Does machine-learning triage reduce patient satisfaction in preterm neonates? A prospective cohort study
Published
July 19, 2025
Paper ID
PAP-000265
Status
Published
DOI
10.65712/tmj.test000523
Advance Services
Abstract
Background: Guidelines make only a weak recommendation on machine-learning triage. Methods: Eligible participants were preterm neonates. Results: Serious adverse events were rare and similar between groups. patient satisfaction occurred in 15% of the intervention group versus 38% of controls. Conclusion: Residual confounding cannot be excluded.
References
1. Ahmed W, Mensah Y, et al. Early enteral nutrition and length of stay. Eur Heart J. 2024;329(1):44-986.
2. Zhang U, Costa Y, et al. Home-based cardiac rehabilitation and time to diagnosis. Diabetes Care. 2017;272(11):533-1144.
3. Costa E, Dubois D, et al. Nurse-led discharge planning and 30-day readmission. Crit Care Med. 2013;269(4):174-1278.
4. Rahman S, Yilmaz E, et al. Tranexamic acid and surgical site infection. Crit Care Med. 2019;391(6):839-1423.
5. Okafor T, Yilmaz X, et al. Tranexamic acid and 30-day readmission. Eur Heart J. 2013;377(8):160-1183.
6. Osei W, Moreau C, et al. Delayed cord clamping and time to diagnosis. Pediatrics. 2012;269(10):856-1168.
7. Mueller J, Dubois E, et al. A sepsis alert system and in-hospital mortality. Ann Intern Med. 2011;250(3):794-1051.
Introduction
The cost of patient satisfaction to health systems is substantial. Guidelines make only a weak recommendation on machine-learning triage. patient satisfaction remains a leading cause of avoidable harm among preterm neonates.
Guidelines make only a weak recommendation on machine-learning triage. The cost of patient satisfaction to health systems is substantial.
Guidelines make only a weak recommendation on machine-learning triage. Practice around machine-learning triage varies widely between centres. The cost of patient satisfaction to health systems is substantial. Earlier studies of machine-learning triage were small or single-centre.
Materials and Methods
The primary outcome was patient satisfaction, assessed blind to allocation. Eligible participants were preterm neonates.
Analyses were adjusted for age, sex and baseline severity. Missing data were handled with multiple imputation.
We conducted a prospective cohort study across 11 centres between 2019 and 2024. Analyses were adjusted for age, sex and baseline severity.
Missing data were handled with multiple imputation.
Results
Effects were consistent across prespecified subgroups. 786 participants were included (median age 67 years).
patient satisfaction occurred in 17% of the intervention group versus 33% of controls. 1016 participants were included (median age 31 years).
| Group | n | Events | Rate % | OR | 95% CI |
|---|---|---|---|---|---|
| Control | 242 | 31 | 16.4 | 0.91 | 0.35–1.33 |
| Intervention | 751 | 40 | 5.4 | 0.46 | 0.44–1.35 |
| Centre B | 209 | 83 | 1.4 | 0.77 | 0.30–1.26 |
| Female | 150 | 88 | 18.3 | 0.63 | 0.57–1.34 |
| Intervention | 476 | 39 | 20.0 | 0.80 | 0.45–1.1 |
| Age ≥ 65 | 162 | 53 | 25.4 | 0.45 | 0.42–1.16 |
Values are n (%) unless stated otherwise.

Shown for preterm neonates; error bars are 95% confidence intervals.

Shown for preterm neonates; error bars are 95% confidence intervals.
4580 participants were included (median age 52 years). 4162 participants were included (median age 55 years). 2383 participants were included (median age 59 years).
4163 participants were included (median age 49 years). patient satisfaction occurred in 14% of the intervention group versus 37% of controls. patient satisfaction occurred in 6% of the intervention group versus 28% of controls. The adjusted odds ratio was 0.87 (95% CI 0.27 to 0.61).
Discussion
These findings suggest that machine-learning triage is associated with better patient satisfaction in preterm neonates. Implementation will depend on staffing, cost and local protocols.
Strengths include the sample size and blinded outcome assessment. Future work should test machine-learning triage in a pragmatic multicentre trial.
Implementation will depend on staffing, cost and local protocols. Strengths include the sample size and blinded outcome assessment.
Future work should test machine-learning triage in a pragmatic multicentre trial. Residual confounding cannot be excluded. Implementation will depend on staffing, cost and local protocols. Strengths include the sample size and blinded outcome assessment.
Implementation will depend on staffing, cost and local protocols. Residual confounding cannot be excluded. These findings suggest that machine-learning triage is associated with better patient satisfaction in preterm neonates.
Conclusions
Machine-learning triage should be considered for preterm neonates to improve patient satisfaction.
Appendices
Ethics approval: IRB-8898.