Fluid therapy and outcome: a prospective observational study in 65 German intensive care units between 2010 and 2011

Aim

The aim of this study is to gather data on the practice of fluid therapy and associated outcomes in order to assess the impact of specific colloids and colloids in general on 90-day survival, ICU mortality and acute kidney injury (AKI).

Study design

RaFTinG (Rational Fluid Therapy in Germany) is a prospective, observational, multicentre database. It assessed the characteristics of unselected ICU patients, focusing on fluid therapy and related outcomes. For recruitment, all German ICUs received an invitation letter and a notification in “Deutsches Ärzteblatt”.

Setting

Sixty-five German ICUs participated in this registry.

Study population

Patients with an indication for fluid therapy (judged by the attending physician) and presumed length of ICU stay > 24 h were included. Exclusion criteria were age < 18 years, psychological disorders, reasonable doubt regarding the patient’s discernment and institutionalisation upon court or other official order. Inclusion started 01.06.2010 and ended 31.05.2011. Centres were offered four inclusion schemes to avoid selective inclusion: 1. all patients, 2. all patients admitted on a specific weekday, 3. all patients admitted in one week per month and 4. first 10 consecutive patients per month.

Study protocol and collected data

No specifications regarding diagnostics, medication or procedures were made. All relevant decisions were performed as part of usual care. Only routine records were documented for the study starting at admission to the ICU. Basic biometrical data, admission diagnoses, haemodynamic and laboratory parameters and severity scores (Acute Physiology And Chronic Health Evaluation, APACHE II; Simplified Acute Physiology Score, SAPS II; Sequential Organ Failure Assessment score, SOFA) were documented. On each ICU day, new diagnoses, haemodynamic and laboratory variables, fluid balance and therapeutic interventions were assessed. Documentation was completed by the medical condition at discharge.

ICU survivors were contacted by postal mail to retrieve survival status 90 days after ICU admission. If no reply was returned, survival status was attempted to be retraced via telephone calls and the residents’ registration offices.

Data entries were possible in electronic or paper forms (Additional file 1: Supplemental digital content 1) as preferred by the centres. Automatic enquiries for values outside of pre-specified limits ensured data validity. All data were continuously checked for formal and content-related errors. Missing and inconsistent information was reassessed.

Outcome parameters and pre-defined subcohorts

Main outcome parameters were 90-day mortality (death within 90 days after first ICU admission), ICU mortality and AKI (RIFLE [13] “failure”). Renal replacement therapy (RRT) was analysed for completeness, despite the fact that, without protocol, it is an inaccurate parameter.

Patients were a priori stratified as having received crystalloids and colloids or exclusively crystalloids. Patients having received crystalloids and colloids were further substratified a priori as having received gelatine, HES 130/0.4, HES 130/0.42), HES 200/0.5 or human albumin. Patients who were treated with more than one type of colloid were excluded from subcohort analyses.

Data were also a priori stratified for surgical or medical patients and patients with or without severe sepsis on admission.

Statistical analysis

Data are presented as median (25th; 75th percentiles) for numeric variables or percentages for categorical variables, if not otherwise specified. Crude results cover the whole study population. Univariate data comprise only patients eligible for multivariable analysis. The predicted individual risk of mortality on admission was calculated from SAPS II and APACHE II scores [14, 15]. We used the highest calculated risk for further analyses because a relevant subset of patients only had entries for one of these scores. In order to assess the maximum mortality risk for each patient, the score that predicted a higher risk was used for the regression analyses. Comparison of predicted vs. actual mortality shows that both are well correlated (Additional file 1: Supplemental digital content 2).

For all multivariable tests, covariables with a clear clinical relevance on the investigated outcome were chosen as cofactors and restricted to those with a p < 0.10 in univariate analysis in the model. Colloid dose is entered as a continuous variable in mL into the statistical models.

For AKI, RRT and ICU mortality, a multiple logistic regression model was fitted with adjustment for the predicted mortality risk in percentile steps of 10% (resulting in an ordinal covariable stratifying the predicted risk from 0–10 up to 90–100%), gender, chronic kidney disease (CKD, according to KDOQI [16]) and severe sepsis (according to ACCP/SCCM [17]) on admission as categorical cofactors (model referred to as “baseline adjustment” in the tables).

