Introduction

Chronic kidney disease (CKD) is the 11th cause of death and is classified as the 18th source of disability-adjusted life years according to the 2019 Global Burden of Disease Study [1]. The worldwide prevalence of CKD is around 8.6% [1]. In South Asia, the prevalence of CKD ranges from 10.6% in Nepal to 23.3% in Pakistan [2]. According to a study, the overall prevalence of CKD among Pakistani adults was reported to be 21.2% [2]. According to other studies, the highest prevalence noted was 29.9% [3] and the lowest prevalence noted was 12.5% [4]. Glomerular filtration rate (GFR) is the most beneficial single index for evaluating kidney function and for diagnosing and staging CKD according to the Kidney Disease Improving Global Outcomes (KDIGO) guidelines [5]. Measuring GFR directly using radioisotope or inulin clearance is, however, not feasible in clinical practice, therefore, GFR based on estimating equations (eGFR) using endogenous filtration markers, such as serum levels of creatinine (SCr) and cystatin C (SCysC), is commonly employed [6].

The Lund–Malmö equations were revised by adding more complex terms such as lean body mass. The revised Lund–Malmö equations (LMRevised) showed superior performance in a Swedish population [7,8,9,10,11]. A few studies have also assessed its performance in Asians [12, 13], but no study has been carried out specifically in South Asians. The new Asian-Modified CKD-EPI (CKD-EPIAsian-Modified), a four-level race variable (Black, Asian, Native American and Hispanic, and White and other) was developed that significantly refined bias in Asians and in Chinese [14,15,16,17,18]; it has never been validated or compared in a South Asian population.

An influential and groundbreaking study from Karachi, Pakistan evaluated the performance of the present GFR estimation formulas, modified the present equations and developed a new equation for implementation in South Asians [19]. Modification factors of slope and intercept to the CKD-EPI formula led to a Pakistani CKD-EPI equation (CKD-EPIPK) which significantly reduced bias and improved accuracy.

Previously, the CKD-EP IPK equation was used to evaluate the prevalence, determinants and management of CKD in this region [4]. Moreover, it has recently been used to assess the variation in CKD in relation to demographics, comorbidities and outcomes [20] but there are only a few studies that compared this equation with other eGFR equations [21, 22]. Furthermore, there is no study that evaluated LMRevised and CKD-EPIAsian-Modified equations in comparison to CKD-EPIPK in a South Asian population. In this report, we aim to assess the performance of the LMRevised, CKD-EPIAsian-Modified and CKD-EPIPK equations in the Pakistani CKD population and to investigate the prospective implications on CKD classification and End-Stage Kidney Disease (ESKD) prevalence across these eGFR equations in this population.

Methods

Participant’s characteristics

We performed a cross-sectional analysis using estimating equations including the CKD-EPIAsian-Modified, LMRevised equations and CKD-EPIPK on a CKD cohort of 385 patients from December 2021 to February 2023. Patients sent by nephrologists for laboratory testing of kidney function who were diagnosed with CKD at Allama Iqbal Medical College, Jinnah Hospital, Lahore were included in this study. Chronic kidney disease was diagnosed according to the criteria of the KDOQI practice guidelines (GFR ≤ 60 ml/min/1.73 m2 and albumin to creatinine ratio (ACR) > 30 mg/g for ≥ 3 months) [23]. The inclusion criteria were subjects aged > 18 years, with complete kidney function tests including BUN levels, ACR and serum creatinine values and with a confirmed diagnosis of CKD. Exclusion criteria were incomplete laboratory findings, diagnosis of kidney disease other than CKD, on-going dialysis, acute kidney failure or severe heart failure, severe malnutrition or edema, abdominal or pleural effusions, ketoacidosis, amputation or skeletal muscle atrophy. Patients taking cimetidine, angiotensin-converting enzyme inhibitors and/or angiotensin-receptor blockers and trimethoprim, or recently undergoing hemodialysis and glucocorticoid treatment were also excluded. Written informed consent was provided by each patient before participation. The study was approved by the Ethical Review Board of Allama Iqbal Medical College, Jinnah Hospital (ERB No. 167/23/12/2021/S2 ERB) in its 108th meeting dated 23/12/2021.

