CORRESPONDENCE| VOLUME 7, ISSUE 10, PE31, OCTOBER 01, 2019

Secondary re-analysis of the FEAST trial – Authors' reply

Michael Levin, Aubrey J Cunnington, Clive J Hoggart

Lancet Respir Med 2019; 7: e31 DOI:https://doi.org/10.1016/S2213-2600(19)30264-4

Kathyrn Maitland and colleagues have voiced concerns about the difference between findings in our re-analysis of the FEAST trial¹ and their 2013 report² on which one of us (ML) was also an author. The 2013 study was a rigorous effort to explain the increased mortality associated with bolus fluids in FEAST, which found that a composite of shock or severe acidosis was the most common presenting clinical syndrome and cardiovascular collapse was the most common terminal clinical event (TCE), both of which were associated with the excess mortality in bolus recipients.² These conclusions differed from those in our analysis, which found that bolus improved cardiovascular function but produced six adverse physiological changes: worsening respiratory function, worsening neurological function or intracranial pressure, lower haemoglobin, lower bicarbonate, larger base deficit, and increased chloride.¹ We believe there might be two reasons for this. First, the 2013 study included lactate concentrations greater than 5 mmol/L in the definition of severe shock.² Our new analysis showed that lactate was negatively correlated with haemoglobin,¹ and thus elevated lactate identified patients with severe anaemia, who were incorrectly (inadvertently) considered to have cardiovascular shock in the 2013 study. Second, the TCE definition of cardiogenic or cardiovascular collapse at the point of demise (severe tachycardia or bradycardia plus one of prolonged capillary refill time >2 s, cold peripheries, or low blood pressure) also specified that when both shock and respiratory failure were present, the review panel had to choose either event as the sole TCE. This choice is likely to have been influenced by the earlier assignment of presenting syndrome. For example, a patient assigned as having shock or severe acidosis at presentation, who continued to deteriorate, was unlikely to be reassigned to a different TCE even if hypoxia was present as well as extreme tachycardia or bradycardia pre-terminally. Furthermore, severe tachycardia or bradycardia and poor perfusion are almost inevitable before cessation of cardiac activity (which defines death), and do not necessarily imply that worsening cardiovascular function was the preceding cause. Our analysis describes the physiological changes that are likely to set individuals on the course to their terminal clinical events, rather than findings nearest the terminal event.

Matteo Quartagno and colleagues state that major flaws in our analysis invalidate the conclusions that “hyperchloraemic acidosis and respiratory and neurological dysfunction induced by saline or albumin bolus explained the excess mortality due to bolus”. The quoted phrase is reported in the Summary as a finding in Cox survival models, not a conclusion, and as stated in the manuscript the main purpose of the study was to describe the physiological effects of fluid bolus on organ systems, haematological, biochemical, and acid-base parameters¹. The main conclusion of the study is based on the description of these changes, which we believe would be considered universally by clinicians to be detrimental to health, and in combination they are likely to be more detrimental. We have clearly explained that the Cox models are part of a post-hoc exploratory analysis of whether these changes could explain the excess mortality. The findings of these models are not the central conclusions of the paper, but do suggest that the combination of changes induced by bolus could explain the increased mortality.

Maitland and colleagues and Quartagno and colleagues criticise our imputation of the early changes in plasma biochemistry. They believe our approach of allocating the same increase in chloride to each bolus recipient is statistically flawed, because the imputed concentrations at 1 h are confounded by the treatment assignment. However, the rationale for this approach is not purely statistical. Adding a fixed volume per kilogram of high chloride bolus solution to a fixed volume per kilogram of blood (which has a lower chloride concentration) causes a predictable rise in chloride concentration in the blood. Regardless of the initial chloride concentrations at baseline in FEAST participants (median [10–90th centiles] 104 [98–110·7] mmol/L) a bolus of albumin or saline (chloride concentrations 154 mmol/L) would cause similar increase in plasma chloride because both the volume of distribution (plasma volume plus extracellular water) and volume of infused fluid are related to bodyweight. Thus randomisation to fluid bolus is indeed analogous to randomisation to a fixed increase in chloride concentration in blood. The Stewart equation then predicts the ensuing changes in acid-base status.^1, 3 We agree that the statistical models that use these imputed values cannot provide proof of causation. However, our analysis, and the models we reported, do support a biologically plausible hypothesis on how bolus fluids increase mortality in FEAST, which was previously lacking. This hypothesis can now be tested in clinical trials.

We declare no competing interests.