Participants and procedure
This cross-sectional study included 72 patients (48 women and 24 men) aged between 40 to 70 years old. They were selected among renal patients who referred to the Shohadaye Tajrish Hospital to check their kidney function between 2012 and 2103. The selected volunteers had no chronic illness such as cardiovascular and liver diseases. Furthermore, the patients with the stable renal disorder had their creatinine and urine in normal levels (29).
At the beginning of the study, the volunteers’ blood samples were collected in order to measure biochemical parameters. The samples were stored at -20°C until the plasma NGAL and Cys-C were evaluated by enzyme-linked immunosorbent assay (ELISA) method. The eGFR was assumed as the gold standard method in which the fixed value 78 ml/min/1.73 m2 was chosen as a cut-off between patients with and without CKD. In this study, the NGAL and Cys-C were used as predictors and eGFR was used as a response variable. This study was approved by the Medical Ethics Committee of Shahid Beheshti University of Medical Sciences, Tehran, Iran. (IR.SBMU.RET ECH.REC.1397.1013)

Statistical analysis
The statistical analysis was performed using the IBM SPSS Statistics, Version 25.0 (IBM Corp., Armonk, NY, USA). For univariate purposes, the Chi-square and independent samples t-tests were utilized. In addition, to investigate the predictive power of the biomarkers (NGAL and Cys-C), the univariate and multiple logistic regression model and receiver operating characteristics (Roc) curve analysis were used. In this process, the following logistic regression models were applied in which the binary outcome was eGFR (with the cut-off 78 ml/min/1.73 m2).
   (1)- Model with only "NGAL" as the predictor
   (2)- Model with only "Cys-C" as the predictor
   (3)- Model with both "NGAL" and "Cys-C" as the predictors
In all of these models, the predicted probability of having CKD (Π_i (x)) for each person was calculated using the general formula π(x)=exp⁡(α+X'*β)/(1+exp⁡(α+X'*β)), Where α is the intercept, and  β is vector of regression parameters. The ROC curve and the Youden Index were used to evaluate the diagnostic power of markers and determine the optimal cut-off points. This cut-off point was used to compare the predicted probabilities and classify each individual as patient or non-patient. According to the obtained cut-off points, then, the sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV) and area under ROC curve (AUC) were calculated. 

 

The A total number of 72 patients who referred to the Shohadaye Tajrish Hospital for checking their renal function were studied. In this sample, the mean age 66.7% (48 patients) were female and 34.7% (25 patients) had kidney damage. The Mean (SD) age and weight for the entire sample were 54.51 (8.06) and 79.28 (13.65) years, respectively. The mean (SD) NGAL and Cys-C levels of the sample were 78.72 (105.44) ng/ml and 423.72 (744.96) ng/ml, respectively.
Table 1 shows the results of logistic regression models which were described in the methods section. In the firs model, (model with only NGAL as the predictor), one can observe that one unit increase in NGAL value can raise the odds of having renal damage by 14.2% (P = 0.001). For the model with only Cys-C as the predictor, it can be seen that one unit increase in Cys-C value has only 0.4% increase in odds of CKD resulted (P = 0.002). Finally, in the third model (model with both NGAL and Cys-C as the predictors), only NGAL showed a significant relationship with the kidney damage (P = 0.007). According to this model results, one unit increase in NGAL value could lead to about 12.5% increase in odds of kidney damage.
Table 2 summarizes the findings from Roc curve analysis for the three fitted models in Table 1. Moreover, ROC curve for each model was shown in Figure1. Regarding these findings, one can conclude that the model with only NGAL as the predictor could result in a rather perfect prediction of renal disorder.  
 

