Perioperative acute kidney injury in hip and knee arthroplasties. Incidence, risk factors, diagnostic methods and machine learning
Thesis event information
Date and time of the thesis defence
Place of the thesis defence
Oulu University Hospital, Auditorium 4.
Topic of the dissertation
Perioperative acute kidney injury in hip and knee arthroplasties. Incidence, risk factors, diagnostic methods and machine learning
Doctoral candidate
Licentiate of Medicine Okke Nikkinen
Faculty and unit
University of Oulu Graduate School, Faculty of Medicine, Research Unit of Translational Medicine
Subject of study
Faculty of Medicine, Research group of Anesthesiology.
Opponent
Docent Maija Kaukonen, Medbase
Custos
Docent Merja Vakkala, Oulu University Hospital, Pohde
Acute kidney injury in hip and knee arthroplasties. Incidence, risk factors, diagnostic methods and machine learning
Perioperative acute kidney injury is a common adverse condition. It is a multifactorial process and has multiple risk factors. Its incidence has been approximated to be between 0.3 and 15% in elective hip and knee arthroplasties and 8.4–24% in emergency arthroplasties. It increases mortality and affects the postoperative care. Chronic kidney disease is closely related to acute kidney injury, and they share similar risk factors. Traditionally, perioperative acute kidney injury has been defined by increased serum creatinine. The other important diagnostic criteria, decreased urine output, has been almost completely omitted in medical research due to difficulties assessing it.
Machine learning methods are increasingly being utilized in medical research. Acute kidney injury is one of the challenges that have been tried to be resolved by building predicting machine learning models. As a whole, the research is still in its early stages.
The aim of this doctoral thesis was to study perioperative acute kidney injury in hip and knee arthroplasties. The study focused on incidence, risk factors, outcomes and assessing the effect of using urine output as an acute kidney injury definition. The aim was also to develop a machine learning model predicting acute kidney injury.
Two retrospective cohort materials were collected and used. The first one included hip and knee arthroplasty patients operated in Oulu University Hospital and Oulaskangas Hospital in 2014. These patients were over 65 years of age. The second material had hip and knee arthroplasty patients operated in Oulu in 2016 and 2017.
The main outcomes of the study were that the risk factors of acute kidney injury are the same as previously studied: preoperatively decreased kidney function, age, diabetes, high body mass index, cardiovascular diseases and emergency arthroplasty. The risk factors of elective and emergency hip patients seem to differ. Using urine output as a diagnostic criterion significantly increases acute kidney injury incidence and isolated oliguria is associated with mortality. The main conclusion of the machine learning development was that the algorithms we used, RUSBoost and Naïve Bayes, worked well on our material with promising results, but at the same time, we were only able to build preliminary models. This is in line with the current early stage of machine learning development in medicine.
Machine learning methods are increasingly being utilized in medical research. Acute kidney injury is one of the challenges that have been tried to be resolved by building predicting machine learning models. As a whole, the research is still in its early stages.
The aim of this doctoral thesis was to study perioperative acute kidney injury in hip and knee arthroplasties. The study focused on incidence, risk factors, outcomes and assessing the effect of using urine output as an acute kidney injury definition. The aim was also to develop a machine learning model predicting acute kidney injury.
Two retrospective cohort materials were collected and used. The first one included hip and knee arthroplasty patients operated in Oulu University Hospital and Oulaskangas Hospital in 2014. These patients were over 65 years of age. The second material had hip and knee arthroplasty patients operated in Oulu in 2016 and 2017.
The main outcomes of the study were that the risk factors of acute kidney injury are the same as previously studied: preoperatively decreased kidney function, age, diabetes, high body mass index, cardiovascular diseases and emergency arthroplasty. The risk factors of elective and emergency hip patients seem to differ. Using urine output as a diagnostic criterion significantly increases acute kidney injury incidence and isolated oliguria is associated with mortality. The main conclusion of the machine learning development was that the algorithms we used, RUSBoost and Naïve Bayes, worked well on our material with promising results, but at the same time, we were only able to build preliminary models. This is in line with the current early stage of machine learning development in medicine.
Last updated: 8.9.2025