In-line filtration of intravenous infusion may reduce organ dysfunction of adult critical patients

This single-centre, retrospective controlled cohort study covering the period between February 2013 and April 2015 assessed the effects of in-line filtration of intravenous fluids on the reduction of complications in critically ill adult patients of a tertiary German hospital.

Approval of the study protocol was obtained from the local ethics committee (Goethe-University Frankfurt am Main, ref. 8/14 from 30 January 2014). The study was registered at ClinicalTrials.​gov (NCT02281604). Funding was exclusively provided by internal departmental funding.

All adult (age ≥ 18 years) critical patients who were admitted to the University Hospital Frankfurt between February 2013 and March 2015 and had at least 1 admission to the 34-bed intensive care unit (ICU) were included. Admission to the ICU for more than one time during the same hospital stay was not an a priori exclusion criteria, and the information of these ICU stays was cumulated. Patients admitted to the hospital more than one time during this period (ranging from returns after weeks to months or years) were not excluded from the study, but only the first hospital stay was used.

Until February 2013, no in-line filters were used routinely on the ICU, intraoperatively or on the regular wards. To reduce the risk of particle contamination, larger 5 μm in-line filters have been routinely introduced intraoperatively and on all normal care stations from February 2013 on up to the present. On the ICU instead, in the period from February 2013 to January 2014, finer 0.2 or 1.2 μm in-line filters were routinely used (0.2 μm Sterifix for aqueous solutions and 1.2 μm Intrapur for lipid-containing mixtures) (Fig. 1). In February 2014, filter systems on the ICU were again switched to the larger 5 μm in-line filters during a cost-reduction initiative (about 15 € per 0.2 or 1.2 μm in-line filter and doubts regarding the efficacy of the finer filters), and the remaining 0.2 or 1.2 μm in-line filters were consumed in parallel. Thus, since April 2014, only larger 5 μm in-line filters have been used on the ICU. The peripheral lines were protected equally as the central lines by the respective in-line-filters of each time period and ward. Usually, the filters were placed directly behind the three-way cock of the (central) venous line (Fig. 2).

While patients admitted to the ICU between February 2013 and January 2014 were assigned to the fine filter cohort (finer 0.2 or 1.2 μm in-line filters; n = 1621), patients admitted between April 2014 and March 2015 were assigned to the control filter cohort (larger 5 μm in-line filters; n = 1594).

No dropout occurred as routine data could be followed up until hospital discharge and no patient consent was needed.

Microparticles are suspected to be responsible for inflammation of any kind that could lead to different organ manifestations. Inflammation on its own can cause vasodilatation via increased NO production, possibly leading to severe vasoplegia, also known as vasodilatory shock. As this complex syndrome can in fact only be assessed clinically by the vasopressor dosage necessary to maintain an acceptable mean arterial blood pressure, we tried to identify patients suffering from this syndrome by choosing high dose norepinephrine as one of the most appropriate potential surrogate parameters [21].

Primary endpoint of this epidemiological study was the rate of severe vasoplegia [22‐24] defined by a continuous norepinephrine rate ≥ 0.3 μg/kg/min for at least 1 min. Patients that did not receive any administration of norepinephrine reported in the data files were counted as patients with dose zero due to the fact that only records exist if patients did receive norepinephrine. Values of norepinephrine rates > 1.0 μg/kg/min were discarded as unrealistic on our ICU. In addition, we analysed the rate of patients receiving norepinephrine and also methylene blue (which is an additional option to treat vasoplegia) to obtain a more comprehensive overview on this endpoint.

The following secondary endpoints were recorded until hospital discharge:

Patients who need higher norepinephrine rates might also suffer from sepsis or septic shock, but especially, inflammation after cardiopulmonary bypass triggers vasoplegia as well. In contrast, the diagnosis sepsis is reasonably different from inflammation and vasoplegia; therefore, we analysed these two endpoints independently.

Routine data were obtained from the hospital information systems ORBIS (AGFA HealthCare, Bonn, Germany) [30] and the electronic patient data management system Metavision (iMDsoft, Tel Aviv, Israel) [31], which was first established in 2013 on the ICU. No additional study documentation on CRFs/eCRFs was performed. The documentation of the patient-related medical data was performed by using a pseudonymised patient ID or case number. No characteristics that would allow the direct identification of certain patients during the data analysis were transmitted. The study data were protected from foreign access, and only members of the study team had access to it.

The exported data were screened for apparent inconsistencies (e.g. inverted start/end dates, values being clearly out of the normal reference range by many orders) and missing information (e.g. type of discharge, measurements, medication) to detect documentation errors. Missing information was corrected manually, where possible, based on the original full patient file from the source hospital information systems AGFA ORBIS or Metavision. Patients with missing laboratory or score values for an endpoint were only excluded from that endpoint but not from the study. For patients not mentioned in the therapeutic intervention data (e.g. norepinephrine, dialysis), it was assumed that those patients had no intervention (e.g. zero doses, no dialysis). For patients with cardiac surgery, the potential event of myocardial infarction was not taken into account, as well as for patients with neurosurgery not the potential event of ischemic stroke (such patients were treated as having no event) to exclude potential pre-existing diseases.

