How much furosemide should be administered to prevent transfusion‐associated circulatory overload? Results of a dose‐finding study

Abstract

Background and Objectives

Transfusion‐associated circulatory overload (TACO) is a common and life‐threatening transfusion complication. Because of uncertainty regarding dosing, pre‐transfusion furosemide prophylaxis is not widely endorsed. The aim of this study was to generate a furosemide dose–response curve in TACO‐susceptible patients using the multiple comparisons procedure and modelling (MCP‐Mod) methodology.

Materials and Methods

Inpatients aged ≥50 years receiving intravenous (IV) furosemide were screened for eligibility at two academic hospitals. Exclusion criteria were active bleeding, haemodynamic instability, glomerular filtration rate (GFR) < 30 mL/min/1.73 m² and diuretic therapy administered within 24 h or albumin administered within 8 h. The primary outcome measure was 6‐h urine output post furosemide administration. After incorporation of age, sex, chronic diuretic use, mean arterial pressure, GFR and serum albumin as covariates of diuretic response, MCP‐Mod was applied after every 50th enrolment until a weight‐adjusted dose–response curve was identified with 100 mL precision.

Results

One‐hundred forty‐nine patients were enrolled. Urine output varied widely at each furosemide dose. Because of the presence of outliers and a paucity of patients receiving higher doses, only those receiving doses up to 0.6 mg/kg (n = 132) were included. After incorporating covariates, linear‐log was identified as the best fitting model. Application of this formula revealed that, depending upon patient characteristics, 10–40 mg of furosemide IV would be required to achieve a diuresis volume of 400 mL, which is sufficient to offset 1 red blood cell unit.

Conclusion

We report a novel furosemide dose–response model for TACO‐susceptible patients. Once validated, this model will guide furosemide dosing for a planned controlled trial evaluating furosemide for TACO prevention.

Highlights.

• There is significant patient variability in diuretic response to intravenous furosemide.

• By incorporating key clinical variables, a precise dose–response curve can be generated for patients at risk of transfusion‐associated circulatory overload.

• The dose–response formula can be applied to clinical trials of pre‐transfusion furosemide as a means of safely and effectively decreasing adverse transfusion reactions.

INTRODUCTION

Transfusion‐associated circulatory overload (TACO) is the most commonly reported serious transfusion reaction, complicating up to 10% of transfusion exposures, with 18% of cases requiring care escalation, and death occurring in 2% [1, 2]. TACO is characterized by pulmonary oedema due to increased hydrostatic pulmonary capillary pressure and is also likely associated with inflammation [3, 4]. A two‐hit model has been proposed for TACO development, with recipient risk factors such as advanced age, history of cardiac and renal disease and pre‐transfusion positive fluid balance constituting the first hit, and factors associated with the transfusion itself (such as transfusion volume, speed and product storage lesion) constituting the second [4].

In addition to single‐unit transfusion and slow transfusion rates, pre‐transfusion administration of the loop diuretic furosemide is often used as a TACO mitigation strategy [5]. Pre‐emptive furosemide administration has been shown to prevent post‐transfusion increases in left ventricular filling pressure [6], providing a rationale for its use in this context. However, furosemide administration has not received widespread endorsement, in part due to uncertainty regarding safe and effective dosing [7, 8]. A pilot double‐blind, randomized controlled trial comparing 20 mg intravenous (IV) pre‐transfusion furosemide to placebo prior to single‐unit red blood cell transfusion for TACO prevention amongst in patients over the age of 65 showed similar post‐transfusion changes in fluid balance between the two treatment groups, and one case of TACO occurred in each arm [9]. It is possible that the preselected 20‐mg IV dose of furosemide was insufficient to effect a difference between the two arms. However, a wide range in post‐transfusion fluid balance was also observed, and some physicians declined to enrol their patients in this trial owing to the concern that even 20 mg might result in excessive diuresis.

