Net ultrafiltration intensity and mortality in critically ill patients with fluid overload

Background Although net ultrafiltration (UFNET) is frequently used for treatment of fluid overload in critically ill patients with acute kidney injury, the optimal intensity of UFNET is unclear. Among critically ill patients with fluid overload receiving renal replacement therapy (RRT), we examined the association between UFNET intensity and risk-adjusted 1-year mortality. Methods We selected patients with fluid overload ≥ 5% of body weight prior to initiation of RRT from a large academic medical center ICU dataset. UFNET intensity was calculated as the net volume of fluid ultrafiltered per day from initiation of either continuous or intermittent RRT until the end of ICU stay adjusted for patient hospital admission body weight. We stratified UFNET as low (≤ 20 ml/kg/day), moderate (> 20 to ≤ 25 ml/kg/day) or high (> 25 ml/kg/day) intensity. We adjusted for age, sex, body mass index, race, surgery, baseline estimated glomerular filtration rate, oliguria, first RRT modality, pre-RRT fluid balance, duration of RRT, time to RRT initiation from ICU admission, APACHE III score, mechanical ventilation use, suspected sepsis, mean arterial pressure on day 1 of RRT, cumulative fluid balance during RRT and cumulative vasopressor dose during RRT. We fitted logistic regression for 1-year mortality, Gray’s survival model and propensity matching to account for indication bias. Results Of 1075 patients, the distribution of high, moderate and low-intensity UFNET groups was 40.4%, 15.2% and 44.2% and 1-year mortality was 59.4% vs 60.2% vs 69.7%, respectively (p = 0.003). Using logistic regression, high-intensity compared with low-intensity UFNET was associated with lower mortality (adjusted odds ratio 0.61, 95% CI 0.41–0.93, p = 0.02). Using Gray’s model, high UFNET was associated with decreased mortality up to 39 days after ICU admission (adjusted hazard ratio range 0.50–0.73). After combining low and moderate-intensity UFNET groups (n = 258) and propensity matching with the high-intensity group (n = 258), UFNET intensity > 25 ml/kg/day compared with ≤ 25 ml/kg/day was associated with lower mortality (57% vs 67.8%, p = 0.01). Findings were robust to several sensitivity analyses. Conclusions Among critically ill patients with ≥ 5% fluid overload and receiving RRT, UFNET intensity > 25 ml/kg/day compared with ≤ 20 ml/kg/day was associated with lower 1-year risk-adjusted mortality. Whether tolerating intensive UFNET is just a marker for recovery or a mediator requires further research. Electronic supplementary material The online version of this article (10.1186/s13054-018-2163-1) contains supplementary material, which is available to authorized users.


S1: Study Population
For each patient the following information were extracted: age, sex, primary diagnosis and patient comorbidities (coded according to the International Statistical Classification of Diseases and Related Health Problems, Ninth Revision ). For patients with multiple ICU admissions, the first ICU admission was considered as the reference point for our analyses. Severity of illness was computed from electronic abstraction of all physiologic variables comprising the Acute Physiologic and Chronic Health Evaluation (APACHE)-III score [1]. Because sepsis is underreported with ICD-9 codes, we defined "suspected sepsis" as the ordering of blood cultures and intravenous antibiotics within 24 hours of each other, as defined previously [2]. Following initiation of RRT, we extracted hourly mean arterial pressure (MAP) and vasopressor type and dose, and daily fluid balance for the duration of RRT.
We excluded patients with no available baseline hospital weight since fluid balance is expressed as a body weight percentage ( Figure E1). We excluded patients discharged within 48 hours of ICU admission since most were either post-operative patients or in whom the critical illness resolved rapidly. We excluded those who died within 72 hours of ICU admission since any immediate death after ICU admission is more likely due to severity of illness. Since patients with end-stage renal disease and those with chronic kidney disease have different prognosis, we excluded patients on chronic dialysis, serum creatinine ≥3.5mgs/dl, and renal transplantation. We excluded patients with missing fluid balance and UF NET data. Since the goal of this study is to examine the association between UF NET intensity and mortality in patient with FO, we excluded patients with ≤5% FO and those who never received RRT.

S2: Determination of Cumulative Fluid Balance
We calculated cumulative fluid balance expressed as percentage of body weight for each patient prior to initiation of RRT and included only patients with ≥5% cumulative fluid balance as it was associated with long-term mortality in our prior work [3]. For each patient, we extracted input fluid including all intravenous and enteral fluids including resuscitation and maintenance fluids in the form of colloids and crystalloids, blood products, drug infusions, and enteral and parenteral nutrition.
Output fluid included all body fluids (output from drains, rectal, orogastric and nasogastric output) including urine. We determined the cumulative FB expressed as percentage (%) from ICU admission until initiation of renal replacement therapy (RRT) using the following equation [3,4]: Cumulative Fluid Balance (%)=(Cumulative fluid input -fluid output) in litres X 100 Hospital admission weight (kg) For inclusion in the study, we identified patients who only had cumulative fluid balance ≥5% of body weight from the time of ICU admission prior to initiation of either intermittent hemodialysis (IHD) or continuous renal replacement therapy (CRRT). We used ≥5% cumulative fluid balance as a threshold as it was associated with both shortand long-term mortality in our prior work [3]. We excluded patients from our analysis if they had a cumulative fluid balance <5% of body weight before initiation of RRT. In order to determine the patient's cumulative fluid balance after the initiation of RRT, we used the following approach. For patients receiving CRRT, daily cumulative fluid balance was calculated as input minus output excluding the net volume of fluid removed during CRRT (i.e., UF NET , which is the exposure variable). For patients receiving IHD, cumulative fluid balance including the inter-dialytic period was calculated as daily input minus output excluding UF NET during each IHD session.

