Rapid, effective, and affordable randomisation for emergency neonatal research in a low-resource setting: a feasibility randomised controlled trial

Abstract

Background

Randomisation is an essential component of any clinical trial to eliminate selection bias and ensure similar distribution of confounders between the treatment groups. For randomisation to be successfully implemented, there must be adequate allocation concealment. Many low-cost randomisation and allocation concealment methods have the potential to introduce bias by ineffectively concealing the allocation sequence from the researcher and are now rarely used in high-resource settings. In such settings, centralised randomisation services have been developed to prepare the randomisation sequence, conceal the sequence, and provide a secure mechanism to acquire the allocation. Such services are often prohibitively expensive for researchers in low-resource settings or for small/pilot trials. We describe our experience of adopting a low-cost third-party randomisation and allocation concealment process for emergency neonatal research in a low-resource setting.

Methods

This was a single-site feasibility trial in Eastern Uganda, which randomly assigned neonates in a 1:1 ratio at birth to receive either early continuous positive airways pressure (intervention) or standard of care (control). Centralised third-party randomisation and allocation were used, with two researchers not involved in the trial serving as randomisation coordinators. The lead coordinator prepared a randomisation schedule using http://www.randomization.com/. This was stored securely on a server with exclusive access granted to the randomisation coordinators. Randomly permuted block randomisation was used to ensure balance between the two arms and enhance concealment in group allocation. Block sizes were varied and kept confidential from the investigators and research assistants to prevent prediction of upcoming assignments, and the randomisation sequence was managed by an independent third party. After identifying and obtaining consent, the research assistant requested an allocation from the randomisation coordinators by SMS. Upon receiving the SMS request, the randomisation coordinators sent the allocation to a study tablet device by email. The time from sending the SMS request to the time the allocation email was received was used to evaluate the suitability of this method.

Results

One hundred participants were successfully randomised. Group allocation was received for 91/100 (91.0%; 95%CI 84–96%) of participants within 15 min of the research assistant sending the SMS request, and for 98/100 (98.0%; 95%CI 93–100%) within 30 min.

Conclusion

This study demonstrates that a low-cost, third-party randomisation method is a feasible and effective approach for emergency neonatal research in low-resource settings. It enabled timely and concealed allocation without the high costs associated with centralised systems. This method presents a scalable solution for small to medium-sized trials where rapid and unbiased allocation is critical.

Trial registration

Study is registered on Pan African Clinical Trials Registry (PACTR) PACTR202208462613789. Registered 08 August 2022. https//pactr.samrc.ac.za/TrialDisplay.aspx?TrialID=23,888.

Background

Various research designs are available for scientific research, but the most credible level of evidence comes from well-conducted and adequately powered randomized controlled trials (RCTs). Randomisation is an essential component of the research design, helping to balance both known and unknown confounders between the intervention and control groups, and allowing the effectiveness of a given intervention to be estimated [1]. Randomisation involves the generation of a random sequence by which to assign participants [2]. However, for randomisation to be successfully implemented, it relies on adequate allocation concealment. Allocation concealment prevents the investigator or research assistant from knowing or being able to predict the allocation group of their participant. Without allocation concealment, even properly generated randomisation sequences can be undermined, leading to selection bias and potentially misleading estimated differences between the groups apart from the intervention of interest [3]. Selection bias has the potential to influence the trial outcome and can both exaggerate and underestimate the effectiveness of an intervention [4].

Many low-cost randomisation and allocation concealment methods, have the potential to introduce bias by ineffectively concealing the allocation sequence from the researcher [5–8]. Methods like allocation by day of the week, or by hospital inpatient numbers may allow researchers to confidently predict the allocation of their next participant, and are therefore no longer recommended [7]. More robust methods have been used including the generation of a randomisation sequence by an independent researcher, with placement of the codes into envelopes to conceal the allocation sequence from the research team. However, this approach is now rarely used in high-resource settings [5, 8]. Even this method carries a potential for subversion if fixed or very small block sizes are used, as these can make upcoming allocations predictable, or if envelopes are unsealed, translucent, or otherwise inadequately secured, allowing staff to visualise or tamper with assignments [6]. Subversion of allocation concealment can compromise the internal validity of a trial by introducing selection bias. Investigators aware of upcoming allocations may consciously or unconsciously assign participants to a preferred treatment arm, potentially inflating or obscuring treatment effects. Such subversion may occur if staff believe one intervention is superior or if the procedural weaknesses make the allocation sequence predictable. To minimise these risks, randomisation in our trial was undertaken by an independent third party, who securely held the allocation list and revealed each assignment only after participant enrolment, thereby maintaining full allocation concealment and preventing potential subversion.