Diuretic response to furosemide is difficult to predict, with previous studies [10, 11, 12] showing wide variability in urine output for a given dose: for example, a 40‐mg IV dose of furosemide in the placebo‐controlled randomized study of the selective A1 adenosine receptor antagonist rolofylline for patients hospitalized with acute heart failure and volume overload to assess treatment effect on congestion and renal function trial produced urine outputs with an interquartile range of 130–800 mL in acute heart failure patients [10]. Several factors have been identified as predictors of poor diuretic response, including low blood pressure, high blood urea nitrogen, chronic kidney disease, diabetes, chronic diuretic use and atherosclerosis [10, 11, 12]. In light of known urine output variability, titrating furosemide until the desired urine output is achieved has been recommended as a dosing strategy [13]. However, this approach is not applicable for use in TACO prophylaxis, where a one‐time dose is given before transfusion.

The ideal dose of furosemide for the prevention of TACO is one that will prevent circulatory overload without causing harm such as hypotension; presumably, this is best accomplished by effecting a volume of urine output that approximates the volume of blood products administered. Selecting the optimal dose to achieve this target urine output is necessary to determine furosemide efficacy in TACO prevention. Traditional phase II dose‐finding studies randomize patients to different drug doses and determine the best dose through a multiple comparisons procedure (MCP); that is, statistically comparing clinical outcomes following different dose regimens. This approach has several limitations, including the requirement for large sample sizes to adequately power multiple dose comparisons and failure to consider the pharmacokinetic and pharmacodynamic properties of the study drug when selecting doses for the trial [14]. High attrition rates in phase III clinical trials have been attributed to these limitations [14].

To overcome some of the limitations associated with MCP, modelling has been recommended as an alternative strategy for dose selection in phase II trials. This approach assumes a dose–response relationship for the study drug and estimates a dose–response curve that can be used to select the dose required to produce the desired effect [14, 15]. A limitation of the modelling approach is that the initial selection of an appropriate dose–response curve is essential for the validity of the selected model [15]. An E _max or sigmoid E _max dose–response relationship was previously reported for loop diuretics, suggesting a ceiling effect for this drug class due to saturation of the drug binding sites [16, 17, 18]. However, it is not known whether this dose–response curve can be extrapolated to patients at risk for TACO, given that several TACO risk factors, such as chronic kidney disease, are known to decrease responsiveness to furosemide [18].

The aim of our dose‐finding study was to establish a reliable furosemide dose–response curve in patients at risk for TACO. Here, we used the combined approach of MCP and modelling (MCP‐Mod) to overcome some of the limitations of these individual strategies, with the goal of mathematically modelling a statistically valid IV furosemide dose–response curve in hospitalized patients requiring diuresis. MCP‐Mod is performed by pre‐selecting a set of candidate parametric dose–response models, followed by multiple contrast testing to determine whether a dose–response relationship exists for any of the selected models [15]. The dose–response model that best fits the data is then used to estimate drug doses for clinical trials. Once validated, this model will be used to determine individualized furosemide dosing for a future randomized controlled trial evaluating pre‐transfusion furosemide as a TACO prevention strategy.

MATERIALS AND METHODS

This observational cohort study was conducted at two academic hospitals in Toronto, Canada: the University Health Network (UHN) and Sunnybrook Health Sciences Centre (SHSC). Research ethics board approval was obtained at both sites. All participants or their substitute decision makers provided written informed consent.

The charts of inpatients who had been ordered IV furosemide were screened for the same eligibility criteria used in a planned randomized controlled trial of pre‐transfusion IV furosemide for TACO prevention: inpatients 50 years of age or older, without the following exclusion criteria: (a) active bleeding (active visible bleeding, transfusion requirement of two or more red cell units in the preceding 24 h, or drop in haemoglobin greater than 20 g/L in the preceding 24 h); (b) haemodynamic instability (defined as systolic blood pressure < 90 mmHg or receiving vasopressors) or (c) glomerular filtration rate (GFR) < 30 mL/min/1.73 m² or need for renal replacement therapy; (d) non‐furosemide diuretic administered less than 24 h prior to or 6 h following index furosemide therapy, or an additional dose of furosemide administered less than 12 h prior to or 6 h following the index furosemide therapy; or (e) lack of serum albumin measurement in the 7 days preceding the index furosemide therapy, or albumin administered less than 8 h prior to index furosemide therapy. After June 2022, patients receiving an index furosemide dose of ≤40 mg were also excluded to favour enrolment of patients receiving higher furosemide doses.

