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

Rationale There remains significant controversy regarding the optimal approach to fluid resuscitation for patients in shock. The magnitude of care variability in shock resuscitation, the confounding effects of disease severity and comorbidity, and the relative impact on sepsis survival are poorly understood. Objective To evaluate usual care variability and determine the differential effect of observed and predicted fluid resuscitation volumes on risk-adjusted hospital mortality for mechanically ventilated patients in shock. Methods We performed a retrospective outcome analysis of mechanically ventilated patients admitted to intensive care units using the 2013 Premier Hospital Database (Premier, Inc.). Observed and predicted hospital mortality were evaluated by observed and predicted day 1 fluid administration, using the difference in predicted and observed outcomes to adjust for disease severity between groups. Both predictive models were validated using a second large administrative database (Truven Health Analytics Inc.). Secondary outcomes included duration of mechanical ventilation, hospital and ICU length of stay, and cost. Results Among 33,831 patients, observed hospital mortality was incrementally higher than predicted for each additional liter of day 1 fluid beginning at 7 L (40.9% vs. 37.2%, p = 0.008). Compared to patients that received expected (± 1.5 L predicted) day 1 fluid volumes, greater-than-expected fluid resuscitation was associated with increased risk-adjusted hospital mortality (52.3% vs. 45.0%, p < 0.0001) among all patients with shock and among a subgroup of shock patients with comorbid conditions predictive of lower fluid volume administration (47.1% vs. 41.5%, p < 0.0001). However, in patients with shock but without such conditions, both greater-than-expected (57.5% vs. 49.2%, p < 0.0001) and less-than-expected (52.1% vs. 49.2%, p = 0.037) day 1 fluid resuscitation were associated with increased risk-adjusted hospital mortality. Conclusions Highly variable day 1 fluid resuscitation was associated with a non-uniform impact on risk-adjusted hospital mortality among distinct subgroups of mechanically ventilated patients with shock. These findings support closer evaluation of fluid resuscitation strategies that include broadly applied fluid volume targets in the early phase of shock resuscitation.

. Age groups and admission diagnoses associated with risk of death 3 Table E2. Factors associated with prdicted day one fluid resuscitation 4 Table E3. Day one fluid volume resuscitation by hospital demographic factors 5 Table E4.

Data Sources
The Premier Hospital Database is an electronic, administrative-level data repository of patients discharged in the USA, representing approximately 40% of patients discharged nationally including more than 5 million patient visits per year (1). The database contains International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) diagnoses including present on admission information and ICD-9-CM procedures by date. Information is collected on all patients within all hospital treatment areas, additionally including patient demographic data and a date stamped log with quantity, cost, and charge details for all invoiced items, such as medications, laboratory orders, diagnostic and therapeutic services. Participating hospitals are geographically diverse, primarily nonprofit and non-governmental, and include community and teaching hospitals from rural and urban areas (1). Fewer than one percent of patient records have missing data for most components, while fewer than 0.01 percent have missing data for key components, such as demographics and diagnostics information (2). Currently, more than 375 studies utilizing the Premier Hospital Database have been published in peer-reviewed journals across multiple disciplines (1). The Truven Health MarketScan Hospital Drug Database contains clinical data on inpatient episodes for all patients discharged from approximately 600 United States Hospitals. These records include admission and discharge dates, patient demographics and provider specialty, and specific codes for diagnosis, procedures, drug administration, and facility descriptors. This database has been shown to be representative of acute care hospitals in the United States (3)(4)(5).

Case Mix Adjustment
We constructed binary response models for hospital mortality and a linear analysis of variance model for fluids on day one, using patient age and codes for acute conditions present on admission including acute organ dysfunctions, anemia, electrolyte disorders and altered awareness (280.

Statistical Analysis
Predicted hospital mortality model fit was assessed using likelihood r-square, chi-square dispersion, area under the ROC curve, and the Hosmer-Lemeshow C statistic. The predicted DOF resuscitation model was assessed using sum of squares, R-squared, and Fratios. Continuous data were compared by Mann-Whitney U test and categorical data by Chi-square or Fisher's exact test as appropriate. The databases were constructed in FoxPro (Microsoft Corp., Redmond WA, USA) and analyses were conducted in Data Desk (Data Description, Ithaca NY, USA).

Model Development
The initial models were developed using a systematic review process looking at all recorded patient factors. The initial development set from the Premier database was split into two random sets of hospitals. An iterative review of a series of models looked at the residuals for each condition as factors were added and removed from the model. This process examined all possible factors that were present in at least 2.5% of the population and included those that were at least as large a factor as the difference between being 40 and 50 years of age. The same set of factors was then identified in the control set of the other half of the hospitals. The model coefficients from the two models were within the statistical uncertainty for each factor. Because of the size of the dataset and the small number of predictors, there was no significant risk of over fitting. This same set of model predictors was then applied to the Truven data to properly calibrate for differences that may occur overall from a different set of hospitals and different years of care.

Predicted Mortality
Our predicted mortality model identified 11 admission diagnoses that were statistically significantly associated with increased mortality, including five that overlapped with covariates within the predicted DOF model (Supplementary Table E1). The model accounted for 65.61% of variation and also demonstrated good calibration and performance (HL=14.81, p=0.063, AUROC=0.80). Risk of death increased with age, and the oldest age group (75-99 years of age) also had the highest observed frequency (24.5%). Of the admission diagnoses, the neoplastic diseases, including pulmonary and CNS neoplasm followed by metastatic neoplasm, conferred the highest risk of death. Of the conditions comprising each segment, a circulatory diagnosis with shock conferred the highest risk of death. This was followed by other (non-infectious) diagnoses with shock and septic shock. Conditions without shock, as anticipated, conferred lower risks of death (Supplementary Table E1).

Predicted Day One Fluid Resuscitation
The predicted day one fluid resuscitation model accounted for 18.9% (R 2 ) of variation. Septic shock patients were predicted to receive the most additional fluid ( FRF = Fluid Reductive Factor * Less-Than-Expected = difference between observed and predicted day one fluids is less than -1.5L † Expected = difference between observed and predicted day one fluids is between -1.5L and 1.5L ‡ Greater-Than-Expected = difference between observed and predicted day one fluids is more than 1.

C) Cases
Observed Predicted Figure E5. The effect of day one fluid volume under or over-resuscitation in patients with septic shock. Risk adjusted observed vs. predicted hospital mortality by less-than-expected (Δ < -1.5L), expected (|Δ| <1.5L), or greater-than-expected (Δ > 1.5L) resuscitation as determined by the difference between predicted and observed day one fluid resuscitation for septic shock patients. Risk adjustment performed by adding the predicted hospital mortality difference between expected-resuscitation and under or overresuscitation groups to their respective observed hospital mortality. * Indicates statistically significant difference in observed hospital mortality when compared to the expected-resuscitation group

Less-Than-Expected
Greater-Than-Expected