The trial was designed as multicenter, patient-randomized, double-blind placebo-controlled phase-2 superiority trial testing effectiveness of 91 days bumetanide treatment followed by a 28-day wash-out period. The trial was initiated and conducted by the UMC Utrecht in the Netherlands with Jonx Groningen as participating center. Participants were recruited through outpatient clinics and advertisement on websites of the Dutch ASD parent association (NVA), epilepsy expert association (SEIN), and the Dutch ADHD parent association (Balans). The medical ethical committee of the UMC Utrecht approved the trial protocol and the study was conducted according to the principles of the Declaration of Helsinki and Good Clinical Practice (ICH-GCP). Written informed consent was obtained from all parents or legal representatives and participants received no financial compensation. The trial was registered on 25/09/2016 with registration number 2016-002875-81. The full trial protocol is available at www.umcutrecht.nl/nl/ziekenhuis/wetenschappelijk-onderzoek/de-bascet-studie.

Children with a current ASD, ADHD (according to DSM-IV-TR or DSM-5 criteria), and/or epilepsy diagnosis, aged 5–15 years and IQ ≥ 55 were eligible as participants. Children were enrolled when a diagnosis was accompanied by altered sensory reactivity, defined as a deviant score (>1 SD deviant) on the Sensory Profile for parents or teachers (SP-NL or SP-SC) (48, 49). Use of concomitant psychoactive and antiepileptic drugs (AED) was allowed, when being taken on an unadjusted dosage at least 2 months prior to baseline measures. Exclusion criteria were renal or liver insufficiency, serious unstable illnesses (including gastroenterological, respiratory, cardiovascular, endocrinologic, immunologic, hematologic disease, dehydration or hypotension, electrolyte disturbances), treatment with NSAIDS, aminoglycosides, digitalis, antihypertensive agents, indomethacin, probenecid, acetazolamide, lithium, other diuretics, stimulants (like methylphenidate and dexamphetamine, due to it assumed diametrical effects), and drugs known to have a nephrotoxic potential. Children were allowed to receive care as usual when it was initiated minimally 2 months prior to baseline measures. Amendments to eligibility criteria were made to further include patients with ADHD and/or epilepsy besides patients with ASD with or without epilepsy. Consequently, the SP-NL was used as inclusion criterion to select patients based on sensory reactivity problems rather than diagnosis. The sensory profile was chosen since it is commonly used to detect SPD and used in several case reports and drug trials. A cut-off score of 1 SD in both directions was applied since the SP-NL detects not only sensory sensitivity.

Detailed descriptions of randomization and blinding practice are described in an earlier RCT with similar study design (44). In brief, participants were randomly allocated (1:1) to receive bumetanide or placebo treatment, which was provided by Tiofarma. Sequence generation, concealment, and treatment allocation was overseen by a third-party not involved in the study (i.e., Julius Center, a consultant support agency for clinical research and trials located in the UMC Utrecht). Restricted randomization was used with permuted block design randomly varying between two, four, and six participants. Treatment allocation was done automatically using minimization with a probability of 0.75 on the participant factors active epilepsy (y/n), IQ (55–75; 76–110; >110) and study center (UMC/Jonx). Participants, parents, healthcare providers, and outcome assessors were blinded for randomization, by organizing safety checks at the pediatric nephrology department of the nearby Wilhelmina Children's Hospital.

The study procedures adhere to the study procedures described by Sprengers et al. (44) and to which we refer for details on the procedures. The first study visit included clinical history taking by a medical doctor or psychologist, the administration of an abbreviated WISC-III intelligence test, and medical screening by a medical doctor. Besides, study outcomes were measured at baseline (D0) and repeated after treatment (D91) and 28-day wash-out (D119), similar to previous studies (42–44).

