Impact of blood glucose levels on the accuracy of urinary N-acety-β-D-glucosaminidase for acute kidney injury detection in critically ill adults: a multicenter, prospective, observational study

Three general ICUs was included in this prospective observational study. Patients admitted to our ICUs between October 2014 and July 2016 were enrolled. Exclusion criteria were known the refusal of consent, age < 18 years, pregnancy, unavailability of the urine sample, hemodialysis or peritoneal dialysis prior to enrollment or end-stage renal disease (ESRD), nephrectomy or renal transplantation. The protocol was followed according to that of the Strengthening the Reporting of Observational Studies in Epidemiology [24] and Standards for Reporting Diagnostic Accuracy [25] criteria. The current study received the approval of the local institutional review board. Written informed consent was obtained from each participant or a family member at the time of enrollment.

At the time of ICU admission, urine and blood samples were collected immediately. For patients from the participating hospital, urine samples were shipped by cold chain transportation and examined at the clinical laboratory of the Guangdong Provincial People’s Hospital within 24 h after collection. We measured the level of serum creatinine (sCr), uNAG, serum glucose, and glycosylated hemoglobin (Hb1Ac) at the time of ICU admission. uNAG and HbA1c were measured upon ICU admission. sCr and serum glucose were measured at ICU admission and daily as a part of routine clinical care during ICU stay.

We prospectively collected the demographic and clinical characteristics of each patient, including sex, age, body mass index (BMI), preexisting chronic conditions, sepsis, the categories of diseases, Acute Physiology and Chronic Health Evaluation II score (APACHE II), previous application of antidiabetic drugs, baseline serum creatine, baseline estimated glomerular filtration rate (eGFR), and the outcomes. The eGFR of patient was calculated using the abbreviated Modification of Diet in the Renal Disease formula [26]. The hourly urine output was also recorded.

Creatinine, uNAG, and serum glucose were analyzed using the UniCel DxC 800 Synchron System (Beckman Coulter, Brea, CA, USA). The values of uNAG were normalized to that of the urinary creatinine concentrations. The coefficients of interassay and intraassay variation for uNAG were both ≤10%. HbA1c was measured using the D-10™ Hemoglobin Analyzer (Bio-Rad, Hercules, CA, USA). The normal range for HbA1c was 3.8–18.5%.

AKI was determined based on the criteria of Kidney Disease Improving Global Outcomes Clinical Practice Guidelines [27]: sCr levels increased exceed 0.3 mg/dL (26.5 μmol/L) from baseline within 48 h, or more than 1.5-fold increase in sCr levels from baseline within 7 days, or urine output < 0.5 mL/kg/h for 6 h. KDIGO stage 2 or stage 3 within 7 days after ICU admission were considered as severe AKI. The baseline sCr was defined as following rules in sequence as described in previous study [28]: if patients had serum creatine value before ICU admission, (1) the most recent pre-ICU value (between 30 and 365 days before ICU admission); (2) for patients aged < 40 years, a stable pre-ICU value > 365 days before ICU admission (stable defined as within 15% of the lowest ICU measurement); (3) pre-ICU value (> 365 days before ICU admission) and less than the initial sCr at ICU admission; (4) a pre-ICU value (between 3 and 39 days before ICU admission) ≤ initial sCr at the time of admission to ICU and not distinctly in AKI; if patients did not have serum creatine value before ICU admission, (5) the lowest sCr upon initial admission value, the final ICU value, or the minimum value at follow-up unto 365 days.

All of the groups were preset. To evaluate the performance of uNAG for AKI detection with respect to the admission serum glucose in various degrees, the patients were stratified into five groups as described previously [29]: < 110 mg/dL, 110 to < 140 mg/dL, 140 to < 170 mg/dL, 170 to < 200 mg/dL, and ≥ 200 mg/dL. Patients were also divided into five quintiles based on the levels of serum glucose at the time of admission.

In order to detect the impact of HbA1c levels and history of diabetes on the accuracy of uNAG, patients were additionally stratified into four groups as described previously [30]: recognized diabetes (previous diagnosis of diabetes), undetected diabetes (with HbA1c ≥6.5% but no previous diagnosis of diabetes), prediabetes (5.7% ≤ HbA1c < 6.5% but no previous diagnosis of diabetes), and normal glycemic status (HbA1c < 5.7% but no previous diagnosis of diabetes). Patients were classified as having diabetes if they were diagnosed with diabetes or at least 1 prescription for insulin or an oral antidiabetic agent. Patients without a known history of diabetes and an HbA1c < 6.5% were defined as non-diabetic patients. The recognized and undetected diabetes were categorized as diabetic patients. Stress hyperglycemia was defined as serum glucose levels exceeding the threshold of 200 mg/dL in non-diabetic patients, in accordance with the consensus statement [31]. Likewise, diabetic patients were divided into two subgroups according to abovementioned threshold of serum glucose. The diagnosis of sepsis was defined according to the American College of Chest Physicians/Society of Critical Care Medicine Consensus Conference Committee guidelines [32].

