K TAN 1,2,3, S MCDONALD 4,5,6, J ZHANG 1,2, Z WANG 1,2, W HOY 1,2
1NHMRC CKD CRE and CKD.QLD, Brisbane, Australia, 2Faculty of medicine, University of Queensland, Brisbane, Australia, 3Renal unit, Logan hospital & Metro South health service, Brisbane, Australia, 4Faculty of medicine, Adelaide University, Adelaide, Australia, 5ANZDATA, Adelaide, Australia, 6Renal unit, CNAHS, Adelaide, Australia
BACKGROUND: The ability to predict future kidney failure (KF) even in the absence of data on eGFR or proteinuria might allow machine learning to prompt clinicians to intervene and avoid “missed opportunities” to prevent further CKD progression.
AIMS: Determine if hospital admissions with a kidney disease related discharge diagnosis predict KF independent of eGFR and proteinuria.
METHODS: CKD,QLD registry patients with DM were studied (none on kidney replacement therapy at enrolment). Primary discharge diagnoses (ICD-10 v 2016) for all hospital admissions between 23/3/2011 and 30/6/2016 inclusive were determined from Queensland Hospital Admission patient data collection. Follow-up was censored by death, KF and relocation interstate/overseas. Logistic regression was performed firstly to determine if different classes of primary discharge diagnosis (≥1 admission during follow-up versus no admissions) predicted future KF and whether this effect was independent of eGFR and proteinuria,
RESULTS: 2355 patients experienced 17145 hospital admissions over 10873 patient-years follow-up. In a multivariable model incorporating age, gender, ethnicity, GFR and proteinuria, admissions with a renal primary diagnosis independently predicted future KF with an odds ratio 3.35 (95% CI 2.41-4.68, p<0.0001). In contrast, admissions for cardiovascular disease, infection, diabetes control, respiratory disease and gastro-intestinal disease did not predict future KF.
CONCLUSIONS: In patients with DM and CKD receiving kidney specialist care, hospital admissions for renal disease predicted future KF independent of eGFR and proteinuria. Admissions for other discharge diagnoses did not have this effect. If these findings also occur in other patients with CKD (or indeed without known prior CKD), then there is potential for machine learning to predict future KF even in the absence of eGFR and proteinuria.
Nephrologist and Clinical pharmacologist. Current PhD Candidate, University of Queensland,