Skip to main content

Sop Acute Kidney Injury (SAKI): Predictive models in the management of acute kidney injury

Bedford, Michael (2016) Sop Acute Kidney Injury (SAKI): Predictive models in the management of acute kidney injury. Doctor of Philosophy (PhD) thesis, University of Kent,. (KAR id:62611)

Language: English

Click to download this file (15MB) Preview
[thumbnail of 227Thesis Submission - Michael Bedford - 11905027 - Final.pdf]
This file may not be suitable for users of assistive technology.
Request an accessible format


The overarching aim of this PhD thesis is to develop methods, which will ultimately improve the management of patients with acute kidney injury (AKI). Over the years one inherent problem in both diagnosing AKI clinically and reviewing and comparing studies published in the literature has been the numerous definitions used to define AKI. 87 With now accepted definitions of AKI, the first question raised was to determine the true impact of AKI, in terms of incidence and outcomes, for both the patient (morbidity and mortality) and the healthcare economy.

A retrospective observational database study was performed from secondary care in East Kent (adult catchment population of 582,300). All adult patients (18 years or over) admitted between 1st February 2009 and 31st July 2009, were included. Patients receiving chronic renal replacement therapy (RRT), maternity and day case admissions were excluded. AKI was defined by the acute kidney injury network (AKIN) criteria. A time dependent risk analysis with logistic regression and Cox regression was used for the analysis of in-hospital mortality and survival.

The incidence of AKI in the 6 month period was 15,325 pmp/yr (adults) (69% AKIN1, 18% AKIN2 and 13% AKIN3). In-hospital mortality, length of stay and ITU utilisation all increased with severity of AKI. Patients with AKI had an increase in care on discharge and an increase in hospital readmission within 30 days.  In comparison with patients with no AKI those with AKI stage 1 had a 52% longer length of stay (LOS) in hospital, a 2.8-fold increased risk of admission to the intensive therapy unit (ITU), a 39% longer ITU stay (in those who went to ITU), and a 2.4-fold greater in-hospital mortality. Furthermore, patients with AKI stage 1 had twice the long-term risk of death, a 33% higher likelihood of an increase in care, and a 42% higher risk of re-admission within 30 days. In those patients with AKI stage 3 (the subject of the NCEPOD report) 100 hospital LOS doubled, there was a 22 times higher risk of admission to ITU and ITU LOS was also doubled, consistent with national data from the Intensive Care National Audit and Research Centre.

A further study using this data in collaboration with Marion Kerr (health economist) at the Department of Health, suggested the annual number of excess inpatient deaths, with AKI in England may be greater than 40,000, 106 and the annual cost of AKI-related inpatient care in England is estimated at £1.02 billion. 106

With the problem now evident and clearly defined, the first stage in improving management was to alert clinicians to the presence of AKI as soon as possible to allow early recognition and intervention. Here the development of a static AKI alert report delivered to the critical care outreach team and specialist renal team is documented.

A qualitative analysis was then used to explore the effect of professional interactions, information sharing, and personal and professional characteristics on the use of electronic clinical information and clinical decision support. Key areas highlighted in the qualitative analysis included real-time delivery of AKI alerts, clear responsibility of care to be with the clinical teams with advice from the critical care outreach nurses and renal consultants as required, and improved communication with the clinical teams looking after the patients. This work informed a development partnership with a commercial company (Careflow Connect Limited) to deliver real-time alerting of acute kidney injury to clinicians at the point of care and allow collaboration within the clinical team and also with the specialist renal and critical care outreach teams.

However, in any disease process, while we can optimise our measures in place (as above) to alert to the presence of a disease (in this case acute kidney injury (AKI)) and manage it effectively and efficiently at recognition, the ultimate form of treatment is the prevention of the disease occurring in the first place. Hence, in order to achieve this we need to determine the patient at risk.

Firstly, potential risk factors were explored. Three time points were also defined where significant clinical decision making takes place and at which points the use of risk models would have greatest impact on clinical care and patient management. These were the point of admission to hospital to guide renal function testing and inform admission planning, and secondly, at 24 hours after admission, often on the post-take ward round to highlight patients who are likely to develop new or worsening AKI if already present, in the first 72 hours of hospital admission so that appropriate management decisions can be made on the ward round.

The study population included hospital admissions to the three acute hospitals of East Kent Hospitals University NHS Foundation Trust (EKHUFT) in 2011, excluding maternity and elective admissions. For validation in a second population the study included hospital admissions to Medway NHS Foundation Trust.

The study developed and assessed traditional methods to provide risk models for the prediction of new or worsening AKI in patients presenting to hospital and in their management within the first 24 hours of admission. Ordinal logistic regression with uni-variable analyses were used to inform the development of multi-variable analyses. Backward selection was used to retain only statistically significant variables in the final models. The models were validated using actual and predicted probabilities, Area Under the Receiver Operating Characteristic (AUROC) curve analysis and the Hosmer Lemeshow test.

The analysis identified key variables which predict AKI both at admission and 72 hours post admission. Validation demonstrated area under ROC of 0.75 and 0.68 respectively. Predicting worsening AKI during admission was unsuccessful.

These models were also re-defined with use of the NHS England algorithm to define AKI which produced similar results with area under ROC of 0.73 and 0.67 respectively.

The work reported here has demonstrated the significant morbidity and mortality both long and short term of patients who experience acute kidney injury managed in hospital and has developed methods of alerting the presence of AKI to the point of care in real-time to ensure efficient intervention with an aim to improve these outcomes. Qualitative work has also highlighted the complexity regarding the implementation and delivery of alerting systems to the clinical front line. The work reported in this thesis has also demonstrated that routinely available data can be used to highlight patients at risk of acute kidney injury both at the point of admission to hospital and following admission.

Item Type: Thesis (Doctor of Philosophy (PhD))
Thesis advisor: Farmer, Chris
Thesis advisor: Coulton, Simon
Uncontrolled keywords: Acute Kidney Injury, Risk, Prediction, Intervention, Alert, Incidence, Outcomes, Communication, Point of Care, Real-Time, Socialisation.
Divisions: Divisions > Division for the Study of Law, Society and Social Justice > School of Social Policy, Sociology and Social Research
SWORD Depositor: System Moodle
Depositing User: System Moodle
Date Deposited: 11 Aug 2017 09:12 UTC
Last Modified: 11 Dec 2022 14:28 UTC
Resource URI: (The current URI for this page, for reference purposes)
  • Depositors only (login required):

Total unique views for this document in KAR since July 2020. For more details click on the image.