Antibiotics are a mainstay of the treatment of lower respiratory infections in young children even though most episodes are caused by self-limiting viruses. Innovative child friendly tools that improve the diagnosis of respiratory illnesses, safely reduce the unnecessary use of antibiotics, and are suitable for implementation in resource-constrained settings are urgently required to safely improve antibiotic stewardship and stem the rising rates of antibiotic resistance globally.

In the Bangladesh Lung Auscultation Artificial Intelligence for Antibiotic Stewardship Trial (BLAAAST) the investigators aim to determine whether treatment failure frequency among children in rural Bangladesh managed by clinical guidelines enhanced by a commercially available, artificial intelligence (AI)-enabled digital stethoscope is non-inferior to guidelines alone. The investigators hypothesize treatment failure frequency among 'enhanced IMCI' participants will be no worse than standard care by a +/-2% margin, safely reducing antibiotic use by 50-60%. The investigators will also evaluate if a diagnostic strategy enhanced by an AI-enabled digital stethoscope is a sustainable alternative to standard care for children in rural Bangladesh. The investigators hypothesize that care augmented by an AI-enabled digital stethoscope will have additional benefits via reduced antibiotic use that will outweigh digital auscultation costs resulting in cost-effectiveness compared to current practice.

BLAAAST affords a unique opportunity to evaluate the efficacy of clinical guidelines enhanced by an AI-enabled digital stethoscope on child pneumonia outcomes in Bangladesh, if digital auscultation may be instrumental in the wider antibiotic stewardship strategy, and whether a digital stethoscope diagnostic tool is cost-effective in the care of children with respiratory illnesses.

Participants will be screened for eligibility by study physicians. Children who screen positive will then be assessed for eligibility. Written informed consent will be obtained. All enrolled children will have a lung sound recording obtained. Study physicians will sequentially record lung sounds using the digital stethoscope from four chest positions (two anterior and two posterior) on participants. Each chest position recording is 15 seconds, and the overall procedure ~2 minutes. Lung sound recordings will be transferred to an encrypted tablet and analyzed by the convolutional neural network model to automatically detect normal and abnormal lung sounds. Blocked randomization will occur after the recording has been completed. Children who consent and meet eligibility criteria will be randomized into one of two arms.

Children randomized to the intervention will receive amoxicillin syrup (250mg/5ml concentration) or placebo based on the automated algorithm's lung sound classification. Children receiving a positive digital stethoscope result will receive amoxicillin while children receiving a negative digital stethoscope results will receive placebo. All children in the control arm (standard care) will receive amoxicillin. Treatment duration will be five days per Integrate Management of Childhood Illnesses (IMCI) guidelines, administering the study product twice daily. Children randomized to the control group will receive amoxicillin syrup (250mg/5ml concentration) for five days as all controls will meet IMCI pneumonia criteria. Study physicians and caregivers in the IMCI clinic will be masked to the randomization group. Masking will be achieved by applying the digital stethoscope to all enrolled children and concealing the automated algorithm result on the device.

Trained community healthcare worker paramedics equipped a pulse oximeter and paracetamol will conduct home visits for all participants on study days 2, 3, 5, and 14 after enrollment (defined as Day 1). A study physician will conduct the day 7 study visit and assign the primary outcome. A 24-hour window period will be permitted for each home visit in the event that the participant is not available on the scheduled visit day.

Any child that meets treatment failure criteria will be immediately transported to the study facility (Upazila Health Complex) for study physician confirmation and further treatment planning. Similarly, an oxygen saturation measured as 90-94% will prompt referral to the study physician for assessment for treatment failure. Study physician confirmation is required for a treatment failure outcome. Caregivers will be provided a trial telephone hotline for communications between scheduled visits (by text message or phone call), should parental concerns arise between household visits. If caregivers prefer, the caregiver may also return to the study clinic for a study physician reassessment between community health worker (CHW) household visits.