To enhance structural equality of the cohorts, the covariables obtained at ICU admission and variables derived during ICU stay (AUC of SOFA score until event or end of stay, cumulative volume of red blood cell products, cumulative volume of other blood products, cumulative fluid balance, application of vasopressor equivalent > 0.6 mg/h and daily crystalloid infusion) were included as covariables in a second analysis (referred to as “full model”).

Association of covariates was tested for significance and removed when not significant to reduce the number of model parameters to be estimated.

Norepinephrine equivalent was defined as 1 mg norepinephrine being equivalent to 1 mg epinephrine or 100 mg of dopamine [18].

90-day survival was analysed by Cox regression with the cofactors and covariables given above. To account for missing follow-up data, the following approaches were chosen: (1) include only patients with complete follow-up data; (2) assume that all patients with unknown status die 1 day after discharge (“worst case”); (3) assume that 10% of all patients die after discharge (unknown status is extrapolated based on post-ICU mortality); (4) assume that 28% of all patients die after discharge (unknown status is extrapolated based on mortality in the cohort without 90-d follow-up); (5) assume that all patients with unknown status survive until 90 days post-discharge (“best case”).

For the multivariable analyses, hazard ratios (HRs) were used to describe point estimates of the instantaneous risk ratio between cohorts in Cox regression analysis. Odds ratios (ORs) were used to quantify the cumulative risk estimate presented derived from logistic regression.

Factors with an exploratory p value of less than 0.05 in the regression equation were considered relevant for the event under investigation. The impact of specific fluids in each analysis is shown as the adjusted odds ratio or hazard ratio and the respective 95% confidence intervals [19], keeping all other covariables constant.

All statistical analyses were done with SAS version 9.3.

Participating centres and patient recruitment

Sixty-five study centres documented 4545 patients. 70.9% (3223) of patients were admitted from the operating theatre. In total, 3902 (85.9%) had records for each adjustment variable and were valid for multivariable analysis (Fig. 1). Baseline characteristics are presented in Table 1. Patients with higher severity of illness (SAPS II and APACHE II scores) or with severe sepsis on admission were more likely to have received colloids later on. An overview about primary diagnoses of the patient populations is presented in Additional file 1: Supplemental digital content 3.

	All patients entered into database				Only patients eligible for multivariable analysis				p value
	All fluids	Crystalloids only	Crystalloids + colloids	Crystalloids + HES 130/0.4	All Fluids	Crystalloids only	Crystalloids + colloids	Crystalloids + HES 130/0.4	Crystalloids vs. colloids (all/eligible for multivariable analysis)
N (n [%])	4545 [100]	2482 [54.6]	2063 [45.4]	1175 [25.9]	3902 [100]	1888 [48.4]	2014 [51.6]	1128 [28.9]
Age (years)	68 [55; 76]	68 [55; 76]	68 [56; 75]	68 [55; 76]	68 [55; 76]	68 [55; 76]	68 [56; 75]	68 [55;76]	0.784/0.641b
Gender, male (n [%])	2788 [61.3]	1491 [60.1]	1297 [62.9]	761 [64.8]	2392 [61.3]	1123 [59.5]	1269 [67.2]	730 [64.7]	0.058/0.026c
Admission type: surgical (n [%])	3223 [70.9]	1714 [69.1]	1509 [73.1]	986 [83.9]	2782 [71.3]	1316 [69.7]	1466 [72.8]	955 [84.7]	0.003/0.036c
Cardiac surgery (n [%])	686 [15.1]	310 [12.5]	376 [18.2]	351 [29.9]	659 [16.9]	290 [15.4]	369 [18.3]	348 [30.9]	<0.001/0.015c
Severe sepsis on admission (n [%])	394 [8.7]	184 [7.4]	210 [10.2]	71 [6.0]	370 [9.5]	163 [8.6]	207 [10.3]	71 [6.3]	0.001/0.090c
History of CKD (n [%])	828 [18.2]	449 [18.1]	379 [18.4]	161 [13.7]	622 [15.9]	312 [16.5]	310 [16.4]	144 [12.4]	0.837/0.356c
APACHE II	20 [14, 25]	18 [13, 23]	22 [17, 27]	24 [19, 30]	20 [14, 25]	18 [13, 23]	22 [17, 27]	24 [19, 30]	<0.001/< 0.001b
SAPS II	21 [9; 36]	17 [7, 31]	25 [11; 41]	34 [18; 52]	21 [9; 36]	17 [8, 31]	25 [11; 41]	34 [18; 52]	<0.001/< 0.001b
Probability of mortality [%]a	39.0	33.3	44.7	50.7	39.1	33.1	44.7	51.0	<0.001/< 0.001b
AKI (RIFLE “failure”) (n [%])	560 [12.3]	174 [7.0]	386 [18.7]	179 [15.2]	549 [14.1]	167 [8.8]	382 [19]	178 [15.8]	<0.001/< 0.001c
Renal replacement therapy in ICU (n [%])	361 [7.9]	78 [3.1]	283 [13.7]	133 [11.3]	358 [9.2]	76 [4]	282 [14]	133 [11.8]	<0.001/< 0.001c
ICU mortality (n [%])	408 [9.3]	129 [5.5]	279 [13.8]	119 [10.2]	367 [9.4]	92 [4.9]	275 [13.7]	118 [10.5]	<0.001/< 0.001c
Length of ICU stay (days)	3 [1, 9]	2 [1, 5]	6 [3, 12]	5 [2, 11]	4 [2, 8]	2 [1, 5]	6 [3, 13]	5 [3, 11]	<0.001/< 0.001b
90-day mortality (n [%])	707 [20.1]	284 [15.4]	423 [25.2]	191 [18.8]	613 [22.4]	198 [14.6]	415 [30.1]	187 [21.9]	<0.001/< 0.001c