Within 45 days of the baseline visit participants were randomized (D0) and received bumetanide or placebo tablets (0.5 mg) matched for taste, smell, and viscosity, albeit without diuretic properties. The tablets were taken orally twice-daily with minimally 6 h between the administrations (e.g., typically with breakfast and dinner). Children ≤33 kg started with halved tablets (i.e., twice-daily 0.25 mg). Children >33 kg received twice-daily 1 tablet (0.5 mg), similar to Lemonnier et al. (42) with best optimal dose-response tradeoff. When blood analysis showed no abnormalities at D7, the dosage was doubled. All participating children were supplemented with 0.5 mmol/kg potassium chloride when <30 kg, or twice-daily 8 mmol potassium chloride when ≥30 kg to avoid hypokalemia.

Safety visits were scheduled at D4, D7, D14, D28, D56, D91, and D119 at the department of pediatric nephrology of the Wilhelmina Children's Hospital, with the purpose of blinding the researchers, and included blood analysis (D4, D7, D14, D28, and D56), medical evaluation, and documentation of adverse events (AEs) (adhering to NCI-CTCAE and MedDRA methodologies). Participants returned for outcome evaluations at the end of the 91-day medication phase and at the end of the 28-day wash-out period. Parents were interviewed at the last study visit (D119) about their experiences (i.e., treatment, AE, and wash-out evaluations) and were asked to predict which treatment their child had received. A schematic overview of the trial is depicted in Figure 1.

The primary outcome was severity of irritable behaviors measured by the ABC-I (range 0–45; higher score is more affected) after 91 days of treatment. This measure was chosen because it is a commonly used outcome scale in (neuro)behavioral trials (50) and because we hypothesized that a beneficial clinical effect of bumetanide in this population would become evident by reducing behavioral reactivity to sensory stimuli. The SP-NL questionnaire was added to assess changes in sensory reactivity as a secondary outcome (range 125–625, lower score is more affected). However, this questionnaire was primarily developed as a screening and characterization scale and not developed to detect change. Other secondary outcomes were chosen to cover broad NDD core symptom domains; severity of restricted and repetitive behaviors, measured by the Repetitive Behavior Scale-Revised (RBS-R; range 0–129, higher score is more affected), symptom severity of social communication and social interaction, measured by the SRS-2 (range 0–195; higher score is more affected), and severity of behavioral executive functioning, measured by the Behavior Rating Inventory of Executive Function (BRIEF-parent; range 72–216; higher score is more affected) total scores at D91. To assess executive function and sensory behaviors in the school environment, the BRIEF-teacher (range 73–219; higher score is more affected) total score and domain scores of the SP-School Companion (SP-SC) were included, respectively. When participants were diagnosed with epilepsy, frequency, and type of seizures were registered with an epilepsy diary. Adverse events were passively (spontaneous report) and actively (evaluation of known side effects) collected by the nurse practitioner and physicians of the pediatric nephrology department. Incomplete individual clinical questionnaires were imputed as “no change” when less than four questions were missing (RBS-R: n = 1; SP-NL: n = 3; SP-SC: n = 2; BRIEF-T: n = 4). When four or more questions were missing, the outcome measures were excluded from analysis (n = 4).

This study was initially powered at 90% to detect an effect size of 0.5 on the primary outcome measure (ABC-I) with a standard deviation of 9.3 (i.e., mean change difference of 4.6 points), assuming two-sided alpha level of 0.05. Allowing for 10% attrition rate, 190 participants had to be randomized. Due to lower-than-expected inclusion rates, the sample size was reevaluated allowing for 80% power resulting in an intended sample size of 124 participants.

Due to the explorative nature of the study, we analyzed outcomes by modified intention-to-treat on allocated participants (see Results section for details). Screening differences between randomized and non-randomized participants were analyzed with appropriate t-statistics or Fisher's exact tests for dichotomized variables.

