Early Net Ultrafiltration during Continuous Renal Replacement Therapy: Impact of Admission Diagnosis and Association with Mortality

Abstract

Introduction

Continuous renal replacement therapy (CRRT) is common in the intensive care unit (ICU) but a high net ultrafiltration rate (UF_NET) calculated with daily data may increase mortality. We aimed to study early UF_NET practice using minute-by-minute CRRT machine recordings and to assess its association with admission diagnosis and mortality.

Methods

We studied CRRT treatments in three adult ICUs over 7 years. We calculated early UF_NET rates minute-by-minute and categorized UF_NET into tertiles of mean UF_NET in the first 72 h and admission diagnosis. We applied Cox-proportional hazards modelling with censoring of patients who died within 72 h.

Results

We studied 1,218 patients, 154,712 h, and 9,282,729 min of CRRT (5,702 circuits). Mean early UF_NET was 1.52 (1.46–1.57) mL/kg/h. Early UF_NET tertiles were similar to, but somewhat higher than, previously reported values at 0.00–1.20 mL/kg/h, 1.21–1.93 mL/kg/h, and >1.93 mL/kg/h. UF_NET values were similar whether evaluated at 24 or 72 h or for the entire duration of CRRT. There was, however, significant variation in UF_NET practice by admission diagnosis: higher in respiratory diseases (pneumonia p = 0.01, other p < 0.0001) and cardiovascular disease (p = 0.005) but lower in cardiothoracic surgery (p = 0.04), renal (p = 0.0003) and toxicology-associated diagnoses (p = 0.01). Higher UF_NET was associated with an increased hazard of death, HR 1.24 (1.13–1.37), independent of admission diagnosis, weight, age, sex, presence of end-stage kidney disease, and severity of illness.

Conclusion

Early UF_NET practice varies significantly by admission diagnosis. Higher early UF_NET is independently associated with mortality. Impacts of UF_NET on mortality may vary by admission diagnosis. Further work is required to elucidate the nature and mechanisms responsible for this association.

Introduction

Continuous renal replacement therapy (CRRT) is a common intervention in critically ill patients with acute kidney injury, a condition associated with significant mortality [1]. Intensity of CRRT has been studied in detail with randomized trials showing no benefit with increased intensity [2], and timing of RRT trials have suggested that delayed intervention may be safest [3]. However, other aspects of CRRT beyond intensity and timing may impact outcomes, such as those related to fluid management [4–7]. Beyond overall fluid balance, a particularly important aspect of CRRT may relate to the rate of fluid removal with the CRRT machine, the so-called net ultrafiltration rate (UF_NET). This is distinct from the patient net fluid balance (all fluid input/gains less all fluid output/losses) as it is purely the volume removed during CRRT.

UF_NET has been of recent increasing interest [8–13] with studies suggesting potential harm associated with high (>1.75 mL/kg/h) UF_NET. Such harm may be particularly strong early during treatment when many patients have the capillary leak syndrome [14–16] and are unable to achieve capillary refill with high UF_NET. However, all multicentre studies of UF_NET have relied on trial databases, which reported the UF_NET as a daily value. This required the assumption that the UF_NET was applied uniformly throughout the day and was based on dividing the daily value by 24 to estimate its hourly rate. This approach, however, is flawed because the UF_NET is frequently adjusted according to perceived clinical needs and the patient’s haemodynamic state, as well as the underlying disease. Such deficiencies in our understanding of hour-to-hour or even minute-to-minute variations in early UF_NET and of the impact of the underlying disease on its values and associations suggest the need for more detailed investigations. We therefore aimed to measure early UF_NET practices by admission diagnosis at a minute-to-minute resolution. The goal was to more robustly evaluate whether an association exists between early UF_NET and mortality. We also sought to assess the impact of admission diagnosis on UF_NET intensity. Finally, we aimed to test the primary hypothesis that a high early UF_NET would still be independently associated with increased mortality risk even after the scrutiny of such detailed analysis and consideration of admission diagnosis.

Materials and Methods

Study Design

We conducted a retrospective cohort study of adult patients treated with CRRT at three university-associated tertiary referral ICUs. Because of the noninterventional nature of the study, the requirement for informed consent was waived by the Local Ethics Committees (Ethical approval: HREC LNR/16/Austin/400).

Participants and CRRT Protocol

Patients undergoing CRRT from July 2014 to October 2021 were included. Patients were excluded if they did not have an electronically recorded treatment or were undergoing both ECMO and CRRT. Patients with end-stage kidney disease (ESKD) were included but accounted for in analyses.

CRRT was provided via the Prismaflex^® device (Baxter, Chicago, IL, USA) as continuous veno-venous haemodiafiltration, an approximate 1:1 dialysis to filtration ratio and predominant pre-dilution (200 mL/h post-dilution only), or as continuous veno-venous haemodialysis. Circuits had a range of anticoagulation including citrate, heparin, low molecular weight heparin, heparin-protamine, bivalirudin, and no anticoagulation. Blood flow protocols ranged from 120 mL/min to 250 mL/min and effluent rates were targeted to 25–50 mL/kg/h.

UF_NET Measurements

UF_NET was established from machine data card events and volumetric scales data. UF_NET measurements were calculated every minute as the volume removed from the patient: UF_NET = effluent - (pre-dilution + post dilution + dialysate). UF_NET was defined in this way, as has previously been described [8–13], and is distinct from the patient net fluid balance which includes all fluid volumes the patient receives (input), less all fluid losses (output). UF_NET was corrected for patients’ actual body weight at admission, or if not available, from the machine-recorded body weight at the start of CRRT. UF_NET was calculated: (1) for each minute of treatment while on CRRT (total UF_NET); (2) for each minute in the first 24 and 72 h from the commencement of CRRT (where CRRT downtime was either excluded or accounted for as 0 UF_NET); (3) in 10-min intervals for graphical representation by admission diagnosis; (4) hourly for the first 6 h, followed by 6-h segments, for the purposes of tabulation and heatmapping.

Subsequently, UF_NET 72-h tertiles were established by censoring all patients dying within 72 h of commencement of CRRT and splitting the remaining patients into thirds by mean UF_NET in the first 72 h, as previously reported [9–11, 13]. We include analysis of UF_NET as delivered during CRRT (excluding downtime) but also present data including downtime, where UF_NET will necessarily be 0. Unless otherwise stated, UF_NET values in results are from while the CRRT circuit was running, consistent with previous studies.

Other Variables and Data Sources

Patient data including demographics (age, sex, and weight), severity of illness (APACHE III), and ICU mortality were extracted from existing databases including the ANZICS Centre for Outcome and Resource Evaluation Adult Patient Database, and hospital electronic medical records.

Statistical Analysis and Data Presentation

Patient characteristics and UF_NET data were summarized as proportions, mean (95% confidence interval [CI]) or median (interquartile range [IQR]) as appropriate. Where reported, group-wise comparisons were made by χ², analysis of variance, or Kruskal-Wallis testing. Survival analysis was undertaken using Cox proportional hazards models with cluster-robust standard errors for site. In addition, modelling was performed with a penalized smoothing spline applied to UF_NET as a predictor to establish the presence of a nonlog linear relationship. Cox models were also performed for each individual admission diagnosis to establish whether UF_NET showed a different impact on mortality by underlying disease process. Significance was considered at p < 0.05. Data were curated, analysed, and presented using R (Vienna, Austria, R version 4.3.1) and R Studio (Build 421, Boston, MA, USA), and R packages are listed in online supplementary Table S1 (for all online suppl. material, see https://doi.org/10.1159/000535315).

Results

Patient and CRRT Details

We studied 1,218 patients undergoing 5,702 CRRT treatments totalling 154,712 patient-hours, resulting in 9,282,729 min of recordings for analysis (STROBE diagram online suppl. material Fig. S1). 92% of circuits were in continuous veno-venous haemodiafiltration mode with the remainder continuous veno-venous haemodialysis and 63.8% of circuits utilized regional citrate anticoagulation. Patient weight was taken from admission weight in 92% of cases and machine-recorded weight at the start of CRRT in 8% of cases.

UF_NET Tertiles

Mean UF_NET in the first 72 h was 1.34 (1.28–1.39) mL/kg/h, or 108 (104–112) mL/h when including downtime, or 1.52 (1.46–1.57) mL/kg/h and 123 (119–127) mL/h when excluding CRRT downtime. After censoring patients that died within 72 h of CRRT commencement, when including CRRT downtime (UF_NET 0), early UF_NET tertiles were very similar to the previously reported values of 0.00–1.00 mL/kg/h, 1.01–1.73 mL/kg/h, and >1.73 mL/kg/h. These values corresponded to weight unadjusted values of 0–71 mL/h, 72–139 mL/h, and 140–390 mL/h, respectively. When excluding time when CRRT was not running, UF_NET was higher, with tertiles 0.00–1.20 mL/kg/h (0–102 mL/h), 1.21–1.93 mL/kg/h (103–164 mL/h), and >1.93 mL/kg/h (165–390 mL/h). Patients in the higher UF_NET tertiles were sicker (APACHE III risk of death 38, 42, and 43%, respectively), had a lower weight (96, 90, and 75 kg, respectively), and had higher ICU mortality (13, 15, and 21%, respectively), as demonstrated in Table 1.

Table 1.