Data are given as absolute number and percentage or median [25th; 75th percentiles], as appropriate

AKI acute kidney injury, APACHE acute physiology and chronic health evaluation score, CKD chronic kidney disease, ICU intensive care unit, SAPS simplified acute physiology score

^aProbability of mortality was calculated from APACHE II and SAPS II scores as given in methods

^bWilcoxon–Mann–Whitney test

^cChi-squared contingency table test

Fluid therapy

During ICU stay, 54.6% (2482) of patients received exclusively crystalloids, whereas 45.4% (2063) received crystalloids and colloids. 57.0% (1175) of the latter collective received 6% HES 130/0.4 (Table 1). Considerably less patients were treated with other colloids (6% HES 130/0.42, gelatine and human albumin) and were therefore excluded from subgroup analyses (Additional file 1: Supplemental digital content 4). Sixteen centres used exclusively crystalloids in all documented patients (207).

Fluid balances are presented in Additional file 1: Supplemental digital content 5 and 6. Colloid dose is given in Additional file 1: Supplemental digital content 7. Cumulative fluid balance and fluid balance on day 1 were more positive for patients receiving crystalloids and colloids as compared to sole crystalloid therapy (Additional file 1: Supplemental digital content 6). Among patients treated with colloids, 77.2% (1555 patients) received colloids on day 1 (Fig. 2), with a median amount of one unit of 500 mL [500; 1000]. Amounts of infused blood products significantly differed between cohorts, but absolute differences were very small (Additional file 1: Supplemental digital content 6).

90-day survival

77.5% of patients had 90-day follow-up data (3115 plus 408 who died on ICU). 55.0% (1713) of patients with follow-up data received only crystalloids and 45.0% (1402) received colloids, a ratio that is similar to the total database (54.6 vs. 45.4%).

Unknown status for 90-day follow-up was associated with lower SAPS II and APACHE II scores on admission (SAPS II median 13 [6, 26] vs. 24 [11, 38], p < 0.001; APACHE II median 19 [13, 24] vs. 20 [15, 26], p < 0.001), as well as significantly lower cumulative crystalloid infusion and red blood cell transfusions. In contrast, there was no significant association of loss to 90-day follow-up with length of ICU stay, CKD or sepsis on admission.

ICU mortality

Overall ICU mortality was 9.3% (408 of 4392, 153 missing data). Treatment with colloids was associated with higher crude ICU mortality compared to crystalloids only (13.8% (279 of 2029) versus 5.5% (129 of 2363), crude OR in the overall population 2.761 [2.221; 3.433]). The risk of ICU mortality associated with colloids or 6% HES 130/0.4 decreased progressively after multiple logistic regression analysis using baseline and progress covariables (Additional file 1: Supplemental digital content 10 and 11). Independent risk factors were predicted mortality, female gender, vasopressor use, severe sepsis and CKD on admission, as well as cumulative fluid balance. Notably, for patients in the subcohort without severe sepsis on admission, the adjusted risk of ICU mortality was lower for patients treated with colloids in general (crude OR 2.367 [1.539; 3.643]; multivariable adjusted OR 0.923 [0.874; 0.974]) or 6% HES 130/0.4 (crude OR 2.179 [1.223; 3.882]; multivariable adjusted OR 0.905 [0.833; 0.983]) compared to crystalloids only.

