Large evaluation within routine primary care
The study enrolled 50,580 children across 39 facilities in two countries, providing implementation evidence beyond a tightly controlled pilot.
↳ Results, Population; Discussion
Abstract
BACKGROUND: Acute illnesses are leading causes of death among children under-five, who often receive antibiotics unnecessarily, contributing to antimicrobial resistance. Pulse oximetry and digital Clinical Decision Support Algorithms (CDSAs) can strengthen the detection and management of severe childhood illnesses, and support antibiotic stewardship in primary care, but lack evidence for scale-up. This study sought to understand the real-world impact of these tools on urgent referrals and antibiotic prescription for children under-five.
METHODS: A quasi-experimental pre-post study of the implementation of pulse oximetry and CDSAs for healthcare providers (HCPs) managing sick children at primary care level was conducted in Kenya and Senegal. Sick children 0-59 months attending study facilities were eligible. Trained research assistants collected data from caregivers and facility records on Day 0, with a follow-up phone call at Day 7. Providers were advised to use pulse oximetry for all sick children in Kenya, and in Senegal for all 1-59 days, and for 2-59 months with cough or difficulty breathing, or a moderate to severe illness. Urgent referral was recommended for SpO2 <90% in Kenya and SpO2 <92% in Senegal. Primary outcomes were antibiotic prescription and urgent referral rates at Day 0. They were assessed using generalised estimating equations for logistic regression. Results were estimated in terms of odds ratios and risk differences (RDs), adjusted where computable. The study is registered with clinicaltrials.gov (NCT05065320).
FINDINGS: A total of 50,580 sick children (1-59 days: 979 pre, 1748 post; 2-59 months: 16,782 pre, 31,071 post) were enrolled from September 13, 2021 to February 8, 2023 in Kenya and August 16, 2021 to March 31, 2023 in Senegal. In the pre-intervention period, urgent referrals were rare (0.6% in 1-59 days; 0.4% in 2-59 months), while antibiotic prescriptions were common (53.9% in 1-59 days; 74.9% in 2-59 months). Intervention uptake was 75% in Kenya and 40% in Senegal where a protracted HCP strike affected the intervention. The prevalence of SpO2 values prompting an urgent referral recommendation was 1.3% in 1-59 days and 0.8% in 2-59 months, but few of them resulted in actual referrals (26.1% in 1-59 days; 11.4% in 2-59 months). There was no change in overall urgent referrals (RD 0.2% [-0.5%, 0.9%] in 1-59 days; adjusted RD 0.2% [-0.2%, 0.5%] in 2-59 months). Antibiotic prescription rate was reduced by 14.6% [8.7%, 20.6%] in 1-59 days and by 22.6% [18.3%, 26.9%] in 2-59 months in the post-intervention period while caregiver-reported recovery rates at Day 7 remained stable.
INTERPRETATION: When implemented in routine health systems at primary care level in Kenya and Senegal, pulse oximetry and CDSAs were not found to be associated with an increase in urgent referrals but likely mediated antibiotic prescription reductions. The absence of referral increase may stem from limited severe illness detection due to low hypoxaemia prevalence and barriers to referral, also affected in Senegal by a protracted post-intervention HCP strike. Strengthening the referral system and implementing broader antibiotic stewardship strategies are likely to be needed to improve the effectiveness of the intervention and its impact on child health outcomes.
FUNDING: Unitaid grant n°2019-35-TIMCI: Tools for Integrated Management of Childhood Illness.
Key Claims
Primary findingSupported
Antibiotic prescription rate was reduced by 14.6% [8.7%, 20.6%] in 1-59 days and by 22.6% [18.3%, 26.9%] in 2-59 months in the post-intervention period
uncontrolled pre–post models support a temporal reduction, although abstract and Results estimates differ
Abstract, Findings
SecondaryTentative
Urgent referral rates remained largely unchanged in the post-intervention period
the uncontrolled comparison was underpowered at observed referral rates and cannot establish absence of a clinically important change
Results, Primary outcomes
SecondaryTentative
Despite antibiotic prescription reduction in the post-intervention period, caregiver-reported recovery rates at D7 remained statistically stable in 1–59 days
the uncontrolled comparison, incomplete Day-7 follow-up, and wide confidence interval do not establish equivalent recovery
Results, Additional secondary outcomes
SecondaryTentative
When implemented in routine health systems at primary care level in Kenya and Senegal, pulse oximetry and CDSAs were not found to be associated with an increase in urgent referrals but likely mediated antibiotic prescription reductions
causal attribution of prescribing reductions is not identified by the uncontrolled pre–post design
Abstract, Interpretation
Derived from the full evaluation — not a separate score.
