Comorbidity recording and predictive power of comorbidities in the Australia and New Zealand dialysis and transplant registry compared with administrative data: 2000–2010. Issue 11 (11th October 2016)

• Record Type: Journal Article
• Authors: Kotwal, Sradha; Webster, Angela C; Cass, Alan; Gallagher, Martin
• Is Part Of: Nephrology. Volume 21:Issue 11(2016)
• Page Start: 930
• Page End: 937
• Publication Date: 2016-10-11
• Subjects: comorbidity -- end‐stage renal failure -- epidemiology and outcomes -- medical record linkage -- risk adjustment; Nephrology -- Periodicals; Kidneys -- Diseases -- Periodicals; Nephrologists -- Periodicals; 616.61; 616.61
• DOI: 10.1111/nep.12694
• Languages: English
• ISSNs: 1320-5358

Abstract

Abstract: Aim: To compare comorbidity recording and predictive power of comorbidities for mortality between a clinical renal registry and a state‐based hospitalisation dataset. Methods: All patients that started renal replacement therapy (dialysis or transplant ‐ RRT) in New South Wales between 1/07/2001 and 31/7/2010 were identified using the Australia and New Zealand Dialysis and Transplant Registry (ANZDATA) and linked to the State Admitted Patient Data Collection (APDC) and the Death Registry. Comorbidities (diabetes mellitus, coronary artery disease (CAD), chronic lung disease, peripheral vascular disease and cerebrovascular disease) were identified at the start of RRT in both datasets and compared using kappa statistics (κ). Survival was calculated using cox proportional hazards models from the start of RRT to death date or end of study (31/07/2011). Four multivariable models were adjusted for age, gender and comorbidities to estimate the predictive power of the comorbidities as recorded in ANZDATA, APDC, either or both datasets Results: We identified 6285 people (23, 845 person‐years follow‐up). Diabetes recording had excellent agreement (94.5%, κ = 0.88), CAD had fair to good agreement (80. 6, κ = 0.56), with poor agreement between the two datasets for the other comorbidities. Deaths totalled 2594 (41.3%). Median follow up time was 3.3 years (IQR 1.7 to 5.4). All five comorbidities were powerful predictors of poor survival in all four models. All models had a similar predictive ability (Harrell's c = 0.71‐0.72). Conclusion: Variable agreement exists in comorbidity recording between the ANZDATA and APDC. The comorbidities have a similar predictive ability, irrespective of dataset of origin in an End Stage Kidney Disease (ESKD) population. Summary at a Glance: This study aimed to compare comorbidity recording and predictive power of comorbidity for mortality between a clinical renal registry and a state‐based hospitalisation dataset. The results showed that variability agreement between the APDC and ANZDATA in the coding of five clinically important comorbidities (DM, CAD, chronic lung disease, PVD and cerebrovascular disease). Despite these differences, the ability of these reported comorbidities to predict mortality in an ESKD population is similar, irrespective of the dataset of origin and whether used individually or in combination. This suggests that existing administrative data is likely to be a reliable tool for risk adjustment of mortality outcomes in research and health service planning analysis.