A Pilot Randomized, Controlled, Double-Blind Trial of Bumetanide to Treat Neonatal Seizures

Abstract

Objective:

In the absence of controlled trials, treatment of neonatal seizures has changed minimally despite poor drug efficacy. We tested bumetanide added to phenobarbital to treat neonatal seizures in the first trial to include a standard-therapy control group.

Methods:

A randomized, double-blind, dose-escalation design was employed. Neonates with postmenstrual age 33 to 44 weeks at risk of or with seizures were eligible. Subjects with electroencephalography (EEG)-confirmed seizures after ≥20 and <40mg/kg phenobarbital were randomized to receive additional phenobarbital with either placebo (control) or 0.1, 0.2, or 0.3mg/kg bumetanide (treatment). Continuous EEG monitoring data from ≥2 hours before to ≥48 hours after study drug administration (SDA) were analyzed for seizures.

Results:

Subjects were randomized to treatment (n = 27) and control (n = 16) groups. Pharmacokinetics were highly variable among subjects and altered by hypothermia. The only statistically significant adverse event was diuresis in treated subjects (48% vs 13%, p = 0.02). One treated (4%) and 3 control subjects died (19%, p = 0.14). Among survivors, 2 of 26 treated subjects (8%) and 0 of 13 control subjects had hearing impairment, as did 1 nonrandomized subject. Total seizure burden varied widely, with much higher seizure burden in treatment versus control groups (median = 3.1 vs 1.2 min/h, p = 0.006). There was significantly greater reduction in seizure burden 0 to 4 hours and 2 to 4 hours post-SDA (both p < 0.01) compared with 2-hour baseline in treatment versus control groups with adjustment for seizure burden.

Interpretation:

Although definitive proof of efficacy awaits an appropriately powered phase 3 trial, this randomized, controlled, multicenter trial demonstrated an additional reduction in seizure burden attributable to bumetanide over phenobarbital without increased serious adverse effects. Future trials of bumetanide and other drugs should include a control group and balance seizure severity.

Seizures are a frequent manifestation of acute cerebral injury and are refractory to standard antiseizure drugs in 40 to 60% of neonates,^1,2 and higher neonatal seizure burden is likely to be associated with worse neurologic outcome.^2–5 Despite these findings, little has changed in neonatal seizure treatment over the past decades, owing in large part to a lack of neonatal antiseizure drug trials.^6,7 The few published trials of drugs for neonatal seizures either have employed an open label design^8,9; have allowed subjects to cross over,^1,10 limiting evaluation of drug safety and efficacy; or were retrospective studies with historical controls.^11,12 There is increasing interest in conducting anticonvulsant trials,^13–15 motivated also by elucidation of unique characteristics of neonatal neurons that facilitate seizures. The Na-K-Cl cotransporter NKCC1 is highly expressed in many developing cortical neurons, and this expression is correlated with a depolarizing rather than inhibitory action at γ-aminobutyric acid type A (GABA_A) receptors.¹⁶ Experimental data showed that by blocking NKCC1, the diuretic bumetanide could be an effective adjunctive therapy to GABAergic anticonvulsants.^16–20 We proposed to test bumetanide as add-on therapy to the commonly used GABAergic drug phenobarbital to treat neonatal seizures, in the first controlled trial of a drug with an age-specific mechanism of action. We employed a sequential dose-escalation design to test bumetanide, because experimental data suggested that the effective dose for seizure reduction was higher than that used previously for diuresis but did not indicate an optimal human dose.^21–23 Furthermore, bumetanide safety and pharmacokinetics/pharmacodynamics (PK/PD) had not been evaluated in neonates with seizures. We hypothesized that, compared with standard phenobarbital monotherapy, bumetanide would be safe and reduce seizure burden, and that the reduction in seizure burden would be proportionate to bumetanide exposure.

Patients and Methods

Population and Trial Design

We enrolled neonates at postmenstrual age 34 to 44 weeks if they had clinically suspected or electroencephalography (EEG)-proven (ie, confirmed) seizures, or were at high risk for developing seizures caused by hypoxic–ischemic encephalopathy (HIE), focal stroke, intracranial hemorrhage (ICH), acute meningoencephalitis, brain malformation, or a suspected/known genetic disorder. Although some genetically based neonatal onset epilepsies may not be best treated by bumetanide, it is typically impossible to make a definitive genetic diagnosis in the first hours to days after onset of acute seizures, when decisions about eligibility and randomization need to be made. We excluded neonates with seizures caused by transient metabolic abnormalities or inborn errors of metabolism; neonates who had received bumetanide, furosemide, phenytoin, or ≥40mg/kg phenobarbital; neonates with total bilirubin > 15mg/dl; and neonates treated with extracorporeal membrane oxygenation or at risk of imminent death. Written informed consent was obtained from one or both parents/guardians. The study was conducted with institutional review board approval at 4 participating neonatal intensive care units in Boston, Massachusetts, and monitored by an external data and safety monitoring committee (DSMC). This trial was prospectively registered with ClinicalTrials.gov (NCT00830531); neonates were enrolled from 2010 to 2017.

