This validation study compared patient identified by codes for kidney replacement therapy (KRT) in primary and secondary care data to cohorts from the gold standard UK Renal Registry data. The study was conducted using OpenSAFELY, a secure health-analytics platform linking multiple datasets which covers 40% of the English population.
Both prevalent (receiving KRT on 1st January 2020) and incident (receiving KRT for the first time in 2020) cohorts were examined. Accuracy was measured using sensitivity (“are routine data good at picking up true KRT patients?”) and positive predictive value (“are people identified by routine data likely to be true KRT patients?”).
Less than one in ten prevalent KRT patients in the UK Renal Registry were missed by routine health data (in both primary and secondary care), however this increased to over 1 in 3 for incident KRT patients. Around 1 in 5 people were falsely identified as a prevalent KRT patient by primary care codes, increasing to 1 in 3 for secondary care codes. In particular, around half of the dialysis patients identified in routine data were not chronic dialysis patients in the UK Renal Registry. Sensitivity of the primary care codes was higher in younger patients and those from less deprived areas, while the converse was true for secondary care. Half of the incident dialysis patients had no KRT code in the primary care data within three months of starting KRT.
This study shows that routine electronic healthcare records alone cannot accurately capture chronic KRT patients, and highlights the value of linkage with UK Renal Registry data for researchers using these data. Patients starting KRT who become eligible for targeted interventions, such as vaccination or health policy responses might not be identified promptly in routine data, causing delays in treatment. This work demonstrates the extent of acute kidney care that is performed (and not reported in registries of chronic kidney failure), which is particularly relevant for settings where financing of kidney services is driven only by chronic need.
The analysis was restricted to people in the OpenSAFELY database. London has high rates of KRT but is under-represented in the OpenSAFELY database. In routine data, KRT was defined as the presence of one of a list of codes to reflect the approach taken in previous studies. Higher accuracy could perhaps be achieved if more complex definitions (e.g. combinations of codes) were applied.
Other studies examining routine data for people chronic kidney disease found an opposite pattern of low sensitivity and high positive predictive value, supporting the need for further studies to look specifically at kidney replacement therapy. The large numbers of acute dialysis, which is currently not part of any statutory reporting in the UK, and the associated healthcare burden, also warrants further investigation.
Routine electronic health data are a valuable data source but may not be suitable for identifying chronic dialysis and transplant patients, whether for research purposes or for care planning. Linkage of routine data with gold-standard registries gives the best of both worlds: well-defined cohorts combined with the richness of their wider health record. Platforms such as OpenSAFELY demonstrate that it is possible to bring a variety of data sources together efficiently and securely, with huge potential for research.