Patient characteristics by UF_NET tertiles in the first 72 h of treatment

Variable	Died within 72 h of CRRT	Tertile 1, 0.00–1.20 mL/kg/h	Tertile 2, 1.21–1.93 mL/kg/h	Tertile 3, >1.93 mL/kg/h	p value
N	174	350	344	350	–
Age, years (95% CI)1	63.4 (61.1–65.7)	59.1 (57.5–60.8)	58.6 (57.1–60.1)	55.7 (53.9–57.5)	0.01*
Apache III risk of death, %1	68.1 (63.6–72.6)	37.9 (34.8–41)	42.1 (39.0–45.2)	43.1 (39.9–46.4)	0.01*
Weight, kg (95% CI)1	79.5 (76.7–82.4)	95.6 (92.3–98.8)	89.9 (87.6–92.1)	75.2 (73.3–77.1)	<0.0001*
Males, n (%)	97 (55.7)	256 (73.1)	237 (68.9)	204 (58.3)	<0.0001*
ICU length of stay, days (IQR)2	1.9 (1–3.5)	7.3 (3.5–16)	10.6 (6–19)	12.9 (7–22.3)	0.19
ICU mortality, n (%)	174 (100)	44 (12.6)	51 (14.8)	73 (20.9)	0.009*
Admission diagnosis – surgical, n (%)
Cardiothoracic	9 (5.2)	48 (13.7)	39 (11.3)	27 (7.7)	0.04*
Trauma	12 (6.9)	25 (7.1)	18 (5.2)	20 (5.7)	0.55
Other surgery	5 (2.9)	23 (6.6)	17 (4.9)	11 (3.1)	0.11
Admission diagnosis – medical, n (%)
Cardiovascular	58 (33.3)	66 (18.9)	98 (28.5)	97 (27.7)	0.005*
Sepsis-shock	49 (28.2)	49 (14)	56 (16.3)	52 (14.9)	0.7
Sepsis-other	13 (7.5)	27 (7.7)	27 (7.8)	24 (6.9)	0.86
Respiratory-other	5 (2.9)	9 (2.6)	19 (5.5)	38 (10.9)	<0.0001*
Renal	2 (1.1)	37 (10.6)	13 (3.8)	16 (4.6)	0.0003*
Pneumonia	7 (4)	13 (3.7)	13 (3.8)	28 (8)	0.01*
Haematology	4 (2.3)	8 (2.3)	6 (1.7)	7 (2)	0.88
Liver including transplant	3 (1.7)	2 (0.6)	8 (2.3)	5 (1.4)	0.15
Toxicology	1 (0.6)	13 (3.7)	4 (1.2)	3 (0.9)	0.01*
Other medical	6 (3.4)	30 (8.6)	26 (7.6)	22 (6.3)	0.52
ESKD, n (%)	2 (1.1)	12 (3.4)	13 (3.8)	16 (4.6)	0.73
CRRT total time, hours (IQR)2	18.1 (8.4–32.2)	49.2 (25.9–96.8)	86.2 (54.7–166.2)	124.2 (63.6–244.1)	<0.0001*
Time to CRRT, days (IQR)2	0.6 (0.2–1.8)	0.7 (0.2–3.1)	1.4 (0.4–5.1)	2.2 (0.6–6)	<0.0001*

Note: admission diagnosis percentage is a percentage of all the patients in that tertile. p value represents comparisons between the three tertiles. Time to CRRT is the time from ICU admission to commencement of CRRT.

ESKD, end stage kidney disease.

¹Mean (95% confidence interval); ²Median (interquartile range). *p < 0.05.

A higher proportion of males were present in the first and second tertile, than the third (73, 69, and 58%, respectively). Higher UF_NET tertile was associated with later commencement of CRRT with the first tertile commencing after 0.7 (0.2–3.1) days, second tertile after 1.4 (0.4–5.1) days, and third tertile after 2.2 (0.6–6) days.

UF_NET Evaluation over Time by Tertile

UF_NET mean values were similar in each tertile whether evaluated over the first 24 h, 72 h, or for the entire duration of CRRT (Table 2), although patients showed the greatest separation in UF_NET between tertiles at 48–72 h and in the second and third tertile, patients had similar UF_NET by 120 h (online suppl. material Fig. S2). Within patient variability measured by standard deviation was higher in the third tertile. However, when measured by the coefficient of variance (i.e., variation in relation to the magnitude of the mean), this effect was reversed.

Table 2.

