Fibroblast Growth Factor 23 Levels Associate with AKI and Death in Critical Illness

Abstract

Elevated plasma levels of the osteocyte-derived hormone fibroblast growth factor 23 (FGF23) have emerged as a powerful biomarker of cardiovascular disease and death in patients with CKD. Whether elevated urinary or plasma FGF23 levels are prospectively associated with AKI and death in critically ill patients is unknown. We therefore conducted a prospective cohort study of 350 critically ill patients admitted to intensive care units at an academic medical center to investigate whether higher urinary FGF23 levels associate with the composite end point of AKI or in-hospital mortality (AKI/death). We measured urinary FGF23 levels within 24 hours of admission to the intensive care unit. In a subcohort (n=131) we also measured plasma levels of FGF23, calcium, phosphate, parathyroid hormone, and vitamin D metabolites. Urinary and plasma FGF23 levels, but not other mineral metabolites, significantly associated with AKI/death. In multivariate analyses, patients in the highest compared with the lowest quartile of urinary FGF23 had a 3.9 greater odds (95% confidence interval, 1.6 to 9.5) of AKI/death. Higher urinary FGF23 levels also independently associated with greater hospital, 90-day, and 1-year mortality; longer length of stay; and several other important adverse outcomes. In conclusion, elevated FGF23 levels measured in the urine or plasma may be a promising novel biomarker of AKI, death, and other adverse outcomes in critically ill patients.

AKI is a common complication of critical illness and is associated with markedly increased hospital length of stay, mortality, and cost.^1–3 AKI is frequently complicated by disordered mineral metabolism,^4–6 with elevations in circulating levels of the osteocyte-derived phosphaturic hormone fibroblast growth factor 23 (FGF23) and decreased levels of 25-hydroxyvitamin D (25D) and 1,25-dihydroxyvitamin D (1,25D).

Elevated FGF23 levels are strongly associated with adverse outcomes in patients with AKI. In a pilot study of 30 patients with established AKI, we reported that higher plasma FGF23 levels are associated with an increased risk of RRT or death (RRT/death).⁴ In a more recent study we reported that plasma FGF23 levels rise early and predict AKI and death in patients undergoing cardiac surgery.⁵ Mechanistically, these associations may be related to “off-target” effects of FGF23, such as impairment of neutrophil function⁷ or increased hepatic production of inflammatory cytokines.⁸ Alternatively, these associations may relate to FGF23-mediated inhibition of renal and monocyte conversion of 25D to 1,25D and enhanced catabolism of 25D and 1,25D.^9,10 FGF23 thereby lowers 25D and 1,25D levels, which are in turn associated with heightened risk of infection.^11,12

In contrast to the above studies, limited data exist on the prospective association between FGF23 and incident AKI among critically ill patients. Additionally, although plasma FGF23 has emerged as a powerful biomarker of cardiovascular disease and death in patients with CKD,^13,14 few studies have measured urinary FGF23 levels, and none, to our knowledge, have done so in AKI. We therefore measured urinary FGF23 levels, as well as plasma FGF23, parathyroid hormone (PTH), vitamin D metabolites, vitamin D–binding protein (DBP), and other markers involved in mineral metabolism, to evaluate their prospective association with AKI and death in a cohort of critically ill patients.

Results

Patient Characteristics

We enrolled 350 patients admitted to intensive care units (ICUs) at Brigham and Women’s Hospital (Boston, MA) into a prospective cohort study. We collected and stored urine and EDTA plasma aliquots within 24 hours of arrival to the ICU. Samples were available as follows: urine, n=338 (97%); plasma, n=131 (37%); both urine and plasma, n=122 (35%), respectively.

Median (interquartile range [IQR]) age was 62 (53–71) years, and 45% were women. The most common comorbidities were hypertension (55%), active malignancy (38%), chronic obstructive pulmonary disease (28%), and diabetes mellitus (24%). Median (IQR) Acute Physiology and Chronic Health Evaluation (APACHE) II score was 20 (14–26), 78% required mechanical ventilation, and 21% had undergone urgent surgery. Urinary C-terminal FGF23 (cFGF23) levels ranged from 0.8 to 2049 RU/ml, and urinary cFGF23 levels normalized to the urinary creatinine concentration (FGF23/Cr) ranged from 2 to 19,727 RU/mg. Additional baseline characteristics are shown in Table 1.