Measurement of reference GFR (rGFR)

Reference GFR (rGFR) was measured by employing the urinary clearance of inulin which was considered the gold standard. Inulin clearance was estimated from urine and serum concentrations and the flow rate of urine. A continuous infusion of 1% inulin was administered intravenously in all 385 patients for 2.5 h after 12 h of overnight fasting. Patients were hydrated orally with 65 ml of water after 30, 60, 90, and 120 min. During the infusion of inulin, serum samples were collected four times (0, 40, 70, and 100 min), whereas urine samples were collected three times (30–60, 60–90, and, 90–120 min) after completely voiding the bladder 30 min after inulin infusion was initiated. Samples of inulin were assayed according to the enzymatic method by utilizing a kit. Reference GFR was expressed in terms of body surface area as per 1.73 m2 by multiplication of measured values 0.007184 × W0.425 × H0.725. The mean value of three measurements was expressed as the rGFR which served as a gold standard for comparison with eGFR equations.

Estimation of GFR

Estimation of GFR (ml/min/1.73 m2) was carried out using the CKD-EPIPK, LMRevised and CKD-EPIAsian modified equations. The list of included equations is shown in Table 1.

Table 1 Equations included in the study
Full size table

Laboratory methods

Blood samples of each patient collected for inulin measurement were measured for serum creatinine by Jaffe reaction carried out on a Siemens analyzer, ADIVA 2120. Urine samples were collected for measurement of ACR by A1Care™ HbA1c and ACR analyzer (Precision; Albumin: ≤ 8% CV, Creatinine: 8% CV).

Calibration of serum creatinine assays

Calibration of the assay was carried out daily by two-point calibration with the help of calibrators provided by the manufacturer and traceable to isotope dilution mass spectrometry (IDMS) employing the National Institute of Standards and Technology (NIST) creatinine standard reference material (SRM 967). The system was standardized by routine internal quality control procedures and by involvement in external quality assurance surveys by the College of American Pathologists.

GFR categories

Chronic kidney disease was categorized as G3a: 45–59 ml/min/1.73 m2, G3b: 30–44 ml/min/1.73 m2, G4: 15–29 ml/min/1.73 m2 and G5: < 15 ml/min/1.73 m2 according to the KDIGO 2012 guideline [5, 23]. The prevalence of ESKD, defined as < 15 ml/min/1.73 m2, was compared among the equations on the basis of the serum creatinine.

Statistical analysis

Data were analyzed using IBM-SPSS version 26.0. We assessed the performance of three equations with benchmarks of bias, precision, and accuracy as suggested by KDOQI guidelines along with a percentage of GFR category misclassification. Bias was expressed as median difference between rGFR and eGFR. A negative value of bias showed overestimation of rGFR by equations and vice versa. Precision was expressed as interquartile range (25th percentile; 75th percentile) of difference between rGFR and eGFR. P30 accuracy was defined as the percentage of participants within ± 30% of rGFR. Bland–Altman Plots were made to show the mean differences and limit of agreements between each equation. Scatter plots were made and regression equations were derived by linear regression method for each equation. Pearson’s correlation coefficients (r) were calculated to describe the relationship between all equations and rGFR. Negligible correlations were attributed to r coefficients ≤ 0.30; low correlations were attributed to 0.30–0.49; moderate correlations were attributed to 0.50–0.69; high correlations were attributed to r coefficients 0.70–0.89; and very high correlations were attributed to r coefficients ≥ 0.90 [24]. Categorical agreement rates were estimated when rGFR and eGFR by each equation fell within similar GFR categories. Weighted kappa (κ) value was calculated to evaluate the degree of categorical agreement. The κ values were interpreted as follows: < 0.20, poor; 0.21–0.40, fair; 0.41–0.60, moderate; 0.61–0.80, good; and > 0.81, very good [25]. Counts with percentages were reported for the prevalence of ESKD. P-values < 0.001 were considered statistically significant. A lower significance level was chosen so that stronger evidence can be demonstrated before the null hypothesis is rejected.