More recently, web-based randomisation services have been developed to generate customised randomisation sequences. Typically, these are automated, password-protected digital platforms used to allocate participants and are often managed by an independent organisation. These systems typically support features such as blocking, stratification, dynamic randomisation methodology, ensuring allocation concealment and real-time access to group assignments via Short Messaging Services (SMS) or email. While these tools provide simplicity, flexibility, and scientific rigor, their advanced features can be expensive. This expense is often prohibitive for studies funded by small grants or for pilot studies conducted in low-income countries (LICs), restricting equitable access to robust randomisation methods. For instance, one web-based service, Sealed Envelope, (https://www.sealedenvelope.com) charges £2,070 for setup plus £70 per month thereafter, while Randomize.net (https://randomize.net) offers its service at a flat rate of $2,500 per trial regardless of duration. These methods can also be difficult to implement in areas with poor network access [9].

We focused on emergency neonatal research because many life-saving interventions for preterm and critically ill neonates must be initiated within minutes of birth. The Delivery Room Continuous Positive Airways Pressure (DR-CPAP) trial was a single-centre feasibility and acceptability study evaluating the use of immediate continuous positive airways pressure (CPAP) among preterm infants with a birthweight of 800–1500 g at a referral hospital in Uganda [10]. Infants were randomised to one of two arms: (1) application of CPAP +/- oxygen in the delivery room within 15 min of birth, or (2) supplementary oxygen at delivery when indicated. Preterm birth is often rapid and unpredictable, and in such situations, conventional consent and randomisation processes can delay treatment and reduce study feasibility. Rapid randomisation methods are therefore essential to ensure timely interventions while maintaining scientific rigour. Notably, studies conducted in Kenya and Canada have explored alternative low-cost randomisation methods using mobile phone-based systems: an SMS and Android application in Kenya, and email and SMS in Canada [11, 12]. The Kenyan study successfully randomised all participants before 35 min, while the Canadian study randomised 95% of participants within 15 min.

Building on this evidence, we developed and tested a low-cost, third-party allocation process for randomisation in our small feasibility trial [10]. Key considerations included: ensuring that the randomisation sequence was generated independently from the research team; guaranteeing that the sequence would remain inaccessible to the research team throughout the study; developing a reliable means for research assistants to request the allocation code; and ensuring timely receipt of the allocation code to avoid delays in the intervention (our intervention was required within 15 min of birth). We describe our experience of adopting such a method for emergency neonatal research in a low-resource setting.

Methods

The trial randomly assigned neonates in a 1:1 allocation ratio at birth to receive either early CPAP (intervention) or standard of care (control) [10]. Centralised “third-party” randomisation and allocation were used, where independent individuals carried out the randomisation process manually. Two researchers not directly involved in the study serving as our randomisation coordinators. The lead coordinator prepared a randomisation schedule using http://www.randomization.com/ and stored it securely on the Mbale Clinical Research Institute (MCRI) server with exclusive access granted to the randomisation coordinators only. Randomly permuted block randomisation, using blocks of varying sizes (4, 6 and 8) was used to ensure balance between the two arms. The randomisation co-ordinators used their personal smart phones and wireless internet access was freely available whilst the research co-ordinators were at work. As the study was 24/7, additional data bundles were purchased for the randomisation co-ordinators to be used out of office hours e.g. night duties, weekends. The total cost of our randomisation and allocation procedure for the 12-month trial was USD 500 (additional data for two randomisation co-ordinators, purchase of trial phone for sending SMS, SMS packages, additional data for the trial tablet).

