The Disagreement Among Methods to Evaluate Glomerular Filtration Rate in HIV-Infected Patients Naive to Antiretroviral Therapy

E ditor: In response to the rising prevalence of chronic kidney disease among HIV-positive subjects, a strict follow-up of the glomerular filtration rate (GFR) is strongly recommended. ^1,2 Several equations based upon serum creatinine were developed to estimate GFR, such as the Cockcroft–Gault ³ and the Modified Diet in Renal Disease (MDRD). ⁴ However, the Chronic Kidney Disease Epidemiology Collaboration (CKD-EPI) formula is more accurate than the others in HIV-uninfected subjects with GFR ≥60 ml/min/1.73 m². ^5,6 Although none of the these formulas has yet been validated in HIV-infected patients, the CKD-EPI was recommended by the Italian guidelines for HIV disease management. ⁷ Noteworthy, GFR estimated through cystatin C clearance is suggested to be more reliable in HIV-infected patients because it may be less influenced by muscle mass or liver function; however, variability due to inflammatory status cannot be excluded. ^{8 –12} Therefore, it is important to conduct analyses of the agreement between the available methods.

Overton et al. compared MDRD with plasma cystatin C concentration (but not with cystatin C clearance) in 670 HIV-infected patients, obtaining a statistically significant correlation with r=−0.35, accounting only for 12.3% of the variability between the two measures. ¹³ Also in this study a significant direct correlation between MDRD and CKD-EPI was found using Spearman correlation (r _s=0.94, p<0.001). In another paper, ¹⁴ Dabrowska et al. used paired t-test to demonstrate that GFR estimated by CKD-EPI was consistently lower than MDRD. The same approach was employed by Ravasi G et al. ¹⁵ In both papers, ^14,15 differences between methods were attributed to a number of clinical and laboratory variables using linear regression analysis. Although Ravasi et al. used the kappa test to evaluate concordance between categorical measures, ¹⁵ no proper agreement analyses have been conducted so far. Importantly, significant linear correlation between two measures does not necessarily mean their agreement and paired Student's t-test is a flawed approach to assess concordance. ^16,17

We conducted a proper agreement analysis, comparing not only the most frequent formulas to estimate GFR by creatinine (i.e., Cockcroft–Gault, ³ MDRD based on four variables, ⁴ and CKD-EPI ⁵ ), but also by cystatin C clearance using the CKD-EPI formula (eGFR=76.7×cystatin C^−1.19). ¹⁸ Estimated GFRs were also compared with the actual creatinine clearance (CrCl) obtained through 24-h urine collection.

A cohort of 85 patients naive to antiretroviral therapy enrolled in a controlled clinical trial (EudraCT number: 2007-007934-21) was used. ¹⁹ Patients had actual CrCl>50 ml/min as per inclusion criterion in the trial. The majority of patients (78.8%) were males; 87.1% had acquired HIV infection through sexual intercourse and 20% had CD4⁺ T cell count<200/mm³. Six patients were affected by hypertension, two by diabetes, and 17 took thrimethoprim/sulfamethoxazole as prophylaxis.

Serum creatinine was measured by the Roche enzymatic assay on a Roche/Hitachi P module automated analyzer (COBAS INTEGRA 400/700/800 Creatinine plus ver.2, Roche Diagnostics GmbH); isotope-dilution mass spectrometry was also used as standard. Cystatin C was measured in frozen plasma samples (−70°C) using a particle-enhanced immunonephelometric assay (BN II nephelometer system). All GFR estimates by Cockcroft–Gault and CKD-EPI cystatin C were corrected for body surface area by the DuBois method. ²⁰ Measured values were expressed as means and standard deviations (SD).

Agreement analysis was first conducted on categorical data (GFR<or≥90 ml/min) by Cohen's kappa for qualitative variables. A value lower than 0.00 implies poor agreement; 0.00–0.20, slight agreement; 0.21–0.40, fair agreement; 0.41–0.60, moderate agreement; 0.61–0.80, substantial agreement; and 0.81–1.0, almost perfect agreement. ²¹ Two patients had severe renal dysfunction (GFR<60 ml/min/m²) using the CKD-EPI creatinine formula and one patient using the cystatin C-based equation. Noteworthy, 45.9% patients showed moderate renal dysfunction (GFR<90 ml/min/m²) using MDRD, but only 9.4% using actual CrCl, accounting for a slight agreement between the two methods (Cohen's kappa=0.0670). A similar result was obtained by Ravasi et al. ¹⁵ Furthermore, when we assessed the agreement among the other methods a substantial agreement was found only between MDRD and CKD-EPI creatinine formulas (Table 1).

To complete the agreement analysis, data as continuous measures were analyzed. First, the Bland and Altman method was used to estimate the global bias throughout the mean difference between GFR values and plot the differences against their means. ²² The width of the 95% limits of agreement (mean of the difference−1.96 times the SD of the differences) has to be lower than the 15–20% of the mean of the averages of the two methods to decree the agreement. ²² Second, to provide a synthetic agreement index, we calculated the concordance correlation coefficient (CCC) by which the value of the within-sample variability can be considered as an agreement criterion. ²³ For claiming an agreement, the within sample variability between methods should be ≤15%, leading to a CCC equal to 0.9775. ²³ Third, we employed the nonparametric Passing and Bablok regression to obtain nonparametric estimates of the intercept (to be interpreted as a measure of the presence of a systematic measurement error between two methods) and of the regression coefficient (to be interpreted as a measure of the presence of a proportional measurement error between two methods). ²⁴

The results of these analyses are illustrated in Table 2. Mean actual CrCl was 143.1 (SD: 45.2) ml/min, greater than those obtained with the other methods (mean incremental difference: 31 to 47 ml/min). The Bland and Altman analysis demonstrated a substantial disagreement. In fact, the percentage of the agreement region width ranged from 33.7% (between CKD-EPI creatinine and MDRD) to 153% (between actual CrCl and CKD-EPI cystatin C), far more than <20% to decree the agreement. ²² CCC confirmed the disagreement, but a moderate agreement was found between CKD-EPI creatinine and the MDRD equation (CCC=0.906). Interestingly, both with Bland and Altman and CCC analyses, actual CrCl showed the worst agreement with each of the other methods. Notably, the proportional measurement errors obtained by nonparametric Passing and Bablok regression were significantly different from 1, indicating that discordance between methods increased with increasing GFR values.

In conclusion, the present analysis confirms that a significant proportion of HIV-infected patients suffer from moderate renal dysfunction (GFR<90 ml/min/m²) even though they are not exposed to antiretroviral therapy. ²⁵ However, GFR varied according to the employed formulas, showing a substantial disagreement when a proper statistical analysis was performed. Noteworthy, we demonstrated that the problem of a disagreement was more evident in patients with higher GFR values, but even at lower GFR a significant discordance was found and this may be clinically relevant. For instance, only 2.3% of patients had GFR<90 ml/min/m² with all four methods, 27.1% with three methods, 29.4% with two methods, and 58.8% with at least one method. Among the methods used, MDRD and CKD-EPI creatinine formulas appeared to be stricter than the others because about 40% of patients had eGFR<90 ml/min/m² with these two methods, in comparison with only 9% with actual CrCl. Since the clinical accuracy and the predictive values for severe renal insufficiency are unknown for the available formulas, it is urgent to validate GFR-estimating equations (either alone or combined ¹⁸ ) for assessment and monitoring of renal function in HIV-infected patients.