Results

Baseline subject characteristics analysis

The baseline characteristics of the participants are shown in Table 2. Among 385 patients, 201 (52.2%) were females. Mean ± SD age (years) was 61.99 ± 16.66, weight (kg) was 80.14 ± 12.98, height (cm) was 168.19 ± 9.53, BMI (kg/m2) was 28.56 ± 5.47, and body surface area (m2) was 1.89 ± 0.16, respectively. Mean ± SD serum creatinine and rGFR were 3.72 ± 2.03 and 15.73 ± 10.59. Mean eGFR by CKD-EPIPK was closest to rGFR while other equations yielded higher mean eGFR than the rGFR, and the degree of variation differed with the equations and age groups. The eGFR by CKD-EPIPK was 16.91 ml/min/1.73 m2, by CKD-EPIAsian-Modified it was 21.36 ml/min/1.73 m2 and by LMRevised it was 19.79 ml/min/1.73 m2.

Table 2 Baseline characteristics of studied samples (n = 385)
Full size table

Out of 385 participants, 0.52% had CKD Stage 3a, 12.20% had CKD Stage 3b, 29.35% showed CKD Stage 4 and 57.92% showed CKD Stage 5 according to rGFR.

There were 40.78% of patients with unknown cause of CKD, 21.56% of patients had diabetic nephropathy, 20.26% had hypertensive nephropathy, 8.31% had chronic glomerulonephritis, 5.45% had polycystic kidney disease and 3.64% had chronic interstitial nephritis (Table 2).

Serum creatinine levels, reference and estimated glomerular filtration rates were stratified by age groups and are shown in Table S1. A gradual reduction in the estimated GFR was noted in elderly patients by all the equations in this study.

Performance of GFR estimating equations in comparison to rGFR

CKD-EPIAsian-Modified and LMRevised showed the highest bias (median difference, − 5.32 ml/min/1.73 m2; − 4.14 ml/min/1.73 m2), respectively. Whereas, the CKD-EPIPK equation showed the lowest bias (− 1.33 ml/min/1.73 m2) in comparison to rGFR. CKD-EPIPK also showed higher precision [IQR, 2.33 (− 2.36, − 0.03)], and P15 and P30 accuracy (70.39% and 89.35%) than the CKD-EPIAsian-Modified (11.69% and 40.78%) and LMRevised equations (24.94% and 54.55%) (Table 3).

Table 3 Performance of GFR estimating equations as compared to reference GFR by urinary inulin clearance
Full size table

Concordance between GFR estimating equations and rGFR

Mean differences and 95% limit of agreement

The mean differences between rGFR and eGFR by the three equations are illustrated by Bland–Altman plot. The mean difference of − 1.18 ml/min/1.73 m2, 95% limit of agreement of − 6.14 ml/min/1.73 m2 was shown between the CKD-EPIPK equation and rGFR (Fig. 1). The mean difference of − 5.98 ml/min/1.73 m2, 95% limit of agreement of − 13.03 ml/min/1.73 m2 was shown between CKD-EPIAsian-Modified and rGFR (Fig. 2). The mean difference and 95% limit of agreement was − 4.06 ml/min/1.73 m2 and − 8.13 ml/min/1.73 m2 between the LMRevised equation and rGFR (Fig. 3). Bland–Altman plots also show that the data points for CKD-EPIPK are evenly dispersed across the mean difference which is quite similar to the dispersion shown by LMRevised.

Fig. 1
figure 1

Bland–Altman plot of CKD-EPIPK and rGFR (ml/min/1.73 m2). The mean of CKD-EPIPK plus rGFR is located on the x-axis, and the value of rGFR minus CKD-EPIPK is located on the y-axis. Solid blue line represents mean difference between CKD-EPIPK and rGFR and dark red lines represent 95% limits of agreement of the mean difference between them

Full size image
Fig. 2
figure 2

Bland–Altman plot of CKD-EPIAsian-Modified and rGFR (ml/min/1.73 m2). The mean of CKD-EPIAsian-Modified plus rGFR is located on the x-axis, and the value of rGFR minus CKD-EPIAsian-Modified is located on the y-axis. Solid blue line represents mean difference between CKD-EPIAsian-Modified and rGFR and dark red lines represent 95% limits of agreement of the mean difference between them

Full size image
Fig. 3
figure 3

Bland–Altman plot of LMRevised and rGFR (ml/min/1.73 m2). The mean of LMRevised plus rGFR is located on the x-axis, and the value of rGFR minus LMRevised is located on the y-axis. Solid blue line represents mean difference between LMRevised and rGFR and dark red lines represent 95% limits of agreement of the mean difference between them