We employed a randomized, double-blind, dose-escalation design to test bumetanide doses of 0.1, 0.2, and 0.3mg/kg in comparison to a control group (normal saline), given in conjunction with 5 to 10mg/kg phenobarbital (bumetanide+phenobarbital vs saline+phenobarbital). The choice of 5 or 10mg/kg phenobarbital was at the discretion of the treating physician; doses and levels were the same in the control and bumetanide groups. A 2:1 ratio of treatment:control subjects was planned with up to 8 treated subjects per dose. Stratification by hypothermia treatment was employed to balance any potential effect of therapeutic hypothermia on bumetanide PK/PD and safety. All study investigators, clinicians, and family members were masked to treatment assignment, except study statisticians.

Procedures and Endpoints

Continuous video-EEG monitoring (cvEEG) with 10–20 montage was initiated after study enrollment, if not already started (Fig 1).²⁴ Subjects were randomized if an EEG-proven seizure (confirmed by a study pediatric neurophysiologist) occurred at least 30 minutes after a loading dose of ≥20 to <40mg/kg phenobarbital. Randomization was assigned by research pharmacists using stratified block randomization with random block sizes. Persistent seizures after study drug administration (SDA) were managed per treating clinicians; cvEEG monitoring was continued for ≥48 hours after SDA. Start and stop times of each seizure and cvEEG recording were determined to derive seizure burden in min/h. Total seizure burden (in min/h) for each subject was defined as the sum of all seizure activity from onset of first suspected/confirmed seizure to end of last EEG-confirmed seizure. A single pediatric neurophysiologist (A.M.B.) with expertise in neonatal EEG interpretation supervised all cvEEG data analysis.²⁵ We defined baseline and drug response periods as 0 to 2 hours prior to SDA (pre-SDA) and 0 to 4 hours after SDA (post-SDA), respectively, based on expected bumetanide PK/PD. The first 15 minutes post-SDA were excluded to account for bumetanide distribution and crossing of blood–brain barrier. Because the trial employed an add-on therapy design, with randomization only if seizures persisted after the phenobarbital load, neonates could have EEG monitoring initiated either before seizures started (mostly neonates with HIE) or after presenting seizures but before the loading dose of phenobarbital. Hence, even if the qualifying seizure occurred at the minimum time interval of 30 minutes after the phenobarbital load, EEG monitoring was started ≥2 hours before SDA in all subjects.

FIGURE 1:

Study algorithm. EEG = electroencephalography; HIE = hypoxic–ischemic encephalopathy.

Bumetanide levels were measured up to 6 times per subject at prespecified times from 2 minutes to ~16 hours post-SDA (Fig 2).²⁶ Individual bumetanide concentration–time data were analyzed first by noncompartmental analysis (WinNonlin, v6.3), with bumetanide exposure (area under the curve [AUC]) for the drug response period calculated by the trapezoidal rule. A population PK analysis was then performed incorporating a nonlinear mixed effects modeling approach (NONMEM, v7.2)²⁷ to determine bumetanide PK parameters. Several structure models were tested, and demographic parameters including sex, race, weight, and postmenstrual age were evaluated in a covariate analysis for potential association with PK parameters. The final PK model was selected based on goodness-of-fit plots and a simulation-based visual predictive check.²⁸

FIGURE 2:

Data collection and laboratory drawing schedule. ALT = alanine aminotransferase; AST = aspartate aminotransferase; BTN = bumetanide; BUN = blood urea nitrogen; EEG = electroencephalography.

Predefined clinical and laboratory data were collected daily for 7 days after enrollment to evaluate bumetanide safety (see Fig 2). Measures of vital signs, fluid status, electrolytes, liver and renal function, doses of vasopressors/inotropes, and fluid boluses were compared between 24-hour periods before and after SDA and between treatment and control groups. Safety was also evaluated by comparing the rate of serious adverse events (SAEs) and minor adverse events between treatment and control groups, with SAEs as defined by the US Food and Drug Administration. The DSMC reviewed the safety data at each meeting (every 6 months, after the completion of each dose group, and as needed) to determine whether there were significant differences between treatment and control groups that indicated a significant safety concern. Notably, enrollment of subjects was put on hold after completion of each dose group so that the DSMC could evaluate safety and other data before proceeding with enrollment at the next higher dose of bumetanide.

The primary endpoints were determination of the pharmacokinetics and safety of bumetanide as add-on therapy to treat neonatal seizures. An exploratory endpoint was the effect of bumetanide dose and exposure on seizure burden.