UF_NET details by UF_NET tertile

Variable	Died within 72 h of CRRT	UFNET tertile 1, 0.00–1.20 mL/kg/h	UFNET tertile 2, 1.21–1.93 mL/kg/h	UFNET tertile 3, >1.93 mL/kg/h	p value
UFNET - all CRRT, mL/h (95% CI)	62.8 (52.5–73.1)	73.2 (67.3–79)	135.1 (131–139.3)	168.3 (163.3–173.3)	<0.0001*
UFNET first 24 h, mL/h (95% CI)	66.4 (55–77.7)	51 (45.6–56.3)	132 (125.5–138.6)	192.3 (184.6–199.9)	<0.0001*
UFNET first 72 h, mL/h (95% CI)	65.7 (55–76.5)	58.8 (54.2–63.5)	141.7 (137.6–145.8)	197.1 (191.7–202.6)	<0.0001*
UFNET - all CRRT, mL/kg/h (95% CI)	0.8 (0.7–1)	0.8 (0.7–0.8)	1.5 (1.5–1.6)	2.3 (2.2–2.4)	<0.0001*
UFNET first 24 h AT, mL/kg/h (95% CI)	0.8 (0.7–1)	0.5 (0.4–0.5)	1.4 (1.3–1.4)	2.4 (2.3–2.6)	<0.0001*
UFNET first 24 h ITT, mL/kg/h (95% CI)	0.9 (0.7–1)	0.5 (0.5–0.6)	1.5 (1.4–1.5)	2.6 (2.5–2.7)	<0.0001*
UFNET first 72 h AT, mL/kg/h (95% CI)	0.8 (0.7–1)	0.5 (0.5–0.5)	1.4 (1.3–1.4)	2.4 (2.3–2.5)	<0.0001*
UFNET first 72 h ITT, mL/kg/h (95% CI)	0.9 (0.7–1)	0.6 (0.6–0.6)	1.6 (1.6–1.6)	2.7 (2.6–2.8)	<0.0001*
UFNET variability first 72 h, SD (95% CI)	0.65 (0.54–0.75)	0.53 (0.49–0.57)	0.94 (0.9–0.97)	1.25 (1.2–1.31)	<0.0001*
UFNET variability first 72 h, CoV (95% CI)	145.3 (74.3–216.3)	156.8 (126.9–186.7)	74.2 (70–78.5)	62.9 (52.3–73.4)	<0.0001*
Max UFNET, mL/h (95% CI)	151 (128.9–173.1)	212.7 (197.4–227.9)	344.2 (327–361.4)	385.8 (372.7–398.8)	<0.0001*
Max UFNET first 24 h, mL/h (95% CI)	138.8 (117.5–160.1)	120.1 (109.5–130.6)	245.7 (232–259.4)	303.1 (292.9–313.3)	<0.0001*
Max UFNET first 72 h, mL/h (95% CI)	151 (128.9–173.1)	164.8 (153.4–176.2)	302.2 (289.3–315.1)	349.7 (339.5–359.9)	<0.0001*
Max UFNET, mL/kg/h (95% CI)	2.0 (1.7–2.3)	2.3 (2.1–2.5)	3.9 (3.7–4.1)	5.3 (5.1–5.5)	<0.0001*
Max UFNET first 24 h, mL/kg/h (95% CI)	1.9 (1.6–2.2)	1.3 (1.2–1.4)	2.8 (2.6–2.9)	4.2 (4–4.4)	<0.0001*
Max UFNET first 72 h, mL/kg/h (95% CI)	2 (1.7–2.3)	1.8 (1.6–1.9)	3.4 (3.3–3.5)	4.8 (4.7–5)	<0.0001*
Time to 90% maximum UFNET first 72 h, h (IQR)	2.5 (0–10.3)	12.9 (1.2–30.3)	18.0 (4.7–39.9)	14.0 (5.1–36.1)	<0.0001*
Time on CRRT first 72 h, % (95% CI)	31.4 (27.8–35.1)	54.1 (51–57.1)	75.2 (72.6–77.9)	76.5 (73.6–79.3)	<0.0001*

All values are mean (95% confidence interval) excluding time to 90% maximum UF_NET which is presented as the median (interquartile range). SD and CoV values are calculated for each patient and the mean and 95% CI calculated of these values as a summary of variability.

Higher SD with increasing UF_NET tertile indicates there is greater dispersion of UF_NET values; however, the coefficient of variance decreases – suggesting the variation in relation to the mean is lower. Lower fluid removal may be associated with more changes in the fluid removal amount due to patient instability, thus increasing the coefficient of variance.

SD, standard deviation; CoV, coefficient of variation; Max, maximum – represents maximum value in a single minute; AT, as treated (includes UF_NET 0 when CRRT not running); ITT, intension to treat (includes only values of UF_NET when circuit running, i.e., prescribed UF_NET).

As expected, the mean maximum UF_NET in the first 72 h was higher with increasing UF_NET tertile (1.8, 3.4, 4.8 mL/kg/h, or 165, 302, 350 mL/h, respectively). However, time to maximum UF_NET showed a different pattern with the first and third tertile achieving the most rapid acceleration (time to 90% maximum UF_NET 12.9 and 14.0 h, vs. 18.1 h for the middle tertile). Patients in each tertile showed a gradual increase in UF_NET over the first 6–12 h, as shown in Figure 1 and online supplementary Material Figure S3. Those in the lowest tertile had proportionally less time on CRRT compared to higher tertiles (54%, 75%, 77%, respectively).

