The comparison between patient-adjusted and BSA-normalized versions of the equations demonstrated a small underestimation of GFR by the BSA-normalized versions. However, the limits of agreement, calculated as 1.96 times the standard deviation, consistently remained below 20 mL/min. This indicates that the differences between the two versions of the equations were within the recommended 20% precision limit [32, 33].

In contrast, when comparing the results of the five original equations, notable differences were observed, with limits of agreement close to ± 40 mL/min. This suggests that the selection between a normalized or adjusted version of an eGFR equation may be of less significance for dose adjustments compared to the choice of one of the five equations themselves.

We conducted a comparative analysis of the different equations used to estimate GFR using the Bland–Altman plot, which is recommended for such comparisons. In addition to the visual analysis of the plot, we calculated additional statistics to assess the distribution of differences between the paired instruments. These statistics helped determine the limits of agreement, which represent the average discrepancy between the two instruments across the entire range of measured values in the analyzed patients. Furthermore, we calculated the regression line of the difference distribution, which precisely identifies the GFR value at which the measurements obtained with the two instruments diverge in magnitude. This information is crucial in assessing the clinical relevance of the discrepancy and its potential impact on dose adjustment procedures. Moreover, by employing regression analyses of the discrepancies with anthropometric measures, we were able to establish the range of BSA where adjusted and normalized equations can be interchangeably used.

A key strength of our study was the utilization of a real-world group of patients comprising aged individuals attending a primary care center in central Portugal. This approach allowed us to evaluate the practical relevance of the different equations in a real-life setting bearing in mind the typical limitations of a primary care setting with routine care procedures. In this environment, clinicians must make decisions in a limited amount of time, with limited resources available, where clear recommendations contribute to enhanced patient safety. Additionally, studies reported that using clinical decision support systems (CDSS) in drugs with definite dose recommendations produce better results than in drugs without definite dose recommendations [34]. However, it is important to note that these patients may not be fully representative of all aged patients globally or even within Portugal. One characteristic of studies that employ secondary data obtained from medical records is the reliance on healthcare providers outside the research team for data reliability. Furthermore, it is important to acknowledge that some data was missing, which is a common challenge in studies of this nature. It is essential to clarify that our study did not aim to identify the best possible equation, as this was not within the scope of our research. Additionally, we did not validate the results against a gold standard method for measuring GFR.

Another strength of the study is the use of simple formulas to fit in primary care setting needs. Quetelet and Du Bois are the accepted gold standard equations to estimate BMI and BSA, respectively. In the absence of an accepted gold standard to calculate IBW [35], we preferred using Friesen’s formula, instead of Devine’s [36], because they are equivalent for practical purposes [29] but much easier to use [37].

Regulatory agencies recommend adjusting drug doses based on absolute (patient-adjusted) GFR rather than BSA-normalized GFR [38]. This recommendation is grounded on the understanding that renal drug clearance is proportional to an individual's GFR [8, 17]. However, for the diagnosis and staging of renal disease, BSA-normalized GFRs are commonly used to categorize the degree of renal impairment [5]. It is worth noting that computer systems often display calculators that provide both patient-adjusted and BSA-normalized estimated GFR equations. This flexibility is intended to accommodate situations where weight and height data are not available [23]. Our study revealed that the discrepancies between patient-adjusted and BSA-normalized equations for estimating GFR are not highly significant. The limits of agreement, which represent the overall discrepancies across the entire range of measurements, were below the commonly accepted ± 20% precision limits considered acceptable [32, 33]. Importantly, the differences between each adjusted-normalized pair were smaller in the clinically relevant range of the GFR interval (below 60 mL/min) than in areas where dose adjustments are not expected (Supplementary File 1) [5].

Normalizing GFR calculations may be valuable for comparative population studies, but it may not be as relevant for dose adjustment practices. This is because higher body weight does not necessarily correspond linearly to greater muscle mass, kidney volume, or the number of functioning nephrons [8]. The literature suggests that considering the difference between the two equations becomes more important when patients deviate significantly from the average body size (BSA around 1.73 m2) [17, 23, 39]. To establish the precise limits where patient-adjusted and BSA-normalized eGFR can be interchangeably used, we conducted two regression analyses of the discrepancies between the two equations using two simple anthropometric measures: BSA and BMI. While BMI is commonly used in clinical practice, BSA exhibited a stronger and more reliable association with discrepancies between the instruments. Consequently, using BSA allowed us to identify the specific limits where differences between the two equations remained below recommended ± 10 mL/min [40], that varied from 1.39 to 1.50 m2 (lower limit) and from 1.96 to 2.07 m2 (upper limit) depending of the equation used (Fig. 1).