For any participant receiving amoxicillin or placebo who is confirmed to have a World Health Organization (WHO)-defined emergency sign, WHO-defined clinical danger sign, or WHO-defined hypoxemia, and the child will be hospitalized and receive parenteral antibiotics according to WHO-guidelines and aligned with Bangladesh national guidelines. Further treatment and clinical management will be at the discretion of the treating physician. For any participant on amoxicillin or placebo who is confirmed to have lower chest wall indrawing and/or very fast breathing for age on day 7 and who is confirmed to be without any WHO-defined emergency signs, to be without any WHO-defined clinical danger signs, and to be without WHO-defined hypoxemia, then the child will be initiated on oral amoxicillin-clavulanic acid over a minimum of 5 days. Further treatment and clinical management will be at the discretion of the treating physician. Unblinding will occur only when knowledge of treatment allocation is essential for the clinical management of a participant or when required to address a serious adverse event or unexpected problem posing risk to participant safety.

Participants will be enrolled into the trial for a period of 2 weeks each. There is a minimum of 6 study visits. The use of a placebo in this study is justified to maintain participant adherence to the assigned treatment and ensure comparability between the intervention and control groups. Without a placebo, participants who do not receive antibiotics are likely to be dissatisfied and seek treatment from another provider, thereby introducing a protocol violation, observer bias, and compromising the study's validity by leading to an over- or underestimation of the true treatment effect. By ensuring blinding, the placebo is also crucial for reducing bias in outcome assessment by study investigators. In sum, by using a placebo, the study can accurately differentiate between the effects of appropriately withholding antibiotics and any potential treatment effects, thereby providing robust and reliable results that are most likely to impact on health policy.

With a sample size of 2500 and one-week (7 day) treatment failure rate in the placebo arm of the trial to be approximately 4% based on prior data, and the margin of non-inferiority was chosen to be 2% based on the largest clinically acceptable difference between treatment failure of the intervention compared to the standard care. Given these parameters, the investigators will have at least 80% power to reject the null hypothesis of inferiority given that the treatment failure rate in the intervention arm is equal to the placebo arm, at 4%, assuming a one-sided type one error of 0.05.

Requiring medical care, requiring hospitalization, and mortality among study participants will be recorded by the study as the 3 primary markers of safety. In addition to these 3 parameters, the investigators will collect data on all unexpected adverse events (AEs) that occur during the study period. AEs will be actively collected from enrollment through Day 14 post-enrollment, or until completion of all follow-up procedures, whichever occurs later.

Rigorous safety measures will be in place including eligibility restrictions to 'low risk' children with non-severe clinical pneumonia, a hotline for 24/7 communication, close household follow-up (at least 5 household visits, including daily over the first 3 days of illness), and facilitating participant clinic transportation if needed. The trial will have a Data Safety and Monitoring Board (DSMB) and Technical Advisory Group. If unscheduled healthcare facility visits are necessary, then participants will be reimbursed for any travel expenses incurred. Participants will not be directly compensated for study visits conducted in the home and for study visits coordinated with routine healthcare visits at hospitals.

Inclusion Criteria:

• 2 to 59 months of age (60 days to 59 months 29 days),
• illness duration 14 days or less,
• meeting non-severe pneumonia criteria, and
• from the trial catchment areas.

Exclusion Criteria:

• prior participation,
• age <2 months or >59 months,
• illness duration >14 days,
• not meeting IMCI-defined non-severe pneumonia criteria,
• oxygen saturation (SpO2) <90%,
• moderate or severe malnutrition,
• known chronic disease,
• history of prematurity,
• one or more high risk clinical signs (any WHO-defined clinical danger sign and/or WHO-defined emergency sign),
• known antibiotic exposure in the past 1 week,
• non-respiratory illness requiring systemic antibiotic treatment according to IMCI guidelines.

The study will generate multiple gigabytes of de-identified individual-level data, shared as non-proprietary formats (.csv for tabular data; .wav for audio). Outcomes include treatment failure at Days 7 and 14, time to treatment failure, relapse by Day 14, modified treatment failure, outcomes stratified by age, sex, and clinic, and correct lower respiratory infection diagnosis. No participant identifiers will be shared. Metadata will include participant identification (ID) and a data dictionary. Associated documentation will include data collection instruments, methods, and final protocols. Data will be provided in formats accessible without specialized software. Common data elements, best practices, and standardized reporting approaches will be used where feasible. De-identified data, statistical analysis plans, and analytic code will be shared without restriction via public download.

Publicly shared data will be deposited into the open-source research data repository The Dataverse Project.

Data will be shared by unrestricted download from The Dataverse Project open source research data repository.