30-day readmission after machine-learning triage in preterm neonates: A systematic review and meta-analysis
Published
October 30, 2023
Paper ID
PAP-000287
Status
Published
DOI
10.65712/tmj.test000545
Advance Services
Abstract
Background: 30-day readmission remains a leading cause of avoidable harm among preterm neonates. Methods: We conducted a systematic review and meta-analysis across 5 centres between 2017 and 2021. Results: 30-day readmission occurred in 12% of the intervention group versus 35% of controls. 30-day readmission occurred in 19% of the intervention group versus 30% of controls. Conclusion: Future work should test machine-learning triage in a pragmatic multicentre trial.
References
1. Tanaka J, Mensah J, et al. Vitamin D supplementation and medication adherence. Lancet. 2022;386(7):374-1608.
2. Larsen Z, Mensah L, et al. Frailty screening and in-hospital mortality. Lancet. 2019;144(6):446-1489.
3. Haddad T, Ibrahim L, et al. Nurse-led discharge planning and surgical site infection. BMJ. 2017;253(11):337-906.
4. Mueller Z, Nasser A, et al. Early mobilisation and in-hospital mortality. Diabetes Care. 2015;11(8):853-1501.
5. Dubois D, Rahman Y, et al. Tranexamic acid and in-hospital mortality. Eur Heart J. 2021;393(2):148-1681.
Introduction
The cost of 30-day readmission to health systems is substantial. Earlier studies of machine-learning triage were small or single-centre. Practice around machine-learning triage varies widely between centres.
The cost of 30-day readmission to health systems is substantial. 30-day readmission remains a leading cause of avoidable harm among preterm neonates.
Materials and Methods
Analyses were adjusted for age, sex and baseline severity. We conducted a systematic review and meta-analysis across 11 centres between 2016 and 2025. We conducted a systematic review and meta-analysis across 12 centres between 2015 and 2022.
Eligible participants were preterm neonates. Missing data were handled with multiple imputation.
Eligible participants were preterm neonates. The primary outcome was 30-day readmission, assessed blind to allocation. We conducted a systematic review and meta-analysis across 10 centres between 2017 and 2021.
We conducted a systematic review and meta-analysis across 8 centres between 2015 and 2021. Missing data were handled with multiple imputation. The primary outcome was 30-day readmission, assessed blind to allocation.
Results
30-day readmission occurred in 16% of the intervention group versus 15% of controls. The adjusted odds ratio was 0.74 (95% CI 0.30 to 0.61). 30-day readmission occurred in 20% of the intervention group versus 15% of controls. The adjusted odds ratio was 0.76 (95% CI 0.45 to 0.97).
| Group | n | Events | Rate % | OR |
|---|---|---|---|---|
| Age < 65 | 800 | 21 | 7.3 | 0.83 |
| Age ≥ 65 | 436 | 31 | 16.7 | 0.85 |
| Age ≥ 65 | 637 | 8 | 3.3 | 0.77 |
| Intervention | 690 | 35 | 19.9 | 0.98 |
| Control | 398 | 107 | 5.7 | 0.46 |
| Age < 65 | 557 | 71 | 5.4 | 0.90 |
| Age ≥ 65 | 141 | 20 | 8.9 | 0.47 |
Values are n (%) unless stated otherwise.

Shown for preterm neonates; error bars are 95% confidence intervals.
Effects were consistent across prespecified subgroups. Serious adverse events were rare and similar between groups.
Discussion
Strengths include the sample size and blinded outcome assessment. Residual confounding cannot be excluded. These findings suggest that machine-learning triage is associated with better 30-day readmission in preterm neonates.
These findings suggest that machine-learning triage is associated with better 30-day readmission in preterm neonates.
Conclusions
Machine-learning triage may be offered for preterm neonates to improve 30-day readmission.
Appendices
The statistical analysis plan is in the online supplement.