Indication Bias Explains Some of the Observed Increased Mortality Associated With Use of Prehospital Intravenous Fluids in a Pediatric Trauma Population.

Abstract

Traditionally, in both pediatric and adult trauma patients, management of hemorrhage and shock has included early rapid intravenous fluid (IVF) replacement at the scene or during transport to a definitive care facility. Because prehospital resuscitation can be considered as a lifesaving intervention, severely injured patients are more likely to receive IVF. Observational studies not adequately adjusting for this confounding by indication (indication bias) while evaluating the impact of prehospital IVF on mortality in clinically heterogeneous patient populations are likely to find an increased mortality associated with the use of prehospital IVF, an association that may be spurious even after traditional multivariable risk adjustment. Propensity scores can be used to mitigate the impact of this selection bias on the estimated effect. The authors hypothesized that the effect of IVF on mortality will differ based on whether propensity scores (based on a set of prehospital indications for IVF) are adjusted for in a multivariable outcome model. This was a retrospective cohort study of severely injured pediatric (<18 years) patients consecutively evaluated and treated between January 1, 2008, and June 30, 2011, at Oklahoma's only Level I pediatric trauma center. Patients were divided into those receiving 250mL or more (GE250 group) and those receiving less (LT250 group) of prehospital IVF based on area under curve (AUC) analysis (AUC=0.7, 95% confidence interval [CI]=0.6 to 0.80, sensitivity=0.81 and specificity=0.56). Propensity scores were used to minimize confounding by indication of the mortality estimate and were calculated based on measurable prehospital factors. Using Cox's regression to minimize survival bias, the independent effect of prehospital IVF on the risk of 30-day in-hospital mortality was evaluated with and without adjusting for the propensity to receive 250mL of prehospital IVF. A total of 482 patients met study criteria. Of these, 46.3% (223 of 449) were in the GE250 group. After adjusting for Injury Severity Score, presence of a severe head injury, shock, and a penetrating injury, all of which were significant predictors of mortality, receiving 250mL or more of prehospital IVF was significantly associated with an almost threefold increase in the risk of 30-day in-hospital mortality (hazard ratio [HR]= 2.96, 95% CI=1.1 to 8.2). However, further adjusting for the propensity to be in the GE250 group, in addition to the aforementioned variables, attenuated the effect estimate and resulted in a nonsignificant (p=0.3408), more precise association between prehospital IVF and mortality (HR= 1.9, 95% CI=0.6 to 6.6). Propensity-adjusted survival analysis suggests that the observed increased risk in mortality associated with use of prehospital IVF replacement may be a spurious association resulting from inadequate control of confounding by indication inherent in observational studies. In the absence of patient subgroup-specific results from well-controlled studies, IVF resuscitation should not be a reason to delay patient transport to a definitive care facility. Randomized trials evaluating the effect of prehospital fluids are warranted in the pediatric trauma population, as such studies have shown clinical significance in the adult trauma population.