Table 1. - Baseline characteristics

Characteristic	All (n=350)	AKI on Enrollment (n=98)	No AKI on Enrollment (n=252) uFGF23/Cr Quartilesa				P Value
			Q1 (n=61)	Q2 (n=61)	Q3 (n=61)	Q4 (n=60)
Urinary cFGF23/Cr, RU/mg
Range (min, max)	2, 19727	14, 19727	2, 21	21, 44	44, 134	135, 12084
Median (IQR)	67 (28–248)	267 (68–2133)	14 (11–19)	30 (27–36)	81 (57–105)	344 (186–1349)	<0.001
Urinary cFGF23, RU/ml
Range (min, max)	0.8, 2049	13, 2049	0.8, 39	9, 55	15, 148	18, 1957
Median (IQR)	33 (22–112)	114 (39–1720)	19 (15–23)	23 (19–28)	29 (25–43)	141 (38–513)	<0.001
Demographic
Median age (IQR), yr	62 (53–71)	65 (58–73)	60 (51–65)	58 (48–69)	65 (58–71)	62 (54–68)	0.03
Women, n (%)	156 (45)	49 (50)	18 (30)	24 (39)	33 (54)	32 (53)	0.02
White race, n (%)	308 (88)	92 (94)	51 (84)	55 (90)	54 (89)	56 (93)	0.39
Baseline renal function
Median SCr (IQR), mg/dlb	0.8 (0.6–1.0)	0.9 (0.7–1.1)	0.7 (0.6–0.9)	0.7 (0.5–0.9)	0.7 (0.6–1.0)	0.8 (0.7–0.9)	0.09
Median eGFR (IQR), ml/min per 1.73 m2c	93 (75–105)	83 (66–95)	101 (89–111)	103 (85–115)	93 (75–105)	89 (75–100)	<0.001
CKD, n (%)d	36 (10)	23 (23)	1 (2)	1 (2)	6 (10)	5 (8)	0.08
Comorbidities, n (%)
Hypertension	191 (55)	29 (30)	34 (56)	31 (51)	33 (54)	27 (45)	0.65
Diabetes mellitus	83 (24)	8 (8)	12 (20)	11 (18)	18 (30)	13 (22)	0.43
Congestive heart failure	42 (12)	22 (22)	3 (5)	6 (10)	8 (13)	3 (5)	0.28
Active malignancy	134 (38)	43 (44)	25 (41)	21 (34)	24 (39)	24 (40)	0.91
COPD	98 (28)	31 (32)	11 (18)	20 (33)	22 (36)	14 (23)	0.10
Surgical ICU, n (%)	128 (37)	22 (22)	36 (59)	27 (44)	20 (33)	23 (38)
Severity of illness
Median APACHE II score (IQR)e	20 (14–26)	26 (20–32)	16 (12–20)	17 (14–23)	17 (14–22)	19 (14–24)	0.14
Urgent surgery, n (%)	75 (21)	20 (20)	18 (30)	14 (23)	8 (13)	15 (25)	0.79
Mechanical ventilation, n (%)	272 (78)	76 (78)	56 (92)	48 (79)	42 (69)	50 (83)	0.01

P values were calculated using Kruskal Wallis and chi-squared tests for continuous and categoric variables, respectively, across the four quartiles of uFGF23/Cr. uFGF23, urinary FGF23; Q, quartile; COPD, chronic obstructive pulmonary disease.

^aUrine samples were available in 243 out of the 252 patients who did not have AKI on enrollment.

^bBaseline SCr was defined as the lowest value within 365 days before hospitalization. If no SCr values were available before hospitalization, the lowest SCr during hospitalization (excluding values obtained during RRT) was used as the baseline.¹⁵

^ceGFR was determined using the Chronic Kidney Disease Epidemiology Collaboration (CKD-EPI) equation.¹⁶

^dDefined as baseline eGFR<60 ml/min per 1.73 m².

^eAPACHE II, an ICU severity of illness scoring system ranging from 0 to 71, with higher scores indicating more severe disease.