Our trial used a two-step consent process. The research assistant first screened the mother for pre-delivery eligibility criteria and then approached her for verbal consent [10]. Verbal consent, including agreement to randomisation, was obtained from the mother and/or the father if a preterm delivery was anticipated during active labour or in obstetric conditions requiring urgent delivery. Upon obtaining verbal consent, a research assistant was present at the delivery and initiated timing with a stopwatch at birth. Participants had to be spontaneously breathing before the inclusion criteria could be met, and up to 5 min were permitted before the participant was excluded. If the entry criteria were fulfilled within five minutes after birth, and after determination of the birthweight, the neonate was randomised. If verbal consent had been given, the research assistant then requested a treatment allocation from the two randomisation coordinators by SMS. Due to the nature of the intervention, it was not possible to blind mothers, caregivers, or hospital staff. Once the mother and infant were stabilised and eligibility was confirmed, full written informed consent was obtained.

The study phone was dual sim, allowing two different networks to be functional on the study phone. This helped ensure that if the first cellular network was not functional, then an SMS could be sent from the alternative cellular network. To save time and ensure that the SMS requests were sent only to, and always to, the two randomisation co-ordinators, the study phone had only the numbers of the two randomisation co-ordinators saved in it, then the research assistant who was requesting an allocation code, would select the option “send to all”. The SMS was a standard message with fixed syntax that read, “Please send the next allocation code for DR-CPAP trial”. Upon receiving the SMS request, the randomisation coordinators sent the allocation to a study tablet device by email. In order to ensure that the correct allocation code was sent and received, the randomisation co-ordinators included the study code in the email subject e.g., DR0045. The email response contained only the allocation code, either Code DR-CPAP or S-bCPAP. No personal information was included in either the SMS request or the email allocation.

Having two randomisation coordinators helped ensure that at least one of them would be connected to a phone service and internet network. Both randomisation co-ordinators had access to the same secure server, ensuring that there was no risk of inconsistent or conflicting allocations. A unique notification sound was implemented on the smart phones of the randomisation co-ordinators to aid their recognition and timely response. The time on the study phones and the study tablets was synchronised and the time lapse between the SMS request and the receipt of the allocation was recorded. Any failures in communication such as internet or phone service unavailability, were documented. Upon receipt of the allocation, the research assistant recorded it in the RedCap database, with no subsequent communication with the randomisation coordinator.

The investigators acknowledged that this method of randomisation was novel, and although it was expected to be a more robust method, it was possible that it might have limited the feasibility of initiating the trial intervention within 15 min of birth. According to our trial protocol, after enrolling 20 participants, the timings of receiving the randomisation code and initiating the intervention were reviewed by the Trial Steering Committee and the Trial Management Group. Our protocol dictated that if the receipt of the randomisation code was found to be limiting the timeliness of the intervention, the randomisation coordinator would transition to using serially numbered, sealed and opaque envelopes.

The duration (in minutes from the time an SMS requesting group allocation was sent, to the time the randomisation allocation was received on the study tablet) was used to evaluate the suitability of a third-party allocation process for randomisation. We computed the median and interquartile range of the duration to randomisation. We also evaluated third-party randomisation feasibility by determining the proportion of infants randomised within 15 min, with the associated two-sided 95% confidence interval.

For a feasibility trial, it is not necessary to conduct sample size calculations to power the study [13]. A randomised sample size of n = 100 (n = 50 per arm) was considered appropriate for this feasibility and acceptability study. Allocation accuracy was defined as the proportion of participants who were correctly assigned to their intended study arm according to the pre-determined randomisation sequence.

The project was reviewed by three research ethics committees. Initially, local approval was obtained from Mbale Regional Referral Hospital Research Ethics Committee (MRRH-REC) for approval to undertake this research study (Approval ID: MRRH REC 123). MRRH-REC is mandated by the Uganda National Council of Science and Technology (UNCST) to offer Ethical Approval for research involving human subjects in Uganda. Subsequently, the trial was approved by UNCST (Approval ID: HS2605ES). Lastly, in alignment with the funder’s requirements, ethical opinion was sought from a UK Research and Ethics Committee, resulting in a favourable ethical opinion from Liverpool School of Tropical Medicine Research Ethics Committee.