Statistical Analysis

Comparisons of clinical characteristics, laboratory data, adverse events, drug levels, and seizure burden between combined treatment and control groups and pre-SDA comparisons across control and all 3 bumetanide dose groups used Fisher exact tests for categorical measures and exact Wilcoxon rank sum or exact Kruskal–Wallis tests for continuous measures. Trend comparisons of post-SDA variables across control and increasing bumetanide dose groups used exact Cochran–Armitage trend tests for binary measures and exact Jonckheere–Terpstra tests for continuous measures. To further compare the combined treatment and control groups, risk differences with exact 95% confidence intervals were calculated for binary measures and median differences with bootstrapped 95% confidence intervals were calculated for continuous measures. Associations between seizure burden variables were assessed with Spearman rank correlations. In post hoc analyses, linear regression was used to compare change in seizure burden from baseline to post-SDA periods either between treatment and control groups or related to bumetanide exposure (AUC), adjusting for total seizure burden as a potential effect modifier.¹⁴ To confirm these findings, we also ran these interaction models using linear regression with robust variances based on generalized estimating equations valid under non-normality of the residuals, robust linear regression using Huber weights to downweight observations with larger residuals, and the inclusion of quadratic total seizure burden effects and both linear or quadratic interactions. We also explored the effects of time from first suspected/confirmed seizure to SDA, and whether seizure burden was decreasing, stable (seizure slope less than 10s/h), or increasing in the 2 hours pre-SDA. All p values are 2-tailed. Statistical analyses were performed in R, version 3.5.1,²⁹ and SAS, version 9.4.

Results

Study Participants

Of 539 screened neonates, 111 (58%) of 191 eligible neonates were enrolled; many enrolled neonates had no further seizures after phenobarbital load, and 10 met exclusion criteria after enrollment. Of 43 randomized neonates, 27 subjects were randomized to treatment (bumetanide) and 16 to standard therapy (Fig 3, Table 1). The DSMC requested additional subjects be enrolled in the 0.2mg/kg dose group because of variability in seizure burden. The DSMC stopped the trial when they determined that sufficient PK and safety data were obtained to satisfy the primary endpoints. No subjects were withdrawn from the trial, and all prespecified neonatal data were collected.

FIGURE 3:

Screening, enrollment, and randomization of the study subjects. EEG = electroencephalographic.

TABLE 1.

Characteristics of the Study Subjects, according to Study Group

Characteristic	All Treatment, n = 27	Control, n = 16	pa	0.1 mg/kg Bumetanide, n = 7	0.2 mg/kg Bumetanide, n = 15	0.3 mg/kg Bumetanide, n = 5	pb
Male sex, n (%)	14 (52)	7 (44)	0.75	3 (43)	10 (67)	1 (20)	0.31
Gestational age at birth, wk, median [IQR]	39 [38, 40]	39.5 [39, 41]	0.56	39 [38, 40]	40 [38, 41]	39 [39, 39]	0.86
Birth weight, kg, median [IQR]	3.4 [3.0, 3.8]	3.3 [3.0, 3.5]	0.44	3.4 [3.1, 3.8]	3.4 [3.1, 3.8]	3.0 [2.9, 3.3]	0.39
Race, n (%)
Caucasian	23 (85)	12 (75)	0.70	6 (86)	13 (87)	4 (80)	0.95
Asian	1 (4)	1 (6)		0	1 (7)	0
Unreported	3 (11)	3 (19)		1 (14)	1 (7)	1 (20)
Hispanic or Latino ethnicity, n (%)	2 (7)	4 (25)	0.17	1 (14)	1 (7)	0	0.53
Seizure etiology, n (%)			0.07				0.09
Hypoxic ischemic encephalopathy	14 (52)	8 (50)		3 (43)	7 (47)	4 (80)
Stroke	7 (26)	0		4 (57)	3 (20)	0
Intracranial hemorrhage	3 (11)	4 (25)		0	2 (13)	1 (20)
Otherc	3 (11)	4 (25)		0	3 (20)	0
Therapeutic hypothermia, n (%)	10 (37)	5 (31)	0.75	3 (43)	4 (27)	3 (60)	0.60
Baseline EEG background grade			0.29
Normal for age, n (%)	6 (22)	7 (44)
Excessively discontinuous, n (%)	14 (52)	6 (38)
Depressed and undifferentiated, n (%)	4 (15)	0
Burst suppression, n (%)	1 (4)	1 (6)
Extremely low voltage, invariant, and unreactive, n (%)	2 (7)	2 (12)

^aProbability values for the comparison between treatment and control groups were determined by Fisher exact tests or exact Wilcoxon rank sum tests, as appropriate.

^bProbability values for the comparison of dose groups (ie, control and 0.1, 0.2, and 0.3mg/kg groups) for pre–study drug administration variables were determined by Fisher exact tests or exact Kruskal–Wallis tests, as appropriate.

^cOther etiologies included brain malformation, group B streptococcal meningitis, herpes simplex virus encephalitis, and neonatal onset genetic epilepsy.

EEG = electroencephalographic; IQR = interquartile range.

There were no significant differences in race/ethnicity or any clinical characteristics among groups (see Table 1). Seizure etiologies differed among groups but were not quite statistically significantly different (p = 0.07). Most subjects (36/43, 84%) had acute symptomatic seizures from ischemic/hemorrhagic etiologies, as 22 (51%) had HIE, 7 (16%) had stroke, and 7 (16%) had ICH. Although there appeared to be a trend to more abnormal EEG background in the bumetanide group, there was no significant difference in EEG background pattern between treatment and control groups. There were no statistically significant differences in phenobarbital doses or levels at or within 24 hours of randomization, and no statistically significant difference in hypothermia treatment among groups (Table 2).

TABLE 2.