The differences observed in eGFR values obtained using the original versions of the equations were greater than those between the adjusted and normalized versions of each equation. In the overall comparisons across the entire interval, the CG equation exhibited the largest discrepancies, as indicated by wider limits of agreement, with MDRD, CKD-EPI, and BIS1 (limits ranging from 25 to 35 mL/min). These differences could be clinically relevant for dose adjustment considerations, especially when the eGFR approaches the nearest cutoff value [23]. While CG was the first equation considered in official prescribing information, there is no consensus on which equation to prefer when discordance is observed [17, 41]. On the other hand, as previously reported, the MDRD and CKD-EPI equations produced similar eGFR results [42, 43]. However, some of these comparisons present an additional challenge that complicates simplistic corrections based solely on overall bias. For instance, the overall bias of the comparison between MDRD and BIS1 is 10.4, suggesting that, on average, MDRD overestimates eGFR by approximately 10 mL/min (Supplementary File 2). However, the regression line intercepts the null difference at 30 mL/min (CI 95% 25:35), indicating that below that value, MDRD underestimates eGFR when compared to BIS1. It is important to pay special attention when the regression line of the discrepancies intercepts the null difference in an area where dose adjustments are expected. Many drugs have dose adjustment recommendations that can lead to different dosing regimens based on different eGFR equations, resulting in shifting estimations between 30 and 60 mL/min. For example, European SmPCs recommend that the normal dose of 2000 mg of metformin should be reduced to 1000 mg when GFR is less than 45 mL/min, or the recommended dose of 5 mg of apixaban is reduced to 2.5 mg when GFR is less than 30 mL/min, and the normal dose of rosuvastatin (10–20 mg) should be reduce to 5 mg if GFR < 60 mL/min. Literature reported than differences in eGFR calculated using the MDRD and BIS1 equations could lead to doubling the dose depending on the equation used [17, 23].

Considering the wide variation in eGFR values among the different equations, our study does not provide sufficient evidence to advocate for the use of one SCr-based equation over another. Instead, a comprehensive evaluation of clinical outcomes resulting from discordant dose recommendations would be necessary to gain further insights and make informed decisions [44]. Alternatively, the use of serum cystatin C–based equations (e.g., CKD-EPI cystatin C [45]) or the combined equations (e.g., CKD-EPI Scr-Scys combined formula [46]) could improve the eGFR calculations, especially at early stages of kidney dysfunction [47]. However, these equations are less likely to be useful in primary care where Serum cystatin C is not routinely measured.

Literature demonstrated that different equations were more accurate than others in different situations. For example, MDRD is preferred to CKD-EPI in low eGFR [8]. Or MDRD and CKD-EPI have a better performance in obese patients than CG [48]. Also, MDRD and CKD-EPI without any race coefficient performed well in sub-Saharan black populations [49]. But considering a primary care setting and aiming to estimate GFR to adjust the dose of renally excreted drugs, using different equations according to the differential characteristics of each patient may not be appropriate. CG was created using isotope dilution mass spectrometry (IDMS)-nontraceable creatinine data and should not be used with IDMS-traceable SCr tests, where results may vary between 10 and 20% [50]. Additionally, CG is the most affected equation in low eGFR due to weight of the creatinine tubular secretion. Subsequently, KDIGO recommends against CG equation for dose adjustments [13]. BIS1 was created and validated to be used in patients over 70 years of age, which limits its generalizability.

MDRD was originally validated using IDMS-nontraceable SCr tests, but then was adapted to IDMS-traceable SCr tests by modifying the coefficients. CKD-EPI was created to be used with IDMS-traceable SCr tests. Thus, both equations may underestimate GFRs if used with IDMS-nontraceable SCr tests. Between these two equations, CKD-EPI is more accurate than MDRD in GFR > 60 mL/min [10], with official bodies suggesting never reporting eGFR > 60 with MDRD [51]. Below 60 mL/min, estimates obtained with MDRD could be slightly more accurate than CKD-EPI, but discrepancies were reported to be negligible [10]. Our results confirm this pattern, with discrepancies below 10 mL/min in the lowest part of the interval, while substantial discrepancies in high GFR (Supplementary File 2). It seems that, if we want to use only one equation in primary care, coincidently with KDIGO, the CKD-EPI could be the best choice.