#3434 A FREE OPEN-SOURCE FRAMEWORK TO INFER PATIENT-LEVEL EGFR SLOPES FOR ASSOCIATION WITH MOLECULAR DATA

Abstract

Abstract Background and Aims Unfolding over years or decades, patient kidney disease trajectories can be characterised using routinely collected longitudinal data. One frequently measured marker of kidney function is serum creatinine, used in clinical settings to estimate the glomerular filtration rate (eGFR). Change in eGFR over time has been put forward as a proxy endpoint for clinical trials, and most work on inferring eGFR slopes has been done in this context. However, 'omics-based studies of the molecular mechanisms of kidney disease have so far largely neglected to make use of longitudinal data linked to tissue samples. Here, we build on existing approaches [1, 2] and provide an open-source, R framework to infer eGFR slopes from routinely available clinical data linked to biobank samples. Method In our framework, we calculate eGFR from serum creatinine measurements using the CKD-EPI 2021 formula. To aid data adjudication, we implement criteria outlined by Hapca et al. [1] for automatic flagging of acute kidney injury (AKI) events, and extend the criteria to detect potential in-patient stays. Since the biopsy procedure may arise as a result of loss of kidney function and/or coincide with a change in treatment regime, we use a two-spline linear mixed effects model similar to that proposed by Vonesh et al. [2] to differentiate between immediate-post-biopsy (acute) and chronic eGFR slopes. Additionally, to account for non-random dropout due to the onset of dialysis or occurrence of kidney transplant, we jointly model the longitudinal and dropout processes. Our framework is made available as free, open-source software in R. Results From routinely collected longitudinal clinical data, we generate patient-specific values characterising the post-biopsy disease trajectory. These values can then be incorporated into the analysis of 'omics data to find molecular patterns associated with treatment response or prognosis. Conclusion We have developed an open-source R framework for inferring patient-level eGFR slopes from serum creatinine measurements linked to biobank samples. Leveraging this data could improve 'omics studies of the molecular mechanisms of kidney disease, allowing us to incorporate information about disease trajectory and treatment response and thus ask questions of the molecular data which have real clinical relevance.