Subject Adverse Event and Drug Data, according to Study Group

	All Treatment, n = 27	Control, n = 16	Effect Size (95% CI)a	pb	0.1 mg/kg Bumetanide, n = 7	0.2 mg/kg Bumetanide, n = 15	0.3 mg/kg Bumetanide, n = 5	pc
Adverse events after SD, n (%)
Serious adverse events
Neonatal death, all unlikely to be related to SD	1 (4)	3 (19)	−15 (−41, 5)	0.14	0	1 (7)	0	0.24
Hearing impairment among survivors, all possibly related to SD	2 (8)	0	8 (−15, 25)	0.54	0	1 (7)	1 (20)	0.21
Minor adverse events
Excess diuresis, any	13 (48)	2 (13)	36 (7, 59)	0.02	6 (86)	4 (27)	3 (60)	0.18
Possibly/probably related to SD	9 (33)	1 (6)	27 (1, 48)	0.06	6 (86)	2 (13)	1 (20)	0.87
Electrolyte imbalance, any	9 (33)	5 (31)	2 (−28, 30)	>0.99	4 (57)	4 (27)	1 (20)	0.65
Possibly/probably related to SD	4 (15)	4 (25)	−10 (−39, 16)	0.44	3 (43)	1 (7)	0	0.10
Drug data, median [IQR]
Bumetanide
Exposure (AUC 0–4 h), μg h/l	3,118 [2,321, 4,507]				1,598 [1,283, 2,007]	3,050 [2,708, 4,374]	4,790 [4,556, 5,365]	<0.001
Half-life (t1/2), h	16.0 [9.1, 18.8]				15.6 [9.4, 16.8]	13.4 [8.8, 17.8]	19.1 [18.4, 35.0]	0.15
Clearance, ml/min/kg	0.10 [0.06, 0.17]				0.12 [0.06, 0.18]	0.12 [0.06, 0.17]	0.05 [0.05, 0.07]	0.29
Phenobarbital
Total phenobarbital loading dose prior to randomization, mg/kg	20.0 [19.9, 29.8]	20.1 [19.8, 24.6]	−0.1 (−1.1, 8.8)	0.91	19.9 [19.7, 20.0]	25.5 [20.0, 30.0]	20.0 [19.9, 29.0]	0.25
Phenobarbital level at randomization, μg/ml	25.8 [21.1, 34.6]	24.3 [22.0, 34.2]	1.5 (−11.4, 13.6)	0.90	21.9 [18.8, 29.9]	30.7 [20.2, 34.6]	25.8 [23.2, 33.0]	0.91
Phenobarbital dose at randomization, mg/kg	9.9 [5.0, 10.0]	9.9 [5.0, 10.0]	0.0 (−4.8, 4.9)	0.91	9.9 [4.9, 10.0]	5.2 [5.0, 10.0]	10.0 [5.2, 10.0]	0.94
Phenobarbital level within 24 h after SD, μg/ml	41.4 [30.6, 47.1]	34.3 [32.2, 42.3]	7.1 (−8.1, 12.8)	0.54	30.1 [28.7, 38.0]	47.1 [41.4, 51.6]	39.0 [34.2, 43.8]	0.23

^aEffect sizes are reported as risk differences as a percentage with exact 95% CIs or median differences with bootstrapped 95% CIs, as appropriate.

^bProbability values for the comparison between treatment and control groups were determined by Fisher exact tests or exact Wilcoxon rank sum tests, as appropriate.

^cProbability values for the comparison of dose groups (ie, control and 0.1, 0.2, and 0.3mg/kg groups) for pre-SDA variables were determined by Fisher exact tests or exact Kruskal–Wallis tests, as appropriate. Probability values for the comparison of dose groups for post-SDA variables were determined by exact Cochran–Armitage trend tests or exact Jonckheere–Terpstra tests, as appropriate.

AUC = area under the curve; CI = confidence interval; IQR = interquartile range; SD = study drug; SDA = SD administration.

There were no statistically significant differences in the incidence of minor or serious adverse events among groups, except for a higher rate of diuresis in treated versus control subjects (48% vs 13%, p = 0.02). Despite the higher rate of diuresis, there were no significant differences between groups with regard to measures of electrolyte abnormalities or cardiorespiratory status (data not shown). There was 1 neonatal death in the treatment group and 3 in the control group, occurring 6 to 34 days after SDA; all deaths were related to multiorgan failure and/or poor neurologic prognosis (with withdrawal of life-sustaining support) and unrelated to study drug. Three of the surviving, enrolled subjects developed hearing impairment; all 3 had HIE with severe multiorgan dysfunction, but only 2 were randomized, whereas the other was a nonrandomized subject. Thus, hearing impairment occurred in 2 of 26 surviving treated subjects (8%; both received gentamicin, 1 with a supratherapeutic level), 0 of 13 surviving control subjects, and 1 of 54 surviving nonrandomized subjects (that subject did not receive gentamicin). Among surviving subjects, gentamicin was administered to 23 of 39 (59%) randomized subjects and 19 of 54 (35%) nonrandomized subjects, including 10 of 20 (50%) randomized subjects and 10 of 39 (25%) nonrandomized subjects where HIE was the seizure etiology. The lower rate of hearing impairment and gentamicin administration in nonrandomized subjects was related to a milder severity of illness (ie, most subjects had mild HIE and/or few or no seizures).