Acute kidney injury

Further information about renal function and failure (incidences of AKI and RRT in patients with and without underlying CKD on admission; creatinine and diuresis) is given in Additional file 1: Supplemental digital content 13 and 14.

Renal replacement therapy

During ICU stay, 7.9% (361) of the patients received RRT (see also Table 1). The crude incidence of RRT was considerably higher in patients treated with colloids compared to those without (13.7% (283 of 2063) vs. 3.1% (78 of 2484), crude OR for the overall population 4.904 [3.789; 6.348]). With multivariable logistic regression analysis, the risk associated with colloids or 6% HES 130/0.4 decreased after adjustment for baseline covariables. In the full model adjusted for baseline and progress covariables, colloids and 6% HES 130/0.4 were associated with a reduced risk of RRT (Additional file 1: Supplemental digital content 15). Independent risk factors for RRT included predicted risk of death, severe sepsis and CKD, AUC of the SOFA score and cumulative fluid balance. In contrast, vasopressor use was negatively associated with RRT.

Limitations of the present study

Our study has several limitations. Since it is observational, unblinded outcome assessment is unavoidable. Patients were not randomised and the cohorts are heterogeneous, with significant imbalances at baseline (e.g. severity scores, prevalence of sepsis) and many potential confounders on outcome. Even though we are confident that we were able to identify most of them and perform an appropriate adjustment, several approaches to adjusting the data are possible. It appears virtually impossible to account for the different severity of illness in the colloid cohort versus the crystalloid cohort without using adjustment parameters that are also influenced by disease progress. Thus, both risks of residual confounding and overadjustment exist. As a consequence, the adjusted results should be interpreted with caution. However, the small adjusted confidence intervals around or below 1 strongly suggest neutrality of the investigated colloids in terms of mortality and AKI. Unfortunately, although the analysis plan for the major endpoints (AKI, 90-day and ICU mortality) was designed a priori, no statistical analysis plan has been pre-published, which would have further strengthened the present results.

The present study did not investigate pre-admission fluid therapy or haemodynamics and can therefore not provide an estimate of the indication and effectiveness, which would have required a different study design [31]. Nevertheless, the timing of colloid infusion suggests colloids were used predominantly for initial or post-operative resuscitation (with pre-ICU fluid therapy, e.g. in the OR, being a blind spot).

Subcohorts having received other colloids than HES 130/0.4 were small compared to HES 130/0.4 or crystalloids. Any comparison between these subgroups must be done with caution, if at all.

Furthermore, the incidence of RRT may be a unreliable outcome measure, since most centres had no clear protocol for it. Any conclusion based on RRT should therefore be drawn with great caution.

It also needs to be acknowledged that an estimated mortality risk calculation by combining APACHE II and SAPS II (as available) may be less accurate than having a full set of both scores.

Finally, the follow-up might be considered incomplete. Besides that, missing information was not evenly distributed between both analysed groups with more incomplete baseline data with crystalloids only and more incomplete 90-day follow-up with colloids. The reason for this imbalance in missing data remains unknown, although we found that patients lost to follow-up were less severely ill on ICU admission. It is unclear how this finding may have led to a lower follow-up rate in the colloid cohort. Therefore, it is possible that unknown confounders influenced follow-up rates in the two cohorts. Nevertheless, a 90-day follow-up rate of 77.5% appears to be reasonably high when compared to 50–80% in other epidemiological cohort studies [32].

We also checked the effect of missing data on our full multivariable analysis with a best-/worst-case scenario. As expected, all risks were “diluted” by the assumption that all patients without follow-up had died. However, the risk associated with colloid use did not increase by this approach although colloid use was associated with greater loss to 90-day follow-up. Thus, the best-case/worst-case analysis suggests that missing follow-up data did not substantially affect the overall result.