S3: Vasopressor Standardization to Norepinephrine Equivalents
All vasopressors were standardized according to the following conversion scale below.

The conversion scale was developed based on the cardiovascular Sequential Organ
Failure Assessment score and the medical literature [5,6]. Vasopressin equivalence to norepinephrine was developed with the use of the Vasopressin and Septic Shock Trial data set [7].

S4: Gray's Survival Model
The Cox multivariable regression model relies on the assumption that the proportionality of hazards remains constant over the length of duration that it's used to estimate the conditional hazard rate. However, in most real clinical scenarios, especially in acute illness, wherein multiple clinical factors affect a patient simultaneously, this may not hold true. To address this issue, models that allow for non-proportionality of the conditional hazards by introducing covariate effects have been proposed. The Gray's model, proposed by Gray [8] is one such model that employs products of the covariates of interest with the spline functions of time [8].
The advantage of the Gray's model is that it retains most of the mathematical simplicity of the Cox model since the proportional hazards assumption is only required for each of the time intervals between the successive knots (i.e., the time points within the duration over which treatment effects are studied) [9]. Gray's model may therefore be viewed as a piecewise Cox proportional hazards model for the conditional hazard rate. We used Gray's model in this manuscript, as well as prior work [3,10], for two reasons: a.) Cox models failed proportionality assumptions for several covariates, and b.) to assess the variation in AHRs associated with UF NET intensity on mortality over time.

S5: Propensity Score Estimation and Matching
We constructed a propensity score to account for indication bias associated with UF NET using multinomial logistic regression with UF NET intensity categorized as low (≤ 25ml/kg/day) and high (>25ml/kg/day) as an outcome. The variables used in the model for propensity score estimation included age, sex, race, body mass index, admission under surgical service, admission for liver transplantation, baseline estimated glomerular filtration rate (eGFR), APACHE-III score, mechanical ventilation, suspected sepsis, first RRT modality, cumulative fluid overload before initiation of RRT, total duration of RRT, time-to-initiation of RRT, mean arterial pressure on day 1 of RRT, cumulative vasopressor dose and cumulative fluid balance after initiation of RRT. We then matched the low (≤25ml/kg/day) with that of high intensity UF NET (>25ml/kg/day) using propensity scores on 1:1 basis without replacement creating 258-matched pairs.
Matches were created without replacement using computational geometry based on distance between propensity scores. The matching was based on a 1:1 ratio along with a distance set by a caliper of 0.025. The analysis was carried out using source code and SAS macros that are available online [11].

Confounder
To estimate the potential impact of an unmeasured confounder [12], we made the following assumptions: 1) only one unmeasured confounder was present (or a combination of confounders that can be described as one), 2) the unmeasured confounder is binary, 3) the unmeasured confounder is independent of measured confounders, and 4) the exposure UF NET intensity is not an effect modifier for the unmeasured confounder's effect on outcome.
The impact of the unmeasured confounder was determined by the following: 1) the prevalence of the unmeasured confounder in the exposed (i.e., high UF NET group) vs.
the unexposed group (i.e., low intensity UF NET group) and 2) the association between the unmeasured confounder and mortality, independent of the measured confounders, expressed as an odds ratio (e.g., an OR=0.8 indicates that the odds of death is 20% lower than those with the unmeasured confounder compared to those without the unmeasured confounder). We independently varied these 2 parameters to assess their influence on the adjusted odds ratio for high intensity UF NET on mortality and present the results in graphical form.

Figure E1
Study Population and Analysis Cohort

Figure E2
Association Between Intensity of Net Ultrafiltration and Crude Hospital Mortality X -axis represents the UF NET expressed in ml/kg/day. Y-axis corresponds to the crude hospital mortality. The first arrow on the left represents the point on the distribution curve that corresponds to UF NET 25ml/kg/day and the second arrow on the right corresponds to the point on the distribution curve that corresponds to UF NET 20ml/kg/day. A. High UF NET (>25ml/kg/day) as compared with low UF NET (≤20ml/kg/day), was associated with decreased risk for death in the first 39 days after ICU admission, however subsequently, there was no association with mortality.

Association between Net Ultrafiltration Intensity and Time-to-Mortality Using Grays model
B. Whereas, moderate UF NET (>20-≤25ml/kg/day), as compared with low UF NET (≤20ml/kg/day), was not associated with mortality.
C. High UF NET (>25ml/kg/day), compared with moderate UF NET (>20-≤25ml/kg/day), was associated with decreased risk of death only up to 15 days after ICU admission.

14
Quantitative bias sensitivity analysis of a hypothetical unmeasured confounder. [12] The above plot shows the strength of a hypothetical confounder (X axis, univariate odds ratios of confounder in the high intensity UF NET group on 1-year mortality) versus the odds ratio for 1-year mortality from adjusted model for high intensity UF NET , if the hypothetical confounder was included (Y axis). Panels A, B, C, D, and E, correspond to varying prevalence of unmeasured confounder among patients in the high intensity UF NET group of 10%, 20%, 30%, 40% and 50%, respectively, compared with patients in the low intensity UF NET group.
To abrogate the odds ratio for high intensity UF NET from our primary analysis (i.e., OR=0.61, 95% CI, 0.41-0.93), the hypothetical unmeasured confounder must be at least twice as common (prevalence of 20% or more, panels B, C, D, and E) among patients with high intensity U FNET , and have an odds ratio for 1-year mortality that is lower than 0.7 (i.e., associated with more than 30% lower risk of death). Stronger confounders are required if the prevalence among patients who received high intensity UF NET is lower than 10%.     Table S3) [5][6][7].