 24 h does not contribute to mortality and therefore, the association between these parameters and outcome increases after the first day.
However, because we did still confirm the statistically significant but limited improvement in the predictive accuracy of the VR found in other studies, this is actually an advantage of the estimation of impaired ventilation as it increases the external validity of such measurement. Second, we were not able to quantify the correlation between VR and the estimated dead space fraction measurements, due to complexities secondary to mathematical coupling . Third, the association between the studied variables was attenuated after inclusion of the missing data. Because we were aware of this, we reported the lower effect size. On the other hand, the “true” predictive accuracy was statistically significantly higher but with limited improvement after imputation for missing data, as shown by the NRI and IDI, which estimate the added value of the VD/VT phys and the VR for the second day of ARDS on top of the base model. However, although the IDI and the NRI are used to measure and evaluating the improvement in prediction performance of a prognostic marker, we suggest that their results and interpretation should be treated with caution as the interpretation is debated.
Besides these limitations, our study has some strengths. We studied a large prospectively enrolled cohort and, therefore, had sufficient power to detect significant differences in the primary outcome. This was an observational study that included all consecutive patients who developed ARDS in two ICUs over a 3-year period, thereby increasing the generalizability of the findings of the study. Despite using different equations for estimating dead space ventilation fraction, they yielded similar results, suggesting validity of the findings. Furthermore, we used a predefined analysis in line with the latest recommendations for predictive research [33, 34].
The statistically significant but limited improvement of the added predictive value of the variables (dead space estimations and ventilatory ratio) shown in this study indicate that they may be useful bedside indices to monitor impaired ventilation in critically ill patients, and may offer clinicians information about ventilatory failure at the bedside. When direct measurements of dead space fraction are unavailable, dead space estimations may be used in place of direct measurement in clinical practice. Furthermore, the relative simplicity of calculation allows for the incorporation of dead space fraction surrogates in future clinical trials. Given the simplicity of ventilatory ratio to be easily calculated using routine bedside variables, albeit, with known inherent limitations, it may be used as a tool to adjunct mortality estimation in patients with ARDS on day 2.
We believe that rather than a value in itself to be applied in isolation, dead space and its estimates should be understood as a marker to assess ARDS severity and therefore, it could be used to build a model based on oxygenation abnormalities (PaO2/FiO2) and respiratory mechanics (PEEP) to track respiratory changes to prevent ventilator-induced lung injury or avoid further lung deterioration. Thus, a modified lung injury score may aid in decision-making therapies and risk stratification for future clinical trials.