FC078: Impact of Removing Race from the CKD-EPI Equation: Analysis of 1.6 Million Swedish Adults

BACKGROUND AND AIMS

The Chronic Kidney Disease Epidemiology Collaboration (CKD-EPI) recently developed a novel creatinine-based eGFR equation without a race coefficient. While American nephrology societies recommend using this novel equation, its implications are unknown.

METHOD

We included 1.6 million adult individuals with routine outpatient serum creatinine testing during 2007–2018 in Stockholm, Sweden. First, we calculated reclassification across KDIGO eGFR categories when changing from the 2009 to 2021 CKD-EPI equation. Second, for both equations, the association between eGFR and (1) kidney failure with replacement therapy (KFRT), (2) all-cause mortality, (3) cardiovascular mortality and (4) major adverse cardiovascular events was estimated with Cox regression. Third, prognostic accuracy of both eGFR equations within the Kidney Failure Risk Equation was assessed with discrimination and calibration.

RESULTS

Compared with the 2009 equation, the 2021 equation yielded a higher eGFR by a median (IQR) of 3.9 (2.9–4.8) mL/min/1.73 m2, decreasing prevalence of CKD G3–G5 from 5.1 to 3.8%. The 2021 equation reclassified 9.9% of the total population and 36.2% of the CKD G3–G5 population to a less severe eGFR category. Individuals who were reclassified to less severe eGFR categories were older and therefore exhibited higher crude risks of all-cause/cardiovascular death and major adverse cardiovascular events, and lower risk of kidney replacement therapy compared with nonreclassified participants of similar eGFR. eGFR by both equations strongly predicted study outcomes, with similar discrimination and calibration for the Kidney Failure Risk Equation.

CONCLUSION

Implementing the 2021 CKD-EPI equation in predominantly white European populations raises eGFR by a modest amount (larger at older age and men) and shifts a major proportion of CKD patients to a higher eGFR category, with eGFR by both equations strongly predicting outcomes.

P0811CONFUSION BETWEEN ETIOLOGY AND PREDICTION IN CLINICAL OBSERVATIONAL STUDIES; HOW OFTEN DOES IT ARISE AND HOW CAN WE AVOID IT? A SCOPING REVIEW

Background and Aims

In etiological research the aim is to uncover the causal effect of a specific exposure on an outcome. The aim in prediction research is to predict an outcome with the best accuracy, irrespective of possible causality. Although in observational research both types of studies use similar statistical methods (generally multivariable modelling), the interpretation differs. For example, when researching the causal effect of BMI on kidney failure, it is important to correct for confounders such as age (i.e. factors that causally affect both BMI and kidney failure). Judgement of confounders must be made on pre-existing knowledge. If we select confounders based on the data, we might also correct for hypertension and atherosclerosis (as they are associated with BMI and kidney failure). As this correction is in the causal pathway we would erroneously conclude that BMI does not affect the risk of kidney failure. Vice versa, a mortality prediction model for dialysis patients may include antihypertensive medication use as a predictor. It would be wrong to conclude that patients should discontinue this medication to improve prognosis; the medication use is a marker for a certain health status and we cannot interpret it in a causal manner. Similar to these examples, we’ve found that characteristics from etiology and prediction are often confused, leaving us with studies that may be misinterpreted. The aim of the current study is to quantify the amount of confusion between etiology and prediction in clinical observational studies and identify common mistakes that lead to this confusion.

Method

Studies published in January 2018 in the top journals of four distinct medical fields: General & Internal medicine, Surgery, Cardiology and Nephrology were screened for inclusion. Original research studies on observational cohorts of humans were included. A list of key study characteristics of etiological and prediction research was developed by CLR and MvD in an iterative fashion. From these characteristics signaling questions for confusion were developed to score included studies.