Pharmacokinetic analysis from 116 bumetanide levels showed that bumetanide PK was best described by a 2-compartment model. Bumetanide had a median half-life of 16.0 hours, clearance of 0.10ml/min/kg, and volume of distribution of 0.12l/kg and 0.15l/kg for the central and peripheral compartments, respectively. There was a large intersubject variability in clearance (coefficient of variation [CV] = 76.2%) and volume (CV = 37.8%). Allometric scaling was applied on clearance and volumes to account for size differences in bumetanide PK.³⁰ Race, sex, and hepatic or renal dysfunction had no effect on bumetanide PK, but higher postmenstrual age was associated with higher clearance (p < 0.01). Hypothermia was associated with significantly prolonged half-life (median = 17.4 vs 8.8 hours, p = 0.01) and lower clearance rate (0.06 vs 0.14ml/min/kg, p = 0.03). Bumetanide exposure was naturally higher with higher doses (see Table 2).

There was a very broad and continuous distribution of total seizure burden among subjects (range = 0.05–20.8-min/h, 1–676 minutes; Fig 4). There was a striking, significant difference in total seizure burden among groups (Table 3, Fig 4), which occurred by chance. The 0.3mg/kg group had highest total seizure burden (median = 6.1 min/h), and the control group had the lowest seizure burden (1.2min/h, Jonckheere–Terpstra trend p < 0.001). There was a similar significant difference among groups with respect to seizure burden pre-SDA (p = 0.03, Fig 5A). Total seizure burden was strongly correlated with 0 to 2 hours pre-SDA seizure burden (r = 0.57, p < 0.001) and also the entire pre-SDA EEG monitoring period (r = 0.78, p < 0.001). The differences in total and pre-SDA seizure burden among groups occurred by chance; these differences were not associated with any change in enrollment strategy or subject characteristics. Additionally, the time from first suspected/confirmed seizure to SDA was shortest in the 0.3mg/kg dose group, but was not statistically significant different among treatment groups.

FIGURE 4:

Total seizure burden (min/h) by subject (ranked).

TABLE 3.

Seizure Burden of the Study Subjects, according to Study Group^a

Variable	All Treatment, n = 27	Control, n = 16	Effect Size (95% CI)b	pc	0.1 mg/kg Bumetanide, n = 7	0.2 mg/kg Bumetanide, n = 15	0.3 mg/kg Bumetanide, n = 5	pd
All subjects
Total seizure burden, median [IQR]	3.1 [1.3, 5.1]	1.2 [0.3, 2.7]	1.9 (−0.1, 3.3)	0.006	1.6 [0.9, 3.9]	3.1 [1.3, 4.0]	6.1 [4.6, 6.2]	<0.001
Total pre-SDA seizure burden, median [IQR]	2.5 [1.0, 5.4]	1.1 [0.4, 5.7]	1.4 (−2.1, 3.6)	0.25	1.1 [0.7, 2.8]	2.3 [0.7, 3.6]	7.3 [6.7, 8.7]	0.03
Time from first seizure to SDA, h, median [IQR]	17.5 [9.0, 27.7]	15.5 [8.7, 18.5]	2.0 (−5.5, 10.6)	0.26	21.0 [10.5, 23.8]	18.6 [10.5, 30.3]	8.8 [7.8, 13.0]	0.19
Baseline period: 0–2 h pre-SDA seizure burden, median [IQR]	3.3 [0.2, 9.9]	1.6 [0.0, 4.6]	1.6 (−1.5, 4.6)	0.20	3.3 [1.1, 9.9]	1.4 [0.0, 6.4]	26.5 [4.2, 32.7]	0.03
Trend in seizure burden 0–2 hr pre-SDA, n (%)				0.84				0.64
Increasing	9 (33)	3 (19)			3 (43)	5 (33)	1 (20)
Stable	9 (33)	9 (56)			2 (29)	6 (40)	1 (20)
Decreasing	9 (33)	4 (25)			2 (29)	4 (27)	3 (60)
Differences in seizure burden, median [IQR]
0–4 h post-SDA vs 0–2 h pre-SDA	−1.2 [−6.4, 0.0]	−0.1 [−1.5, 0.0]	−1.1 (−4.3, 0.7)	0.14	−1.4 [−7.3, −1.1]	−0.2 [−4.3, 0.0]	−16.6 [−23.0, −1.8]	0.17
2–4 h post-SDA vs 0–2 h pre-SDA	−1.1 [−5.1, 0.0]	0.0 [−1.7, 0.8]	−1.1 (−4.3, 0.5)	0.09	−3.0 [−4.8, −1.1]	0.0 [−4.3, 0.4]	−14.5 [−16.0, −1.8]	0.10
Subjects with HIE, stroke, or ICH, median [IQR]	n = 24	n = 12			n = 7	n = 12	n = 5
Total seizure burden	3.4 [1.5, 5.6]	1.2 [0.6, 2.8]	2.1 (−0.1, 3.4)	0.03	1.6 [0.9, 3.9]	3.4 [2.3, 6.1]	6.1 [4.6, 6.2]	0.003
Total pre-SDA seizure burden	2.7 [1.6, 5.5]	1.6 [0.4, 5.7]	1.1 (−3.0, 3.9)	0.24	1.1 [0.7, 2.8]	2.4 [2.1, 4.3]	7.3 [6.7, 8.7]	0.02
Baseline period: 0–2 h pre-SDA seizure burden	3.8 [1.3, 11.5]	2.6 [0.1, 5.9]	1.3 (−2.7, 5.5)	0.34	3.3 [1.1, 9.9]	1.9 [0.7, 6.8]	26.5 [4.2, 32.7]	0.10
Differences in seizure burden
0–4 h post-SDA vs 0–2 h pre-SDA	−1.3 [−6.9, 0.0]	−0.6 [−2.3, 0.0]	−0.7 (−4.9, 1.1)	0.22	−1.4 [−7.3, −1.1]	−0.5 [−4.9, 0.0]	−16.6 [−23.0, −1.8]	0.21
2–4 h post-SDA vs 0–2 h pre-SDA	−1.7 [−5.5, 0.0]	−0.1 [−2.3, 0.8]	−1.6 (−4.8, 1.1)	0.13	−3.0 [−4.8, −1.1]	−0.4 [−5.1, 0.2]	−14.5 [−16.0, −1.8]	0.13

