Use this URL to cite or link to this record in EThOS:
Title: Prediction modelling in acute hospital care : a case study of acute kidney injury
Author: Hodgson, Luke Eliot
ISNI:       0000 0004 7972 1192
Awarding Body: University of Southampton
Current Institution: University of Southampton
Date of Award: 2018
Availability of Full Text:
Access from EThOS:
Full text unavailable from EThOS. Please try the link below.
Access from Institution:
Background: acute kidney injury (AKI) is a global healthcare priority, with strategies to improve associated outcomes desirable. Prediction models or clinical prediction rules (CPRs), utilise multiple predictors to provide objective estimates of future risk and their use has been suggested in the field of AKI. A rapid expansion in derived CPRs has rarely been followed by external validation, with even fewer models undergoing impact analysis on patient outcomes. Using e-alert systems to identify AKI has become commonplace in the UK, though few studies have assessed their efficacy. Combining an inhospital e-alert system with an AKI CPR has not previously been described. Methods: a systematic review of CPRs to predict hospital-acquired AKI (HA-AKI) in acute hospital settings was performed, followed by an external validation of one - the AKI prediction score (APS). Thirdly an impact analysis study using a controlled, before-and-after design on acute medical admissions to two adult non-specialist hospital sites was conducted (2014-16). At admission, the CPR highlighted patients at risk of HA-AKI in conjunction with an e-alert which triggered care bundles of interventions. Primary outcome was incident HA-AKI using a difference-in-differences analysis. Secondary outcomes in those developing HA-AKI included: in-hospital mortality, AKI progression, intensive care unit (ICU) escalation and effects on process measures. Patients with established community-acquired AKI (CA-AKI) were also highlighted only at the intervention site. Results: I. A systematic review found 53 CPR studies, the majority in specialised areas; of the 11 in general hospital settings five had external validation. Significant shortcomings in design and reporting were found. II. External validation of the APS model found modest discrimination with area under the receiver operating characteristic curves (AUROCs) ranging 0.65-0.71 and acceptable calibration plots. III. An impact analysis combining the CPR with an AKI e-alert found a reduction in HA-AKI at the intervention site (odds ratio, OR 0.990 (0.981-1.000), P=0.049). Of cases who developed HA-AKI mortality significantly reduced on unadjusted (27.46% pre vs 21.67% post intervention, OR 0.731 (0.560-0.954) P=0.021) and difference-in-differences analysis (OR0.924 [95% CI 0.858-0.996] P=0.038). Process measures significantly improved at the intervention site. In contrast no outcome improvements were found in patients presenting with CA-AKI following introduction of an AKI e-alert. Conclusions: Few validated AKI CPRs have been described in general hospital settings. An external validation of one model, the APS, led on to an impact analysis of this CPR. This innovative study utilised IT to integrate the CPR and an AKI e-alert in an acute hospital setting. The major findings were a reduction in de novo hospital-acquired AKI and reduced mortality in those who developed AKI. Future research should assess if these findings are generalisable and sustainable. Integration with primary care IT and the employment of biomarkers to apply precision care are further avenues following on from this case study.
Supervisor: Roderick, Paul ; Dimitrov, Borislav Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID:  DOI: Not available