Results

The developed key study characteristics of etiology and prediction are shown in the table. In total, 286 studies were screened of which 123 were included. Overall, 27% (n=33) of included articles contained some form of confusion between etiology and prediction. In the figure the journal impact factor is mapped against proportion of confusion per included journal. We can see a trend that as impact factors increase the amount of confusion decreases. In etiological studies, the most frequent (n=15) form of confusion was adjustment for variables based on predictive performance in the data, instead of known causal structure. The majority selected ‘confounders’ purely based on p-values and therefore potentially adjusted for intermediate variables, resulting in incorrect effect estimates. Another mistake in etiological studies (n=5) was the reporting of predictive performance measures such as the C-statistic for an etiological model. In prediction studies the most confusion occurred in the discussion (n=14). Seven studies interpreted predictors causally, for example by concluding that these predictors should be modified in order to improve patient outcomes. As the effect estimates of these studies do not account for confounding, this interpretation is invalid. Lastly, seven studies mention residual confounding as a limitation, which is only a problem in etiological research.

Conclusion

Confusion between etiology and prediction is a wide-spread methodological flaw in medical observational studies, particularly those published in lower impact clinical journals. As confusion may lead to erroneous conclusions, the distinction between causal and predictive research deserves more attention in medical research and scientific education.

MO074RENAL FUNCTION DECLINE IN OLDER MEN AND WOMEN WITH ADVANCED CKD - RESULTS FROM THE EQUAL STUDY

Background and Aims

Understanding the mechanisms underlying the differences in renal decline between men and women may improve sex-specific clinical monitoring and management. To this end, we aimed to compare the slope of renal function decline in older men and women in CKD 4-5, taking into account informative censoring related to the sex-specific risks of mortality and dialysis initiation.

Method

The EQUAL study is an observational prospective cohort study in stage 4-5 CKD patients ≥65 years not on dialysis. Data on clinical and demographic patient characteristics were collected between April 2012 to December 2018. eGFR was calculated using the CKD-EPI equation. Linear mixed models were used to model the eGFR trajectory by sex, and joint models were applied to deal with informative censoring.

Results

We included 7801 eGFR measurements in 1682 patients over 2911 years of follow-up. Renal function declined 14.0% (95% CI 12.9%-15.1%) on average each year. Renal function declined faster in men (16.2% per year, 95% CI 15.9%-17.1%) compared with women (9.6% per year, 95% CI 6.3%-12.1%), which remained largely unchanged after accounting for various mediators, and for informative censoring due to mortality and dialysis initiation. We identified effect modification by diabetes, with faster declines in renal function found especially in women with diabetes.

Conclusion

In conclusion, renal function declines faster in men compared with women, which remained similar after adjustment for mediators, and despite a higher risk of informative censoring in men. We demonstrate a disproportional negative impact of diabetes specifically in women.

P1618DEVELOPMENT AND EXTERNAL VALIDATION OF A PREDICTION MODEL FOR ADVERSE OUTCOME FOLLOWING KIDNEY TRANSPLANTATION FROM OLDER DECEASED DONORS

Background and Aims

With rising demand for kidney transplantation and the kidney donor pool lagging behind, the acceptance criteria for donor kidneys are expanding. Hence, reliable pre-transplant assessment of organ quality has become a top priority. Estimating the risk of adverse outcomes at the time of kidney allocation is challenging and particularly relevant for recipients of kidneys from older donors. The existing kidney donor risk index (KDRI) has been criticized for heavily depending on donor age. Therefore, the aim of the current study was to develop and validate a prediction model for adverse outcome after kidney transplantation from deceased donors aged 50 years or older and compare this model’s performance to the KDRI.