Results

A total of 100 participants were successfully randomised between 10th April 2023 and 4th April 2024. Allocation accuracy was 100%. 52 participants were randomised to the intervention arm and 48 to the control, their characteristics are described in Table 1. No significant differences observed were observed in the baseline characteristics between the two groups, indicating that randomisation was balanced.

Table 1.

Characteristics of participants

	Intervention group	Control group	Total	p-value*
	N = 52	N = 48	N = 100
Birthweight (g)				0.21
Median (Interquartile range)	1218 (1030, 1387)	1275 (1083, 1403)	1235 (1066, 1400)
Sex				1.00
Male, n (%)	25 (48.1)	23 (47.9)	48 (48.0)
Female, n (%)	27 (51.9)	25 (52.1)	52 (52.0)
Gestational age by Ballard (weeks)				0.68
Median (Interquartile range)	32 (31, 33)	32 (30, 33)	32 (30,33)
Multiple birth				0.66
No, n (%)	38 (73.0)	33 (68.8)	71 (71.0)
Yes, n (%)	14 (26.9)	15 (31.3)	29 (29.0)
Birth order of participant if multiple birth				0.70
First born, n (%)	9 (64.3)	11 (73.3)	20 (69.0)
Second born, n (%)	5 (35.7)	4 (26.7)	9 (31.0)
Person who assisted with delivery
Nurse/midwife, n (%)	32 (61.5)	28 (58.3)	60 (60.0)	0.87
General doctor / intern / medical officer, n (%)	19 (36.5)	18 (37.5)	37 (37.0)
Specialist doctor / obstetrician, n (%)	1 (1.9)	2 (4.2)	3 (3.0)
Type of delivery
Vaginal normal, n (%)	33 (63.5)	22 (45.8)	55 (55.0)	0.11
Breech vaginal delivery, n (%)	2 (3.8)	3 (6.3)	5 (5.0)
Planned c-section, n (%)	2 (3.8)	0 (0.0)	2 (2.0)
Emergency c-section, n (%)	15 (28.8)	23 (47.9)	38 (38)
Need for resuscitation
Yes, n (%)	16 (30.8)	19 (39.6)	35 (35)	0.41
No, n (%)	36 (69.2)	29 (60.4)	65 (65)

*P-values for categorical variables were calculated using the χ² test or Fisher’s exact test, as appropriate; P-values for continuous variables expressed as medians (IQR) were calculated using the Mann–Whitney U test

The median duration from the time of delivery of the infant to the time the SMS request is sent to the randomisation coordinators was 7 min (IQR, 5–10 min) as shown in Table 2. The median time from the SMS request to the time the randomisation allocation email was received from the randomisation coordinator on the study tablet was 2.5 min (IQR 2–4 min). Overall, the median total time from birth to randomisation was 11 min (IQR 8–16 min).

Table 2.

Time to randomisation in minutes

Group	Intervention	Control	Total
	N = 52	N = 48	N = 100
Time from requesting to receiving randomisation code, n (%)
Within 15 min	50 (96.1)	41 (85.4)	91 (91)
15 to 30 min	2 (3.9)	5 (10.4)	7 (7)
Beyond 30 min	0 (0.0)	2 (4.2)	2(2)
Time from birth to randomisation, n (%)
Within 15 min	42 (80.8)	32 (66.7)	74 (74)
15 to 30 min	9 (17.3)	11 (22.9)	20 (20)
Beyond 30 min	1 (1.9)	5 (10.4)	6 (6)
Continuous variables, median (IQR) minutes
Time from birth to requesting for randomisation code	6.5 (5, 9)	7 (4, 13)	7 (5, 10)
Time from requesting to receiving randomisation code	2 (2, 3)	3 (2, 5)	2.5 (2, 4)
Time from birth to randomisation	9 (7.5, 13.5)*	14 (9, 19)*	11 (8, 16)

*p-value <0.01 comparing control and intervention arm using the Mann-Whitney U tes

Figure 1 illustrates the duration from the research assistant sending the SMS request to receipt of the allocation email. Most randomisations occurred promptly, with relatively few instances of delay. Group allocation was received for 91.0% (95% confidence interval [CI], 84–96%) of participants within 15 min of the research assistant sending the SMS request, and for 98.0% (95% CI, 93–100%) or participants within 30 min. The proportion of participants randomised within 15 min of sending the SMS request did not differ significantly between the two arms (p = 0.08, Fisher’s exact test).

Fig. 1.

A dot plot showing the time from sending SMS to randomisation with each point representing an individual randomisation event including the median (solid black line), 25th and 75th centile (grey dotted lines), and reference lines at the pre-defined threshold of 15 min (solid red line) and 30 min (dashed red line)

Time from birth to receipt of the allocation email is illustrated in Fig. 2. Group allocation was received within 15 min of birth for 74.0% of participants and within 30 min of birth for 94.0% of participants. Of note, the time from birth to randomisation was significantly longer in the control group compared to the intervention group (14 min vs. 9 min, p < 0.01); in fact, 80.8% (42/52) of participants in the intervention arm were randomised before 15 min of age, and 98.1% (51/52) were randomised before 30 min of age. In comparison, in the control arm 66.7% (32/48) of participants were randomised before 15 min of age, and 89.6% (43/48) were randomised before 30 min of age. The proportion of infants randomised before 15 min of age in the intervention arm was significantly higher than in the control arm (80.8% vs. 66.7%, p = 0.004, χ²).

Fig. 2.

A dot plot showing the time from birth to randomisation with each point representing an individual randomisation event including the median (solid black line), 25th and 75th centile (grey dotted lines), and reference lines at the pre-defined threshold of 15 min (solid red line) and 30 min (dashed red line)

Discussion

The low-cost randomisation and allocation method we evaluated was easy to implement, and effective, with 91% (95%CI, 84–96%) of participants being randomised within 15 min of the SMS request being sent, and 98% (95%CI, 93–100%) being randomised within 30 min. This supports the feasibility of the third-party randomisation approach. As expected, the time from birth to randomisation was longer, and the frequencies of participants being randomised within 15 and 30 min of birth were lower, 74% and 94%, respectively. Interestingly, the time from birth to randomisation was significantly shorter in the intervention group than the control group (9 min vs. 14 min, p < 0.01). The shorter time from birth to randomisation in the intervention group may reflect variability in coordinator response times, or chance imbalance. Although the need for resuscitation or caesarean delivery could have delayed requests for allocation codes, there was no difference in resuscitation rates or type of delivery between groups. While the reason for this difference remains unclear, the finding demonstrates that timely randomisation and intervention delivery were feasible within the early neonatal period.

As shown in Fig. 1, allocation information was not received within the 15-minute of age threshold for 9% of participants. In the allocation of the first three participants, the randomisation co-ordinators had not set a specific notification tone on their devices. After this was initiated, the co-ordinators were more easily alerted to the SMS request. In the other 7 instances where the code was received after 15 min, these all occurred during weekends, nights or when the randomisation co-ordinator was out of station for training, the challenges included: technical issues with the smart phone of the randomisation co-ordinator; unreliable power supply when out of workstation to charge phone; when the co-ordinator had stepped away from their smart phone when SMS request sent; and SMS sent during morning and evening commute while in transit.

A comparable low-cost approach (approximately CAD $250 ) using email and SMS was employed in a smaller 48-participant randomised controlled trial [11], where 96% of participants were randomised within 15 min of the SMS request - a level of efficiency similar to that observed in our study. In our trial, 91.0% of participants were successfully randomised within 15 min of the SMS being sent. Although this proportion remains high, it is slightly lower than that reported in the aforementioned study, which only recruited during daytime hours. The modest difference is likely explained by our 24/7 enrolment approach, which required teams to respond at all hours, including overnight periods when staff availability and immediate phone access may be reduced.