Strengths · Limitations
Strengths
Prespecified analysis with transparent sensitivity checks
The Methods specifies confounder adjustment and facility-clustered GEE models, and distinguishes prespecified sensitivity analyses from post-hoc analyses.
↳ Methods, Statistical analysis
Referral cascade identifies practical implementation barriers
Table 2 follows children whose SpO2 values prompted referral recommendations and documents how few were referred. The Discussion connects these findings to mixed prior evidence and referral-system barriers.
↳ Table 2; Results, Care cascade and severe complications; Discussion
Limitations
Uncontrolled comparison cannot identify causal effectiveness
The pre- and post-intervention periods differed in season, pandemic exposure, clinical presentation, and Senegalese strike conditions. These differences prevent the temporal prescribing reduction from establishing that the package caused it.
↳ Methods, Study design; Results, Population; Discussion, limitations
Referral null remains statistically inconclusive here
Observed baseline urgent referral rates were 0.4–0.6%, rather than the assumed 3.0%. The resulting confidence intervals permit important changes, so unchanged referrals should not be read as evidence of no effect.
↳ Methods, Statistical analysis; Results, Primary outcomes
Recovery and prescribing reports need qualification
Day-7 outcomes were available for 70.2% of children, limiting stability and safety interpretations. The abstract reports a 22.6% antibiotic reduction in 2–59 months, whereas the Results report 22.1 percentage points.
↳ Abstract, Findings; Results, Population, Primary outcomes and Additional secondary outcomes
Nabu Commentary
The large routine-care sample, prespecified analyses, and referral cascade make this evaluation useful for understanding implementation. The main constraint is the uncontrolled pre–post design, particularly when the abstract attributes prescribing reductions to the tools. Low referral event rates make the referral result inconclusive, while incomplete Day-7 follow-up limits recovery and complication interpretations. Engagement with mixed prior evidence supports the contextual assessment, but does not resolve causal attribution.
Check another paper
Author Note / Field Feedback
Nabu’s assessment, alongside the field’s view.
Are you an author of this paper?
Dimension breakdown
Study Quality
Sound3.3
Confidence mediumContribution
3.6
The 50,580-child evaluation adds routine-care evidence on prescribing, implementation, and the referral cascade across two countries. Its advance is bounded by the uncontrolled comparison and an underpowered referral outcome.
Research in Context; Results, Primary outcomes; Table 2
Methodological Rigour
2.8
Registration, prespecified adjustment, facility-clustered models, and labelled sensitivity analyses support execution merit. The abstract's causal attribution alongside an uncontrolled pre–post design activates the 2.8 ceiling.
Methods, Study design and Statistical analysis; Abstract, Interpretation
Reporting
3.4
The study's outcomes, procedures, and numerical results are traceable. Central causal wording, null-as-stability phrasing, and the abstract–Results antibiotic estimate difference reduce reporting precision.
Abstract, Findings and Interpretation; Results, Primary outcomes and Additional secondary outcomes
Positioning
3.8
The Discussion engages mixed CDSA results, the companion RCT, referral barriers, and SpO2 thresholds. Limitations are substantive, but competing explanations for the antibiotic association are not fully traced through the interpretation.
Research in Context; Discussion
Reliability
Caveats4 of 4 checks
Results checkNoted
- Antibiotic prescription rate was reduced by 14.6% [8.7%, 20.6%] in 1-59 days and by 22.6% [18.3%, 26.9%] in 2-59 months in the post-intervention period: "by 22.6% [18.3%, 26.9%] in 2-59 months" (Abstract, Findings) vs "22.1 [17.8, 26.3] percentage points in 2–59 months (adjusted models)" (Results, Primary outcomes). Compare the abstract's 2–59-month reduction of 22.6 with confidence limits 18.3–26.9 against the Results' adjusted reduction of 22.1 percentage points with limits 17.8–26.3. → Conclusion holds
Reporting and transparencyClean
Ethics approvals, registration, funding, competing interests, and named data repositories are declared. Actual repository contents and subsequent release of promised materials were not independently verified.
Conflict of interest: declared · Ethics approval: declared · Preregistration: declared · Data availability: declared
Reference IntegrityClean
46 references in manuscript 43 of 43 checkable references found in an index 3 are books, websites or datasets — counted, but not index-checkable
Post-Publication RecordClean
No retraction or expression of concern found in Crossref (includes Retraction Watch).
Last checked: 20 Oct 2026
Sources: Crossref (incl. Retraction Watch) ✓
Readiness
Where this paper’s evidence sits on the path from initial observation to real-world use.
Validated
Implementation in government primary-care facilities provides a practical application pathway. Causal uncertainty, incomplete uptake, and unresolved referral effectiveness require further work before broad action on effectiveness claims.
Methods, Settings and participants; Results, Intervention uptake; Discussion
AI-generated, human-governed. Something look off? Contact us to request a review.