Primary and secondary outcomes at all available time points were analyzed with a linear mixed model. A random intercept was included to correct for multiple follow-up measurements per participant. Treatment and treatment by time interaction were included to assess the difference between placebo and bumetanide. In a second step, sex, age, and baseline measurement of the corresponding outcome measures were included to correct for potential confounding and optimize the statistical analysis for power (51, 52). Statistical assumptions of the models (i.e., distributional assumptions, homoscedasticity) were assessed by examining residuals (53). From these models, we derived estimated means for each treatment arm as well as a mean difference (MD) between treatment groups at 91 days with 95% confidence intervals (CI) and p-values. Additional analyses were performed for treatment interactions with sex, age, and total IQ and were evaluated with likelihood ratio tests (LRT). Safety was analyzed in all allocated subjects (i.e., ITT) with Fisher's exacts tests. Agreement of predictions by parents of the allocated treatment arm vs. the actual treatment allocated to children was analyzed with Cohen's kappa. All analyses were performed with SPSS v25 (IBM, Corp., Armonk, NY) and SAS v9.4 (SAS, Cary, NC).

Study safety was overseen twice a year by the Data Safety Monitoring Board (DSMB) of the UMC Utrecht. This study was registered with the EudraCT trial registry (2016-002875-81) and Dutch trial registry (NL6178).

Participants were enrolled between June 20th 2017 and June 26th 2019, the end of planned recruitment and funding. The study was finished without meeting the intended study population (n = 124), since inclusion rates were not met. Based on the initial power calculation the actual power of the study reached 27%. Multiple attempts were undertaken to increase recruitment, including adding research staff, advertisements on Dutch ASD, ADHD, and epilepsy parent associations and presentations during their meetings. However, various strategies to increase inclusions failed, which rendered extension of the trial to meet the sample size not feasible. The participating center in Groningen was unable to follow the study procedures and the few participants randomized at this site (n = 5) could not be included for analysis in the study due to incomplete questionnaires. No serious AEs were reported in this participating center. As a consequence, the study is reported as a single center trial.

As shown in the CONSORT diagram in Figure 2, a total of 158 caregivers contacted the research team and obtained a study information folder. After information was sent, 53 potential participants gave informed consent and 52 were assessed for eligibility. Fourteen participants did not progress to randomization for reasons of non-eligibility (n = 7), requirement of immediate other therapy (n = 4), inability to adhere to study protocol (n = 1), resistance to blood withdrawal (n = 1), and participation in another study (n = 1), resulting in 38 participants that were randomly allocated to bumetanide or placebo treatment (Supplementary Table 1). There was no difference in baseline characteristics and outcomes between eligible participants who did and who did not advance to randomization (p ≥ 0.153).

Of the 38 randomized participants, 19 (4 female participants) were allocated to the bumetanide arm and 19 (6 female participants) to the placebo arm. Five participants discontinued treatment prior to collecting outcomes. Two were allocated to placebo: one required immediate psychiatric intervention with other therapy and one withdrew consent because of the high burden, absence of benefits, and perseverance of mild potential side effects (i.e., dermal abnormalities). The other three participants that discontinued treatment had been allocated to bumetanide: one required immediate psychiatric intervention with other therapy, one due to repeating hypoglycemia, and one due to palpitations. During the trial, none of the participants, parents, healthcare providers, or outcome assessors were unblinded. On completion of the trial and before unblinding, one participant was excluded from analyses as questionnaires were not reliable (i.e., were filled out by a different parent; placebo) and for two participants multiple D91 questionnaires were missing (1 placebo, 1 bumetanide).

Table 1 depicts baseline characteristics of the analyzed sample including (previous) medication use and diagnoses.

Of all analyzed participants, 22 (73.3%) were classified with ASD with or without comorbidities, 6 (20%) with ADHD, and two (6.7%) had epilepsy (Table 1). A total of 16 (53.3%) participants were naïve for the use of psychoactive medication. At the start of (and during) the trial three participants (10%) were taking antipsychotics, two (6.7%) were taking AEDs, one participant used a selective serotonin reuptake inhibitor (SSRI) (3.3%), and one antipsychotics together with benzodiazepines. Twenty-three (76.7%) were not taking any medication.