Data collection

Intravenous furosemide dose (as determined by the ordering clinician), patient weight, post‐diuresis 6‐h urine output and several predicted covariates of diuretic response were recorded. Covariates identified from previous studies as modulators of diuretic response to furosemide were selected on the basis of feasibility and included: (a) age; (b) sex; (c) history of chronic diuretic use, defined as having received IV or oral diuretic during more than 50% of days in the 3 months prior to enrolment; (d) mean arterial pressure (MAP); (e) GFR or creatinine clearance (CrCl), with GFR calculated using the chronic kidney disease epidemiology collaboration (CKD‐EPI) creatinine [19] formula (UHN) and CrCl calculated using the Cockroft–Gault equation [20] (SHSC); and (f) serum albumin. While no attempt was made to approximate patient volume status at the time of diuresis, the MAP and serum albumin levels were used as surrogates for the competing effects of renal perfusion and colloid oncotic pressures on urinary natriuresis.

Patients were enrolled prospectively at UHN, where they were approached by study personnel once a bolus dose of IV furosemide was ordered in the health information system (HIS). At SHSC, patients were enrolled retrospectively through chart review of intensive care unit (ICU) and furosemide administration records obtained from the pharmacy. Clinical and demographic data were obtained from the HIS at both sites, including total urine output at 6 h post furosemide administration.

Statistical considerations

Patient demographics and clinical characteristics were reported using descriptive statistics, including mean, median and range for continuous variables, and proportions for categorical variables. For comparisons between study sites and between prospective and retrospective cohorts, the Wilcoxon rank‐sum non‐parametric test and Fisher's exact test or Chi‐square test were applied for continuous and categorical variables, as appropriate. A two‐sided p‐value < 0.05 was considered statistically significant. All analyses, including MCP‐Mod, were performed using R software, version 4.3.0 (R Foundation for Statistical Computing).

Modelling a dose–response curve for furosemide with MCP‐Mod

Six standard candidate dose–response curve models (E _max, linear‐log, exponential, sigmoid E _max, logistic and quadratic) were selected for this study, with parameter estimates of the expected maximum diuretic effect at a given weight‐adjusted furosemide dose grouped into increments of 0.1 mg/kg. Optimal contrast coefficients were assigned, assuming a balanced allocation of 20 patients per dose group. Following the collection of furosemide dosing and urine output data and incorporation of clinical covariate data, multiple contrast testing was performed to determine which of the six candidate models yielded statistically significant dose–response signals. The best fitting candidate model was determined using the Akaike information criterion (AIC). MCP‐Mod was applied to the collected data after every 50th enrolment. This process was continued until a statistically significant dose–response curve was identified (p < 0.05) with 100 mL precision.

RESULTS

Patient demographics and baseline clinical characteristics

Patient enrolment took place over 77 weeks, from 3 June 2021 to 14 November 2022, starting with UHN and with SHSC added as a second study site on 14 March 2022. Patients with incomplete measurement of urine output within the 6‐h post‐diuretic period were excluded from analysis. MCP‐Mod identified a statistically significant dose–response curve after data from 149 patients were analysed (see below), and further enrolment was therefore held. The majority of these 149 patients were male (n = 96, 64%), normotensive, moderately hypoalbuminaemic and without a history of chronic diuretic use or renal impairment. Detailed patient demographics and pre‐furosemide clinical and laboratory data are outlined in Tables 1 and 2. Comparison of patient cohorts enrolled at each study site, and between those patients enrolled prospectively versus retrospectively, are shown in Supplement 1.

TABLE 1.

Patient demographics and clinical characteristics.

