Variability in usual care fluid resuscitation and risk-adjusted outcomes for mechanically ventilated patients in shock

In the early phases of resuscitation for undifferentiated shock, uncertainty regarding the underlying etiology and patient comorbidities are common reasons for non-patient-centered resuscitation practices, conflicting treatments, unwarranted practice variability, and failure to meet targeted end points for initial resuscitation [1]. A positive association between fluid balance and mortality has been demonstrated in observational studies [2‐8]. In smaller retrospective studies, however, comparisons of administered fluid volumes between resuscitation groups are frequently confounded by disease severity. Additionally, the extent and impact of variable fluid resuscitation decisions in common usual care practice is poorly characterized. Consequently, assessments of actual administered fluid volumes alone may not effectively address the clinical appropriateness of fluid resuscitation or the effects of relative under- and over-resuscitation on patient outcomes.

In a novel attempt to both characterize the nature and frequency of variable fluid administration practices in the initial phase of shock resuscitation and to adjust for confounding by disease severity, we conducted a retrospective outcome analysis of day 1 fluid (DOF) resuscitation in high-risk mechanically ventilated adult patients resuscitated for shock. We sought to determine the impact of variable DOF volumes on hospital mortality in the most severely ill patients, including those with undifferentiated shock and septic shock. Additionally, we sought to identify hospital- and patient-level contextual factors that contribute to irregular practice and evaluate whether these predict or explain differences in prescribed fluid volumes and hospital mortality. We hypothesized that mechanically ventilated patients with shock are frequently prescribed greater- or less-than-expected DOF volumes resulting in adverse impacts on standardized hospital mortality after risk adjustment.

In this retrospective outcome analysis using the 2013 Premier and 2013–2016 Truven databases, we found highly variable day 1 fluid resuscitation practices among mechanically ventilated patients with hemodynamic shock from diverse etiologies. Our results are consistent with prior observations associating over-resuscitation with adverse outcomes. However, we also found that lower-than-expected day 1 fluid resuscitation, while beneficial among patients with certain diagnoses predictive of receiving less fluids (e.g., heart failure, end-stage renal disease), was associated with increased risk-adjusted hospital mortality among a relatively healthier patient cohort without such diagnoses; a U-shaped relationship.

Initial fluid resuscitation practices for patients with hemodynamic shock remain controversial. Recently published data from a prospective randomized control pilot study of patients with severe sepsis and septic shock demonstrated no increase in adverse outcomes using a restrictive fluid resuscitation strategy in the first 72 h of care [11]. Larger multicenter trials comparing fluid-restrictive and fluid-liberal resuscitation are currently underway [12]. Nevertheless, the non-uniform response to fluid resuscitation between patients with and without an FRF observed in our study suggests widely inclusive fluid resuscitation strategies in usual care and outside of study protocols may not be appropriate for all patients. Moreover, while over-resuscitation is increasingly recognized as harmful to patients with shock [2‐8] and fluid-restrictive strategies are currently being evaluated in large multicenter trials, our findings stress that under-resuscitation may also be harmful particularly for a subgroup of patients without FRFs.

Only half of shock patients respond positively to a volume challenge, indicating that many may have effective circulating fluid volumes despite macrocirculatory shock [13]. Diagnoses identified as FRFs frequently include volume overload as a primary or secondary feature, potentially explaining the observed mortality benefit of receiving less-than-expected fluid volumes among patients with at least one of these conditions. Patients without an FRF were generally younger and had fewer comorbidities and acute organ dysfunctions present on admission. They also had lower predicted hospital mortality compared to patients with an FRF. Nonetheless, hospital mortality for patients without an FRF was increased with higher and lower fluid volumes and with greater- and less-than-expected fluid resuscitation. These findings challenge the assumption that younger and presumably healthier patients are more tolerant of a wide range of fluid resuscitation volumes. Likewise, we observed significantly lower ICU and hospital resource utilization and lower hospital costs among patients without an FRF despite higher-than-predicted risk-adjusted hospital mortality. This incongruity may reflect relatively less attentive care-titration for a population presumed to have lower risk for further clinical decompensation.

To our knowledge, this study represents the largest novel analysis of fluid resuscitation outcomes for a wide spectrum of shock states. We specifically focused on patients requiring mechanical ventilation as they have the highest comorbidity burden and risk for hospital mortality. To adjust for confounding by disease severity, we developed predicted mortality and predicted DOF models. By uniquely evaluating the impact of greater- or less-than-expected fluid resuscitation, we expand on previous studies limited to comparisons of absolute fluid volumes in septic shock alone [8, 14]. Furthermore, we validated our predictive models and findings in a second large database that includes patients discharged as recently as 2016 and, pertaining to septic shock, subsequent to the ProCESS [15], ARISE [16], and ProMISe [17] trials.