Medication adherence was monitored via several approaches: interview, inspection of returned medication packages, and a drug diary. We found no evidence of non-adherence in either treatment group. The mean provided bumetanide dosage was 0.0430 mg/kg/day (range: 0.0182–0.0637). Treatment dose was increased at D7 in all but two participants (due to hypokalemia and a postponed safety visit and which were both increased at D14). In three participants, the target dose had to be temporarily halved for 11, 12, and 13 days, respectively, due to hypokalemia (n = 3, bumetanide).

We documented parent predictions of the treatment their child had received once the last study visit for the participant was completed. In the bumetanide group (n = 15), 12 parents expected allocation to bumetanide and three parents expected allocation to placebo. In the placebo group (n = 15), one parent expected allocation to bumetanide, 14 parents expected allocation to placebo, and one parent was not assessed. A substantial accordance between expected and actual treatment allocation was found (κ = 0.737 [with 0 indicative of effective blinding and 1 indicative of a potential failure of blinding], p = 0.000).

Bumetanide showed a superior treatment effect on severity of irritability symptoms, the primary outcome of this study (ABC-I MD: −4.78, 95%CI: −8.43 to −1.13, p = 0.0125; Figure 3; Table 2). No effects were found on the secondary outcomes. There was no superior effect of bumetanide on core ASD symptomatology measured with RBS-R (model adjusted for heteroscedasticity, MD: −4.90, 95%CI: −10.97 to 1.17, p = 0.109) and SRS-2 (MD: −6.61, 95%CI: −16.51 to 3.28, p = 0.181), indicating no effect of bumetanide on repetitive behaviors and social communication and social interaction (Figure 4; Table 2). Moreover, no superior effects were found on sensory symptoms measured with the SP-NL (MD: 3.14, 95%CI: −29.3 to 35.6, p = 0.844) nor on executive behavior (BRIEF) measured by parents (MD: −7.88, 95%CI: −17.6 to 1.8, p = 0.105) or teachers (MD: −3.08, 95%CI: −19.7 to 13.5, p = 0.698). Analyses showed no wash-out effects on any outcome. Descriptive results of the subscales are presented in Supplementary Table 2 as the study was not sufficiently powered to statistically test subscales.

Sub-analyses on treatment-by-sex, age and IQ interaction showed only a significant treatment-by-IQ interaction effect on the BRIEF-teacher (MD: −0.65, 95%CI: −1.33 to 0.02, LRT = 3.9, p = 0.0483), indicating that within the bumetanide group, a higher IQ was associated with higher scores after treatment, whereas in the placebo group a higher IQ was associated with lower scores.

Mean treatment dose showed no association with change in ABC-I in the bumetanide group (ρ = 0.157, p = 0.591) i.e., there was no dose-response relationship. Lastly, change in ABC-I showed no association with baseline SP-NL total scores in the bumetanide group (respectively, r = 0.438; p = 0.117).

Adverse events that occurred in more than 5% of participants are shown in Table 3. All events were mild to moderate according to CTCAE rating-scale and all resolved. Three serious AEs occurred in two patients (both bumetanide group): anaphylactic reaction to incidental cow milk ingestion in a child with preexisting cow milk allergy, blood loss after elective adenoidectomy/tonsillectomy requiring prolonged hospital observation; and exaggeration of preexisting palpitations by sinus tachycardia. Cardiac evaluations showed no abnormalities. The serious AEs were registered as probably unrelated to study treatment, except for palpitations, which was possibly related due to hypovolemia, although no signs of hypovolemia were found on echocardiography. Common cold, myalgia, orthostatic hypotension, and hypokalemia were the most frequently occurring AEs. A total of 32% of participants in the bumetanide group experienced increased diuresis compared to none in the placebo group (p = 0.020). In addition, 26% of participants in the bumetanide group developed hypokalemia against none in the placebo group (p = 0.046). Hypokalemia occurred in one patient at D10. In the other four patients hypokalemia occurred only after D14 and potassium levels did not drop below 3.0 mmol/L and normalized with increased oral potassium chloride (Supplementary Table 3).