Demographics	Total (N = 149)
Age (years)
Median (Q1, Q3)	69 (62, 77)
Range	50, 99
Sex
Female	53 (35.57%)
Male	96 (64.43%)
Site of admission
UHN	72 (48.32%)
SHSC	77 (51.68%)
Study cohort
Prospective	91 (61.07%)
Retrospective	58 (38.93%)
Location of admission
Ward	66 (44.30%)
ICU/CCU/high‐intensity unit	83 (55.70%)
Congestive heart failure	19 (12.75)
Cirrhosis	4 (3.36%)
Renal impairment (GFR < 60 mL/min/1.73 m2)	36 (24.16%)
Chronic diuretic therapy	19 (12.75%)
Height (cm)
N	124
Mean ± SD	168.5 ± 11.7
Median (Q1, Q3)	170 (160, 178)
Range	134, 196
Height measured
Measured	99 (66.44%)
Self‐reported	18 (12.08%)
Not documented	32 (21.48%)
Weight (kg)
N	133
Mean ± SD	78.35 ± 19.69
Median (Q1, Q3)	76.0 (65.0, 93.8)
Range	40.3, 155.4
Weight measured
Measured	124 (83.22%)
Self‐reported	7 (4.70%)
Not documented	18 (12.08%)
BMI (kg/m2)
N	119
Mean ± SD	27.66 ± 6.05
Median (Q1, Q3)	26.7 (23.0, 32.4)
Range	15.9, 49.6
ECOG status
1	4 (2.55%)
2	19 (12.10%)
3	49 (31.21%)
4	77 (49.04%)
Undetermined	8 (5.10%)

Abbreviations: BMI, body mass index; CCU, cardiac care unit; ECOG, Eastern Cooperative Oncology Group; GFR, glomerular filtration rate; ICU, intensive care unit; SHSC, Sunnybrook Health Sciences Centre; UHN, University Health Network.

TABLE 2.

Pre‐furosemide vital signs and laboratory data.

	Total (N = 149)
Pre‐furosemide vital signs
Pulse (beats/min)
N	149
Mean ± SD	87.2 ± 19.5
Median (Q1, Q3)	86 (74, 95)
Range	50, 196
Blood pressure systolic (mmHg)
N	149
Mean ± SD	128.6 ± 20.3
Median (Q1, Q3)	127 (112, 144)
Range	91, 178
Blood pressure diastolic (mmHg)
N	149
Mean ± SD	68.8 ± 13.1
Median (Q1, Q3)	69 (61, 75)
Range	42, 145
Mean arterial pressure (mmHg)
N	149
Mean ± SD	88.7 ± 12.7
Median (Q1, Q3)	89 (80, 96)
Range	62, 145
Respiratory rate (breaths/min)
N	144
Mean ± SD	21.2 ± 7.2
Median (Q1, Q3)	18 (18, 24)
Range	9, 64
SpO2 (%)
N	149
Mean ± SD	94.7 ± 10.1
Median (Q1, Q3)	96 (94, 98)
Range	1, 100
Temperature (°C)
N	148
Mean ± SD	36.89 ± 0.54
Median (Q1, Q3)	36.8 (36.5, 37.2)
Range	35.8, 38.3
Supplemental oxygen
No	54 (36.24%)
Yes	95 (63.76%)
Positive pressure ventilation
No	87 (58.39%)
Yes	62 (41.61%)
Details (n = 62)
Face mask	1 (1.61%)
Mechanical ventilation	54 (87.10%)
Nasal prongs	1 (1.61%)
Nasal high‐flow therapy	2 (3.23%)
N/A	4 (6.45%)
GFR formula
Chronic kidney disease epidemiology collaboration	72 (51.80%)
Cockroft–Gault	67 (48.20%)
Pre‐furosemide labs
Plasma creatinine (mmol/L)
N	149
Mean ± SD	82.2 ± 40.9
Median (Q1, Q3)	71 (55, 98)
Range	27, 240
GFR (mL/min/1.73 m²)
N	132
Mean ± SD	87.8 ± 40.7
Median (Q1, Q3)	87 (59, 101)
Range	30, 242
Albumin (g/L)
N	149
Mean ± SD	27.4 ± 5.2
Median (Q1, Q3)	27 (24, 31)
Range	12, 38
Sodium (mEq/L)
N	149
Mean ± SD	140.0 ± 5.2
Median (Q1, Q3)	140 (137, 144)
Range	127, 152
Potassium (mEq/L)
N	149
Mean ± SD	3.98 ± 0.47
Median (Q1, Q3)	3.9 (3.7, 4.3)
Range	2.7, 5.1

Abbreviations: GFR, glomerular filtration rate; N/A, not applicable.

