Predictive value of serum cystatin C for acute kidney injury in adults: a meta-analysis of prospective cohort trials

The role of serum cystatin C (Scys) for the detection of acute kidney injury (AKI) has not been fully discussed. This meta-analysis was aimed to investigate the overall diagnostic accuracy of Scys for AKI in adults, and further identify factors affecting its performance. Studies before Sept. 2016 were retrieved from PubMed, Embase, Web of Science and the Cochrane Library. A total of 30 prospective cohort studies (involving 4247 adults from 15 countries, 982 patients occurring AKI) were included. The revised Quality Assessment for Studies of Diagnostic Accuracy (QUADAS-2) tools demonstrated no significant bias had influenced the methodological quality of the included studies. Scys showed a high predictive power for all-cause AKI, that the area under the receiver operating characteristic curve was 0.89. The detailed assessment parameters, such as sensitivity, specificity, positive likelihood ratio, negative likelihood ratio and diagnostic odds ratio for Scys were 0.82, 0.82, 4.6, 0.22 and 21, respectively. Although Scys could be slightly influenced by the following factors: settings, AKI diagnostic criteria, ethnicity, determination method, age and gender, these factors above did not reach statistically significance. In conclusion, Scys could be a vital promising marker to screen out AKI.

Influence factors affecting Scys of AKI. Various Scys blood sampling point-in-time, cut-off value, and determination method resulted in various Scys predictive value for AKI by subgroup analysis (Tables 1,2,3,4). Foremost, Twenty-four hours after AKI occurrence to adopt the blood seems to be an optimal time, with sensitivity of 0.82, specificity of 0.83, DOR of 23, and AUROC of 0.89 (Table 4). Besides, 50% elevated from baseline could be an ideal cut-off value to predict AKI, with AUROC 0.99 (Table 2). Last but not the least, PETIA performed better than other two determination methods, with sensitivity 0.76, specificity 0.87 and AUROC 0.90 (Table 3). In addition, several factors other than Scys assay itself were also analyzed in this study, such as AKI  (Table 3). However, these factors mentioned above were not the origin of possible sources of heterogeneity by meta-regression analysis.
Publication bias. No publication bias and high symmetry of the included studies were proved by Deeks' funnel plot asymmetry test (P = 0.72; Fig. 5).

Discussion
The overall AKI incidence in this study was 23.1% (982/4247), similar to the prevalence reported in Siew' study4, indicating the disease is still not in control and prevented. The mean age in this meta-analysis achieved 61 years old, demonstrating more attention should be taken on the susceptible population: the elderly. Various setting, various AKI incidence. The top three settings prone to AKI were cardiovascular surgery, ICU/CCU and radiology intervention department.   Facing to the severe reality, early diagnosis is crucial to prevent and relieve the prognosis of AKI. Scys has been known to be an ideal marker to assess renal function in CKD patients 46,47 . Whether it is a satisfactory marker to predict AKI is still in debate. Thus, to comprehensively and objectively evaluate the value of Scys predicting AKI, this meta-analysis set a rigorous inclusion and exclusion criteria at the very start. One of the essential selected condition should be prospective cohort studies. After literature searching, 30 studies finally were included. The pooled sensitivity, specificity and AUROC of Scys was 0.82, 0.82 and 0.89, respectively. These diagnostic efficiency demonstrated that Scys would be an excellent bio-marker for the all-cause AKI prediction.
Further subgroups analysis indicated several influence factors should be noted. Different Scys blood sampling point-in-time, cut-off value, and determination method, different Scys predictive value for AKI. If an AKI event would occur, it could be suggested that Scys should be determined by PETIA method at 24-hours after the possible AKI event, referring the diagnostic criteria-50% elevated from baseline. Compared with the previous studies results, this advice is rational and acceptable. The blood sampling point-in-time was another focus. Among the various time point, 24-hours point after AKI might be a preferable selection.
Otherwise, factors potentially influencing Scys were also assessed in this study. Three main criteria to diagnose AKI were presented in Table 3. The newly KDIGO criteria performed an increased diagnostic accuracy in Thomas' study 48 . The subgroup analysis in this study also confirmed its superiority, that the newly KDIGO criteria showed higher specificity and AUROC than the RIFLE criteria and AKIN criteria.
As reported, the most common cause of AKI is acute tubular necrosis (ATN), which could be caused by prolonged hypotension, sepsis, surgery, nephrotoxic medications, and contrast media in hospitalized patients 49,50 . Among the three main causes of AKI in this meta-analysis, Scys performed the best accuracy in CIN-AKI. The probable reason might be that kidney injury and hemodynamic disorder induced by CIN-AKI is less serious and complicate than that by CS-AKI and ICU-AKI. CIN could be the most simple, but also the most important AKI model to ascertain the value of Scys. To our knowledge, CIN is the third leading cause of AKI in hospitalized patients 51 Table 3. Pooled diagnostic accuracy of Scys in various AKI subgroup studies. Abbreviations: AKI, acute kidney injury; AUROC, the area under the receiver operating characteristic curve; CCU, coronary care unit; CIN, contrast-induced nephropathy; CS, cardiac surgery; DOR, diagnostic odd ratio; ELISA, enzyme-linked immunosorbent assay; KDIGO, Kidney Disease: Improving Global Outcomes; PENIA, particle-enhanced nephelometric immunoassay; PETIA, particle-enhanced turbidimetric immunoassay; Pts, patients; RIFLE, riskinjury-failure-loss-end stage renal disease; Scr, serum creatinine; Scys, serum cystatin C.
promising bio-markers 53,54 . However, the former two biomarker determination method have not yet been established in clinical laboratories. Thus, according to the results of this study, Scys could be the optimal marker predicting various AKI.  It should be mentioned in the end, the same as the previous CKD studies proved 55 , Scys was not significantly influenced by gender and age in this AKI-related study, as well. Moreover, although settings, AKI diagnostic criteria, race and assay method might play a little bit of influence on the accuracy of Scys, it did not reach statistical significance. Thus, these results above showed that Scys could be a nice marker, not only for CKD diagnosis, but for AKI prediction.
For all meta-analyses, heterogeneity is a potential problem when interpreting the results. The I 2 statistic was 96% in our meta-analysis, indicating significant heterogeneity across the included studies. One major source of heterogeneity is the threshold effect in which different cut-offs are used in the included studies. The Spearman correlation analysis in our study indicated no threshold effect related heterogeneity exit. Furthermore, meta-regression analysis results revealed that factors potentially affecting Scys did not participate in the heterogeneity (p > 0.05; Table 3). Thus, we considered that the heterogeneity may be related to additional factors, such as specified ethnicity (except from the two race in this study), kidney function, and etc. However, these factors is difficult to unify and analyze.
In summary, this meta-analysis demonstrated that Scys shows a good diagnostic performance for predicting all-cause AKI. Several factors could affect the predictive value of Scys for AKI, but not reach significant differences. More randomized controlled trials in multicenter are in need to further investigate the accuracy of Scys.

