(**B**)

**Figure 2.** Generalized additive model (GAM) plot for the probability of (**A**) 90-day mortality, and (**B**) 90-day composite outcome against serum cFGF-23 levels at initiation of dialysis. Note: The GAM plot was incorporated with the subject-specific (longitudinal) random effects expressed as the logarithm of the odds (logit). The probability of outcome events was constructed with cFGF-23 levels averaging zero over the range of the data, i.e., cFGF-23 = 2050 ng/mL. All the relevant covariates, including characteristics, comorbidities, laboratory data, at intensive care unit (ICU) admission, etiology of acute kidney injury (AKI), indication for dialysis, dialysis modality, SOFA score, and plasma cFGF-23 at dialysis, and some of their interactions, such as interventions listed in Table 1, were put on a selected variable list to predict the outcome of interest.

#### *3.4. Plasma cFGF-23 and Outcome*

Using a cut-off value of 2050 RU/mL, patients were divided in a "high" and a "low" cFGF-23 group. Subjects with high cFGF-23 had lower baseline sCr, but higher phosphate concentrations at dialysis initiation, higher in-hospital and 90-day mortality, lower dialysis weaning rate and higher composite outcome results (Table 2).


**Table 2.** Clinical characteristics of patients with high versus low plasma cFGF-23 levels.


**Table 2.** *Cont.*


**Table 2.** *Cont.*

**Abbreviations:** AKI, acute kidney injury; APACHE; Acute Physiology and Chronic Health Evaluation, BMI, body mass index; CABG, coronary artery bypass graft; Cre, creatinine; BUN, blood urea nitrogen; COPD, chronic obstructive pulmonary disease; CPB, cardiopulmonary bypass; Cre, creatinine; CVA, cerebrovascular accident; CVVH, continuous venovenous hemofiltration; eGFR, estimated glomerular filtration rate; FGF-23, Fibroblast growth factor-23; GCS, Glasgow Coma Scale; IABP: intra-aortic balloon pump; IE, inotropic equivalent; ICU, intensive care unit; IHD, intermittent hemodialysis; KIM-1, Kidney Injury Molecule-1; LVEF, Left ventricular ejection fraction; MDRD, Modification of Diet in Renal Disease; MODS, Multiple Organ Dysfunction Syndrome; NGAL, neutrophil gelatinase-associated lipocalin; SLED, sustained low efficiency dialysis; SOFA, Sequential Organ Failure Assessment; Vit D, vitamin D.

A high cFGF-23 level represented an independent risk factor for in-hospital mortality (OR, 1.80, *p* = 0.049), composite outcome at discharge (OR, 1.80, 95% CI = 1.01–3.24; *p* = 0.043), 90–day mortality (OR, 2.19, 95% CI = 1.20–4.00; *p* = 0.011), and 90-day composite outcome (OR, 2.39, 95% CI = 1.31–4.35; *p* = 0.005) after adjusting for age, gender, baseline eGFR, and factor interaction with cFGF-23 and SOFA score. Importantly, no interaction was observed between the cFGF-23 level and underlying diabetes mellitus, baseline eGFR, age, and AKI risk predicting score at dialysis initiation (all *p* > 0.05) (Table 3).

**Table 3.** Logistic regression model for mortality and composite outcomes at hospital discharge and 90 days after discharge. Significant risks were shown.


**Abbreviations**: cFGF-23, c-terminal fibroblast growth factor-23; CI, confidence interval; HR, hazard ratio; SOFA, Sequential Organ Failure Assessment. All the univariate significant and non-significant relevant covariates, including age, sex, baseline comorbidities, indication for dialysis, etiology of AKI, kidney function profile (e.g., baseline eGFR and candidate biomarkers), cFGF-23 and SOFA score at dialysis initiation, dialysis modality, and some of their interactions were put on the variable lists to be selected (Table 1).

Cox proportional hazard regression analysis revealed that patients undergoing RRT who displayed high cFGF-23 levels had a higher 90-day mortality during the follow-up period with an adjusted HR of 1.76 (95% CI, 1.22–2.53; *p* = 0.020) as compared with patients with lower cFGF-23 values (Figure 3A). There was no interaction of the baseline comorbidities with high cFGF-23 to predict 90-day composite outcome. (Table S2) Taking mortality as a competing risk factor for dialysis, high cFGF-23 levels also predicted less weaning from dialysis in surviving patients (HR, 0.62, *p* = 0.032) (Figure 3B).