AKI Definition and Event Rates

The primary end point was the composite of incident AKI or in-hospital mortality (AKI/death). AKI was defined as an increase in serum creatinine ≥0.3 mg/dl within 48 hours, ≥50% in 7 days, or need for RRT.¹⁷ Among the 350 patients enrolled, 98 (28%) had AKI on enrollment, among whom 41 (42%) had stage 1, 33 (34%) had stage 2, and 24 (24%) had stage 3. Among the 252 patients who did not have AKI on enrollment, 80 (32%) developed incident AKI within 14 days, among whom 40 (50%) had stage 1, 14 (18%) had stage 2, and 26 (32%) had stage 3. The median time from enrollment to development of AKI was 2 days (IQR, 1–5 days).

Urine and Plasma Mineral Metabolite Correlations

Urinary cFGF23/Cr correlated directly with plasma cFGF23 (r_s=0.38, P<0.001), phosphate (r_s=0.22, P<0.001), and PTH (r_s=0.25, P<0.01) (Figure 1A), and correlated inversely with serum calcium (r_s=−0.23, P<0.001). Plasma cFGF23 correlated directly with phosphate (r_s=0.29, P<0.001) and inversely with 25D (r_s=−0.21, P=0.02) and 1,25D (r_s=−0.34, P<0.001) (Figure 1B). Additional correlations are shown in Figure 1C.

Figure 1.: Urine and plasma mineral metabolite correlations. (A) Spearman correlations with urine FGF23/Cr; (B) Spearman correlations with plasma FGF23; (C) Spearman correlations for all metabolites. *P<0.05; **P<0.01; ***P<0.001. PO₄, phosphate.

Urine and Plasma Mineral Metabolite Levels

Urinary and plasma mineral metabolite levels are shown in Table 2 in patients who did or did not develop AKI/death, restricted to those who had both urine and plasma samples available (n=122). Patients who developed AKI/death had higher levels of urinary cFGF23/Cr, as well as higher levels of plasma cFGF23 and phosphate, compared with patients who did not develop AKI/death. Levels of other mineral metabolites were similar between the two groups (Table 2).

Table 2. - Urine and plasma mineral metabolite levels and AKI/death

Variable	All (n=122)a	No AKI/Death (n=75)b	AKI/Death (n=38)b	P Value
Urinary measurements
cFGF23/Cr, RU/mg	37 (20–131)	28 (17–56)	80 (30–217)	<0.001
cFGF23, RU/ml	26.4 (20.2–42.7)	24.1 (19.3–33.7)	28.4 (23.4–111.4)	0.01
Plasma measurements
cFGF23, RU/ml	266 (148–593)	205 (112–480)	476 (257–887)	<0.001
Phosphate, mg/dl	3.3 (2.7–4.1)	3.0 (2.6–3.9)	3.6 (3.1–4.1)	0.02
PTH, pg/ml	41 (28–56)	38 (23–54)	46 (33–55)	0.08
Total calcium, mg/dl	8.1 (7.6–8.6)	8.2 (7.6–8.6)	8.1 (7.8–8.8)	0.90
Ionized calcium, mmol/L	1.1 (1.1–1.2)	1.1 (1.1–1.2)	1.1 (1.1–1.2)	0.83
25D, ng/ml	11.2 (8.2–15.1)	11.9 (8.4–15.2)	10.9 (7.0–15.2)	0.64
1,25D, pg/ml	26.0 (17.9–43.7)	26.4 (19.4–45.2)	22.3 (15.5–45.3)	0.28
24,25D3, ng/ml	0.8 (0.5–1.4)	0.9 (0.5–1.4)	0.7 (0.4–1.2)	0.32
DBP, mg/dl	13.0 (9.4–17.0)	13.4 (9.4–17.7)	12.0 (9.7–16.8)	0.61

Levels are shown as median (IQR).

^aRestricted to patients who had both plasma and urine samples available.

^bFurther restricted to patients who did not have AKI on enrollment.

Urinary cFGF23/Cr (Quartiles) and AKI/Death

The prospective association between quartiles of urinary cFGF23/Cr and risk of AKI/death is shown in Figure 2. In these analyses, patients with established AKI on enrollment were excluded. Patients with urinary cFGF23/Cr levels in the highest compared with lowest quartile had a greater risk of AKI/death (odds ratio [OR], 4.8; 95% confidence interval [95% CI], 2.2 to 10.3). After adjusting for potential confounders (model 1 is adjusted for age, sex, baseline eGFR, hypertension, and diabetes mellitus; model 2 is further adjusted for ICU type, mechanical ventilation, and APACHE II score), the association was slightly attenuated but remained significant (Figure 2). Other quartiles of urinary cFGF23/Cr were not significantly associated with AKI/death.

