Acute kidney injury after machine-learning triage in patients with hip fracture: A case series
Published
May 10, 2025
Paper ID
PAP-000167
Status
Published
DOI
10.65712/tmj.test000425
Advance Services
Abstract
Background: Earlier studies of machine-learning triage were small or single-centre. Methods: We conducted a case series across 13 centres between 2015 and 2022. Results: Effects were consistent across prespecified subgroups. The adjusted odds ratio was 0.76 (95% CI 0.44 to 0.97). Conclusion: These findings suggest that machine-learning triage is associated with better acute kidney injury in patients with hip fracture.
Introduction
Guidelines make only a weak recommendation on machine-learning triage. acute kidney injury remains a leading cause of avoidable harm among patients with hip fracture. Practice around machine-learning triage varies widely between centres.
Practice around machine-learning triage varies widely between centres. acute kidney injury remains a leading cause of avoidable harm among patients with hip fracture. Guidelines make only a weak recommendation on machine-learning triage.
Guidelines make only a weak recommendation on machine-learning triage.
Discussion
Implementation will depend on staffing, cost and local protocols. Residual confounding cannot be excluded.
Strengths include the sample size and blinded outcome assessment. Residual confounding cannot be excluded.
Conclusions
Machine-learning triage merits wider use for patients with hip fracture to improve acute kidney injury.
References
1. Haddad B, Costa K, et al. SGLT2 inhibitors and emergency department revisits. Pediatrics. 2019;85(2):451-1257.
2. Andersen V, Lindqvist D, et al. Nurse-led discharge planning and post-operative delirium. N Engl J Med. 2009;281(4):150-1072.
3. Schmidt U, Okafor W, et al. Machine-learning triage and medication adherence. Pediatrics. 2024;38(5):642-1788.
4. Lindqvist X, Rahman G, et al. Nurse-led discharge planning and surgical site infection. Ann Intern Med. 2016;169(6):157-1148.
5. Mueller Z, Mensah C, et al. An enhanced recovery pathway and patient satisfaction. Crit Care Med. 2016;194(1):448-1319.
6. Ahmed X, Ibrahim J, et al. Telemedicine follow-up and glycaemic control. Lancet. 2009;372(5):774-1337.
7. Moreau W, Larsen I, et al. Machine-learning triage and health-related quality of life. Pediatrics. 2011;252(4):847-1002.
8. Farouk S, Schmidt N, et al. Telemedicine follow-up and emergency department revisits. JAMA. 2008;153(4):657-1449.
9. Novak N, Novak V, et al. SGLT2 inhibitors and ventilator-free days. JAMA. 2008;51(3):771-964.
10. Rahman V, Farouk E, et al. SGLT2 inhibitors and post-operative delirium. N Engl J Med. 2014;11(7):227-1180.
11. Kaur Z, Yilmaz T, et al. Machine-learning triage and emergency department revisits. BMJ. 2021;35(10):104-946.
Appendices
The statistical analysis plan is in the online supplement.
Materials and Methods
We conducted a case series across 4 centres between 2017 and 2023. Analyses were adjusted for age, sex and baseline severity.
We conducted a case series across 6 centres between 2016 and 2023. We conducted a case series across 12 centres between 2015 and 2023.
We conducted a case series across 14 centres between 2020 and 2025. Analyses were adjusted for age, sex and baseline severity.
Missing data were handled with multiple imputation. Eligible participants were patients with hip fracture.
Results
148 participants were included (median age 42 years). acute kidney injury occurred in 12% of the intervention group versus 26% of controls. Effects were consistent across prespecified subgroups.
acute kidney injury occurred in 8% of the intervention group versus 37% of controls. acute kidney injury occurred in 10% of the intervention group versus 24% of controls. 1595 participants were included (median age 28 years). 1724 participants were included (median age 50 years).
| Group | n | Events | Rate % |
|---|---|---|---|
| Intervention | 537 | 92 | 20.8 |
| Age < 65 | 232 | 69 | 14.0 |
| Intervention | 818 | 99 | 33.9 |
| Male | 337 | 74 | 20.3 |
| Age ≥ 65 | 149 | 38 | 27.9 |
Values are n (%) unless stated otherwise.

Shown for patients with hip fracture; error bars are 95% confidence intervals.

Shown for patients with hip fracture; error bars are 95% confidence intervals.
Serious adverse events were rare and similar between groups. 1050 participants were included (median age 81 years). acute kidney injury occurred in 9% of the intervention group versus 23% of controls. The adjusted odds ratio was 0.65 (95% CI 0.41 to 0.66).
The adjusted odds ratio was 0.77 (95% CI 0.48 to 0.60). acute kidney injury occurred in 19% of the intervention group versus 36% of controls. acute kidney injury occurred in 22% of the intervention group versus 19% of controls. Serious adverse events were rare and similar between groups.