Diuretic response is variable between patients

Furosemide dosing and 6‐h urine output are shown in Table 3 and Figure 1. The median (Q1, Q3) administered IV furosemide dose was 40 mg (20, 40), and doses ranged from 20 to 120 mg. The vast majority (95%) of selected doses were 20 and 40 mg. Median (Q1, Q3) urine output at 6 hours was 935 mL (655, 1250). Urine output varied significantly between patients for a given IV furosemide dose: 6‐h urine output following a 20‐mg dose ranged from 340 to 1850 mL, and following a 40‐mg dose it was 170–2800 mL. Median (Q1, Q3) weight‐based dose was 0.41 mg/kg (0.28, 0.57). Sixteen patients (11%) had missing weight data and were assigned the median dose/kg value of 0.4 mg/kg.

TABLE 3.

Diuretic response.

	Total (N = 149)
Hours since the last IV Lasix dose
N	129
Mean ± SD	24.9 ± 28.4
Median (Q1, Q3)	17 (12, 24)
Range	11, 216
Length of hospital stay at time of Lasix administration (days)
N	149
Mean ± SD	21.6 ± 52.1
Median (Q1, Q3)	9 (6, 18)
Range	0, 381
Dose administered (mg)
N	149
Mean ± SD	33.8 ± 15.9
Median (Q1, Q3)	40 (20, 40)
Range	20, 120
Distribution
20	62 (41.61%)
40	80 (53.69%)
60	2 (1.34%)
80	3 (2.01%)
120	2 (1.34%)
Dose/body weight (mg/kg)
N	133
Mean ± SD	0.460 ± 0.243
Median (Q1, Q3)	0.41 (0.28, 0.57)
Range	0.17, 1.81
Urine output measured at 6 h (mL)
N	149
Mean ± SD	1003.7 ± 495.5
Median (Q1, Q3)	935 (655, 1250)
Range	120, 2800

Abbreviation: IV, intravenous.

FIGURE 1.

Scatter plot showing dose/body weight and urine output at 6 h following furosemide administration in all 149 study participants. For 16 patients with missing dose/body weight data, the median weight‐based dose (0.4 mg/kg) was used for imputation. Dose/body weight was rounded to one decimal place.

Furosemide dose–response best approximates a linear‐log model

The six candidate dose–response curves are shown in Supplement 2. After adjustment for covariates, the best fitting candidate model initially identified through MCP‐Mod analysis was an exponential curve, which was physiologically difficult to explain given previous reports of a ceiling effect for furosemide due to saturation of drug binding sites [18]. Upon close review, the best fitting model appeared to be exponential because of one patient with very large diuresis (>2500 mL) after a 1.0 mg/kg furosemide dose (Figure 2). This result was felt to be an outlier and reflected the poor precision of the dose–response curve at doses above 0.6 mg/kg due to the relative paucity of patients receiving doses this high. Improving the precision of the curve, and thereby neutralizing the effect of this presumed outlier, would have required enrolling an additional 5–10 patients receiving doses of 0.6 mg/kg or higher, which was not feasible due to the low enrolment at these high doses and the unlikelihood of using such high doses before transfusion. Therefore, a decision was made to limit the range of the dose–response curve to a maximum dose of 0.6 mg/kg. This allowed each 0.1 mg/kg dose increment to include a minimum of 20 patients.

FIGURE 2.

Multiple comparisons procedure and modelling (MCP‐Mod) showing an exponential model as the best fitting dose–response model, showing the skewing effect of a single outlier with a very large diuresis.

When MCP‐Mod analysis was subsequently applied to this smaller sample of 132 patients receiving furosemide doses of 0.2–0.6 mg/kg, the linear‐log model was identified as achieving statistical significance with a precision of ±100 mL. After adjusting for covariates, the linear‐log model had an AIC of 2013.9, the lowest of the six models evaluated, indicating optimal fit for the data collected (full results are shown in Supplement 2). This dose–response curve is shown in Figure 3, based on a formula for urine output at 6 h (mL) defined as 717.21 + 191.33 ln(dose/weight) + 0.73 × age + 103.37 × sex + 1.63 × GFR + 5.45 × MAP − 9.56 × albumin − 62.31 × diuretic history, where age is represented in years, sex is scored as 1 = male/0 = female, diuretic history is scored as 1 = yes/0 = no, and GFR (mL/min/1.73 m²), MAP (mmHg) and albumin (g/L) are continuous variables.

