Mortality risk over time after early fluid resuscitation in African children

The Fluid Expansion As Supportive Treatment (FEAST) randomised controlled trial in African children with shock found a 3.3% absolute increase in mortality at 48 h and a 3.4% increase by 28 days, in those given boluses of fluid (20–40 ml/kg over 1 h of 5% albumin or 0.9% saline) compared to a control group given maintenance fluid (4 ml/kg/hr 5% dextrose or others following national guidelines) [1]. It has been well documented that child mortality rates shortly following hospital admission in sub-Saharan Africa are high with, for example, 14% of in-hospital deaths occurring within 4 h of admission in a prospective study in Kenya [2]. However, how mortality risk changes over time post-admission in this setting has not been explored. Previous analyses to understand mortality risk in the FEAST trial focussed either on describing the mode of death [3] or on mortality prediction from baseline measures [4]. Currently, the mechanisms by which bolus increased mortality risk remain unclear [3]. Examining how and when mortality risk changed over time in those receiving and not receiving boluses could provide insights into the mechanisms and guide future research, including clinical trials. Different hypotheses could be that the excess mortality risk was restricted to the period around the bolus administration, meaning that the maximum mortality experienced by children receiving boluses was greater than controls. Alternatively, mortality risk could differ by the same amount throughout admission/follow-up, or differences in risk could predominantly be later during admission/follow-up. We therefore used flexible parametric models to characterise changes in mortality risk after randomisation and investigate whether, and for how long, any increased risk from receiving a bolus was maintained.

There were 3141 children randomised between albumin bolus, saline bolus, and control in FEAST (2097 and 1041 to the combined bolus and control groups, respectively). Median age was 24 months; 1795 (57%) had malaria (Table 1). By 28 days, there were 345 deaths (11%), and 87 (3%) children were lost to follow-up: 254/2097 deaths (12%) and 65/2097 lost (3%) in the bolus group and 91/1044 deaths (9%) and 22/1044 lost (2%) in the control group [1]. Adherence to fluid administration was high with 2086/2097 (99%) children randomised to the bolus groups receiving a bolus. Median (IQR) bolus fluid in the bolus groups received by 4 h was 20 (20, 40) ml/kg and was the same at 8 h; total fluids received in bolus groups at 4 h and 8 h were 38 (28, 43) ml/kg and 40 (30, 50) ml/kg, respectively. Total fluids received in the control group by 4 h and 8 h were median (IQR) 10 (6, 15) ml/kg and 10 (10, 26) ml/kg, respectively. In the control group, 7 (< 1%) children received a 20-ml/kg saline bolus within 4 h (Table 1) with all children receiving maintenance fluids at 4 ml/kg/hr.

The FEAST trial showed a 3.3% absolute increase in risk of mortality by 48 h (and a 3.4% increase by 28 days) for those in the bolus groups compared to the control group [1]. The amount of fluid given as a bolus and rate was modest, in contrast to current recommendations, and the majority of children received only one 20 ml/kg bolus given over 1 h. Here, we investigated temporal changes in absolute increased mortality risk post-admission. Whilst it could have been hypothesised that bolus administration directly increased immediate mortality risk compared to controls around the time that the bolus was given, in fact, we found that mortality risk in children in the bolus groups simply declined more slowly than in the control group (Fig. 2). Overall mortality risk was at its maximum at around 30 min post-randomisation in both groups, consistent with children been admitted when acutely sick and at high risk of mortality, but we found no evidence that the excess mortality risk from bolus observed overall in the trial was concentrated in the time period immediately after bolus administration, nor that the maximum mortality risk differed between bolus and control. However, the detrimental impact of receiving a bolus continued beyond this acute phase up to 4 days after randomisation. Thus, the main effect of boluses was to delay ‘normalisation of risk of mortality’ following hospital admission, such that mortality risk in children receiving boluses declined at a slower rate compared with those not receiving boluses. This was a consistent finding with all statistical approaches. This is also in agreement with previous reported additional analyses, and active clinical monitoring of solicited serious adverse event by sites and the blinded ERC indicating cardiovascular collapse was the main TCE which contributed most substantially to the excess mortality in the bolus groups compared to control rather than increasing the fluid overload events (i.e. raised intracranial pressure or respiratory TCE) [3].

The WHO continue to advise that for children with three features of impaired perfusion (WHO definition of shock: all 3 of (i) cold extremities AND (ii) capillary refill time MORE than 3 s AND (iii) weak and a fast pulse (definition of tachycardia not stated)) 10–20 ml/kg over 30–60 min, and a further 10 ml/kg over 30 min repeating two times for uncorrected shock, should be given [9]. In the FEAST trial, which assent/consent processes permitted enrolment of very sick children, only a very small number had WHO shock (n = 65). In this sub-group, 48% of bolus recipients died compared to 20% of non-bolus control group with an increased absolute risk of 28% and relative risk increase of 240% (p = 0.07 (two-sided Fisher’s exact test)) [10]. A recent appraisal of published and unpublished literature to determine the frequency of WHO shock in unselected paediatric cohorts admitted to hospital showed that WHO shock criteria were present in very few cases (101/91,952 (~ 0.1%)) and overall these children have a very high mortality (up to 100%) [11]. In the light of these recommendations, the findings of the current analysis, demonstrating that a conservative fluid bolus volume given over a longer period of time increased the risk of mortality for up to 4 days post-bolus, is of concern.