**Figure 3.** Cox proportional hazard plots stratified by serum cFGF-23 level for assessing probability of 90-day mortality (**A**) and the weaning from dialysis (**B**) by competing analysis and with mortality as a risk factor. Abbreviations: cFGF-23, c-terminal fibroblast growth factor-23; Using a cut-off value of 2050 RU/mL of cFGF-23 at initializing dialysis, patients were divided in a "high" and a "low" cFGF-23 group; all the relevant covariates, including characteristics, comorbidities, laboratory data, at ICU admission, etiology of AKI, indication for dialysis, dialysis modality, SOFA score, and plasma cFGF-23 at dialysis, and some of their interactions, such as interventions listed in Table 1, were put on a selected variable list to predict the outcome of interest.

The relationship of cFGF-23 with these variables was also underscored by a GAM analysis adjusted for SOFA score, gender, and age, which showed that cFGF-23 levels correlated with iFGF-23 (*p* = 0.013) and SOFA score (*p* < 0.001), but not with sCr (*p* = 0.116), phosphate (*p* = 0.591), 25-hydroxy Vit D (*p* = 0.485) and 1, 25 dihydroxy vit D (*p* = 0.638) concentrations at initiating RRT (Figure S1).

#### *3.5. Addition of cFGF-23 to AKI Risk Predicting Score at Start of Dialysis*

Adding cFGF-23 to the AKI risk predicting score at dialysis initiation significantly increased risk stratification (total NRI = 0.148; 95% CI = 0.057–0.239; *p* = 0.002) for detection of 90-day mortality. This effect was primarily determined by death (NRI event = 0.068, 95% CI = 0.043–0.087; *p* = 0.025) and survival (NRI event = 0.069; 95% CI = 0.039–0.097; *p* = 0.029). Similarly, the total IDI was significant. (0.051, 95% CI = 0.024–0.079; *p <* 0.001) (Figures S2 and S3).

## **4. Discussion**

At initializing dialysis, the discriminative power of AKI biomarkers for 90-day mortality is fair. At dialysis initiation, the discrimination of cFGF-23 is better than NGAL, KIM-1, iFGF-23 and creatinine predicting patients' outcome. With mortality as competing risk, higher cFGF-23 levels also predicted lesser kidney recovery in survivors. More importantly, cFGF-23 had better predictive power than creatinine-adjusted urine NGAL and its integration into the AKI risk predicting score significantly enhanced the accuracy of risk stratification. At a cut-off level above 2050 RU/mL, cFGF-23 could predict of AKI mortality after adjusting for different clinical and disease severity parameters. Thus, cFGF-23 could be used as an early determinant of prognosis in ICU patients subjected at initializing RRT and also as an early determinant of the timing of dialysis initiation.

An increasing body of evidence has shown that cFGF-23 levels are increased in patients with AKI [11,14,29–32]. No significant interaction was observed between cFGF-23 and baseline CKD, sepsis grading in predicting mortality. The SOFA score was independently associated with increased cFGF-23 levels, which underpins the potential use of cFGF-23 in a critical care setting. We dare sugges<sup>t</sup> that a higher plasma cFGF-23 not only corresponds with more severe AKI, but also reflects a higher degree of systemic inflammation.

Several mechanisms may explain increased FGF-23 levels in AKI: (1) increased production by osteocytes and possibly osteoblasts, that escapes regulation by parathyroid hormone, vitamin D signaling, and dietary phosphate restriction [33,34]; (2) increased ectopic production of FGF-23 by damaged renal tubules [33,35]; (3) tubular dysfunction resulting in FGF-23 resistance [36]; (4) and decreased clearance of circulating FGF-23 [14]. Whilst circulating FGF-23 levels rise rapidly during AKI [14] and a causal role for FGF-23 in the pathogenesis of left ventricular hypertrophy has previously been unveiled, suggesting that chronically elevated FGF-23 levels contribute directly to cardiac mortality in patients with CKD [37].

The ideal circumstances for whether and when to start RRT remain unclear [4]. We found significantly elevated cFGF-23 levels at the start of dialysis in non-survivors, whilst other structural and functional renal biomarkers failed to discriminate. Elevated plasma cFGF-23 was related to the degree of organ failure at initializing RRT [33]. In fact, high cFGF-23 concentrations predicted worse outcome equally well as the SOFA score in critically ill patients with advanced AKI [38]. Moreover, in patients without AKI, plasma cFGF-23 levels were significantly higher in the more severely ill patients [14]. This underscores that high cFGF-23 levels are correlated with increased systemic inflammation and/or stress secondary to illness or major surgery [33]. Although both serum and urine cFGF-23 could predict AKI mortality after ICU admission [12], many patients were oliguric at initializing dialysis, that will highlight the role of serum cFGF-23. In surviving patients, high cFGF-23 levels also predicted a lesser possibility for RRT withdrawal. Early prediction of renal recovery is likely to be helpful with regard to post-discharge care after critical illness and subsequent progression to CKD and ESRD.

