This multicenter, randomized, controlled trial was conducted between August 2020 and December 2021 and took place at the ICU’s of one academic and two tertiary referral hospitals in the Netherlands. The study was approved by the Medical Ethics Review Board of Erasmus MC (MEC2020-0212) and the local institutional review boards (Ikazia Hospital: IZ/705/SW2037, Haga Teaching Hospital: T20-080). The trial was registered in the Netherlands Trial Register (www.trialregister.nl, ID: NL8595) and the United States National Library of Medicine (www.clinicaltrials.gov, ID: NCT04796389). The study protocol has been previously published [22]. The study is reported according to the Consolidated Standards of Reporting Trials (CONSORT) 2010 statement (Additional file 1) [23].

ICU patients aged 18 years or older, either mechanically ventilated or not, were eligible for inclusion in the study when meeting the following criteria: hemodynamically stable (as assessed by the patient’s direct caregivers), able to communicate (Richmond Agitation and Sedation Scale, RASS > − 3 in the 24 h before inclusion (meaning the patient was at least briefly awakened with eye contact to voice) and was considered to be able to provide information regarding anxiety level, had an expected ICU stay upon randomization of at least 48 h, and a written informed consent was acquired from the patient or legal representative. Exclusion criteria were: severe hearing impairment, neurological condition (e.g. severe stroke, when deemed to interfere with processing of music), insufficient knowledge of the Dutch or English language, and participation in another study that may possibly intervene with the primary outcome (level of anxiety).

Parallel block randomization was used to allocate subjects with an equal allocation ratio in either the intervention or the control group using online web-based randomization program. Subjects were equally allocated by centre. In order to prevent bias due to non-blinding of the outcome assessors (member of the research team or attending nurse), the patient reported outcomes were accompanied by a clear description of how they should be assessed.

Richard-Lalonde et al. [21] found that music interventions of at least 20–30 min significantly reduced pain scores compared to 10–15 min in critically ill patients [21]. Furthermore, Chlan et al. and Fu et al. [24, 25] found that a total of 80–120 min per day music intervention leads to significant reduction in anxiety and sedative and analgesic medication requirement [20, 25]. In addition, several studies suggest the importance of individual music preference of ICU patients in the effectiveness of the music intervention [16, 21]. Based on the previous studies, subjects allocated to the intervention arm were offered to listen to music during three days twice per day, in the morning and evening, during at least 30 min per session in addition to standard care. The intervention was applied as mentioned in order to restore the circadian cycle of the patient, since this is known to be disturbed, by offering music after awakening in the morning and before the lights were turned out in the evening [4]. Music intervention was provided through over-the-ear headphones connected through Bluetooth with a tablet on which a large variety of online music lists were available (based on genre/artist/type, etc., which were pre-arranged or could be re-arranged by investigator, nurse of the patient), from which the patients’ preferred music could be chosen. Music preference was assessed by the patients, or legal representative if the patient was not able to do so, family members, or friends at baseline directly after inclusion and randomization (day before the start of the intervention). We discouraged patients to listen to rock and heavy metal music during the trial, since it is likely that loud and/or rock music may lack the right qualities for this setting [26]. The first session was planned in the morning, between 09.00 and 12.00 AM, the day after inclusion. The evening session was planned before intended sleep, generally between 20.00 and 23.00 PM. In agreement with the direct caregivers, patients were allowed to listen more often or longer to music as requested by the patient or legal representative. Music was only provided when patients were conscious and could reply to the question whether they wanted to listen to music. Additionally, we encouraged nurses to document music being played apart from the music applied with the headphones within the trial protocol, although this was discouraged. Patients in the control group received standard care (no changes in routine care) without structured music intervention.

The primary outcome was level of anxiety as assessed with the visual analogue scale for anxiety (VAS-A), which was assessed directly after the music session in the morning and evening. The VAS-A is validated as a reliable self-rating tool for state anxiety and has been used in the intensive care setting [19, 20, 27]. The VAS-A is a patient reported outcome and ranges from zero to ten, whereas zero is defined as “no feeling of anxiety” and ten as “most anxious ever” on a horizontal scale. The effect of music on anxiety was also assessed using the six-item State-Trait Anxiety Inventory (STAI-6, which only assesses state anxiety, and was added as an additional tool since it assesses anxiety dimensions, such as anxiety about an event, or anxiety level as a personal characteristic). The sum score of the STAI-6 ranges from 20 to 80 and was categorized as low (score of 20–39), moderate (score of 40–59), or high anxiety level (score of 60–80) [28, 29]. Furthermore, we assessed sleep quality (with a Visual Numeric Scale ranging from one to seven, in which one indicates “did not/barely sleep” and seven indicates “slept very well”) [30], pain (using the Critical care Pain Observation Tool (CPOT) in mechanically ventilated patients and the numeric rating scale (NRS)/VAS for pain in non-ventilated) [4], medication requirement (analgesics, sedatives, and antipsychotics, reported as daily administration [yes/no] and dosages), RASS, delirium (measured with the Intensive Care Delirium Screening Checklist [ICDSC]), complications related to agitation, including auto-removal of lines and tubes, time on mechanical ventilation, ICU LOS, physical parameters (heart rate [HR], and mean arterial pressure [MAP] at the time anxiety assessments, and patients’ ICU memory and experiences. Memory was evaluated with the ICU memory tool (ICUMT) [31], which we adapted and shortened to a seven-item questionnaire to avoid overlap with assessment of anxiety or delirium, and with other tools. The patient experience was evaluated in the music group using a five-item and for the control group a three-item self-made questionnaire. Justification and assessment of the tools mentioned above are described in the previously published protocol paper [22].

The baseline characteristics were summarized using means/median (SD/IQR) and number (percentage) for continuous and categorical variables, respectively. Non-normally distributed continuous data were analysed using the non-parametric Mann–Whitney U test, outcomes were presented as median and interquartile ranges (IQR). Normality was assessed with the Shapiro–Wilk test and graphically in Q–Q plots. A sample size of 52 per group was needed to detect a 1.95 point difference in VAS-A, between the groups with a power of 80%, a two-sided alpha of 0.05, and a dropout rate of 10%, which was based on a previous trial [20]. Data analysis was performed using an intention to treat (ITT) approach for all patients who had at least one VAS-A assessed. The total median of the VAS-A was calculated separately for each study day. A two-sided p-value of < 0.05 was considered statistically significant. Our primary outcome, the median VAS-A, was analysed for days one to three separately. A multilevel linear regression with random intercepts was used to compare the change in anxiety over the three study days [32]. In the two-level linear mixed models design (multilevel linear regression model), study day was set at level one, and subjects at level two. Age and sex were included as independent variables in the model. Secondary, a per-protocol analysis was performed. The secondary outcomes were analysed using similar statistical strategy as the primary outcome. Opioid dosages were adapted into fentanyl equivalents (fentanyl intravenous (iv) + remifentanil iv [33] + (fentanyl patch/2.4) [34] + (sufentanil iv/10) [35] + (morphine iv/100) [34] + (oxycodone oral/150) [34]) and intermittent sedatives (benzodiazepines) as lorazepam equivalents (lorazepam + (temazepam/10) [36] + (oxazepam/15) [36] + (diazepam/5) [33] + (bromazepam/5) [37] + (zopiclone/3.75) [33]). Also, each STAI-6 item was analysed separately.