A recent clinical trial conducted in Kenya, researchers implemented a mobile phone-based randomisation system using SMS and an Android application [12]. This approach used a fixed syntax to generate text messages containing participant identifiers, trial site, stratum, and trial name. The system verified the input parameters and obtained an allocation from a central database before returning the response to the sender. The SMS-based method was compared against the master randomisation list, the system demonstrated 100% accuracy, with allocation times ranging from 10 s to 35 min. While the time frame was comparable to that of our study, this method offered the significant advantage of not requiring continuous human oversight - this has the potential to reduce costs further.

The main challenge experienced in our study was during out of office hours when the randomisation co-ordinators sometimes stepped away from their phones or had limited access to power or network for their phones. These challenges could be overcome by ensuring that the personnel selected as the randomisation co-ordinators should be from a team that are also working 24/7 in the research institute to ensure adequate access to wi-fi signal and power. Of note the study phone had a dual sim-card with two different cellular networks, this helped ensure that there were no challenges in sending a SMS request from the research assistant to the randomisation co-ordinators. In addition, the trial tablets had access to both the research institute wi-fi network and an alternative local network provider as a back-up. None of the delays in acquiring the allocation codes occurred on the tablet or phone of the research assistant, suggesting that with adequate power and internet access provision for randomisation co-ordinators, this method could be improved.

This randomisation technique relies on continuous human oversight; therefore, we chose to have two randomisation co-ordinators for our trial so there was always at least one co-ordinator on duty. This however did not overcome the challenges of unstable power supply and network when away from the research institute, and likely the addition of more co-ordinators would not solve this issue. We recommend that if a trial needs to recruit 24/7 like ours, that multiple randomisation co-ordinators are used, so that there is always someone available in-station with reliable power and internet during nights and weekends. It is possible that this randomisation technique might be better suited for smaller studies in terms of sample size or number of sites, as this method will become significantly more resource intensive for larger or more complex studies.

This method cost approximately $500 to run over the 12-month study period. This amount included a small allowance for each independent researcher who served as a randomisation co-ordinator, to cover their communication-related expenses. All co-ordinators were already employed full time in other roles within the research institute, and the additional workload was minimal - averaging only two participant enrolments per week. As such, the randomisation process did not necessitate hiring additional staff or providing supplementary compensation to existing personnel. This was considerably cheaper than the alternative third-party options available although not as cheap as the SMS-method used in the Kenyan trial that did not require 24/7 employment of randomisation co-ordinators [12]. Although this method proved to be affordable for our small feasibility trial, this is in part because the trial was being run in an established research institute with access to reliable wi-fi, and the research co-ordinators were already employed full time in the research institute. If the workload were to be higher, individual payments per participant for the randomisation coordinators, or alternatively dedicated full-time staff may be required. If this method were to be employed in a setting without an established research institute, then additional costs may be incurred for supplementary staff, power, and internet provision. Currently, about 65% of the population in sub-Saharan Africa is covered by 4G networks, rising to 98% among the urban population, suggesting that this would be generalisable to most urban settings in the region [14].

As demonstrated in this pilot study, the third-party randomisation method is not only low-cost but also highly adaptable for rapid deployment. Teams can be trained quickly using simple, well-structured protocols supported by mock scenarios, checklists, and supervision. This is particularly important in low-resource settings, where research capacity may be limited [15]. The approach proved feasible in our context, requiring minimal infrastructure while maintaining allocation concealment and operational integrity. These findings suggest that with basic training and oversight, third-party randomisation can be a practical and scalable solution for future small to medium-sized trials in similar settings.

While our third-party, low-cost randomisation method proved feasible and effective for a simple two-arm trial - and could accommodate limited stratification - its use in more complex trial designs may be limited. Multi-arm trials, particularly those involving adaptive allocation or multiple stratification variables, often require centralised, automated systems to maintain allocation concealment and reduce operational complexity. In such contexts, web-based platforms still offer greater flexibility and real-time coordination. However, such systems may remain inaccessible in low-resource settings due to cost and connectivity constraints, highlighting the need for innovative and context-appropriate alternatives.