Methods
Data sources and search strategy. In accordance with the Preferred Reporting Items for Systematic reviews and Meta-Analyses (PRISMA) guidelines 56 , we searched PubMed, EMBASE, Web of Science and the Cochrane Library from the inception to September 2016.
The following terms were used: "AKI, acute kidney injury, acute renal failure, acute renal insufficiency, acute renal dysfunction and cystatin C". References of the selected studies were further screened manually to identify whether additional eligible articles were available or not.

Study selection.
The inclusion criteria of this study were composed of the following characteristics: (1) prospective cohort study, (2) adults, (3) sample size ≥ 30, (4) original data of sensitivity and specificity, (5) AKI diagnostic criteria. If any disagreement existed, two investigators would check and discuss about the full text.
Authors were contacted when there were incomplete or missing data. Ethics approval and patients consent were not in need for this study.
Data extraction and quality assessment. Two investigators (Z.Z.Y. and X.H.P.) independently extracted information from each article using a standardized collection form. Collected parameters included the first   author, publication year, clinical setting, region, age, gender, AKI diagnostic criteria, Scys determination method, Scys cut-off value, sensitivity and specificity. Differences were resolved by consensus or the third researcher (W.H.Z.). We investigated the methodological quality of the present study using the second version of the Quality Assessment of Diagnostic Accuracy Studies (QUADAS-2) 57 . QUADAS-2 assesses the risk of bias and applicability in four domains: Patient selection (consecutive or random sample enrolled, case-control design and inappropriate exclusions avoided); Index test (blinded interpretation of the Rules); Reference standard (correctly excluded a fracture and blinded interpretation); and Flow and timing (appropriate interval between application of the Rules and reference standard, all patients received the reference standard and were included in the analysis).

Statistical analysis.
A bivariate meta-analytic approach was used to pool sensitivity, specificity, DOR, PLR, and NLR. Subsequently, the respective hierarchical summary receiver operating characteristic (HSROC) curves was constructed to plot sensitivity versus specificity, and then calculate the area under the curve. The highest Youden index (sensitivity + 1-specificity) of every included studies was chosen to end pooled in various Scys measurement times 58 . We used the I 2 statistic to evaluate the heterogeneity 59 , and the I 2 > 75% is supposed of significant heterogeneity, the threshold analysis and meta-regression analysis were further used to identify possible sources of heterogeneity. Publication bias was estimated by Deeks' funnel plot asymmetry test 60 . All the data processing and analysis were performed using the midas and metandi commands of Stata/SE version 12.0 (Stata Corp LP, College Station, TX) and Meta-Disc 1.4 for Windows (XI Cochrane Colloquium, Barcelona, Spain). QUADAS-2 quality assessment was descriptively analyzed using Review Manager 5.3 (The Cochrane Collaboration, Copenhagen, Denmark). P < 0.05 was considered of statistical significance.