There has been intense speculation around the mechanisms underlying the excess mortality from fluid-bolus therapy in the FEAST trial [12]. One hypothesis we suggested was a ‘re-perfusion’ injury where fluid resuscitation acts a ‘microcirculatory flush’ mechanism, mobilising products of catabolism, pro-inflammatory peptides, and toxins, and resulting in multiple downstream effects including myocardial ‘stunning’, organ damage, and poor subsequent outcomes [3, 13]. Our analyses support this hypothesis by showing that mortality risk in the bolus group declined more slowly; the hypoperfused state, involving both macro- and microvascular systems, may thus be serving as a protective physiological response, preventing a vascular systemic spread of pathogens or toxins, which subsequently occurred with re-perfusion of ‘global’ ischaemia [14, 15].

This would also support our previous findings that the adverse effects of bolus fluid resuscitation were mediated via their impact on vascular haemodynamics and myocardial performance [1, 3]. In an ovine endotoxin model of sepsis, which investigated a two-hit model (endotoxin and saline bolus therapy), we reported marked increases in cardiac troponin, a marker of myocyte death, at 16 h in the sepsis group resuscitated with saline [16]. In addition, Atrial Natriuretic Protein (a marker of atrial ‘stretch’) rose significantly in the fluid-resuscitated group, and median ANP was significantly higher than in the no resuscitation group (p = 0.02) over the cross-sectional time series. This was concurrent with an increased vasopressor requirement in the fluid-resuscitated group and increased levels of glycocalyx glycosaminoglycan hyaluronan (a marker of glycocalyx damage of the microcirculation). This aligns with our observation that the largest difference in mortality risk in FEAST was seen in those that had a cardiogenic terminal clinical event (TCE) (i.e. cardiovascular collapse) and that decline in mortality risk from cardiogenic deaths was most delayed in the bolus group [3]. In a multi-centre open-label phase II randomised controlled trial evaluating slower versus rapid rehydration strategies in 122 Ugandan/Kenyan children with severe dehydration secondary to gastroenteritis, we also noted that ANP levels rose substantially by 8 h in both arms and the abnormally high levels of ANP persisted to day 7 [17]. Troponin I (a marker of cardiac muscle injury independent of volume status) also rose following fluid administration at 8 h and 24 h in both groups; however, values remained in normal ranges. Both these studies indicate that volume loading, irrespective of rate, may have adverse consequences on the cardiovascular physiology in severe illness. As this was not anticipated during the FEAST trial, we have not been able to explore biomarkers of cardiac damage post-fluid bolus from the FEAST biobank, as we did not store samples post-bolus administration. Risks of death from neurological and respiratory TCEs were much lower and declined slowly in both groups, whereas deaths from unknown/other TCE predominated very shortly after randomisation (and before completion of fluid bolus administration), thus being unlikely to be due to the detrimental effect of fluid overload.

The maximum mortality risk was similar in bolus and control groups, and occurred very close to randomisation, prior to completion of the majority of boluses. The screening and consent process in clinical trials usually prevents randomisation of children in a pre-terminal state, and this has been hypothesised to cause estimates of increasing risk shortly after randomisation when these type of models have been used in other applications, such as advanced HIV disease [18]. However, FEAST had an emergency assent process, so that critically ill children could be randomised quickly, and full consent obtained when the child had stabilised. Further, our sensitivity analyses suggested that the very early increases in estimated mortality risk prior to the maximum risk were likely to have been due to the statistical modelling process, rather than a clinical increase in risk during the first hour. The estimated mortality risk from the model fitted to the data is also consistent with an underlying high constant mortality risk across the first hour, as suggested by the simulated data, which is clinically feasible.

In the FEAST trial, standard supportive treatments included oxygen therapy, rapid correction of hypoglycaemia, and anti-pyretics; to comply with surviving sepsis recommendations, all children received immediate treatment with parenteral antibiotics (i.e. within the first hour) [19]. The high early mortality raises concern as to whether other aspects of early supportive care may have adverse consequences. Although discussed in the literature [20], prospective evaluation will be challenging as much of early supportive care is an intrinsic part of clinical practice. Literature on toxin release by bactericidal antibiotics, including the Jarisch-Herxheimer reaction and pyrexia control, remains both controversial and debatable topics. Few trials have been prospectively designed to address the possible adverse consequences of early supportive care in the critically sick patient [21].

The main study strength is its novel application of flexible parametric survival models, enabling us to further elucidate potential mechanisms. To our knowledge, no studies have applied such models to critically ill children in resource-limited settings, and our application highlights their potential. By necessity, however, one limitation is that it is a post hoc retrospective analysis of prospectively collected trial data. Other study limitations include the lack of detailed data on causes of death and the absence of longitudinal stored samples for the exploration of explanatory hypotheses. Further, there may be some minor inaccuracies in time of death (recorded to the nearest minute); however, all children were managed by a dedicated trial team, a dedicated trial clock was provided to ensure accuracy of timing of observations, and both randomised groups were monitored closely, for example, for signs of fluid overload or need for other supportive treatments.

This more detailed analysis of the FEAST trial data demonstrates that fluid boluses did not substantially increase mortality risk around the time of administration, but rather delayed the reduction of mortality risk for at least 4 days, preventing mortality risk declining at the same rate as it would without a bolus.