Fig. 1.

UF_NET over time (first 72 h) by UF_NET tertile. Low UF_NET: 0.00–1.20 mL/kg/h; intermediate UF_NET: 1.21–1.93 mL/kg/h; and high UF_NET: >1.93 mL/kg/h. Values are mean, error ranges are 95% confidence interval for each minute in the first 72 h.

UF_NET by Admission Diagnosis

Patient characteristics, outcomes, and CRRT data by admission diagnosis are presented in online supplementary material Table S2, with UF_NET data presented in Table 3. UF_NET varied significantly by admission diagnosis (Figure 2; online suppl. material Fig. S4; p < 0.0001) such that patients with pneumonia or other respiratory diseases had the highest UF_NET at 1.96 mL/kg/h (1.65–2.26) and 2.17 mL/kg/h (1.91–2.42), respectively, similar to those with liver failure or liver transplant at 1.86 mL/kg/h (1.24–2.47).

Table 3.

UF_NET data by admission diagnosis

Admission diagnosis category	n	UFNET 72 h, mL/kg/h1	UFNET 72 h SD, mL/kg/h1	UFNET 72 h CoV, %1	Max UFNET 72 h, mL/kg/h1	Time to 90% Max UFNET 72 h, h2	Died 72 h, n (%)	UFNET Tertile 1, n (%)	UFNET Tertile 2, n (%)	UFNET Tertile 3, n (%)
Cardiothoracic surgery	123	1.39 (1.24–1.54)	0.84 (0.75–0.92)	123.43 (50.23–196.62)	2.98 (2.66–3.31)	9.9 (2.3–35.1)	9 (7)	45 (37)	48 (39)	30 (24)
Trauma	75	1.36 (1.14–1.58)	0.8 (0.7–0.91)	107.02 (82.34–131.71)	2.9 (2.51–3.29)	17.9 (2.6–48)	12 (16)	32 (43)	20 (27)	23 (31)
Other surgical	56	1.28 (1.05–1.52)	0.79 (0.66–0.93)	88.04 (60.75–115.33)	2.63 (2.22–3.05)	18.5 (3.8–43.1)	5 (9)	24 (43)	19 (34)	13 (23)
Cardiovascular	319	1.61 (1.51–1.71)	0.95 (0.9–1.01)	119.59 (89.92–149.25)	3.45 (3.26–3.64)	16.1 (3.2–31.6)	58 (18)	85 (27)	117 (37)	117 (37)
Sepsis-shock	206	1.42 (1.28–1.56)	0.83 (0.75–0.91)	108.81 (75.76–141.87)	2.95 (2.69–3.2)	13.5 (2.7–30.9)	49 (24)	74 (36)	69 (33)	63 (31)
Sepsis/infection-not pneumonia	91	1.42 (1.21–1.63)	0.75 (0.64–0.86)	90.19 (55.23–125.14)	2.9 (2.43–3.37)	7.8 (2–30.3)	13 (14)	33 (36)	31 (34)	27 (30)
Pneumonia	61	1.96 (1.65–2.26)	1.06 (0.9–1.22)	111.85 (30.05–193.65)	4.0 (3.4–4.6)	8.3 (4.6–22.4)	7 (11)	16 (26)	15 (25)	30 (49)
Respiratory-not pneumonia	71	2.17 (1.91–2.42)	1.11 (0.98–1.25)	61.75 (53.7–69.8)	4.24 (3.8–4.68)	13.1 (2.7–41.2)	5 (7)	9 (13)	21 (30)	41 (58)
Renal including transplant	68	1.29 (1–1.59)	0.76 (0.62–0.9)	98.42 (72.73–124.1)	2.72 (2.21–3.22)	13.7 (2.4–32.6)	2 (3)	34 (50)	16 (24)	18 (26)
Liver failure or transplant	18	1.86 (1.24–2.47)	1.07 (0.71–1.44)	81.86 (45.7–118.03)	3.93 (2.86–4.99)	13.1 (4.6–22.7)	3 (17)	4 (22)	8 (44)	6 (33)
Haematological	25	1.61 (1.11–2.12)	0.84 (0.6–1.09)	177.42 (−27.29 to 382.13)	3.16 (2.26–4.06)	7.6 (2.8–28.7)	4 (16)	10 (40)	6 (24)	9 (36)
Toxicology	21	0.86 (0.42–1.31)	0.48 (0.28–0.68)	79.69 (40.42–118.96)	1.74 (1.02–2.45)	4.9 (0.0–15.3)	1 (5)	12 (57)	6 (29)	3 (14)
Other medical	84	1.37 (1.15–1.58)	0.74 (0.62–0.86)	63.22 (50.03–76.4)	2.73 (2.33–3.14)	6.9 (0.3–19.6)	6 (7)	32 (38)	25 (30)	27 (32)

SD, standard deviation; CoV, coefficient of variation. ¹Values are mean (95% confidence interval). ²Values are median (interquartile range).

