Pneumonia is the leading infectious cause of death in children under five years of age worldwide, and most of these deaths occur in low- and middle-income countries. In these settings, frontline health workers diagnose pneumonia using the World Health Organization's Integrated Management of Childhood Illness (IMCI) guidelines, which rely mainly on counting how fast a child is breathing and checking for chest indrawing. This approach has saved many lives, but it is not very specific. As a result, many children who actually have self-limiting viral illnesses that do not require antibiotics are nonetheless treated with antibiotics, contributing to the global rise of antimicrobial resistance.
New digital stethoscopes paired with artificial intelligence (AI) can record a child's lung sounds and automatically detect abnormal sounds such as crackles and wheezes with accuracy comparable to physicians. The LaLeLa Lung Study will evaluate whether adding an AI-enabled digital stethoscope to standard IMCI assessment improves the accuracy of pneumonia diagnosis among children aged 2 to 59 months who present with cough and/or difficult breathing at a primary care clinic in Cape Town, South Africa.
The main component (Objective 1) is a randomized, triple-blinded diagnostic accuracy study that will enroll 350 children, randomly assigned in a 1:1 ratio to either IMCI care enhanced by the AI-enabled digital stethoscope or standard IMCI care. An independent panel of physicians, blinded to the AI results and to study-arm assignment, will review each case and serve as the reference standard for determining whether pneumonia was truly present. The investigators hypothesize that IMCI enhanced by the AI stethoscope will diagnose pneumonia more accurately, and target antibiotics more appropriately, than standard IMCI alone. Nested sub-studies will additionally evaluate a second AI stethoscope for tuberculosis detection, a wearable lung-sound and respiratory-rate patch, an automated respiratory-rate monitor, and a smartphone-connected pulse oximeter.
A separate component (Objective 2) is a mixed-methods implementation study at a second clinic that will assess how easily health workers can use these devices, how acceptable the devices are to health workers and caregivers, and how well the devices fit into routine clinic workflows.
Throughout the study, all AI-generated results will remain concealed from clinic staff, study clinicians, and caregivers, so the AI-generated results will not influence the care any child receives. All children continue to receive standard IMCI care. Findings will help inform whether AI-enabled digital auscultation should be integrated into childhood pneumonia care in South Africa and similar low-resource settings, with the goal of improving diagnosis, strengthening antibiotic stewardship, and reducing antimicrobial resistance and child mortality.
Age 2 to 59 months at the time of screening
Presence of cough and/or difficulty breathing
No WHO-defined emergency/danger signs (e.g., grunting, cyanosis, apnea, convulsions, or altered level of consciousness)
A legal caregiver is present, able to understand the study information, and willing to provide written informed consent
Presence of WHO-defined emergency signs requiring immediate referral or hospital admission (grunting, cyanosis, apnea, uncompensated shock, convulsions, diarrhea with severe dehydration, or altered level of consciousness)
Critical illness or clinical instability judged by the screening clinician or study physician to require urgent medical attention
Age outside the target range (younger than 2 months or older than 59 months)
Previous enrollment in the study
Parallel
StethoMe AI-enabled digital stethoscope system
DeviceA CE-marked (EU Class IIa) wireless electronic stethoscope paired with a mobile application and an on-device deep convolutional recurrent neural network trained on more than 25,000 labeled lung-sound recordings. The device captures high-fidelity respiratory sounds, automatically computes respiratory rate, and classifies sounds in real time as normal or abnormal (fine/coarse crackles, high-/low-pitched wheezes), with ambient-noise detection to flag low-quality signals. Recordings are obtained at four standardized chest positions; the algorithm's classifications are generated automatically but the output display is permanently disabled for field users so results stay concealed and do not inform care.
Standard IMCI assessment
Diagnostic testThe World Health Organization's standardized clinical algorithm for children with cough and/or difficult breathing, in which pneumonia is classified on the basis of age-specific fast breathing and/or chest indrawing in the absence of general danger signs, without digital or AI-assisted auscultation. Conducted by routine clinic health workers using standard equipment, it represents the current WHO-recommended standard of care for outpatient pneumonia assessment.
Proportion of children correctly classified with Pneumonia (Diagnostic accuracy of pneumonia diagnosis - IMCI enhanced by AI-enabled digital stethoscope vs. standard IMCI)
Accuracy of pneumonia classification (pneumonia: yes/no), defined as the proportion of children correctly classified relative to the blinded independent physician reference-panel diagnosis, summarized by sensitivity, specificity, and overall accuracy. Compares IMCI enhanced by the StethoMe AI-enabled digital stethoscope (elevated respiratory rate plus crackles ± wheeze) with standard IMCI care; between-arm differences assessed by tests of proportions and ROC area under the curve. Index clinic visit (Day 1); reference-standard diagnosis assigned retrospectively, incorporating 7-day follow-up.
Accuracy, sensitivity, specificity, and positive/negative predictive values (Diagnostic accuracy relative to routine health care worker (HCW) diagnosis)
Accuracy, sensitivity, specificity, and positive/negative predictive values of AI-enhanced IMCI compared with the routine HCW pneumonia diagnosis, each evaluated against the physician reference panel. Day 1; reference standard incorporates 7-day follow-up.
Proportion of correctly indicated antibiotic decisions (Accuracy of antibiotic decision-making)
Proportion of correctly indicated antibiotic decisions (appropriate vs. inappropriate) for AI-enhanced IMCI compared with (a) HCW antibiotic decisions and (b) IMCI guideline-based decisions, using the physician panel's assessment of antibiotic eligibility (yes/no) as reference.Day 1; reference standard incorporates 7-day follow-up.
Accuracy of pneumonia diagnosis (Expanded lung-sound classification accuracy)
Accuracy of pneumonia diagnosis when any abnormal lung sound detected by the AI algorithm (defined as AI-identified crackle and/or wheeze of any type (coarse or fine crackle, or high or low-pitched wheeze)) is treated as positive, compared with standard IMCI and HCW assessment, against the physician reference panel. Day 1; reference standard incorporates 7-day follow-up
Agreement between AI lung-sound classification and physician auscultation
Agreement between StethoMe AI classifications (crackle, wheeze, or normal) and expert physician auscultation, reported as raw percentage agreement and Cohen's kappa (including prevalence- and bias-adjusted kappa), at the chest-position and patient levels.
Correct chest positions - fidelity of digital devices in routine care (Objective 2)
Device-use fidelity (correct chest positions) from structured observation across the StethoMe, Perin Health Patch, ChARM, and Phefumla 2.0 devices.
Percentage of successful recordings - Technical Performance (Objective 2)
Device-use fidelity (percent successful recordings (workflow adherence) from structured observation across the StethoMe, Perin Health Patch, ChARM, and Phefumla 2.0 devices.
Time per consultation recording -Workflow Integration (Objective 2)
Device-use fidelity (time in minutes per recording) from structured observation) across the StethoMe, Perin Health Patch, ChARM, and Phefumla 2.0 devices.
Usability/acceptability of digital devices in routine care (Objective 2)
HCW-reported usability/acceptability (System Usability Scale. Score range 0-100 higher score better usability/acceptability across the StethoMe, Perin Health Patch, ChARM, and Phefumla 2.0 devices.
Click on the lead sponsor to view all of their trials.
Johns Hopkins University
Lead sponsor
University of Stellenbosch
Collaborator
Eunice Kennedy Shriver National Institute of Child Health and Human Development (NICHD)
Collaborator
This trial is not recruiting at the moment. You can still explore other options:
