Diagnostic Value of Urine Tissue Inhibitor of Metalloproteinase-2 and Insulin-Like Growth Factor-Binding Protein 7 for Acute Kidney Injury: A Meta-Analysis

Background Tissue inhibitor of metalloproteinase-2 (TIMP-2) and insulin-like growth factor-binding protein-7 (IGFBP7) are both involved in renal tubular epithelial cell cycle arrest in acute kidney injury (AKI). Several recent studies showed that urine TIMP-2 times IGFBP7 ([TIMP-2]*[IGFBP7]) is a promising biomarker to predict AKI. Methods The aim of this meta-analysis was to assess the diagnostic value of urine [TIMP-2]*[IGFBP7] for early diagnosis of AKI. Relevant studies were retrieved from the PubMed, EMBASE, and Cochrane Library databases. The sensitivity and specificity were determined, and summary receiver operating characteristic (SROC) curves were constructed. Results Ten full-text prospective studies were included in this meta-analysis. The estimated sensitivity of urine [TIMP-2]*[IGFBP7] for the early diagnosis of AKI was 0.84 (95% CI = 0.80–0.88) and the specificity was 0.57 (95%CI = 0.55–0.60). The SROC analysis showed an area under the curve of 0.8813. Limitation The limited number of included studies, small sample size, unpublished negative results and language limitation might have affected the evaluation. Conclusion Urine [TIMP-2]*[IGFBP7] is a promising candidate for early detection of AKI, especially in ruling-out AKI. However, the potential of this biomarker should be validated in larger studies with a broader spectrum of clinical settings.

Introduction were based on MeSH terms combined with free text terms. In addition, the reference lists of the retrieved articles were reviewed to identify additional studies. The searches were performed independently by two investigators (YS and YW).

Study selection
Clinical studies in English of urine [TIMP-2] Ã [IGFBP7] as biomarkers of AKI were chosen and a 2×2 quadrant table was created. Two reviewers (YS and YW) reviewed the titles and abstracts of all citations and then read the full text of articles deemed relevant.

Data extraction and quality assessment
The following information was extracted by two reviewers (YS and YW): first author, publication year, country of origin, study design, sample size, population setting, patient age, diagnostic criteria for AKI, blinded or not, method for detection of TIMP-2 and IGFBP7, time of detection, cut-off point, true-positive rate, false-positive rate, false-negative rate, and true-negative rate. When there were several cut-off points in one article, the time recommended in the conclusion or that which considered both summary sensitivity and specificity was chosen. If the time of detection was not clearly expressed, the latest time of specimen collection mentioned in the articles was used.
The methodological quality of each study was independently evaluated by two authors (YS and YT). This meta-analysis employed a quality assessment tool, with QUADAS-2 as a standard [22], which included 11 questions (scored as yes, no, or unclear), was specifically developed for systematic reviews of diagnostic accuracy studies to assess study bias.

Data synthesis and analysis
Meta-Disc (ver. 1.4) software for meta-analysis of test accuracy data was employed in this study. Heterogeneity in meta-analysis indicates the degree of variability in results across studies and was appraised using the Q test, p value, and I 2 index, which revealed thresholds for low (I 2 <50%), moderate (50% I 2 <75%), and high (I 2 !75%) heterogeneity. Sensitivity, specificity, positive and negative likelihood ratios, diagnostic odds ratios, and SROC curves were summarized using a random effects model with the DerSimonian and Liard procedure. A useful predictor of AKI risk was defined as an AUC-ROC of>0.7. The Spearman correlation coefficient of logarithmic sensitivity and logarithm (1-specificity) was used to test the presence of a threshold effect.

Results
Search results and study characteristics 10 prospective cohort studies [12][13][14][15][16][17][18][19][20][21] were selected, which were all published from 2014 to 2016 and included a total of 1709 patients, 328 of whom developed AKI. The characteristics of each of the studies are listed in Table 1. The study cohorts included patients accepted for cardiac surgery and ICU patients. One study enrolled infants. As shown in Table 1, the definition of AKI varied among the studies.

Quality assessment
The Spearman correlation coefficient of these 10 articles was 0.116 (p = 0.751), suggesting no significant threshold effect. The quality of the studies according to QUADAS-2 tool is summarized in Fig 2. The quality of studies included was good. As there were only 10 studies included in the analysis, publication bias was not evaluated.

Data synthesis
The data extracted from the 10 eligible studies are shown in

Subgroup analysis
To evaluate the performance of urine [TIMP-2] Ã [IGFBP7] for AKI diagnosis in different population, we did subgroup analysis. Reviewer judgment of methodological quality of each individual study included in the analysis was performed using the Quality Assessment of Diagnostic Accuracy Studies (QUADAS-2) tool. "-" in red and "+" in green mean high risk and low risk respectively. "?" in yellow means unclear risk. In the index test part, some studies [13][14][15][17][18] were valued as high risk, because these studies had not used setting threshold. Two studies [13][14] did not mentioned blind or not so we valued them unclear risk in the reference standard part. Two studies [20][21] was evaluated high risk in flow and timing part because of non-uniform diagnostic criteria used. As shown in Table 1, nine studies included adults and one included infants. Since one of the ten studies enrolled infants, only an adult subgroup was used for analysis. The sensitivity and specificity of the adult subgroup were almost the same as the summary outcomes. The AUC-ROC of the adult subgroup was 0.8477.
The main populations of the included studies were patients who underwent cardiac surgery and those with serious internal medicine diseases. The subgroup analysis results are presented in Table 3. As indicated, the outcome of the cardiac surgery subgroup was much better than that of the ICU subgroup with DOR of 17.76 and AUC of 0.8897.
The 10 studies adopted different diagnostic criteria. Three studies [13][14][15] used all stages of KDIGO/pRIFLE as AKI diagnosis criteria and other studies adopted stage 2-3 AKIN or KDIGO criteria to define AKI. As different AKI definitions might induce discrepancy, these studies were divided into two subgroups (all stages and stage 2-3 AKI), as shown in Table 3.

Discussion
Early treatment can improve the prognosis or even reverse the acute renal damage in AKI. Various biomarkers of AKI have been evaluated in recent years, including NGAL, KIM-1, L-FABP, and CysC, among others. These recently identified biomarkers seem to behave better as diagnostic values during the early stage of AKI than more conventional biomarkers in some studies, but there are some disadvantages. For example, a study by Kooiman et al. [23] found that KIM-1 and NGAL did not fluctuate in 511 patients monitored by post-intravenous contrast-enhanced computed tomography, including those with contrast-induced AKI (CI-AKI). This result raises concern about the accuracy of NGAI and KIM-1 in CI-AKI. In addition, NGAL is reportedly influenced by many other common clinical factors, including baseline  renal function, inflammation, pregnancy, cancer, etc. [24][25][26][27]. IL-18 was reported to be affected by inflammation [28]. Furthermore, the concentrations of NGAL, KIM-1, IL-18 and L-FABP are higher in patients with chronic kidney disease than healthy controls [29][30][31][32][33][34][35], which might interfere in the differentiation between AKI and chronic kidney disease. Hence, these biomarkers are not sufficient to differentially diagnose AKI. TIMP-2 and IGFBP7 both play important roles in blocking stage G1 of the renal tubular epithelial cell cycle in the early stage of AKI. With involvement in the early pathophysiological changes of AKI, TIMP-2 and IGFBP7 are considered as promising candidate early biomarkers of AKI. Kashani et al. [11] evaluated the utility of more than 300 urine biomarkers of AKI performance and found that the most accurate was urine [TIMP-1] Ã [IGFBP7]. At present, several other clinical trials regarding urine [TIMP-2] Ã [IGFBP7] as a diagnostic indicator of AKI by different etiologies are underway. In the present study, a meta-analysis was conducted to further evaluate the diagnostic values of urine TIMP-2 times IGFBP7 in AKI. Part of the work of this meta-analysis has been reported as an abstract [36]. To further investigate the value of [TIMP-2] Ã [IGFBP7] for diagnosis of AKI, among different patient populations and different ages, several subgroup analyses were performed. As shown in Table 3, ICU patients have diverse cause of AKI and more complications, which increases the risk of exposure to potential confounding factors, which may explain why urine [TIMP-2] Ã [IGFBP7] was more accurate in cardiac surgery patients, who usually have a relatively simple pathogenesis.
The AKIN, RIFLE and KDIGO criteria are dependent on serum creatinine and urine output, although neither is considered a gold standard. So, at present, there is no uniform diagnostic criterion for use in clinic trials. In this meta-analysis, each of the 10 included studies used different criteria. The results of this meta-analysis showed that the KDIGO/pRIFLE criteria had an AUC-ROC value of 0.9088 for all stages of AKI with good sensitivity. The specificity was much better than any other subgroup, suggesting that urine [TIMP-2] Ã [IGFBP7] performed well even in the early stage of AKI. However, this subgroup only contained three studies with relatively small sample sizes, thus this result might not be very reliable.
The results of adult subgroup analyses according to patient age showed a very similar AUC to that of the overall analysis. Considering only one study recruited non-adults, this subgroup design might not be appropriate. The time point of detection ranged from the time of enrollment to about 1 day which suggests that urine [TIMP-2] Ã [IGFBP7] was able to predict AKI in the very early period after injury onset.
The summary specificity of this meta-analysis was 0.57, which is not very good. So, [TIMP-2] Ã [IGFBP7] may be more useful to rule-out AKI.
There were some limitations to this study that should be addressed. First, only articles published in English were included, thus language bias inevitable. In the process of searching the databases, several studies without a 2×2 quadrant table were excluded. Among these, one study [37] had less optimistic results (small sample size, patients number 15, and lack of definite information), thus some studies with unpublished negative results were omitted. As shown in Table 1, all of the included studies were conducted in Europe, especially in Germany, or North America. Besides, the study populations were restricted to patients selected for cardiac surgery and those with severe diseases. So, further studies with other etiologies, such as CI-AKI, other races, and larger cohorts are needed to verify the value of [TIMP-2] Ã [IGFBP7] for the diagnosis of AKI. Also, oliguria might impede urine sampling, which prevented the assessment of urine TIMP-2 and IGFBP7. Subsequent studies should also consider other possible interfering factors of urine TIMP-2 and IGFBP7.

Conclusion
The result of this meta-analysis demonstrated that the AUC-ROC for diagnostic accuracy of urine [TIMP-2] Ã [IGFBP7] for AKI was good. The summary AUC-ROC, sensitivity and specificity of this meta-analysis were 0.8813, 0.84 and 0.57, respectively.
Supporting Information S1 Text. PRISMA statement of the meta-analysis. (DOC)