Fig. 2.

UF_NET in the first 72 h by admission diagnosis. Patients admitted with respiratory illnesses, pneumonia, liver failure/transplant, and cardiovascular diseases had higher UF_NET. Patients admitted with sepsis, particularly with shock, trauma, or toxicological diagnoses, had lower rates of UF_NET.

Patients with a toxicological admission diagnosis had the lowest UF_NET at 0.86 mL/kg/h (0.42–1.31), but surgical patients also had relatively low UF_NET. For example, cardiothoracic patients had a UF_NET of 1.39 mL/kg/h (1.24–1.54); those with trauma had a value of 1.36 mL/kg/h (1.14–1.58); and those with other surgical had a value of 1.28 mL/kg/h (1.05–1.52). Patients with renal diagnoses, nonpneumonic sepsis, or with septic shock also showed relatively lower UF_NET in the first 72 h at 1.29 (1.00–1.59) mL/kg/h, 1.42 mL/kh.hr (1.21–1.63), and 1.42 mL/kg/h (1.28–1.56).

When plotted over time by admission diagnosis, UF_NET showed that pneumonia and other respiratory diseases had the most rapid increase in UF_NET, achieving their peak value at 8.3 and 13.1 h, respectively, settling down to a level closer to other patients with sepsis/septic shock only by 24 h (online suppl. material Fig. S5A–D). Patients with liver failure or liver transplant had a similar acceleration to a higher peak by 13.1 h but dropping back by 36 h. UF_NET varied significantly by hospital, with one hospital generally having lower UF_NET values (online suppl. material Fig. S6, S7).

Multivariate Analysis – Associations with ICU Mortality

After adjustment for multiple key variables and using cluster-robust standard errors for site (Table 4), UF_NET in the first 72 h was associated with the increased risk of ICU mortality, HR 1.24 (1.13–1.37), p < 0.0001. Other factors associated with ICU mortality included APACHE III score, age, presence of ESKD, and admission diagnosis.

Table 4.

Multivariate analysis of impacts on ICU mortality, censoring patients who died within 72 h of commencement of CRRT, with clustering by site

	Hazard ratio	p value
UFNET in first 72 h, per mL/kg/h	1.24 (1.13–1.37)	<0.0001*
Weight, per 10 kg	1.01 (0.94–1.08)	0.85
APACHE III risk of death, per 10%	1.07 (1.05–1.09)	<0.0001*
Age, per 10 years	1.15 (1.13–1.17)	<0.0001*
Male sex	1.33 (0.85–2.06)	0.21
Admission diagnosis
Trauma	0.88 (0.56–1.39)	0.59
Other surgical	1.04 (0.81–1.33)	0.77
Cardiovascular	1.25 (1.13–1.38)	<0.0001*
Sepsis-shock	1.08 (0.75–1.55)	0.7
Sepsis/infection-not pneumonia	0.89 (0.52–1.52)	0.67
Pneumonia	0.69 (0.46–1.06)	0.09
Respiratory-not pneumonia	0.57 (0.51–0.64)	<0.0001*
Renal including transplant	0.56 (0.16–1.9)	0.35
Liver failure or transplant	3.24 (1.53–6.85)	0.002
Haematological	3.45 (2.04–5.83)	<0.0001*
Toxicology	0.00 (0.00–0.00)	<0.0001*
Other medical	1.61 (1.29–2.00)	<0.0001*
ESKD	1.76 (1.51–2.06)	<0.0001*

UF_NET only includes delivered UF_NET, excludes when CRRT is off.

Cox-proportional hazards analysis demonstrated an increased hazard of death with higher UF_NET in the first 72 h, increasing APACHE III risk of death, increasing age, presence of ESKD, and admission diagnosis of liver failure/transplant, cardiovascular, haematological, or other medical (vs. cardiothoracic surgery). Nonpneumonic respiratory disease and toxicology admissions had a reduced hazard of death. Note that the model includes cluster-robust standard errors for hospital.

ESKD, end stage kidney disease.

*p < 0.05.