The information about different ICU stays during the same hospital stay was then summarised, and patients were assigned to either the fine filter cohort (if all ICU stays were during the fine filter period), the control cohort (if all ICU stays were during the control filter period), or excluded (if neither of the two).

Statistical analysis was performed with the use of the statistic software R (R Foundation for Statistical Computing, Vienna, Austria) version 3.4.1 [32] with the package “MatchIt” version 3.0.2 for propensity score matching. Additional analysis was performed with BiAS (epsilon-Verlag, Frankfurt/Main, Germany, maintained by Institute of Biostatistics, Goethe University Frankfurt, Germany) for Windows version 11.06 [33]. Although if due to the large number of patients per cohort that depended only on the time period, the patient characteristics were most likely already approximately equally distributed in each cohort, we aimed to minimise potential bias and effects in the outcomes as best as possible. Therefore, in the total cohort of n = 3215 patients, a propensity score matching adjusting pairwise for the main characteristics—surgery group, sex, and age in each cohort—was introduced that lead to a total of n = 3012 patients (n = 1506 each per fine filter and control filter cohort). The propensity score matching was performed completely independent from the later outcome variables (such as Horowitz, creatinine, IL-6), and especially, no patient selection was done. For other minor characteristics (such as scores or baseline values), no matching was performed, as this would need an even much higher amount of patients to match for more characteristics equally in both cohorts. Further pre-tests (Shapiro-Wilk) were performed to determine if any of the endpoints or their logarithmised values followed a normal distribution. As deviations from normal distribution could be confirmed in all tests for the continuous endpoints, the Mann-Whitney U tests were performed for group comparisons.

There was no need to perform any significance correction for multiple testing, as our scope was to analyse the outcomes for different secondary endpoints independently from each other and from the main endpoint.

The main hypothesis was to test if the use of the finer 0.2 or 1.2 μm in-line filters was superior in comparison to the larger 5 μm in-line filter with respect to the incidence rate of severe vasoplegia.

For sample size estimation, a risk reduction of 20% by the usage of the finer 0.2 or 1.2 μm in-line filters from an incidence rate of 10 to 8% for the primary endpoint was considered as clinically relevant. This difference could be detected with a chi-square test with at least 80% power and a significance level of α ≤ 0.05, if about n = 1400 subjects were enrolled in each cohort, as calculated by G*Power (Heinrich Heine University of Duesseldorf, Germany) [34] in the original study analysis plan. The evaluation of the therapy effect was performed by a logistic regression (with regressor variables age, sex, surgery group, and cohort), and the test decision (for difference between the cohorts) was carried out with the Wald test for the coefficients of the therapy effect with R (package “survey” version 3.33). The risk reduction was quantified as odds ratio (OR) with 95% confidence interval.

The secondary hypotheses were to test if the use of the finer 0.2 or 1.2 μm in-line filters was superior to the use of the larger 5 μm in-line filter regarding the incidence of the secondary endpoints. Risk reduction was quantified as risk ratio (RR) with 95% confidence interval. Both cohorts were compared regarding the differences of incidence rates and their 95% confidence interval (CI), of mean ± standard error (SE) for continuous or median values with quartiles for at least ordinal endpoints. Differing from that, median values were also used for continuous endpoints if their distribution was not approximately symmetric.

In addition, appropriate tests (exact Fisher or chi-square for rates and Mann-Whitney U for all other endpoints) for significance were performed. Effect size was reported as risk ratio (range from 0 to infinity) for binary endpoints and by the Mann-Whitney U (MW) estimator (range from 0 to 1) for at least ordinal endpoints, each with corresponding 95% CI. Where appropriate, additionally, also linear or logistic regression models dependent on cohort, age, sex, surgery group, and baseline values were used to perform a Wald test for significance of the cohort and the related odds ratio was reported where appropriate. The latter models provide the advantage of considering potential influence factors (e.g. baseline values) but are less robust than non-parametric tests, on which we primarily based our results.

We found no significant difference in the incidence rate of severe vasoplegia between the fine filter and the control filter cohort (21.0% vs 19.6%; risk ratio (95% CI) = 1.07 (0.93–1.23), P > 0.20 for non-parametric Fisher test; P > 0.20 for Wald test on regression model with odds ratio (95% CI) = 1.05 (0.87–1.27)) (Table 2), although the rate of patients receiving norepinephrine in the fine filter cohort (82.9% vs 65.4%; P <  0.01) and the maximum rate of norepinephrine per patient in the fine filter cohort were significantly higher (median 0.09 vs 0.05; P <  0.01) (Table 2). Interestingly, the calculated rate of vasoplegia in both cohorts was here about twice as high as assumed in the SAP. The additionally analysed methylene blue rate was not significantly different between the two cohorts (9.8% vs 10.2%; P > 0.20) (Table 2).

