Rescaling creatinine makes GFR estimation equations generally applicable across populations – validation results for the Lund-Malmö equation in a French cohort of sub-Saharan ancestry

Introduction

Glomerular filtration rate estimating equations (eGFR) as index test for renal function generally perform worse when validated outside the development population. This may be due to differences in population characteristics including morbidity, creatinine generation [1], methods to measure GFR (mGFR) used as reference test (e.g., renal clearance of the ionic X-ray contrast medium iothalamate vs. plasma clearance of the non-ionic X-ray contrast medium iohexol) [2, 3] and in biomarker assays including differences in standardization [4].

Pottel et al. have introduced the principle of rescaled (normalized) serum creatinine to compensate for population differences in creatinine generation [5], mainly differences in muscle mass and dietary habits concerning e.g. meat [6]. Rescaled serum creatinine (sCr/Q) is a dimensionless ratio where sCr represents the actual serum creatinine level of the individual and Q the expected creatinine level of this individual if in a healthy state. Thus, the Q-value is a surrogate for the creatinine generation determinants in a given population. The European Kidney Function Consortium (EKFC) equation uses this rescaling principle that implies that the same basic expression of the equation can be used for both children and adults, females and males [7] as well as in different ethnic groups [8]. From this rescaling principle, it follows that any GFR estimating equation can be generalized to populations with creatinine generation different from the one where it was developed. Previous work has for example shown how GFR estimating equations developed for adults can be applied in children with satisfactory accuracy by converting their creatinine values to corresponding adult levels [9, 10].

The rescaling principle also implies that any GFR equation can be re-expressed mathematically so its estimation is based on rescaled serum creatinine (sCr/Q). In the current study, we establish the re-expressed Lund-Malmö Revised (r-LMR) equation, an equation that was originally developed in a White European (Swedish) population [11], for use with rescaled serum creatinine. As a proof of concept, we show how the rescaling with population-specific Q-values makes the LMR equation applicable as index test of renal function with satisfactory accuracy in a French Black population of sub-Saharan ancestry. Comparisons are made with the EKFC equation using the same Q-values.

Materials and methods

The present study cohort included 964 Black patients (henceforth designated Black Europeans) with self-reported partial or total ancestry from sub-Saharan Africa (of which about 1/3 were of Caribbean origin), who were referred for GFR measurements between October 2006 and October 2020 at Assistance Publique-Hôpitaux de Paris, Bichat Hospital (Table 1). The cohort included 28 % patients with diabetes mellitus, 7 % with human immunodeficiency virus, 15 % potential renal donors and 46 % renal transplant recipients. Serum concentration of creatinine, weight, height, age and gender were recorded at the time of the GFR examination and used to estimate GFR in cross-sectional analyses.

All procedures involving subjects and data agreed with the ethical principles for medical research involving human subjects established in the World Medical Association Declaration of Helsinki. The study was reviewed and originally approved by the Regional Ethical Board in Lund, Sweden (reg no. 2018/220) with amendments subsequently reviewed and approved by the Swedish Ethical Review Authority (reg no. 2021–04177) and by the Institutional Review Board of Assistance-Publique Hôpitaux de Paris and Paris 7 University, France (IRB 00006477, study 14-051). All patients gave their written consent for scientific use of anonymous data. Relevant items of the 2015 Standards for Reporting of Diagnostic Accuracy checklist (STARD) were considered when preparing this report.

Results

In the current validation cohort, bias moved closer to zero, precision remained unchanged and P₃₀ accuracy increased when appropriate population-specific Q-values were used. The re-expressed LMR equation used with the Q-values for Black Europeans (r-LMR_Q-Black) rather than Q-values for White Europeans (r-LMR_Q-White) led to performance improvements both overall and when stratified by mGFR (Tables 2 and 3). P₃₀ increased overall by +10.4 percentage points (pp, 95 % CI 7.8–13.0 pp, p<0.001) and reached the same level as EKFC_Q-Black. The improvement in accuracy by using population-specific Q-values for Black Europeans was less marked for EKFC with an increase in P₃₀ by only +3.0 pp (95 % CI 0.7–5.3 pp, p=0.01).

The improved performance of r-LMR_Q-Black compared with r-LMR_Q-White was more pronounced among males than among females (Table 4) and was consistently noted when results were stratified by rescaled serum creatinine (sCr/Q; Table S1). The improvement in performance for r-LMR_Q-Black across levels of rescaled serum creatinine was less consistent among females (Table S2). The improvement of EKFC_Q-Black was confined to males only (Table 4).

Improvement in P₃₀ was noted in all age groups (18–39, 40–64 and ≥65 years) for both r-LMR_Q-Black and EKFC_Q-Black (Table 5). When stratifying for BMI (<20, 20–30 and ≥30 kg/m²) bias increased and P₃₀ decreased at BMI<20 kg/m² for both equations based on Q-values for blacks while bias decreased with improved P₃₀ at BMI≥20 kg/m² (Table 6).

