#5620 MAPPING OUT THE RESEARCH FIELD OF PROGNOSTIC MODELS IN NEPHROLOGY: A SCOPING REVIEW

Background and Aims Prognostic models can support individualized care provision and well-informed shared decision-making. Although there has been an upsurge of prognostic research in the field of nephrology, the uptake of prognostic models remains limited. A broad overview of the field is needed to see where we currently stand, and how to progress from here. With this overview we can identify current knowledge gaps, and potential implementation opportunities. Moreover, it will provide guidance as to where future research should be directed. Therefore, the aim of this study was to map the body of existing prognostic models within nephrology and detail the range in outcomes and populations that they cover, as well as their methodological rigour. To do so, we performed a scoping review of studies developing, validating or updating a prognostic model for patients with chronic kidney disease (CKD) or those receiving kidney replacement therapy (KRT). Method A framework for scoping reviews by Arksey and O'Malley was used and the PRISMA extension for Scoping Reviews was adhered to for transparent reporting. A systematic search in PubMed and Embase was performed to identify relevant studies that describe prediction models for patients with CKD or those on KRT. Studies were screened for their relevance based on prespecified criteria, and data was extracted on general characteristics of the included studies and their reporting and methodological quality. For the studies in which a prognostic model was developed, model presentation (e.g., full regression formula or risk score), whether validation strategies were employed, and model type (regression modelling vs machine learning), were extracted. Finally, we extracted data on the predicted outcome definitions, and which models were validated and/or updated most often. Descriptive statistics were used to summarise all findings. Results The systematic search yielded 3728 studies for screening, of which 596 were finally included. Of these, 29.5% concerned a CKD population, 31.4% a dialysis population, and 39.1% a kidney transplantation population. Many studies had a sample size of less than 500 participants (41.4%). Although a measure of discrimination of the model was usually presented (79.5%), a measure of calibration was presented in less than half of the studies (43.5%). Of the 411 studies in which a prognostic model was developed, most performed only internal validation (57.9%) or no validation at all (27.7%). Moreover, in almost half of the development studies (45.3%) no usable version of the model was reported, meaning that insufficient information was reported to apply the model in a new setting. For CKD populations, the majority of models predicted disease progression (n = 78), followed by models predicting mortality (n = 22) and cardiovascular events (n = 13). For dialysis populations, most models predicted mortality (n = 79), cardiovascular events (n = 20), and vascular access related outcomes (n = 15). Finally, models originally developed for kidney transplantation populations mainly predicted graft survival (n = 59), recipient survival (n = 39), and delayed graft function (n = 24). Models for non-traditional clinical outcomes, like health-related quality of life and symptom burden, were scarce. If validated or updated at all (n = 199), most models (n = 123) were externally validated and/or updated only once. The rest (n = 76) were validated and/or updated more, with a median (IQR) of 2 (2-3), 2 (2-3), and 3 (2-4) within the CKD, dialysis, and transplantation populations, respectively. Conclusion A substantial amount of nephrological prognostic research has been performed, but to minimize the gap between research and patient care additional steps have to be undertaken. Methodological rigour, external validation, updating, and impact assessment are of paramount importance. In addition, the current body of literature focusses on traditional clinical outcomes, and models for patient-reported outcomes are scarce. Opportunities to improve implementation of prognostic models in nephrological care are described in Box 1.