The fine filter cohort showed significantly improved lung function by a higher minimal Horowitz index (median 206 vs 191; P = 0.04) and lower ARDS score (median 1 vs 2; P = 0.01) (Table 3). The maximum creatinine value (median 1.09 vs 1.12; P = 0.19) and our modified AKI score rate (11.8% vs 13.7%; P = 0.11) showed no significant difference between the two cohorts. No difference was found for delirium score or multi-organ dysfunction scores in the non-parametric analysis but obtained for delirium score (P <  0.01) and SAPS II score (P <  0.01) in the additional multivariate analysis when taking into account also the baseline values, age, sex, and surgery group (Table 3).

The maximum IL-6 value as marker for inflammation was significantly lower in the fine filter cohort (median of 471.5 vs 540.5 ng/l; P = 0.01) (Table 4). We hypothesised that specific subgroups of patients with a reduced immune response might especially benefit from the use of in-line filtration with finer 0.2 and 1.2 μm filters. Therefore, we conducted an additional subgroup analysis for cardiac surgery group patients only and all other remaining surgery group patients, regarding their IL-6 values. We obtained the following results in the additional subgroup analysis: P = 0.02 and MW = 0.46 (0.43–0.49) for the only cardiac surgery group vs P = 0.17 and MW = 0.48 (0.45–0.50) for the remaining patient group. This showed a lower IL-6 value for the finer filter cohort within the cardiac surgery subgroup. In the other subgroup, the difference was not significant, although the tendency of the MW effect estimator shows again a lower Il-6 value for the finer filter. We compared the effect estimators between the only cardiac and the remaining subgroup (with the appropriate chi-square test for AUCs, using the estimator and its standard deviation) and obtained no statistically significant different result (P > 0.20). Only a higher number of patients might reveal the estimated significant difference here. In addition, we assessed the IL-6 kinetics by comparing the period including IL-6 maximum value. We obtained as typical peak time a median of 3 (0–12) h for the fine filter cohort and 3 (0–13) h for the control filter cohort (Mann-Whitney test, P = 0.19, MW = 0.49 (0.47–0.51)). In this respect, the difference between the two cohorts cannot be explained by different IL-6 kinetics but more likely by the actual maximal value.

The length of both cumulated ICU (median 1.2 vs 1.7 days; P < 0.01) and hospital stay (median 14.0 vs 14.8; P = 0.01) was significantly shorter in the fine filter cohort. Considering the length of stay as a competing risk for death, analysis showed a significantly reduced length of stay in the fine filter cohort also for the survivors (ICU: median 1.0 vs 1.3 days; P < 0.01; hospital stay: 14.0 vs 14.4; P = 0.01) (Table 5).

The overall in-hospital mortality rate was comparable (13.1% vs 14.1%; P > 0.20). Similarly, the 28-day mortality was comparable (8.8% vs 10.3%; P = 0.19) (Table 5).

Analysing routine ICD-10 codes, typical in-hospital complications were significantly lower in the fine filter cohort (pneumonia 11.4% vs 14.4%, P = 0.02, and sepsis 9.6% vs 12.2%, P = 0.03) (Table 5). For rarer complications such as ischemic stroke (1.7% vs 1.7%; P > 0.20) and myocardial infarction (2.3% vs 1.4%; P = 0.06), no significant difference was found (Table 5) between the two cohorts. In addition, a composite endpoint including in-hospital mortality, myocardial infarction, ischemic stroke, pneumonia, and sepsis showed no significant difference between the two cohorts (26.9% vs 29.0%; P >  0.20) (Table 5).

Our study had several limitations. As the study was retrospective and depended on an automated readout system, no additional data could be gathered afterwards. Besides that, being a mono-centre study, composition of patients could vary in other centres but we assume, as many results hold also for various subgroups, that this difference is not essential. Furthermore, our analysis includes mainly surgical ICU patients. The accuracy of the encoded routine ICD-10 diagnoses in general has not been assessed, but we assume that for this short time period of 2 years, the coding pattern did not change relevantly. A simple intraoperative factor we cannot statistically exclude and which might be a potential bias could be seen in different surgical and anaesthesiologic teams performing the procedures. Therefore, especially, fluid management regarding type and amount could differ between the time periods. Unfortunately, we cannot normalise the amount of fluids administered to body weight, as we do not have all information to types and volumes given. But we assume due to equal pair matching in each cohort that the received fluids were approximately the same, given the high number of patients which help avoiding bigger random effects. At any time period, mostly crystalloid fluids were used in general. As we did not analyse data regarding urine output, we have to acknowledge that we lose a significant percentage of AKI cases. Furthermore, we do not have additional data about ventilation length and patients’ oxygenation.

Advantages of our propensity-matched study design are mainly statistical: first, the large number of potentially interesting endpoints being investigated and the high number of patients included that did not need to be enrolled, and second, the retrospective approach using routine data which saved time, staff, and costs. Due to the fact that we matched the cohorts equally in number, age, sex, and surgery discipline, at least no bias due to these patient characteristics should have arisen. However, we cannot exclude mild differences regarding other potentially important characteristics such as comorbidities.