^aAll seizure burden variables are measured in min/h.

^bEffect sizes are reported as median differences with bootstrapped 95% CIs.

^cProbability values for comparison between treatment and control groups were determined by exact Wilcoxon rank sum tests.

^dProbability values for comparison of dose groups (ie, control and 0.1, 0.2, and 0.3mg/kg groups) were determined by exact Kruskal–Wallis tests for pre-SDA variables or exact Jonckheere–Terpstra tests for post-SDA variables.

CI = confidence interval; HIE = hypoxic–ischemic encephalopathy; ICH = intracranial hemorrhage; IQR = interquartile range; SDA = study drug administration.

FIGURE 5:

Effect of bumetanide on seizure reduction. (A) Seizure burden (min/h) for 0 to 2 hours before study drug administration (SDA), then the difference in seizure burden for 0 to 4 hours post-SDA and 2 to 4 hours post-SDA compared with 0 to 2 hours pre-SDA, across control and bumetanide dose groups. (B) Relationship between difference in seizure burden for 0 to 4 hours post-SDA and total seizure burden, across control and combined bumetanide groups. For all subjects receiving bumetanide (red), there was a greater reduction in seizure burden as a function of total seizure burden (slope = −1.12min/h per unit increase in total seizure burden; 95% confidence interval [CI] = −1.71 to −0.53; p < 0.001). For controls (blue), there was no significant change in seizure burden as a function of total seizure burden (slope = 0.27; 95% CI = −0.27 to 0.81; p = 0.34). (C) Relationship between difference in seizure burden for 2 to 4 hours post-SDA and total seizure burden. For all subjects receiving bumetanide (red), there was a greater reduction in seizure burden as a function of total seizure burden (slope = −1.10; 95% CI = −1.61 to −0.58; p < 0.001). For controls (blue), seizure burden increased as a function of total seizure burden (slope = 0.63; 95% CI = 0.06–1.21; p = 0.048).

Table 3 shows the magnitude of seizure reduction in min/h in the periods 0 to 4 and 2 to 4 hours post-SDA as compared to 0 to 2 hours pre-SDA. There were no significant group differences, likely related to the large differences in seizure burden across groups (see Fig 5A). The large and statistically significant difference in seizure burden among groups led to the post hoc decision to analyze the effect of bumetanide versus control as potentially modified by total seizure burden (see Figs 5B, C and 6). The effect of bumetanide (vs control) on reduction in seizure burden increased as a function of total seizure burden (interaction p = 0.008 for 0–4 hours, p < 0.001 for 2–4 hours; see Fig 5B, C). We found comparable interaction effects when using linear regression with robust variances valid under non-normality, robust linear regression that downweights observations with larger residuals, or the inclusion of quadratic terms and interactions. When subjects with only acute symptomatic etiologies (ie, HIE, stroke, and ICH) were analyzed, there was a similar effect of bumetanide on reduction in seizure burden at 0 to 4 hours (p = 0.01) and 2 to 4 hours (p = 0.001) post-SDA (data not shown).

FIGURE 6:

Effect of bumetanide exposure (area under the curve [AUC]) on seizure reduction. The scatterplot shows the relationship between difference in seizure burden for 0 to 4 hours after study drug administration (SDA) compared with 0 to 2 hours pre-SDA and total seizure burden, across control and combined bumetanide groups. The interaction model was based on the effect of bumetanide exposure (AUC) and total seizure burden to predict difference in seizure burden (interaction p = 0.008). The model is graphed for the control group (solid blue; slope = 0.24; 95% confidence interval [CI] = −0.58 to 1.05; p = 0.56) and at the median exposure of the 0.1mg/kg (dashed yellow; slope = −0.20 at median AUC_0–4h = 1,598μg·h/l; 95% CI = −0.79 to 0.40; p = 0.51), 0.2mg/kg (dashed orange; slope = −0.60 at median AUC_0–4h = 3,050μg·h/l; 95% CI = −1.08 to −0.10; p = 0.02), and 0.3mg/kg (dotted red; slope = −1.06 at median AUC_0–4h = 4,790μg·h/l; 95% CI = −1.63 to −0.50; p < 0.001) groups.