A uniform limitation of many administrative database analyses is the lack of patient-level severity of illness scoring. However, comorbidity-based risk-adjustment indices (e.g., Charlson index AUROC = 0.67) have been shown to perform similarly well as severity of illness scores (APACHE II AUROC = 0.805 and SAPS II AUROC = 0.843) in predicting short-term mortality following ICU admission [18‐20]. The discriminative ability of our comorbidity-based mortality model, while still dependent on ICD-9-CM-based coding instead of physiologic variables, compares favorably (AUROC = 0.80). In addition, the linear relationship between prescribed fluid volumes and predicted mortality observed in our model is consistent with prior observational studies suggesting a positive correlation between administered fluids and mortality. The range of predicted DOF accurately correlated with the range of observed DOF, and the ± 1.5-L DOF cutoff was validated by the continuous deviation of observed from predicted mortality rates both above and below this critical value. Nonetheless, we cannot exclude the possibility that variance with respect to predicted versus observed DOF may be related to the performance of our model. Thus, while acknowledging the limitation of the current administrative dataset analyses, future prospective studies should evaluate the temporal relationship between prescribed fluid volumes and mechanical ventilation or other physiologic variables. Such an approach would facilitate further categorization of disease severity, for example by computing Sequential Organ Assessment (SOFA) scores for patients with sepsis, and elucidate the causal relationship between suboptimal fluid resuscitation and adverse outcomes.

Additional potential for confounding arises in administrative database analyses from systematic underreporting by individual hospitals and absent documentation for specific patients. To mitigate this potential bias, we excluded hospitals where the 25th percentile for recorded DOF was below 250 mL and any cases where the DOF was less than 1 L (185 hospitals, 33%). The potential for ascertainment bias, however, exists as a result of excluding patients who may be admitted late in the calendar hospital day and hospitals with systematic underreporting of administered fluid volumes. With respect to identifying patients with shock, we attempted to address this by defining shock according to whether the subject was prescribed a vasopressor on hospital day 1 or 2. Furthermore, while isolating a higher risk population, patients with shock who require mechanical ventilation may represent a phenotypically distinct but heterogeneous subgroup. In separate analyses, we observed comparable associations in acutely resuscitated patients regardless of the need for ventilator support (unpublished data). Categorization of different shock states in administrative data is limited without access to patient-specific physiologic variables. For the purposes of the current analysis, we refer to all shock states as “undifferentiated shock” while acknowledging that bedside clinical assessment is likely to have informed a more precise etiology of shock that contributed to fluid resuscitation decisions. Finally, the present analysis does not distinguish between types of fluid administered that is the focus of ongoing studies. Though resuscitation with crystalloid rather than colloid fluids may have contributed to variability in DOF volumes [21], fluid type has not been consistently associated with differences in shock resuscitation outcomes [21‐23].

Our study focused on early, guideline-informed DOF administration in shock. Future analyses are planned to evaluate DOF volumes for patients with delayed versus direct ICU admission, hospitals where patients receive very small volumes of DOF, the impact of specific shock states, as well as outcomes related to fluid administration beyond day 1 through hospital discharge. Further validation analyses in separate cohorts will be required to confirm the performance our novel risk-adjusted models. Institutional differences, including hospital size, teaching status, location, or geographic region, were not associated with variability in DOF resuscitation. While likely not a function exclusively of organizational or hospital-level factors, these data do not preclude that there may be significant individual- or team-level practice variability. Clinician-level variability in resuscitation decisions could also enhance future analyses of treatment variations. Taken together, these findings compellingly point to the importance of ongoing prospective clinical trials of physiologically informed, patient-centered fluid resuscitation for shock using dynamic measures of circulatory effectiveness rather than uniformly empiric strategies.

Among mechanically ventilated patients with diverse etiologies of shock, this large retrospective analysis highlights extensive treatment variability including frequent greater- or less-than-expected fluid volume resuscitation. There was a non-uniform association with risk-adjusted hospital mortality, which was surprisingly worse in patients without a fluid reductive factor. These findings support further evaluation of which patients may potentially benefit or be harmed by restrictive and liberal fluid resuscitation strategies in usual care.