Table 1. Results of logistic regression model for determining the power of NGAL and Cys-C in predicting kidney damage


Table 2. Results of ROC analysis

Figure 1. ROC curve for model1, model2 and model 3

Discussion

In this study, we used simple and multiple logistic regression models for assessing the power of Cys-C and NGAL for the early prediction of CKD. Our findings showed that the model with NGAL as the single predictor had rather perfect predictive power (all of the diagnostic accuracy measures were close to 1.0). In addition, the Cys-C biomarker revealed no significant relationship with CKD. In other words, adding Cys-C to a model, which includes the NGAL biomarker, did not lead to increase in the value of diagnostic accuracy measures.
Cys-C is a biomarker of CKD, which is produced in all nucleated cells. It is not affected by the muscle mass, gender and age (21). In this study, in order to evaluate the prediction power of Cys-C biomarker, the cut-off point 78 ml/min/1.73 m2  was considered for eGFR. Accordingly, the model which only included Cys-C had low sensitivity, acceptable specificity and AUC 64%, 97.87% and 85%, respectively. However, in a study on 206 patients in Sweden, White et al. aimed to assess the diagnostic accuracy of Cys-C biomarker. The cut-off points of their study were 72 ml/min/1.73 m2 and 1.25 mg/L for eGFR and Cys-C, respectively (30). They found greater sensitivity and less specificity for Cys-C (71.4% and 95.1%, respectively). In addition, Martin et al. assessed 94 patients for evaluating the diagnostic accuracy of Cys-C in advanced stages of CKD. They had CKD as a result of different diseases such as diabetes  (31). Using the analysis of ROC curve, they showed that in sever stages of CKD, Cys-C had AUC of 63%, while, in our study which was performed on the early stage of CKD patients, this value was 85%. According to this, it can be concluded that Cys-C may have more diagnostic accuracy in the early stage of the disease.
NGAL is a promising marker for the kidney damage, which can be easily found in blood and urine. The results of our study also proved high prediction power of NGAL for the early-stage renal damage in CKD patients (32). The models, which include NGAL and defining optimal cut-off point for probabilities, resulted in proper sensitivity, specificity, NPV, PPV and AUC for NGAL (96%, 100%, 100%, 97.92% and 99%, respectively). However, in a prospective cohort study on patients between 18 to 65 years old Basturk et al. evaluated the diagnostic accuracy of NGAL biomarker. Their study showed that NGAL in its optimal cut-off point had less sensitivity and specificity (72.2%, and 72.2%, respectively) (29). Also, in another study on the end stages of CKD patients, they showed that considering best cut-off point for NGAL it had less sensitivity and specificity (83.9% and 53.8%, respectively) (33).
Ghonemy et al. considered the NGAL and Cys-C as the markers of Acute Kidney Injury after cardiac surgery. They showed, even in this condition, the sensitivity and specificity of NGAL were higher than Cys-C (34). In this study, in the early stage of CKD, the model with NGAL showed the highest diagnostic measures as well. Mitsnefes et al. showed that NGAL had a better diagnostic performance than Cys-C for a cut-off point of eGFR <30 ml/min/1.73 m2 and they also concluded that NGAL and Cys-C were similar at GFR levels of ≥30 ml/min/1.73 m2 (35).
None of the above-mentioned studies evaluated NGAL and Cys-C markers simultaneously in order to predict CKD; while the logistic model is an appropriate model for binomial data, none of these studies used it for evaluating the kidney damage. In the field of CKD prediction, Fisher et al. used logistic regression model including 12 covariates such as age, race, hypertension, etc. They performed their study on 11955 adults over 18 years old (36). The results of their study, comparing to our results, showed less sensitivity and specificity (86% and 85%, respectively).
Despite some advantages such as concurrent analysis of biomarkers, our study had some limitations. First, this study had a small sample size and generalization of the results may be questionable. Second, as measuring GFR is costly and time-consuming, we used the eGFR as the gold standard for CKD. As we know, eGFR is an estimate of GFR, thus some imprecise values may be included in the data set. In order to solve this problem, we aimed to assess NGAL and Cys-C markers using a Bayesian approach without considering gold standard in our future study.

 

Conclusion

The findings revealed that the NGAL biomarker resulted in satisfactory power for predicting the early kidney damage. In addition, we found that the predicting power of Cys-C for early diagnosis of kidney damage is less than NGAL biomarker. Further studies with higher sample size are needed to evaluate the prediction role of NGAL and Cys-C using other methods such as machine-learning techniques, support vector machines (SVM) and decision tree analysis . Moreover, we suggest studies with longer follow-up to know the proportion of patients with renal damage.

 

Acknowledgements

T

 

Conflicts of Interest

This The authors declared no conflicts of interest.

Funding

T