Figure 2.: Highest urinary cFGF23/Cr quartile associates with increased risk of AKI/death. Model 1 is adjusted for age, sex, baseline eGFR, hypertension, and diabetes mellitus. Model 2 is further adjusted for ICU type (surgical versus medical), mechanical ventilation, and APACHE II score. This analysis was restricted to patients who did not have AKI on enrollment (n=243, events=105). *P<0.05.

Urinary cFGF23/Cr (Continuous) and AKI/Death and Other Adverse Outcomes

Higher urinary cFGF23/Cr levels, assessed as a continuous variable, were associated with an increased risk of AKI/death in both unadjusted and adjusted analyses (Table 3). Higher urinary cFGF23/Cr levels were also associated with the following secondary end points: AKI, severe AKI (defined as doubling of serum creatinine or need for RRT), RRT/death, severe sepsis or septic shock, hospital mortality, 90-day mortality, and 1-year mortality. Finally, urinary cFGF23/Cr was inversely associated with hospital-free days and ventilator-free days, implying longer duration of hospitalization and mechanical ventilation (Table 3).

Table 3. - Urinary cFGF23/Cr levels and adverse outcomes

End Points	Eventsa	Unadjusted				Model 1				Model 2
		OR (95% CI)	β	SEM	P Value	OR (95% CI)	β	SEM	P Value	OR (95% CI)	β	SEM	P Value
Primary end point
AKI/Death	105	2.12 (1.50 to 3.00)			<0.001	1.85 (1.27 to 2.70)			0.001	1.88 (1.26 to 2.81)			0.002
Secondary end points
AKI	80	2.21 (1.56 to 3.13)			<0.001	1.95 (1.33 to 2.85)			<0.001	1.91 (1.28 to 2.85)			0.002
Severe AKIb	37	2.20 (1.49 to 3.27)			<0.001	2.30 (1.49 to 3.55)			<0.001	2.17 (1.37 to 3.43)			<0.001
RRT/death	63	1.85 (1.31 to 2.60)			<0.001	1.85 (1.28 to 2.68)			0.001	1.82 (1.21 to 2.75)			0.004
Severe sepsis or septic shock	194	1.52 (1.20 to 1.92)			<0.001	1.50 (1.16 to 1.93)			0.002	1.35 (1.01 to 1.80)			0.04
Hospital mortality	90	1.58 (1.25 to 2.01)			<0.001	1.61 (1.25 to 2.09)			<0.001	1.46 (1.09 to 1.97)			0.01
90-d mortality	118	1.51 (1.21 to 1.89)			<0.001	1.45 (1.14 to 1.85)			0.003	1.31 (1.00 to 1.72)			0.05
1-yr mortality	161	1.51 (1.20 to 1.90)			<0.001	1.59 (1.24 to 2.04)			<0.001	1.44 (1.10 to 1.90)			0.01
Ventilator-free days	N/A		−2.28	0.62	<0.001		−2.42	0.67	<0.001		−1.78	0.67	0.01
Hospital-free days	N/A		−2.32	0.44	<0.001		−2.46	0.47	<0.001		−2.27	0.50	<0.001

ORs are shown per 1 unit SD of log-transformed values. Model 1 is adjusted for age, sex, baseline eGFR, hypertension, and diabetes mellitus. Model 2 is further adjusted for ICU type (surgical versus medical), mechanical ventilation, and APACHE II score. RRT/death, RRT or in-hospital mortality; N/A, not applicable.

^aAnalyses that include AKI or RRT as an end point are restricted to patients who did not have AKI on enrollment (n=243). All other analyses include the entire cohort (n=350).

^bSevere AKI was defined as doubling of serum creatinine or need for RRT.¹⁷

In a sensitivity analysis, we examined the above associations without normalizing urinary cFGF23 levels to the urinary creatinine concentration. As shown in Supplemental Table 1, the association between urinary cFGF23 and AKI/death was largely unchanged in both unadjusted and adjusted analyses. Similarly, the associations with most of the secondary end points were also unchanged, although hospital, 90-day, and 1-year mortality were no longer significant in model 2.

In an exploratory analysis, we examined whether urinary cFGF23 may be a nonspecific marker of proteinuria. As shown in Supplemental Figure 1, urinary cFGF23/Cr levels were similar across categories of dipstick proteinuria. Additionally, adjustment for dipstick proteinuria in multivariable models had no effect on the association between urinary cFGF23/Cr and risk of AKI/death (data not shown).