Full size image

The 95% limits of agreement for the CKD-EPIAsian-Modified and LMRevised equations are wider (− 13.03 ml/min/1.7 m2, − 8.13 ml/min/1.73 m2) as compared to the CKD-EPIPK equation (− 6.14 ml/min/1.73 m2). The higher negative values of the mean difference for both the CKD-EPIAsian-Modified and LMRevised equations (− 5.98 ml/min/1.73 m2 and − 4.06 ml/min/1.73 m2) illustrates the overestimation of rGFR by these equations in this CKD cohort compared to the CKD-EPIPK equation.

Correlation and regression analysis

The r coefficients were 0.978 for CKD-EPIAsian-Modified, 0.972 for LMRevised and 0.982 for CKD-EPIPK compared to rGFR (P < 0.001). Hence, all equations showed statistically very high correlation with rGFR.

Figure S1 (supplementary data) represents the scatter plot and regression equation for the CKD-EPIPK formula and rGFR (ml/min/1.73 m2)\(.\) In the regression equation \({\mathrm{eGFR}}_{\mathrm{CKD}-\mathrm{EPI}-\mathrm{PK}}=0.46+1.05*\mathrm{rGFR}\), the intercept is below one and the slope is narrow.

Figure S2 represents the scatter plot and regression equation for the CKD-EPIAsian-Modified formula and rGFR (ml/min/1.73 m2). In the regression equation \({\mathrm{eGFR}}_{\mathrm{Asian}-\mathrm{Modified}}=1.67+1.25*\mathrm{rGFR}\), the intercept is almost two but the slope is narrow.

Figure S3 represents the scatter plot and regression equation for the LMRevised equation and rGFR (ml/min/1.73 m2)\(.\) In the regression equation \({\mathrm{eGFR}}_{\mathrm{LMRevised}}=3.64+1.03*\mathrm{rGFR}\), the intercept is almost four but the slope is narrow. According to Pearson’s correlation and linear regression method, LMRevised and CKD-EPIPK are closer to rGFR than CKD-EPIAsian Modified in this study population.

Categorical agreement rates

The rates of categorical agreement were 90.9%, 73.77% and 66.49% between rGFR and the CKD-EPIPK, LMRevised and CKD-EPIAsian-Modified equations, respectively (Table 4). Upward reclassification was observed from G4 to G3b and from G3b to G3a by both LMRevised and CKD-EPIAsian-Modified equations compared to rGFR but not for G5 stage where a downward reclassification was observed. Categorical agreement and discordant KDIGO GFR categories between rGFR and all equations are shown in Table 4. Weighted kappa values were ‘good’ for CKD-EPIAsian-Modified and ‘very good’ for CKD-EPIPK and LMRevised equations.

Table 4 Categorical agreement and discordant KDIGO GFR categories between rGFR and estimating equations
Full size table

Prevalence of end-stage stage kidney disease

The prevalence of ESKD (≤ 15 ml/min/1.73 m2) stratified by gender and age group are shown in Table S2. Overall, ESKD prevalence is the highest according to the CKD-EPIPK equations (56.3%), whereas the prevalence according to CKD-EPIPK was closer to that shown by rGFR. Of note, high percentages of ESKD were observed in patients in their 80’s according to both the CKD-EPIPK, (80–89, 64.44%; 90–99, 75.00%) and LMRevised (80–89, 62.22%; 90–99, 75.00%) equations. Although ESKD prevailed in all age groups according to all equations, the prevalence was higher in females in all equations.

Discussion

The KDIGO guidelines recommend using CKD-EPI equations unless a substitute equation has been proven to be more reliable in a given population [5]. CKD-EPI equations are recommended in Australia, Europe and North America; different eGFR equations, after sufficient evaluation, are applicable in other regions [5, 6, 14]. This is the first external validation study that has evaluated the CKD-EPIPK, CKD-EPIAsian-Modified and LMRevised equations in comparison to rGFR by urinary inulin clearance in this region. We compared these equations according to the KDIGO guidelines in a Pakistani CKD cohort and assessed their performance, prospective implications on CKD classification and ESKD prevalence.