FIGURE 3.

Multiple comparisons procedure and modelling (MCP‐Mod) showing a linear‐log model (solid line) as the best fitting dose–response model for furosemide, after accounting for covariates and excluding patients at dose levels >0.6 mg/kg. The dotted line represents 95% confidence intervals.

DISCUSSION

The study demonstrated considerable variation between hospitalized patients in the amount of diuresis obtained for a given weight‐adjusted dose of furosemide, reflecting inter‐individual differences in drug sensitivity. After adjusting for the selected covariates of age, sex, history of chronic diuretic use, MAP, GFR, and serum albumin, the MCP‐Mod approach derived a statistically robust dose–response curve, based on a linear‐log formula, with 100 mL precision over a dose range 0.1–0.6 mg/kg.

The linear‐log model is similar to previous reports of E _max/sigmoid E _max relationships identified in human and canine studies [16, 17] and is physiologically plausible. Specifically, it is consistent with the known ceiling effect for furosemide, a result of this drug's mechanism of action as a renal Na‐K‐2Cl transporter inhibitor in the thick ascending limb of the loop of Henle: once all transporters are blocked, no further diuresis can occur, resulting in diminishing responses with increasing doses [16].

To our knowledge, this is the first study to use MCP‐Mod to evaluate furosemide dose–response in humans. The MCP‐Mod approach offers an advantage over methods used in traditional dose‐finding studies as it both confirms whether a dose–response relationship exists for a given drug, and determines the model that best fits this relationship to guide dosing selection for subsequent trials [21]. This approach has been endorsed by the European Medicines Agency (EMA) for dose‐finding studies [14] and has been used to establish a dose–response model for furosemide in canines [17]. MCP‐Mod overcomes some of the limitations of traditional dose‐finding studies: MCP studies compare different drug doses to placebo to determine the minimum effective dose but fail to characterize the dose–response relationship. This makes them susceptible to bias in selecting which doses to study and often results in overlooking the effect of important clinical covariates. The generation of a dose–response curve is therefore preferable and can be achieved most efficiently by presuming that most drugs will follow one of several pre‐defined models, with the most accurate model identified via statistical analysis [14, 21].

Importantly, the dose–response formula generated by this study was based on data from patients who resemble those who will be targeted for enrolment in a future randomized clinical trial of pre‐transfusion furosemide, in whom a baseline incidence of TACO of 3% has been previously documented [22]. Had a different population been studied (i.e., healthy controls) the volume of diuresis predicted by the formula would likely differ significantly from what will be observed in study patients. Finally, by establishing a dose–response curve, an appropriate dose of diuretic which accounts for different volumes of transfusate as well as different patient characteristics can be selected. As an example, the dose–response formula identified in this study would suggest that a 55‐year‐old, 70‐kg woman with a GFR of 60 mL/min/1.73 m², MAP of 60 mmHg, albumin of 40 g/L and no history of chronic diuretic usage would require a pre‐transfusion dose of only 10 mg IV furosemide to achieve a diuresis of approximately 400 mL, which is sufficient to compensate for the fluid challenge of a single unit of red blood cells. Conversely, an 80‐year‐old, 90‐kg man with a GFR of 35 mL/min, MAP of 55 mmHg, albumin of 25 g/L and a history of chronic diuretic usage would require 80 mg of IV furosemide to achieve a diuresis sufficient to offset the fluid challenge of 4 units of plasma (~800 mL).

While MCP‐Mod is an efficient and versatile statistical methodology for dose‐finding studies, it depends upon an a priori assumption that the medication under study will follow a predefined type of pharmacodynamics. Thus, there is a risk that the observed data will be applied to an incorrect model as a result of an inaccurate parameter estimate [21]. This risk underlies some of the limitations of the present study. For example, the decision to exclude 17/149 patients receiving furosemide doses above 0.6 mg/kg means that not only is the validity of the dose response formula generated less certain above this dose threshold but also raises the possibility that further enrolment at higher doses would have demonstrated that the study participant who produced the very large diuresis at a 1.0 mg/kg dose of furosemide was not, in fact, an outlier, but instead was evidence that IV furosemide follows an entirely different dose–response than the linear‐log model derived from patients receiving lower doses. The exponential curve suggested by this outlier, however, would not be consistent with the known mechanism of action of furosemide, whose diuretic effect requires drug binding to a fixed number of receptors and should therefore show diminishing rather than increasing potency with higher dosing.