In an additional exploratory analysis, we examined whether urinary cFGF23 is associated with mortality independently of AKI status/severity on enrollment, because AKI itself is a powerful predictor of mortality. After adjustment for AKI status/severity on enrollment, urinary cFGF23 levels remained significantly associated with hospital mortality (OR, 1.54; 95% CI, 1.18 to 2.01), 90-day mortality (OR, 1.60; 95% CI, 1.24 to 2.07), and 1-year mortality (OR, 1.63; 95% CI, 1.27 to 2.10).

Urinary cFGF23/Cr among Patients with Established AKI

Patients who already had AKI on enrollment had higher urinary cFGF23/Cr levels compared with those who did not (median [IQR] 267 [68–2133] versus 44 [21–133] RU/mg, respectively; P<0.001). Further, among patients with AKI on enrollment, urinary cFGF23/Cr levels were higher in those with severe AKI, defined as doubling of serum creatinine or need for RRT, compared with those with milder AKI, defined as less than doubling of serum creatinine (Supplemental Figure 2). Finally, among patients with AKI on enrollment, urinary cFGF23/Cr levels had a nonsignificant trend toward greater risk of RRT/death (OR, 1.3; 95% CI, 0.9 to 1.9) and progression to a higher AKI stage (OR, 1.3; 95% CI, 0.9 to 2.0).

Characterization of FGF23 Fragments in Urine

To determine the nature of FGF23 fragments in the urine, Western blot characterization of FGF23 in nine urine samples (n=4 without AKI and n=5 with AKI) was performed (Figure 3A). Urine FGF23 was present in all samples as a 28 kD fragment, and the concentration was similar in patients with versus without AKI (Figure 3B). A smaller fragment (12 kD) was also detected in patients with AKI, but was not observed in patients without AKI (Figure 3C). The full-length, glycosylated form (32 kD) of FGF23 was not detected in any of the samples. Intact FGF23 was also undetectable in any of the samples when measured with the intact FGF23 ELISA assay (data not shown).

Figure 3.: Western blot characterization of FGF23 in urine. (A) Western blot of FGF23 immunoprecipitated from human urine samples (n=4 without AKI and n=5 with AKI) using polyclonal antihuman FGF23 antibodies showing nondetectable intact (32 kD) FGF23 and variable concentrations of FGF23 fragments (28 and 12 kD). (B and C) The band intensity of each detected fragment was quantified as an indicator of its relative concentration. Horizontal bars indicate median levels.

Comparison of Urine and Plasma Mineral Metabolite Levels and Risk of AKI/Death

The risk of AKI/death according to each urinary and plasma mineral metabolite biomarker is shown in Figure 4. In univariate analyses, plasma cFGF23 and urinary cFGF23/Cr levels were both significantly associated with AKI/death (OR per 1 SD higher ln[plasma cFGF23], 3.06; 95% CI, 1.71 to 5.47; OR per 1 SD higher ln[urinary cFGF23/Cr], 2.60; 95% CI, 1.51 to 4.49). We further assessed the association between plasma cFGF23 levels and AKI/death within tertiles of phosphate, calcium, PTH, 25D, and 1,25D. In virtually every tertile of each of these mineral metabolites, an increased plasma cFGF23 level was associated with an increased risk of AKI/death (Supplemental Figure 3). Serum phosphate levels were marginally associated with AKI/death (OR for 1 SD higher ln[serum phosphate], 1.68; 95% CI, 1.01 to 2.78). Other biomarkers, including PTH, total calcium, ionized calcium, 25D, 1,25D, 24,25-dihydroxyvitamin D₃ (24,25D₃), and DBP, were not associated with AKI/death (Figure 4).

Figure 4.: Higher urine and plasma cFGF23 levels associate with increased univariate risk of AKI/death. ORs are shown per 1 unit SD of log-transformed values for each biomarker. This analysis was restricted to patients who had both plasma and urine samples available and did not have AKI on enrollment (n=113, events=38).

In an exploratory analysis, we evaluated whether plasma and urinary cFGF23 levels remained associated with AKI/death after adjustment for each other. As shown in Supplemental Table 2, both plasma and urinary cFGF23 levels remained significantly associated with AKI/death even after adjustment for one another.