CKD-EPIPK, the CKD-EPI formula with Pakistani modification factors, showed the best performance in our CKD study population. It has a lower bias [median difference: − 1.33 ml/min/1.73 m2], higher precision [IQR 25th, 75th: 4.94 (− 8.49, − 3.55)], and elevated P30 accuracy [89.35%]. This finding is in line with the development study of this equation [19]. Although the sample population of that study did not have sufficient participants with decreased GFR, and had limited assessment of performance in individuals with different levels of advanced CKD, our study showed ideal performance of this equation in advanced CKD. The categorical agreement with reference GFR was also high (90.9%). Furthermore, CKD-EPIPK was also associated with a high percentage of ESKD which is similar to previous studies [4, 20].

The second closest equation to rGFR identified in this study after CKD-EPIPKwas LMRevised (mean difference − 4.14 ml/min/1.73 m2; 95% limit of agreement  − 8.12 ml/min/1.73 m2) (Fig. 3). Although in our study, LMRevised showed lower bias and higher precision (Table 3) and correlated well with rGFR (Figure S3), accuracy is still poor (Table 3) and there are discrepancies in the assignment of GFR categories and ESKD prevalence (Table 4). Therefore, LMRevised is not suitable in the Pakistani population. This lack of suitability can be explained by the fact that CKD-EPIPK was derived from the linear regression models of natural logarithms of mean GFR (mGFR) versus eGFR calculated by the original CKD-EPI equation in the Pakistani population. The slopes and intercept were back transformed to exponential form and utilized as correction factors with two terms (eGFR = \(0.686\times \mathrm{CKD}-\mathrm{EPI}\)1.059) to modify the equation [19], whereas LMRevised was modified by adding the factor of lean body mass derived from Swedish Caucasians [7]. However, the percentage of categorical agreement was comparatively higher for LMRevised (73.77%) than CKD-EPIAsian-Modified (61.29%) (Table 4) in our study, and both equations showed very high correlation with rGFR (P value < 0.001).

More interestingly, the overestimation by \({\mathrm{eGFR}}_{\mathrm{Asian}-\mathrm{ModifiedCKD}-\mathrm{EPI}}\) was greater than by \({\mathrm{eGFR}}_{\mathrm{LMRevised}}\) compared to \({\mathrm{eGFR}}_{\mathrm{CKD}-\mathrm{EPI}-\mathrm{PK}}\) as shown by the differences among the regression lines and the identity lines on their scatter plots (supplementary data; Figure S1, S2, S3). Although \({\mathrm{eGFR}}_{\mathrm{Asian}-\mathrm{ModifiedCKD}-\mathrm{EPI}}\) has previously been proven to be more appropriate for East Asians [15,16,17,18, 26], according to our results it overestimates GFR in South Asians.

The GFR category distribution for this study population differed according to the formula employed, especially for the new Asian-Modified CKD-EPI formula. The highest percentage of the study population (56.36%) was in the G5 category (< 15 ml/min/1.73m2) according to the CKD-EPIPK equation, whereas it was in G4 (15-29 ml/min/1.73m2) according to the LMRevised equation (43.38%) (Table S1). CKD-EPIPK classified more than 50% of the patients in stage G5 than any of the other equations. This highlights the importance of considering ethnicity factors for different ethnic groups in CKD-EPIAsian-Modified and LMRevised to correctly classify individuals in GFR categories. Furthermore, the Pakistani population is at high risk of ESKD, and implementation of any of these two equations would lead to under-diagnosis and under-treatment of high risk individuals whereas it would result in an over-diagnosis of other CKD stages (G3a, G3b, G4).

The mean rGFR was 15.73 ml/min/1.73 m2 with a standard deviation of 10.59 ml/min/1.73 m2 which was comparatively lower than that estimated by the CKD-EPIAsian-Modified and LMRevised equations in this study (Table 2). Thus, upward reclassification is common in this study. Moreover, CKD-EPIPK, the equation corrected with a Pakistani correction factor, also showed the best performance; thus, this shows that the GFR may be on the lower side for the South Asian population as compared to the Western or other Asian populations of different race/ethnicity. This also highlights the need to re-adjust the GFR categories for individuals in this region and re-define the cut-off value of 60 ml/min/1.73 m2 so as to classify them in the correct CKD category. This finding is identical to one of a previous study [4].