Similarly, analysis by UF_NET tertile showed increasing mortality hazard with each tertile (online suppl. material Table S3). Applying a penalized smoothing spline to UF_NET within the Cox model suggested an increase in hazard ratio up to 0.75 mL/kg/h, followed by a relatively flat HR up to 2.5 mL/kg/h and then a continued log-linear increase thereafter (Fig. 3). Analysis by admission diagnosis demonstrated differential effects of UF_NET (Fig. 4; online suppl. material Table S4), with cardiothoracic surgery demonstrating the reduced risk of death with increasing UF_NET, HR 0.49 (0.29–0.82) per mL/kg/h, p = 0.007, while cardiovascular diseases, HR 1.28 (1.08–1.51), p = 0.005; pneumonia, HR 1.48 (1.27–1.71), p < 0.0001; and other respiratory diseases, HR 2.71 (1.71–4.29), p < 0.0001, showing increases in risk with increasing UF_NET.

Fig. 3.

Hazard ratio for UF_NET as a penalized spline parameter within the multivariable Cox model. A penalized smoothing spline was applied to the continuous variable UF_NET. Note the logarithmic scale of y-axis. Hazard increases up to 0.75 mL/kg/h and then is relatively flat up to 2.5 mL/kg/h where it increases further.

Fig. 4.

Cox models for each admission diagnosis subgroup, hazard of UF_NET. Hazard ratios with 95% confidence intervals are shown for each admission diagnosis. Note the logarithmic scale of x-axis. Patients with respiratory, pneumonia, or cardiovascular diagnoses showed a strong association with increasing hazard of death with increasing UF_NET. Patients admitted following cardiothoracic surgery had reduced hazard with increasing UF_NET.

Discussion

Key Findings

We conducted a detailed analysis of early UF_NET with data from more than a 1000 patients in three academic ICUs over a 7-year period and used electronic CRRT machine data to analyse the characteristics of such early UF_NET and their modulation according to the underlying admission diagnosis. We aimed to test whether, after such detailed analysis, UF_NET was affected by admission diagnosis and whether it remained independently associated with increased mortality. We found that UF_NET tertiles had higher cut-off values than those reported in previous studies and that early values in the first 72 h appear to differentiate patients well, influencing the UF_NET mean values over the whole duration of CRRT. We also found, however, that UF_NET was affected by body weight and sex, implying that its prescription was not adjusted by body weight. Moreover, we found that early UF_NET differed according to the underlying diagnosis, with the highest values in patients with pulmonary disease or liver disease. Additionally, we found that after adjustment for such underlying conditions, the presence of ESKD, age, weight, illness severity on admission, and early UF_NET remained positively associated with increased risk of death. The application of a penalized spline analysis suggested that the risk of death likely increases mostly beyond 2.5 mL/kg/h UF_NET in the first 72 h, rather than a cut-off of the highest tertile. Finally, we found higher UF_NET was associated with increased mortality in patients with pulmonary diagnoses, including pneumonia, as well as (to a lesser extent) cardiovascular diseases; however, increasing UF_NET showed improvements in mortality in patients who were admitted following cardiothoracic surgery, with the caveat that these patients had an overall lower UF_NET.

Relationship to Previous Studies

A higher UF_NET rate has been independently associated with increased mortality in both acute and chronic haemodialysis patients [9, 17, 18]. In CRRT-treated patients, using daily data form a large, randomized trial of dose intensity, Murugan et al. [19] demonstrated that higher UF_NET was associated with both a greater mortality risk and a greater risk of delayed recovery to dialysis independence. Several mechanisms may plausibly mediate the possible effect of higher UF_NET as a contributor to increased mortality. For example, greater UF_NET can cause haemodynamic instability because of insufficient rates of capillary refill and associated intravascular volume [19, 20]. This effect may be particularly strong in the early phases of critical illness when patients experience capillary leak and are therefore unable to achieve capillary refill. Intravascular volume depletion can, in turn, induce myocardial ischaemia [21, 22] and myocardial stunning [23]. In addition, greater UF_NET may increase convective losses of phosphate and thus contribute to low phosphate levels [24–26] and their attendant morbidity. Finally, a combination of these factors may contribute to the increased risk of mortality. Our tertile values were higher than those reported by Murugan et al. [9]; however, this may be due to that study including the UF_NET for all CRRT, rather than the first 72 h as in the present study.

Aligned with our findings, a recent single centre study [27] showed that in CRRT-treated patients and during the first 48 h, an early NUF greater than 1.75 mL/kg/h appeared to have a direct adverse effect on mortality compared to a NUF less than 1.01 mL/kg/h. However, the study suffered from a high risk of type 1 error due to a relatively small sample size. Our study found a high UF_NET to be associated with mortality in concordance with previous work; however, the use of penalized smoothing spline analysis suggested that the cut-off for increasing mortality may be higher than the highest tertile, at around 2.5 mL/kg/h. To the best of our knowledge, no study has previously addressed the impact of admission diagnosis category on the UF_NET and its relationship with mortality.