Discussion

In this prospective cohort study, we report that higher urinary and plasma cFGF23 levels obtained within 24 hours of admission to the ICU are associated with a greater risk of incident AKI/death, as well as several other adverse events including RRT/death, hospital mortality, and fewer ventilator-free and hospital-free days. Moreover, these findings were independent of a number of potential confounders. In contrast to FGF23, other mineral metabolites such as PTH, calcium, and vitamin D metabolite levels were not associated with AKI/death (except for phosphate, which was marginally associated with AKI/death), consistent with prior studies.^5,6 These findings suggest that elevated urinary and plasma cFGF23 levels may be early predictive markers of AKI and other adverse clinical outcomes in critically ill patients.

Multiple studies demonstrated that elevated plasma FGF23 levels are associated with major cardiovascular events and mortality in patients with CKD and ESRD.^{13,14,18–20} However, data on FGF23 levels in AKI are limited. In contrast to our prior study that investigated the kinetics of plasma FGF23 levels in patients at high risk of AKI after cardiac surgery,⁵ this study evaluated FGF23 as a marker of incident AKI and other adverse outcomes in a more heterogenous group of critically ill patients. Thus, the findings from this study are likely more generalizable. Moreover, in contrast to prior studies, which focused on plasma FGF23 levels only, here we measured both plasma and urinary levels of FGF23. Thus, this study not only demonstrates an independent and prospective association between FGF23 and incident AKI, which extends our preliminary findings in established AKI⁴ and cardiac surgery–associated AKI,⁵ but also represents the first comprehensive study of urinary FGF23 in any clinical setting.

The few prior studies that measured urinary FGF23 levels were almost exclusively conducted among patients with ESRD. Larsson and colleagues measured cFGF23 levels in urine samples from four patients with ESRD on hemodialysis, and found the levels ranged from 750 to 10,790 RU/ml. Western blot analysis of the urine from a single patient revealed a band at 32 kD, corresponding to the full-length FGF23 protein, as well as several lower molecular mass bands, corresponding to C-terminal and other FGF23 fragments.²¹ Wesseling-Perry and colleagues characterized urinary FGF23 levels in four patients before renal transplantation. Urinary cFGF23 levels ranged from 6650 to 77,500 RU/ml, and Western blot analysis indicated that FGF23 was present in the urine predominantly as the full-length form.²²

In contrast to these studies, urinary cFGF23 levels in the current study were much lower (median [IQR] 33 [22–112] RU/ml, Table 1), which likely reflects major differences in patient populations. Further contrasting with prior studies, we found that the majority of urinary FGF23 in AKI consists of 12 and 28 kD FGF23 fragments rather than the full-length 32 kD protein. Whereas the 12 kD fragment likely represents the C-terminal fragment of FGF23, the 28 kD fragment may represent either unglycosylated intact FGF23 or alternatively-cleaved FGF23. Whether deglycosylation or alternative cleavage of FGF23 occurs after glomerular filtration (e.g., by brush border enzymes in the proximal tubules) among critically ill patients is an interesting possibility that will require additional study. Alternatively, it is possible that massive production of FGF23 in critical illness is accompanied by under-glycosylation or other post-translational modifications of the peptide.

We could not discern the source of FGF23 in the urine. Renal FGF23 production has been demonstrated in rat models of polycystic kidney disease²³ and diabetic nephropathy,²⁴ but has not been investigated in AKI. Alternatively, FGF23 in the urine could be derived from glomerular filtration of plasma FGF23. However, the very low levels of FGF23 in the urine relative to plasma suggest that FGF23 may be reabsorbed and/or catabolized by the kidney after being filtered, similar to other peptide hormones²⁵ and consistent with several recent studies.^26,27

In addition to AKI/death, we also evaluated the association between urinary cFGF23 levels and several secondary end points and found significant independent associations with RRT/death; severe sepsis or septic shock; hospital, 90-day, and 1-year mortality; and longer duration of mechanical ventilation and hospital length of stay. Whether FGF23 is simply a severity-of-illness marker or directly contributes to adverse outcomes is a key question that could not be answered by this study. Elevated plasma FGF23 levels have been implicated in the pathogenesis of left ventricular hypertrophy,^28,29 immune dysfunction,⁷ and inflammation⁸ in CKD. Thus, it is conceivable that direct effects of FGF23 on the immune system or other “off-target” effects could provide a mechanistic explanation for our findings. Additionally, elevated FGF23 levels could contribute to adverse outcomes via inhibitory effects on vitamin D metabolite activation.^30–32 Indeed, abundant epidemiologic data link decreased vitamin D metabolite levels with adverse outcomes in critical illness,^33–35 and we found inverse correlations between plasma cFGF23 and both 25D and 1,25D levels. Although we did not find an association between any of the vitamin D metabolites and risk of AKI/death, tissue-level paracrine effects of FGF23 on vitamin D activation and degradation can occur in the absence of major changes in circulating vitamin D levels.³⁶ Thus, the clinical importance of FGF23-mediated inhibition of vitamin D metabolite activation in critical illness remains unclear.