In the current study, the prevalence of ESKD was differently reported depending on the GFR estimating equation employed (Table S2). End-stage kidney disease prevailed in all age groups in all equations but regardless of the equation that was employed, the prevalence was higher in women than in men. Moreover, high prevalence percentages of ESKD (≤ 15 ml/min/1.73 m2) were observed in patients in their 80’s according to CKD-EPIPK and LMRevised. These trends are similar to those observed in other studies [12, 27,28,29,30]. KDIGO risk groupings, gender ratios and CKD prevalence also varied widely in Southeast Asian countries depending on the equation employed [31].

According to a prevalence study carried out in Karachi, Pakistan, it was noted that clinically significant decreased kidney function is common in Pakistani adults, and that the overall prevalence of decreased kidney function was 12.5% [4]. Another recent study noted that South Asian subjects with decreased kidney function were younger and had more advanced stages than the White population. The risk of ESKD was high and CKD-EPIPK was linked with the risk of ESKD in the South Asian population [20]. These findings are similar to those of the present study where prevalence of ESKD is observed in the entire sample regardless of age range, and the CKD-EPIPK equation showed the highest percentage (56.36%) of ESKD prevalence in the present population than any other equation.

Here, it is important to note that CKD-EPIPK was associated with a consistent increase in the percentage of patients with kidney impairment (Table 4). This predicts not only the high prevalence of CKD and associated mortality in this region but it also indicates the elevated risk of developing cardiovascular disease, diabetes mellitus, ESKD events, anemia, gout, secondary hyperparathyroidism, and bone disease. The socio-economic status of the region, poor healthcare infrastructure, underdiagnosis of impaired kidney function, and lack of sufficient dialysis facilities for ESKD patients all lead to compromised patient care, poor quality of life and elevated mortality rates, consequently resulting in an added economic burden to the healthcare system. Specified interventional strategies, including the implementation of the CKD-EPIPK equation, at least in Pakistani laboratories, to encourage earlier identification of CKD and timely referral to secondary care for effective management are necessary to reduce CKD risk, slow its progression and impede the associated consequences such as higher ESKD incidence, cardiovascular events and associated mortality rates.

This study has numerous strengths. First, we employed urinary inulin clearance as a gold standard for measuring true GFR. Second, this is the first external validation study of the CKD-EPIPK, CKD-EPIAsian-Modified and LMRevised equations in a Pakistani CKD cohort against reference GFR by urinary inulin clearance. Third, our sample population was derived from the nephrology center of a renowned government hospital with an excellent turnover of kidney patients from all over Pakistan, hence our results can be generalized, at least for this region. Lastly, our estimation of prospective ESKD prevalence based on both albuminuria and eGFR also amplifies the strength of our study [5, 32].

Our study has some limitations. First, the findings of this study are limited to the Pakistani population and have to be confirmed in other South Asian populations, however, it presents a novelty in validating new equations which have not been validated previously in this population. Second, considering that the equations included in this study also have limitations, there is still room for improvement by further optimization and modification of the equations to achieve more accurate eGFR [19, 33, 34]. Third, SCr levels were computed using recommended techniques with sufficient quality control and standardization. However, SCr levels can be affected by biological and analytical changes. Such uncertainties in measurement cannot be eliminated completely in the present study [35]. Lastly, the Jaffe method used for SCr measurement is subject to bias due to susceptibility to interference compared to enzymatic methods [36, 37] and may have increased CKD misclassification. However, the potential for misclassification is lower as CKD staging was based on not only consecutively lower eGFR values but also on albuminuria for ≥ 3 months.

In conclusion, CKD-EPIPK showed the best performance with low bias, high precision and P30 accuracy, whereas both CKD-EPIAsian-Modified and LMRevised showed poor performance and did not offer sufficient advantage in CKD classification and ESKD prevalence estimation over CKD-EPIPK.

CKD-EPIAsian-Modified and LMRevised need to be modified with ethnicity coefficients for better performance in the Pakistani CKD population, whereas CKD-EPIPK seems ideal for South Asian individuals and appropriate measures should be taken for its implementation, at least in Pakistani laboratories.