Prof. RATHIKA KRISHNASAMY1,2,3, MEG JARDINE2, PETA-ANNE PAUL-BRENT3, HAI PHAM3, STEPHEN MCDONALD4, BRENDAN SMYTH2, YEOUNGJEE CHO3, DAVID JOHNSON3, JASON POLE3
1Sunshine Coast University Hospital, , Australia, 2University of Sydney, , Australia, 3The University of Queensland, , Australia, 4ANZDATA Registry, , Australia
Biography:
Prof Krishnasamy is the Director of Nephrology at the Sunshine Coast University Hospital, Professor for the Faculty of Medicine at the Griffith University and Co-Deputy Chair of Australasian Kidney Trials Network. Her research is in the field of pathways to reduce cardiovascular disease in CKD with expertise in trial operations and platform trials.
Aim:
Assess the feasibility and accuracy of identifying clinical events using hospital administrative data compared to manually-collected data for sites participating in the RESOLVE study.
Background:
Onerous data fields and duplicate methods of data collection impose a burden on research activities. Linkage of information held in routinely collected administrative data could be a simple and effective solution.
Methods:
Study data collected through ANZDATA registry for participants enrolled in the RESOLVE study in 13 Queensland haemodialysis units (N=1487) between 2018 to 2022 were linked to state-wide hospital administrative data and death registry. Feasibility was assessed using proportion of records linked to study outcomes [death, hospitalised myocardial infarction (MI), stroke and heart failure (HF)]. Accuracy of administrative data against RESOLVE data were assessed using sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV) and their 95% confidence intervals (95% CI).
Results:
1,402 (94.2%) participants were linked to administrative data. The RESOLVE data identified a total of 415 deaths, 56 MIs, 14 strokes and 3 HF events. Administrative data matched 412 deaths (99.3%), 51 MIs (91.1%), 9 strokes (64.3%) and 2 HF events (66.7%) of the reported data. Administrative data also identified additional 74 MIs, 63 strokes and 223 HF events. Accuracies were highest for mortality [sensitivity 0.99 (0.98-1.00), specificity 1.00 (0.99-1.00), PPV 1.00 (0.99-1.00), NPV 1.00 (0.99-1.00)] and MI [sensitivity 0.91 (0.82-0.98), specificity 0.95 (0.94-0.96), PPV 0.41 (0.36-0.47), NPV 0.99 (0.99-1.00)]. Stroke and HF had lower accuracies with PPVs of 0.12 (0.07, 0.17) and 0.01 (0.00, 0.02), due to the extra events identified in administrative data.
Conclusion:
This project demonstrated agreement between administrative data and manually-collected outcomes, supporting the use of linked datasets for clinical research.
Presentation Slides PDF – Click Here