Implications of Study Findings

Our findings imply that early UF_NET is a reliable descriptor of overall UF_NET during the whole CRRT treatment cycle. Moreover, they imply that UF_NET prescription is typically not adjusted for body weight. Thus, heavier or male patients may be less likely to be exposed to high UF_NET and its associated risks. In addition, they imply that the underlying admission diagnosis significantly influences the UF_NET rate. Finally, they show that even after adjustment for the effect of such diagnosis, a higher UF_NET remains an independent risk factor for mortality, although it may be at a higher extreme (≥2.5 mL/kg/h).

Strengths and Limitations

Our study has several strengths. It was multicentric in nature and involved assessment of UF_NET in more than 1000 patients and more than 5,000 circuits, thus providing confidence in the robustness of its findings. The UF_NET values, dynamics, time to peak level, and relationship to ICU admission diagnosis were analysed to a level of detail, which is unprecedented in the literature. Such detailed analysis provides further confidence in the validity and robustness of our observations. The data represent a contemporary dataset, which overcomes the problem associated with previous studies based on data from more than 15 years ago. Finally, the observation that admission diagnosis modifies the UF_NET is important in helping design trials aimed at comparing usual care with a targeted reduction of UF_NET rates.

We acknowledge several limitations. First, this is not a randomized controlled trial. Therefore, the association demonstrated by our study cannot be inferred to indicate causality and are only hypothesis-generating. Second, although we demonstrated, as in previous studies, that a higher UF_NET is associated with increased mortality even after adjustment for key variables that also predict mortality, unmeasured confounders may be present that explain such findings. In particular, we did not include patient net fluid balance, which is an important confounder, and the later commencement of CRRT observed in higher tertile patients may be a result of greater fluid accumulation and a perceived requirement of higher levels of deresuscitation. However, fluid balance appears to operate in an opposite direction to UF_NET with a more negative balance being associated with reduced mortality [28] and with prior studies demonstrating a similar effect to the present analysis, independent of fluid balance [9, 10, 27]; furthermore, fluid balance is not as accurately measured or accounted for in such granularity as UF_NET. Third, we focused on early UF_NET because we reasoned that this would be the period when patients would be the most vulnerable to intravascular volume depletion and its adverse consequences. However, we did not measure intravascular volume or capillary leak. Thus, further studies are needed to confirm or refute this possible mechanism of injury. We included patients with ESKD; however, this was only 3.5% of the cohort and it did not materially affect the findings when included as a covariate. Additionally, we did not include patient haemodynamics in the analysis and this may be an important area of interest, although it would represent an alternative investigation to the present study. Another limitation relates to the timing of CRRT, given patients in higher UF_NET tertiles showed significantly longer durations between ICU admission and commencement of CRRT. This may be a result of greater fluid accumulation early in admission and a perceived requirement of greater intensity of deresuscitation; however, it remains an important confounder. We also had reduced data capture prior to a machine software update in 2016/2017 and this means not all patients that underwent CRRT had data available for analysis. Finally, it is possible that there is heterogeneity of treatment effect in relation to UF_NET, such that some patients may benefit for a higher UF_NET either when delivered later during the patient’s illness or when fluid overload is severe. Further detailed investigations in specific subgroups of patients are required to confirm or refute such differential effects.

Conclusions

Early UF_NET practice obtained directly from CRRT machines is a detailed and robust representation of subsequent UF_NET values and not dissimilar to tertiles previously reported in the literature. However, it varies significantly according to admission diagnosis, being the highest in patients with a respiratory or liver disease-associated category of admission. After accounting for the impact of such diagnostic categories, however, higher early UF_NET is independently associated with mortality. Additionally, the impact of UF_NET may vary by admission diagnosis. Further work is required to elucidate the nature and mechanism responsible for this association and its modulation by underlying diagnosis.

Acknowledgements

We would like to acknowledge Megan Taylor of Baxter Healthcare (Australia) for assistance with the collection of the CRRT machine data used in this study. We would also like to thank Kathleen Collins and Alison Wells who extracted ANZICS APD data required for this study.

Statement of Ethics

The Human Research Ethics Committee (Austin Health) approved this study and waived the need for informed consent (HREC LNR/16/Austin/400).

Conflict of Interest Statement

The authors have no conflicts of interest to declare.

Funding Sources

Benjamin Sansom has received PhD scholarship funding from the University of Melbourne.

Author Contributions

Study design and final approval of the manuscript: J.P., A.U., R.B., and B.S.; implementation: B.S.; statistical plan: B.S. and J.P.; data collection/curation: B.S.; data analysis: B.S., J.P., and R.B.; and manuscript: B.S. and R.B.

Funding Statement

Benjamin Sansom has received PhD scholarship funding from the University of Melbourne.

Data Availability Statement

Ethical approval did not include permission to make patient data available publicly. On request to the corresponding author, anonymized data may be shared, subject to ethical approval by the relevant healthcare research Ethics Committee.