Next, we analyzed the effect of bumetanide exposure (ie, AUC) as potentially modified by seizure burden. The effect of bumetanide exposure on reduction in seizure burden at 0 to 4 hours increased as a function of total seizure burden (interaction p = 0.008; see Fig 6), and this interaction effect did not change appreciably using robust methods or quadratic terms. This relationship remained significant when we examined only subjects with acute symptomatic etiologies (p = 0.01). Finally, we found no effect of increasing or decreasing seizure burden pre-SDA on bumetanide response. There also was no effect of time from first suspected/confirmed seizure to SDA on bumetanide dose–response at either 0 to 4 hours or 2 to 4 hours, or bumetanide exposure at 0 to 4 hours.

Discussion

This randomized, double-blind trial of bumetanide to treat neonatal seizures demonstrated a dose- and exposure-related benefit of bumetanide on seizure reduction, in comparison with a phenobarbital monotherapy control group, without a significant difference in serious adverse effects. The add-on design with a standard therapy control group was critical to evaluate both bumetanide safety and exposure-response, because many potential adverse effects of bumetanide are also complications of neonatal seizure etiologies and their treatments (particularly neonatal HIE with multiorgan dysfunction).

We found no significant difference in adverse events between treatment and control groups other than a higher rate of diuresis in bumetanide-treated subjects. The rate of hearing impairment among surviving treated subjects was lower in our trial (2/26 = 8%) than the NEMO bumetanide trial (3/11 = 27%).⁹ Notably, all 5 treated subjects with hearing impairment in the 2 trials had HIE. Hearing impairment occurs in ~10% of neonates with HIE treated with hypothermia, related to multiple associated risk factors, particularly gentamicin.³¹ Because 4 of 5 treated subjects with hearing impairment in the 2 trials received gentamicin, the lower rate of hearing impairment in this trial could be related partly to lower use of gentamicin in neonates with HIE in our trial (50% of subjects with HIE, vs 86% in NEMO). Neither bumetanide trial enrolled sufficient subjects to establish definitively any potential contribution of bumetanide to hearing impairment, which is observed experimentally only when combined with aminoglycosides.³² The known occurrence of hearing impairment in neonatal HIE without bumetanide exposure³¹ (eg, in 1 nonrandomized subject in this trial) and the relatively common occurrence of other adverse events (eg, excess diuresis and electrolyte imbalance) in control subjects underscores the importance of including a control group to evaluate drug safety in trials enrolling critically ill neonates. Even with a control group, this early phase trial testing 3 different doses in 27 subjects cannot provide a complete evaluation of bumetanide safety.

Our trial showed a benefit of bumetanide that correlated with exposure, particularly among subjects receiving doses of 0.2 (15 subjects) and 0.3mg/kg (5 subjects). The NEMO trial was stopped early for reported lack of efficacy, evaluated in a total of 14 subjects (no control group),⁹ although others argued that the efficacy target was met.³³ Notably, bumetanide dose in the NEMO trial was lower, including a lower first dose (0.05mg/kg), and only 1 subject received the highest dose of 0.3mg/kg. Additionally, bumetanide exposure-response was not evaluated separately from phenobarbital in the NEMO trial, as there was no comparison control group and the first dose of bumetanide (analyzed for dose–response) was administered together with phenobarbital. In the current trial, subjects receiving bumetanide together with phenobarbital were compared with subjects receiving phenobarbital alone (with saline placebo), so that the effect of bumetanide could be analyzed in comparison with phenobarbital monotherapy. Importantly, phenobarbital doses and levels at randomization were not different between groups in this trial, so differences in seizure reduction between treatment and control groups could be attributed to bumetanide. Comparing the additive effect of bumetanide with phenobarbital to phenobarbital alone is particularly important in light of new trial data showing the high efficacy of phenobarbital as first-line therapy,¹⁰ compared with older data.¹

A significant challenge in analyzing bumetanide response in this trial was the high variability in seizure severity among treatment groups, which occurred by chance. The significant difference in seizure burden among groups necessitated post hoc adjustment to add the interaction between treatment group and seizure burden, because seizure severity is a critical factor in determining drug response.^1,34 The important effect of seizure severity on drug response was demonstrated by the trial comparing phenobarbital to phenytoin, which showed much greater seizure cessation in neonates with low (88%) versus high (10%) seizure severity.¹ In the current trial, subjects received only a single dose of bumetanide and the effect of bumetanide was expected to be brief relative to the period during which neonates were experiencing (or susceptible to) seizures. Thus total seizure burden was judged to be the best indicator of overall seizure severity, so the effect of bumetanide on seizure burden was adjusted by the total seizure burden. Additionally, although seizure burden is the most important variable to consider when assessing drug response, severity of encephalopathy may also influence antiseizure drug response. The treatment group (bumetanide) tended to have more severe encephalopathy (fewer with normal EEG background) than the control group, although this was not a statistically significant difference. Interestingly, a more severe encephalopathy in the treatment (bumetanide) group would be more likely to be associated with reduced drug response, the opposite of what we found.