Discussion

We showed in the present study how the generalizability of GFR estimation equations can be improved by re-expressing the equation formulation using the biomarker rescaling principle and apply population-specific Q-values. We used the creatine-based LMR-equation as an example, but similar rescaling should be possible to apply also for other GFR estimating equations and other biomarkers. Our validation results suggest that rescaling may substantially improve estimation performance when a GFR equation developed in one population is applied to another population with anticipated different creatinine generation. Recently it was also shown how the rescaling principle can be used to generalize a GFR estimation equation developed for one biomarker (creatinine) to another biomarker (cystatin C) by developing biomarker-specific Q-values [15]. This suggests that the GFR-age evolution is generally applicable to all populations, while the biomarker-age evolution is population-specific. Thus, rescaling the biomarker makes the biomarker-GFR relationship less complex, which comes with many advantages for the GFR equations: i) applicable and continuous for all ages, ii) applicable to males and females, iii) applicable for different populations (ethnicities), iv) applicable for different biomarkers and v) stable performance in comparison with equations not based on rescaled biomarkers [7, 15].

The application of the re-expressed LMR-equation with population-specific Q-values resulted in decreased bias and improved accuracy both overall and when stratified by GFR and age. Rescaled creatinine led to larger improvement in performance for r-LMR than for EKFC and after rescaling the two equations had similar performance overall. LMR and EKFC also perform similar in White Europeans with median bias of −0.6 and −3.5 mL/min/1.73 m², respectively, and P₃₀ accuracy of 87.4 and 86.8 %, respectively, (Table S4b and S7a in reference [7]). It should also be noted that imprecision, a major drawback of GFR estimating equations, did not improve by using rescaled creatinine with population-specific Q-values.

The implication of the rescaling was a bit different for the two equations. For EKFC, rescaling creatinine to Black Europeans implied that eGFR was increased with a constant (1.152 for males, 1.065 for females) for all rescaled creatinine values above one. By contrast, the increase in eGFR from rescaling is for r-LMR dependent on how elevated the biomarker is. In the present study the rescaling factor for r-LMR varied between 1.10 and 1.21 for males and between 1.04 and 1.09 for females. It was beyond the scope of the present study to optimize the rescaling at various creatinine levels, but this is a topic for additional investigations.

The improved performance with respect to bias and accuracy was more pronounced and consistent in males, most likely due to a larger difference in Q-values between Black and White male Europeans than between females. However, at BMI<20 kg/m² both r-LMR_Q-Black and EKFC_Q-Black showed increased overestimation resulting in decreased accuracy. This was expected since both LMR and EKFC are known to overestimate mGFR in White Europeans with low creatinine values due to low muscle mass [16], [17], [18]. By using the lower Q-values for White Europeans two different errors cancelled out, thus explaining the lower bias and higher accuracy when not using the Q-values specific for Black Europeans in underweight patients. As this group was small it was not possible to stratify any further.

A similar technique has previously been proposed by Björk et al. [9] for the LMR equation to make it applicable in European children by converting childhood levels of sCr to corresponding adult levels and apply the equation as though the child was 18 years old. Future studies are warranted to establish creatinine growth curves and hence Q-values for children of different ethnicities, thereby further extending the applicability of the rescaling principle for equations like LMR.

Despite the generality of the outlined rescaling principle, the generalizability of the present validation results to other black populations should be considered as a potential study limitation. The validation was based on a single cohort in Paris and the applicability to persons of sub-Saharian origin or other black populations living elsewhere in Europe is a topic for additional investigations [19]. A related limitation is that ethnicity was self-reported in the Paris cohort, although it has been recently shown that genetic ancestry may not be add more accuracy in the GFR estimation than self-reports [1]. It should also be noted that the Q-values utilized for the black population in the present study are subject to statistical uncertainty as the sample size used when they were established was small. Another limitation was that the validation did not include children. While creatinine growth curves and hence Q-values have been established in Swedish children [9], it is yet unclear to what extent they can be generalized with sufficient accuracy to children of other origin. Finally, the choice of reference method may impact the validation results. We considered plasma clearance a coherent choice of reference method in this study, as LMR was developed against plasma clearance (iohexol) and EKFC was initially mainly validated against plasma clearance methods. However, there is some concern that mGFR determined by plasma clearance yields overestimations in patients with low GFR [20, 21]. If this is the case, then the present validation study may have overrated the performance improvement of r-LMR_Q-Black in patients with severe renal impairment. However, it does not affect the generality of the proposed rescaling principle as such.

In conclusion, any well-established eGFR equation may be re-expressed to include rescaled creatinine (sCr/Q) and used in other populations with anticipated different creatinine generation if average creatinine concentration (Q-values) for healthy persons is established. The applied Q-values may either be sex-specific (as in adults) or be both age- and sex-specific (as in children).