Method

We utilized the Dutch kidney transplant registry (NOTR) and identified patients who received a kidney from a deceased donor aged 50 years or older between 2006 and 2019. These recipients were included for model development and temporal validation. The prediction model was externally validated on the United States organ transplantation registry (OPTN), in which we selected patients that were transplanted between 2006 and 2017. Potential pre-transplant predictors were selected by an expert panel of nephrologists and surgeons. The predicted adverse outcome was defined as a composite of graft failure, recipient mortality or CKD stage 4/5 within 1 year of transplantation. A logistic regression model was developed, internally validated and shrunk for optimism through bootstrapping. Missing data were multiply imputed in 10-fold, non-linear continuous predictors were modelled with restricted cubic splines and clinically relevant interaction terms were included. The KDRI was validated on the same NOTR and OPTN cohorts for graft survival within 1 year. The developed model and the KDRI were recalibrated to the baseline risk of outcome in external validation. Model performance was assessed by discrimination and calibration.

Results

The model was developed on 2510 patients of whom 823 experienced an adverse outcome within the first year. The temporal validation cohort contained 837 patients of whom 230 had an adverse outcome and the US external validation cohort consisted of 31987 patients with 6758 adverse outcomes. Selected donor predictors were: age, gender, BMI, cause of death, CPR, inotropes use, serum creatinine, hypertension, hypotension, diabetes, smoking, left/right kidney, warm ischemic time, cold ischemic time and proteinuria. Recipient predictors were: age, gender, BMI, diabetes, cardiovascular comorbidity, primary kidney disease, dialysis duration, number of previous kidney transplantations, HLA mismatches and PRA. Discrimination of the adverse outcome model was moderate, yet considerably better than discrimination of the KDRI (see table). The adverse outcome model’s calibration and distribution of predicted risks were good in both the NOTR and OPTN (see figure).

Conclusion

A prediction model was developed and extensively validated for adverse outcome after kidney transplantation from older deceased donors. Despite the use of advanced and robust methodology, its discriminatory capacity was limited. However, the adverse outcome model showed good calibration and performed considerably better than the KDRI in this population of suboptimal donors. This model could potentially assist nephrologists in deciding whether to accept or decline a specific kidney from an older deceased donor.

Performance of bleeding risk scores in dialysis patients

BACKGROUND

Bleeding risk scores have been created to identify patients with an increased bleeding risk, which could also be useful in dialysis patients. However, the predictive performances of these bleeding risk scores in dialysis patients are unknown. Therefore, the aim of this study was to validate existing bleeding risk scores in dialysis patients.

METHODS

A cohort of 1745 incident dialysis patients was prospectively followed for 3 years during which bleeding events were registered. We evaluated the discriminative performance of the Hypertension, Abnormal kidney and liver function, Stroke, Bleeding, Labile INR, Elderly and Drugs or alcohol (HASBLED), the AnTicoagulation and Risk factors In Atrial fibrillation (ATRIA), the Hepatic or kidney disease, Ethanol abuse, Malignancy, Older age, Reduced platelet count or Reduced platelet function, Hypertension, Anaemia, Genetic factors, Excessive fall risk and Stroke (HEMORR2HAGES) and the Outcomes Registry for Better Informed Treatment (ORBIT) bleeding risk scores by calculating C-statistics with 95% confidence intervals (CI). In addition, calibration was evaluated by comparing predicted and observed risks.

RESULTS

Of the 1745 dialysis patients, 183 patients had a bleeding event, corresponding to an incidence rate of 5.23/100 person-years. The HASBLED [C-statistic of 0.58 (95% CI 0.54-0.62)], ATRIA [C-statistic of 0.55 (95% CI 0.51-0.60)], HEMORR2HAGES [C-statistic of 0.56 (95% CI 0.52-0.61)] and ORBIT [C-statistic of 0.56 (95% CI 0.52-0.61)] risk scores had poor discriminative performances in dialysis patients. Furthermore, the calibration analyses showed that patients with a low risk of bleeding according to the HASBLED, ATRIA, HEMORR2HAGES and ORBIT bleeding risk scores had higher incidence rates for bleeding in our cohort than predicted.

CONCLUSIONS

The HASBLED, ATRIA, HEMORR2HAGES and ORBIT bleeding risk scores had poor predictive abilities in dialysis patients. Therefore, these bleeding risk scores may not be useful in this population.