We acknowledge several limitations, including single-center, observational design, and lack of racial/ethnic diversity. We did not have access to data on urine output, which might have allowed earlier identification of AKI and would have been another clinical biomarker with which to compare urinary and plasma cFGF23. Plasma and urinary FGF23 levels were measured at only a single time point, and plasma mineral metabolites were measured only in a subgroup of patients. Finally, although our analyses were adjusted for a large number of covariates, we cannot exclude residual confounding from unmeasured variables.

In conclusion, urinary and plasma cFGF23 levels measured within 24 hours of ICU admission are independently associated with AKI/death in critically ill patients. Other mineral metabolites, including PTH, calcium, and vitamin D metabolites, did not demonstrate an association with AKI/death. Whether FGF23 has acute toxic effects on nontraditional target organs such as the cardiovascular or immune systems and may thereby contribute directly to adverse outcomes in AKI, as it does in CKD, is an intriguing possibility that will require additional study. Specifically, additional studies are needed to determine the physiologic regulators of FGF23 production and metabolism in critical illness/AKI, to determine whether elevated FGF23 levels in critical illness/AKI are directly pathologic, and to determine whether elevated FGF23 levels could serve as a useful prognostic tool in critical illness, either for clinical purposes or for enrichment in randomized controlled trials.

Concise Methods

Study Design

We conducted a prospective cohort study in 350 patients admitted to ICUs at Brigham and Women’s Hospital (Boston, MA) between September of 2008 and January of 2013. We collected and stored urine and EDTA plasma aliquots within 24 hours of arrival to the ICU. Samples were stored at −80°C within 2 hours of collection. All patients provided written informed consent and all protocols were approved by our hospital’s Institutional Review Board.

Enrollment Criteria

Inclusion criteria were age>18 years and admission to a medical or surgical ICU. Exclusion criteria were: (1) anticipated ICU stay <24 hours; (2) admitted to the ICU for a low risk condition such as airway monitoring or serial neurologic checks; (3) serum creatinine >4.5 mg/dl or receiving dialysis; and (4) pregnancy.

Laboratory Analyses

Urine

We measured urinary levels of cFGF23 in duplicate using a commercial ELISA kit (Immutopics; San Clemente, CA). Because limited data are available on measurement of cFGF23 in urine, we tested the analytic performance of the kit on urine samples through a series of assay validation steps, including linearity of dilution, spike and recovery, and interference studies (Supplemental Figure 4). None of the samples had cFGF23 levels below or above the lower or upper limits of detection, respectively. We also measured urinary creatinine levels in duplicate using the modified Jaffe reaction on a Randox Daytona bench top analyzer. We normalized urinary cFGF23 levels to the urinary creatinine concentration (cFGF23/Cr) to account for the degree of urinary dilution.

Immunodetection of FGF23 (ELISA, Immunoprecipitation, and Western Blotting)

The immunometric assay for measurement of cFGF23 uses antibodies directed against epitopes within the C-terminal portion of FGF23, and thus detects both the intact hormone and C-terminal cleavage products.³⁷ Therefore, to determine the nature of FGF23 fragments in the urine, we also used a second ELISA assay specific for measurement of intact FGF23 only (Immutopics; San Clemente, CA) on urine samples from nine patients (n=5 from patients with established AKI and n=4 from patients without AKI matched on baseline eGFR ±5 ml/min per 1.73 m²). Additionally, we performed immunoprecipitation and Western blotting on the same samples (Supplemental Material).