Taken together, the ability of cFGF-23 to predict adverse outcomes might be related more to the systemic inflammatory status than to tubular damage. Based on our findings, a prognostic model can be constructed that allows to predict individual mortality risk as well as potential kidney recovery in surviving patients before starting RRT. The addition of cFGF-23 to a clinical AKI risk predicting score resulted in greater discrimination, and enhanced the ability to anticipate a higher number of subsequent deaths. Given the lack of appropriate or reliable biomarkers in patients receiving RRT, plasma cFGF-23 tentatively may serve as a novel outcome-specific marker in critical care nephrology. In patients with augmented plasma cFGF-23 concentration to arrive 2050 RU/mL, the clinician should evaluate the traditional AKI risk score or parameters to decide commencing dialysis.

Whether the cFGF-23 assay provides comparable sensitivity to that for iFGF-23 in patients with different stages of AKI or illness severity is still debated [13]. Although measurements obtained with iFGF-23 and cFGF-23 assays reflect the same circulating moiety, it has been suggested that the levels of iFGF23 also increased in patients who developed severe AKI, but the magnitude was lower than cFGF23 [13]. This is also supported by the present study showing that a plasma cFGF-23 concentration exceeding 2050 RU/mL at initializing RRT was significantly associated with worse patient outcome at a higher discriminative power than iFGF-23. The levels of adjusted urine cFGF23 also increased in patients who did not survive, but the magnitude was lower than serum cFGF23.

Several limitations of our study must be highlighted. Our cFGF-23 cutoff value was somewhat higher than that in other AKI studies [11–15], probably because most patients already had advanced AKI when admitted to the ICU. Furthermore, the predicting power of cFGF-23 in patients without AKI but with high inflammation status needs further validation. Finally, the exact mechanism underlying increased cFGF-23 concentrations in AKI patients as well as possible other intrinsic biological effects of cFGF-23 in this particular population remain to be explored. As previously studies few biomarkers were ever validated and they could only modestly predictive of renal recovery [8]; we do acknowledge also that the AUCs of cFGF-23 were relatively modest in AKI-D patients with critical status, however adding cFGF-23 to a parsimonious model augmented prediction of mortality and kidney recovery.

## **5. Conclusions**

At initializing dialysis, the discriminative power of AKI biomarkers for 90-day mortality is fair. Our study showed that cFGF-23, measured at initiation of RRT in critical patients with AKI, may be a novel and distinct marker for predicting 90-day mortality after discharge and less weaning from RRT in survivors. Its predictive discrimination was superior to other established biomarkers of kidney injury, in particular creatinine, NGAL and Kim-1. Adding cFGF-23 to the traditional AKI risk predicting score may allow better risk stratification and enhance prognostic power. cFGF-23 could further be used as a surrogate marker to decide the best timing to initiate RRT.

**Supplementary Materials:** The following are available online at http://www.mdpi.com/2077-0383/7/8/202/s1, Figure S1: Scatter plots with an adjusted spline of cFGF23 with (A) iFGF23 (*p* = 0.013), (B) phosphate (*p* = 0.591) (C) creatinine (*p* = 0.116) (D) 25 OH Vitamin D (*p* = 0.485) and (E) 1,25 OH, Vitamin D (*p* = 0.638) (F) KDIGO-AKI score (*p* = 0.820) (G) SOFA (*p* < 0.001) at initiation of dialysis, Figure S2: Decision curve analysis (DCA) plot to assess 90 day mortality using cFGF-23 in addition to AKI risk prediction score, Figure S3: The correlation of AKI risk predicting score and AKI risk predicting score with cFGF-23 predicting 90 day mortality, Table S1: *p* value comparison of the receiver operating characteristic (ROC) curve for discriminative ability, Table S2: Interaction of baseline co-morbidity with high cFGF-23 to predict 90-day composite outcome.

**Author Contributions:** V.-C.W., C.-C.S., S.-C.J.C., H.-H.L. and T.-S.C. conceived the review topic, analysis and interpretation and wrote the manuscript. H.D.S., N.-H.C., C.-H.W. and P.M.H. revised and approved the final version of the manuscript.

**Funding:** This study was supported by Taiwan National Science Council (grants NSC 102-2314-B-002-140-MY2, 104-2314-B-002-125-MY3, 106-2314-B-002-166-MY3, MOST 106-2321-B-182-002) and NTUH 106-FTN20, 106-P02, UN106-014, 107-T02.

**Acknowledgments:** We also express our sincere gratitude to all staff of the Taiwan Clinical Trial Consortium, TCTC, and extracorporeal membrane oxygenation team in NTUH. All authors read and approved the final version of the manuscript.

**Conflicts of Interest:** The authors declare no conflict of interest.
