| Title |
Predicting mortality and short-term outcomes of continuous kidney replacement therapies in neonates and infants |
| Authors |
Deja, Anna ; Deja, Kamil ; Cappoli, Andrea ; Labbadia, Raffaella ; Baeta Baptista, Rute ; Arslan, Zainab ; Oh, Jun ; Bayazit, Aysun Karabay ; Yildizdas, Dincer ; Schmitt, Claus Peter ; Tkaczyk, Marcin ; Cvetkovic, Mirjana ; Kostic, Mirjana ; Jankauskienė, Augustina ; Viršilas, Ernestas ; Longo, Germana ; Vidal, Enrico ; Mir, Sevgi ; Bulut, Ipek Kaplan ; Pasini, Andrea ; Paglialonga, Fabio ; Montini, Giovanni ; Yilmaz, Ebru ; Correia-Costa, Liane ; Teixeira, Ana ; Schaefer, Franz ; Guzzo, Isabella |
| DOI |
10.1093/ndt/gfaf173 |
| Full Text |
|
| Is Part of |
Nephrology dialysis transplantation.. Oxford : Oxford University Press (OUP). 2025, Early Access, p. [1-11].. ISSN 0931-0509. eISSN 1460-2385 |
| Keywords [eng] |
acute kidney injury ; artificial intelligence ; continuous kidney replacement therapies ; neonates and infants ; outcomes |
| Abstract [eng] |
Background Continuous kidney replacement therapy (CKRT) has emerged as a valuable treatment option in critically ill neonates and infants with acute kidney injury (AKI) requiring dialysis. In this population, we apply artificial intelligence (AI) to identify factors influencing mortality and short-term adverse kidney outcomes. Methods The study involved neonates and infants included in the EurAKId Registry (NCT 02960867), who underwent CKRT treatment. Using the AI XGBoost models, we identified key clinical factors associated with short-term outcomes: mortality before hospital discharge, as well as proteinuria at discharge. We considered the patients’ clinical characteristics, anthropometric features, and CKRT technical settings. Results The study comprised 95 patients: 31.6% neonates and 68.4% infants with a median age at hospital admission of 1 month (interquartile range, IQR 0–7 months). Ten children were born prematurely. The overall mortality rate was 47.3% and did not differ significantly between neonates and infants (53.3% vs 44.4%, respectively, P = .422). The XGBoost model for predicting mortality had the accuracy of 59.53% ± 0.96% and AUC of 0.64 ± 0.11. Lower urine output at CKRT initiation, a greater rise in serum creatinine (SCr), longer time to dialysis initiation, and lower blood pressure were associated with increased risk of mortality. Proteinuria at hospital discharge was present in 30.6% of survivors. The XGBoost model for predicting proteinuria had the accuracy of 79.11% ± 2.46% and AUC (0.74 ± 0.04). Higher SCr concentrations at hospital admission and at CKRT start, as well as primary kidney disease were the most important risk factors for proteinuria. Conclusion We propose the XGBoost models for identifying factors associated with short-term outcomes of CKRT in neonates and infants. Lower urine output at CKRT start, more severe AKI progression and longer time to CKRT initiation might be important risk factors for mortality in infants and neonates. Primary kidney disease and related biochemical parameters are strong predictors of proteinuria at hospital discharge. |
| Published |
Oxford : Oxford University Press (OUP) |
| Type |
Journal article |
| Language |
English |
| Publication date |
2025 |
| CC license |
|