Another limitation of the study was the relatively small number of covariates incorporated into the model. Other modulators of furosemide response that have been identified by previous studies include diabetes mellitus [10], oedema [11] and ethnicity [12]. The inclusion of these additional variables might have improved the precision of the dose–response curve or even suggested a different model than linear‐log. However, the inclusion of more covariates would also have required the enrolment of more patients to achieve sufficient statistical power; as with the decision to limit the dose‐range analysed, further patient enrolment was not deemed feasible with available study resources. In addition, adding too many clinical covariates limits the feasibility of using the model in clinical practice.

The enrolment of patients at different institutions allowed for a greater degree of clinical heterogeneity, which is desirable in the development of a robust model. Subtle differences in the study methodology used at each site, however, raise the theoretical risk of differences in data precision. For example, the accuracy of urine output may have been higher in those patients enrolled prospectively, as nursing staff for these patients were aware that the goal was to develop a model with a degree of precision of ±100 mL. Similarly, estimation of GFR using the Cockroft–Gault equation may be slightly less accurate than what is calculated from the CKD‐EPI formula [23]. While these differences may have introduced some imprecision in the dose–response curve, the consequence is most likely to have been that more patients needed to be enrolled than would have been the case otherwise, rather than the erroneous selection of the linear‐log dose–response curve. Perhaps more significant is the fact that only one‐quarter of enrolled patients had a GFR < 60 mL/min/1.73 m², and patients with a GFR < 30 mL/min/1.73 m² were excluded altogether from analysis, which raises the possibility that the dose–response formula, while statistically precise overall, is less accurate in patients with renal impairment. Future clinical trials with larger enrolment numbers will likely address this concern through subgroup analysis.

A final potential limitation was the assumption that diuresis is an appropriate surrogate for a decrease in hydrostatic pulmonary capillary pressure, which is the proposed pathophysiological mechanism for TACO development [4]. By modelling the dose–response relationship based on urine output as the primary endpoint, it is possible that the doses selected for future studies may fail to prevent TACO despite accurately predicting the volume of urine output. In addition, the known vasodilatory effect of furosemide may mean that TACO can be prevented even with smaller volumes of diuresis than the volume of blood products infused. The invasive and labour‐intensive nature of measuring other capillary pressure surrogates such as left ventricular end‐diastolic pressure precluded their use as a primary endpoint in this dose‐finding study. Conversely, however, as the patients in this study cohort were not transfused blood products prior to being administered furosemide, any vasoregulatory effects of those blood products (i.e., nitric oxide‐scavenging from plasma‐free haemoglobin) would not have been accounted for, although the blood pressure at the time of diuresis can be considered a reasonable surrogate marker of renal perfusion. Confirming the validity of the dose–response curve generated by this study will therefore require its application to a cohort of patients being transfused, to confirm that not only confirm that the is volume of urine output observed similar to what the model predicts, but that achieving a neutral fluid balance is itself sufficient to prevent cardiopulmonary symptoms.

In summary, through the novel application of MCP‐Mod methodology to urine output data following furosemide administration to hospitalized patients with TACO risk factors, we report for the first time a dose–response model for this drug in humans at risk for this type of adverse transfusion reaction. Once validated, the model can be used to inform furosemide dosing in any upcoming randomized controlled trial evaluating furosemide for TACO prevention.

CONFLICT OF INTEREST STATEMENT

The authors declare no conflicts of interest.

Supporting information

Data S1. Supporting information.

Rotin L, Zhang L, Armali C, Malkin A, Massin S, Meirovich H, et al. How much furosemide should be administered to prevent transfusion‐associated circulatory overload? Results of a dose‐finding study. Vox Sang. 2025;120:683–693.

DATA AVAILABILITY STATEMENT

The authors agree to make data and materials supporting the results or analyses presented in their paper available upon reasonable request. It is up to the authors to determine whether a request is reasonable.