Order Substitutions and Education for Balanced Crystalloid Solution Use in an Integrated Health Care System and Association With Major Adverse Kidney Events

This comparative effectiveness research study investigates the association of order substitutions and education for the use of lactated Ringer solution with major adverse kidney events in an integrated health care system.


eFigure 1. Map of Study Sites
Layton Hospital (participating site) is located between Salt Lake City and Ogden, UT and is not depicted on this map. Primary Children's Hospital was excluded from study participation. St. George Medical Center (participating site) changed titles during the study period so is depicted as Dixie Regional Medical.

Moving Forward
• Intermountain Healthcare will soon make LR the default resuscitation fluid.
• NS will become the second-line/alternative option.

Benefit to Patients
Death, Renal replacement therapy, or Persistent kidney injury: • NNT to prevent MAKE in hospitalized patients: 111 Be a model health system by providing extraordinary care and superior service at an affordable cost.

Further Importance
• Long-term health outcome improvements will translate to profound cost savings.
• Meaningful contribution to the field's literature. There were 51 encounters with more than 24,000 mL received in 24 hours that were excluded due to implausible fluid volumes. Baseline serum creatinine values were obtained from up to one year prior to the date of admission for the index encounter. Baseline serum creatinine was calculated in cases where no value was available using the following formula: Normal saline included fluids with 0.83-0.9% sodium chloride with or without dextrose or potassium chloride. Lactated Ringer's (LR) included LR with or without dextrose as well as an alternative proprietary balanced crystalloid solution (Plasma-Lyte). Bolus or maintenance infusions but not diluent or carrier fluids were included when calculating fluid administration volumes. The proportion of fluids received that were LR by volume (mL) was calculated with following formula: Sepsis was identified per Sepsis-3 criteria 17 as the combination acute organ failure (Sequential Organ Failure Assessment score ≥2 points above pre-ED baseline) plus confirmed or suspected infection (based on collection of body fluid cultures and administration of an IV antimicrobial or oral vancomycin, fidaxomicin, or oseltamivir) prior to ED departure using an internally validated electronic data warehouse query.

Data analysis
We used a quasi-experimental analysis strategy, segmented linear regression, to support more robust causal inference from this non-randomized trial. For our effectiveness outcome, we first obtained the weekly adjusted MAKE30 (or other outcome) using binomial regression.

/ ℎ ℎ
Variables in the model were Age (years); Sex (Male or Female); Race/Ethinicity (self-reported race/ethnicity, categorized as Hispanic/Latino, non-Hispanic American Indian/Alaska Native, non-Hispanic Asian, non-Hispanic Black, multiple races, non-Hispanic Native Hawaiian/Pacific Islander, non-Hispanic White, or unknown); Charlson comorbidity score (integers from 0 to 20); Acute physiology score (APS Score, integers from 0 to 52); Baseline Dialysis use (present/absent); and Baseline Creatinine (in mg/dL).  Z represents the vector of coefficients for each variable. Risk adjusted models were then calibrated to predict MAKE30 at the same rate that was observed during the study. Formulas for weekly standardized MAKE30: The Clopper-Pearson method was used to calculate 95% confidence intervals for the weekly risk-adjusted MAKE30 outcome. We then performed segmented linear regression based on fractional binomial regression to obtain interrupted time series estimates for the association between the intervention and the effectiveness outcomes according to the following formula.

30
In this formula, time is the integer count of weeks during study period from 1 to 69; Post Implementation is binary where 0 represents pre-implementation and 1 represents post implementation; timeAfter is the integer count of weeks during the post-implementation phase of the study from 0 for weeks preimplementation and 1 to 33 weeks post-implementation; and rate of Adjusted MAKE30 and # patients are week-specific rates and counts, respectively, from the cohort for a total of 69 datapoints. The week-onweek trend for the outcome pre-and post-intervention is obtained from coefficients  pre and  post , respectively, while the step-off effect is provided by the coefficient  step . Absolute and relative risk difference was calculated with regard to the final week of the study per the following formulas: We used a similar segmented linear regression approach based on beta regression with a logit link to obtain interrupted times series estimates of the association between implementation interventions and the proportion of LR received: 1 2 In this formula, time and Post Implementation are the same as the MAKE30 model above, and time Education1 and time Education2 is the integer count of weeks after the educational interventions from 0 during pre-education time frames and 1 to 47 for time Education1 and 1 to 21 for time Education2.
The binary variable, Post Implementation, allows for analyzing and immediate change, or step-off, of the intervention. The time variables that count whole weeks from a specific time point (e.g., time, timeAfter, time Education1, time Education2) permit an analysis of trend, or slope during the period where the time count is greater than zero. Coefficients provide trend and step-off effect estimates analogous to the prior equation.