We calculated the sample size based on AKI incidence of 26.86%, which we determined in a chart review of 350 critical patients (unpublished). According to the previous studies [33, 34], we estimated that a sample size requirement of 1415 patients was required, with a two-sided test (α error = 5%; power = 80%). With an estimated loss of 10%, the final sample required will be 1557 patients at least. The Kolmogorov–Smirnov test was used to determine the normal distribution of each variable. Continuous variables were presented as means ± standard deviation and medians (interquartile range). The categorical variables were expressed as counts and proportions. The non-normally distributed continuous variables were compared by Wilcoxon rank-sum test or Kruskal–Wallis test for one-way analysis of variance. To compare the categorical variables, the chi-square test or Fisher’s exact test was used. Bivariate correlation analysis was performed to examine the correlation between two variables. Multivariate linear regression was used to search for factors independently associated with uNAG. The performance of uNAG was determined using the receiver operating characteristic (ROC) curve analysis. The differences between AUC were tested by Hanley–McNeil method [35]. The optimal cut-off values for AKI detection were defined for individual biomarkers using Youden’s index based on which the sensitivity and specificity were calculated.

Of 1585 consecutive admitted patients from three general ICUs, 148 patients (9.3%) were excluded according to the exclusion criteria (Fig. 1). Of the remaining 1437 subjects, AKI was diagnosed in 412 patients (28.7%). Table 1 presents the baseline clinical data and outcomes of the enrolled patients. Compared to the non-AKI patients, those with AKI were older with high rates of preexisting clinical conditions, including diabetes mellitus (DM), hypertension, and chronic kidney disease (CKD). Diabetes was presented in 225 patients of the entire cohort, including 127 undetected diabetes patients. Patients with AKI had higher concentrations of sCr and serum glucose at the time of ICU admission. Moreover, the concentrations of uNAG were significantly higher in the AKI group than in the non-AKI group.

Bivariate correlation analysis (Table 2) showed that a high level of uNAG was associated with older age, higher APACHE II score, and a remarkable change between the levels of initial sCr and baseline sCr. In addition, a positive correlation was established between uNAG and diabetes mellitus (r = 0.124, P < 0.001) as well as HbA1c (r = 0.147, P < 0.001) and a positive association between uNAG and admission glucose (r = 0.196, P < 0.001). In multivariate regression (Table 3), age (standardized β = 0.187, P < 0.001), baseline sCr (standardized β = − 0.093, P < 0.001), △sCr (standardized β = 0.089, P < 0.001), and admission glucose (standardized β = 0.147, P < 0.001) were significantly correlated with the levels of uNAG.

ROC analysis revealed that uNAG detected total AKI with poor-to-moderate discrimination capability (Fig. 2). The AUC-ROC value of uNAG for detecting total AKI was 0.664 [95% confidence interval: 0.639–0.689]. To evaluate the impact of admission serum glucose levels on the accuracy of uNAG to detect AKI, an AUC-ROC value was calculated in each group of admission serum glucose: < 110 mg/dL, 110 to < 140 mg/dL, 140 to < 170 mg/dL, 170 to < 200 mg/dL, and > 200 mg/dL (Table 4); no significant difference was observed in AUCs between any two groups. Similarly, no significant difference was observed in AUCs for severe AKI detection between any two groups (P > 0.05). In addition, we evaluated the AUCs of uNAG on the diagnosis of total AKI and severe AKI in each quintile of the level of admission serum glucose (Additional file 1: Table S1). There was no significant difference in the AUC between any two sub- cohorts.

Among the non-diabetic patients, AUCs was calculated as 0.679 and 0.645 in patients with and without stress hyperglycemia, respectively (Table 5). The optimal cut-off value of uNAG for total AKI detection between stress hyperglycemia and non-hyperglycemia patients was markedly different (43.86 and 29.80 U/g Cr, respectively). In addition, the median concentration of uNAG for AKI patients with stress hyperglycemia was 46.15 U/g Cr (29.89–63.13), which was significantly different from non-stress hyperglycemia patients with the value 32.98 U/g Cr (19.10–57.08) (Additional file 2: Figure S1). Furthermore, uNAG exhibited marked difference optimal cut-off value for detecting severe AKI between these two subgroups (44.74 and 31.65 U/g Cr, respectively). However, the optimal cut-off values of uNAG for detecting AKI in diabetic patients between these abovementioned subgroups were similar (Additional file 3: Table S2).

To evaluate the influence of HbA1c levels and history of diabetes on the performance of uNAG in the detection of AKI; the level of HbA1c at the time of ICU admission was measured. Patients were divided into four groups according to the HbA1c levels and history of diabetes (Additional file 4: Table S3). Patients with a known previous history of diabetes were defined as recognized diabetes. Patients without known previous history of diabetes were further divided into three groups. The AUC for total AKI detection was calculated as 0.675 in a sub-cohort of recognized diabetes. In patients without a known previous history of diabetes, the AUC was calculated as 0.649 in the sub-cohort with HbA1c ≥6.5%, 0.645 in the sub-cohort with 5.7% ≤ HbA1c < 6.5%, and 0.659 in the sub-cohort with HbA1c < 5.7%, respectively. The AUC for uNAG in detecting severe AKI in each group were as follows: 0.731 in group of recognized diabetes, 0.704 in group of unrecognized diabetes, 0.700 in group with 5.7% ≤ HbA1c < 6.5% and 0.734 in group with HbA1c < 5.7%. In addition, we also evaluated the AUCs of uNAG on the discrimination ability of AKI in each quartile of HbA1c levels (Additional file 5: Table S4). Similar results were observed, and no significant difference was observed in the AUC for AKI or severe AKI detection between any two groups.