Our feasibility study has several limitations. This randomisation technique relies on continuous human oversight, making it suitable primarily for smaller studies in terms of sample size and number of sites; scaling up would substantially increase resource demands. The tool may also be less effective for more complex or multi-arm trials, as well as for larger studies. Our approach would be most applicable in trials where few prognostic factors influence the outcome. However, when multiple prognostic factors are present, the absence of stratification in this method may increase the risk of chance imbalance, particularly in small to medium-sized trials. Our study also has several notable strengths. Firstly, our findings provide a valuable insight into a novel randomisation approach and lay the groundwork for further research exploring its potential application in other randomised controlled trials conducted in low-resource settings. Secondly, there is limited existing literature on robust, low-cost alternatives for randomisation in low-resource settings, making this contribution particularly relevant. Thirdly, the proposed method is straightforward and easy to implement, which is a considerable advantage in resource-limited environments. Our method is also relatively inexpensive when compared with the sophisticated randomisation systems used in high-resource settings. Finally, the technique could be widely applicable in low-resource contexts under the right operational circumstances.

Conclusion

The selection of an appropriate randomisation and allocation method is critical to the integrity of clinical trials. This study demonstrates the feasibility and effectiveness of a low-cost, third-party randomisation approach for emergency neonatal research. It enabled timely and concealed allocation without the high-costs typically associated with centralised systems, offering a scalable model for small- to medium-sized trials where rapid and unbiased allocation is essential. Although scaling up to larger trials may require additional personnel, potentially impacting cost-efficiency, this feasibility study highlights the potential of low-cost innovations to strengthen trial quality and accessibility, supporting global efforts to expand high-quality research in low-resource settings.

Abbreviations

BCPAP

Bubble continuous positive airways pressure

CPAP

Continuous positive airways pressure

DR-CPAP

Delivery room continuous positive airways pressure

FGD

Focus group discussion

HIC

High income country

IVH

Intraventricular haemorrhage

KII

Key informant interview

LMIC

Low- and middle-income country

MCRI

Mbale Clinical Research Institute

MRRH

Mbale Regional Referral Hospital

NMR

Neonatal mortality rate

NNU

Neonatal Unit

PDA

Patent ductus arteriosus

PI

Principal investigator

RDS

Respiratory distress syndrome

TMG

Trial management group

TMT

Trial management team

TSC

Trial steering committee

VLBW

Very low birthweight

Authors’ contributions

KB conceived the trial. KB, JMS, AN, FO designed the trial and wrote the protocol. KB, JMS, AN, FO, AgN, CH, JN, AHS, KL, AD, JW and POO critically reviewed the study protocol before submission for ethical approval. KB, JMS, FO, MM, GA and AD designed the data collection tools. KB, AN, AgN and MM designed the topic guides for the qualitative research. The study statisticians, JMS and FO, performed the data analysis. KB wrote the first draft of the manuscript and all authors critically revised, read and approved the final manuscript.

Funding

Dr Kathy Burgoine received a Joint Global Health Trials Development Grant (JGHT, grant MR/V004468/1), jointly funded by the Department of Health and Social Care (DHSC), the Foreign, Commonwealth and Development Office (FCDO), the Medical Research Council (MRC) and Wellcome Trust. This UK funded award is part of the EDCTP2 programme supported by the European Union. The trial funders had no role in the study design and will have no role in the data collection, analysis, interpretation of data, nor decision to submit this protocol for publication.

Data availability

The datasets used and/or analysed during the current study are available from the corresponding author on reasonable request.

Declarations

Ethics approval and consent to participate

This study was conducted to the highest ethical and governance standards in accordance with the principles of the Declaration of Helsinki and Good Clinical Practice. Full written informed consent was provided by the mother and/or father for each infant enrolled in the DR-CPAP trial. The study has been given ethical approval by Mbale Regional Referral Hospital research ethics committee (MRRH-REC, 123) and Uganda National Council of Science and Technology (UNCST, HS2605ES). In alignment with the funder’s requirements, ethical opinion was sought from a UK Research and Ethics Committee, resulting in a favourable ethical opinion from Liverpool School of Tropical Medicine Research Ethics Committee.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.