The trial comparing phenobarbital to phenytoin also demonstrated higher seizure cessation in neonates with decreasing (81%) rather than increasing (30%) seizure frequency,¹ suggesting that drugs may appear more effective if administered when seizures are subsiding naturally rather than when seizure burden is increasing. We did not find an effect of decreasing or increasing seizure frequency on bumetanide response, which did not differ by treatment group. Thus we also investigated whether time from seizure onset to SDA might have affected bumetanide response, because this time would reflect how early study drug was administered in the course of seizures. Notably, we found no significant effect of this variable on bumetanide effect, likely because SDA occurred relatively early in the course of seizures for most subjects. This was particularly true for subjects receiving the highest bumetanide dose (0.3mg/kg), who had the highest seizure severity but the shortest time from seizure onset to SDA (median = 8.8 hours). Treatment earlier in the natural history of seizures after brain injury would bias this dose group toward a lesser treatment effect, because seizure frequency would still be increasing (and because of the high seizure severity),¹ but the opposite was seen. These data support that randomization of bumetanide administration was achieved sufficiently early in the course of neonatal seizures, prior to the time when seizures were naturally subsiding.

There are several reasons quantitative change in seizure burden before and after SDA was analyzed instead of percentage seizure reduction, unlike other neonatal seizure trials.^1,9 First, a percentage reduction is limited by the magnitude of seizure burden during the baseline comparison period, which is particularly problematic if there are no seizures during the baseline period.³³ Second, a large percentage reduction (even 100%) for neonates with low seizure burden is likely a less clinically important drug response than a >50% reduction for neonates with high seizure burden (eg, status epilepticus).¹⁴ Although seizure cessation (or even 80% reduction) might be an outcome chosen for phase 3 efficacy trials, the high rate of incomplete response to one or more drugs in neonates makes seizure cessation an unrealistic outcome for early phase, dose-finding trials.^2,11,12 Third, changes in min/h of seizure burden are more meaningful as real world, continuous efficacy measures than fractional reductions, particularly for evaluating the effect of different drug doses or drug exposure in dose-finding trials. Finally, the strong negative effect of seizure severity on drug exposure-response¹ is lost when comparing fractional changes in seizure burden.

Bumetanide pharmacokinetic data showed a similar volume of distribution but lower clearance and longer half-life compared with previous reports, even after accounting for differences in age, illness, and units.^35–37 These differences in pharmacokinetics may be related to the significant effect of hypothermia on clearance and half-life and/or the larger number of subjects and bumetanide levels per subject obtained in the present study, particularly because interindividual differences in clearance and volume were high.^35–37 Concerns have been raised regarding brain penetration of bumetanide based on experiments in animals with less severe brain injury and seizures than reported here.^19,26,38 However, the blood–brain barrier may be compromised in seizing neonates with HIE,^39,40 and bumetanide may reduce the influx of chloride salts into swelling neurons^41,42 by limiting NKCC1 transport at the blood–brain barrier.⁴³ The latter effect would not require blood brain penetration, and should ameliorate cytoplasmic chloride accumulation in injured neurons, the attendant shift in polarity of GABA signaling, the consequent degradation in inhibition,⁴⁴ and the reduced efficacy of anticonvulsants such as barbiturates and benzodiazepines that increase the open time of the GABA_A receptor.⁴⁵ This effect of bumetanide requires reperfusion of injured brain areas. Reperfusion is typical in neonatal brain injury,^46,47 but was not present in the only experiments in which bumetanide was ineffective.⁴⁸

This pilot trial showed a statistically significant benefit of bumetanide over phenobarbital monotherapy for seizure reduction, without serious adverse effects, albeit limited partly by an imbalance of seizure severity among groups. Although seizure severity was highest by chance in the highest bumetanide dose group, the significant bumetanide exposure-response relationship (see Fig 6) supports the greater bumetanide effect with higher bumetanide dose/exposure. Our use of robust estimation methods and quadratic models confirmed the robustness of our interaction modeling, but this small early phase trial was not designed to provide definitive answers to all the questions being tested; more subjects are required to establish bumetanide exposure-response and safety. Based on our findings, subsequent trials of bumetanide should employ a minimum dose of 0.3mg/kg and include more subjects and higher doses to investigate further the magnitude of the bumetanide effect on neonatal seizures. Future trials of bumetanide should also consider limitation of aminoglycosides, particularly for neonates with HIE who have multiple risk factors for hearing impairment. Results from this trial demonstrate the advantage of employing an add-on therapy design with a standard therapy control group to assess both drug safety and exposure-response in an early phase dose-finding trial.³⁴ Our data also illustrate the benefit of measuring a quantitative change in seizure burden rather than percentage seizure reduction to test the magnitude of drug response, at least for early phase trials. Design of future trials of neonatal seizure treatment will need to account for potentially large differences in seizure severity, particularly for early phase trials with typically small sample size, and should require a minimum seizure burden in the period just prior to randomization. Trials of bumetanide and other anticonvulsants should be designed to balance and analyze these important variables that affect evaluation of drug response and safety.