Plasma and Serum

In the subcohort of patients with plasma samples available (n=131), we measured plasma levels of cFGF23; vitamin D metabolites—including 25D, 1,25D, and 24,25D₃; intact PTH; and DBP. cFGF23 was measured in duplicate using the same ELISA as above. Vitamin D metabolites were measured using immunoaffinity enrichment and liquid chromatography–tandem mass spectrometry.³⁸ PTH was measured using a chemiluminescent immunoassay (Beckman Coulter, Fullerton, CA). DBP was measured using a commercial ELISA (R&D Systems, Inc., Minneapolis, MN). Serum levels of creatinine, total calcium, and phosphate, and heparinized plasma levels of ionized calcium, were measured for clinical purposes by the hospital laboratory.

Coefficients of Variation

We calculated interassay coefficients of variation for each assay using blinded replicate samples from study patients. Interassay coefficients of variation were as follows: urinary cFGF23 (12.8%), urinary creatinine (6.4%), plasma cFGF23 (11.5%), 25D (4.1%), 1,25D (11.5%), 24,25D₃ (7.6%), PTH (3.1%), and DBP (11.0%).

Clinical Outcomes

Investigators D.E.L. and A.S. adjudicated all outcomes by reviewing electronic medical records, and were blinded to all study measurements at the time of adjudication. The prespecified primary composite end point was incident AKI or in-hospital death (AKI/death). AKI was defined as an increase in serum creatinine ≥0.3 mg/dl within 48 hours, ≥50% within 7 days, or need for RRT, consistent with the criteria established by the Kidney Disease Improving Global Outcomes (KDIGO) Work Group.¹⁷ To maintain the prospective nature of the study, incident AKI was defined as AKI occurring within 14 days after enrollment. Accordingly, patients who already had AKI on enrollment were excluded from these analyses. Death was included in the composite end point due to its clinical importance and competing risk for incident AKI.

Secondary end points were incident AKI, severe AKI, RRT or in-hospital mortality (RRT/death), severe sepsis or septic shock, and death (assessed in-hospital, at 90 days, and at 1 year). Severe AKI was defined as doubling of serum creatinine or need for RRT, corresponding to stages 2 and 3 of the AKI KDIGO criteria.¹⁷ Severe sepsis and septic shock were defined according to consensus definitions.³⁹

Additional end points included duration of mechanical ventilation and hospital length of stay. To avoid the confounding effect of mortality, we calculated ventilator-free days and hospital-free days as 28 minus the number of ventilator-dependent days or hospitalization days, respectively, assuming survival to 28 days or discharge from the hospital. Patients who died before 28 days were assigned a score of zero.⁴⁰

Finally, among patients with established AKI on enrollment, we evaluated the association between urinary cFGF23/Cr levels and risk of RRT/death and progression to a higher AKI stage. Established AKI was defined using the same KDIGO criteria described above, and in the vast majority of patients was evaluated using sequential daily SCr data available from the same hospital admission. In the rare case in which the enrollment SCr represented the first SCr value during the admission, the closest outpatient SCr value was used for comparison to adjudicate AKI.

Statistical Analyses

Statistical analysis was performed with SAS Version 9.4 (Cary, NC). Data are reported as median and IQR (25th–75th percentiles). Among patients without AKI on enrollment (n=252), baseline characteristics were compared across quartiles of urinary cFGF23/Cr using Kruskal Wallis and chi-squared tests for continuous and categoric variables, respectively.

Comparison of mineral metabolite levels with each other was assessed using Spearman rank correlation coefficient. Comparison of mineral metabolite levels among patients who did or did not develop AKI/death was assessed using the Wilcoxon Rank Sum test. Univariate logistic regression was used to assess the association between mineral metabolite levels and risk of AKI/death. In these models, biomarker levels were natural log–transformed, due to their skewed distribution, and normalized to one SD to allow comparison across biomarkers. For urinary cFGF23 analyses, we used multivariate logistic regression to adjust for covariates using two different models: model 1 was adjusted for age, sex, baseline eGFR, hypertension, and diabetes mellitus; model 2 was further adjusted for ICU type (surgical versus medical), mechanical ventilation on enrollment, and APACHE II score on enrollment. Urinary cFGF23 levels were assessed in quartiles and as a continuous variable.

Logistic and linear regression models were used to assess the association between urinary cFGF23 levels and secondary dichotomous and continuous end points, respectively, adjusting for the same covariates as above. Analyses that included AKI or RRT as end points were restricted to patients who did not have AKI on enrollment. All comparisons are two-tailed, with P<0.05 considered significant.

Disclosures

None.