Longitudinal FGF23 Trajectories and Mortality in Patients with CKD : Journal of the American Society of Nephrology

Journal Logo

Clinical Epidemiology

Longitudinal FGF23 Trajectories and Mortality in Patients with CKD

Isakova, Tamara1,2; Cai, Xuan2; Lee, Jungwha2; Xie, Dawei3; Wang, Xue3; Mehta, Rupal1,2; Allen, Norrina B.2; Scialla, Julia J.4,5; Pencina, Michael J.5; Anderson, Amanda H.3; Talierco, John6; Chen, Jing7; Fischer, Michael J.8,9; Steigerwalt, Susan P.10; Leonard, Mary B.11,12; Hsu, Chi-yuan13; de Boer, Ian H.14; Kusek, John W.15; Feldman, Harold I.3; Wolf, Myles4,5;  on behalf of Chronic Renal Insufficiency Cohort (CRIC) Study Investigators


Appel, Lawrence J.; Go, Alan S.; He, Jiang; Lash, James P.; Ojo, Akinlolu; Rahman, Mahboob; Townsend, Raymond R.

Author Information
Journal of the American Society of Nephrology 29(2):p 579-590, February 2018. | DOI: 10.1681/ASN.2017070772
  • Free
  • Infographic
  • SDC
  • SDC


CKD increases risk of death, but mechanisms are incompletely understood.1 Levels of the bone-derived hormone fibroblast growth factor 23 (FGF23) are elevated in the majority of patients with CKD.2,3 Elevated FGF23 levels in CKD stimulate urinary phosphate excretion and inhibit renal production of calcitriol, which secondarily reduces dietary phosphate absorption.4 Although these adaptations maintain normal serum phosphate levels in CKD, elevated FGF23 is strongly associated with increased risk of death across the spectrum of CKD.5–8 As potential explanatory mechanisms, elevated FGF23 has been implicated mechanistically in development of left ventricular hypertrophy, anemia, impaired leukocyte function, and chronic inflammation that increase risks of death due to cardiovascular disease and infection.9–17

Nearly all previous clinical outcomes studies of FGF23 measured levels at a single time point. How FGF23 levels change over time within individuals with CKD and whether longitudinal assessments are more strongly linked to clinical outcomes than isolated measurements are mostly unknown. Answers to these questions are needed to enhance understanding of the effect of FGF23 excess on clinical outcomes in CKD and inform whether FGF23 testing should be advanced from the research realm into clinical practice. We conducted a prospective, repeated measures, case-cohort study in the Chronic Renal Insufficiency Cohort (CRIC) Study to test the hypotheses that FGF23 levels increase over time and that rising FGF23 trajectories are independently associated with increased risk of death among individuals with CKD stages 2−4.


We measured FGF23 at two to five annual time points in a randomly selected subcohort of 1135 participants, of whom 203 died, and in all remaining 390 participants who died through mid-2013 (Figure 1). Participants underwent an average of 4.0±1.2 annual FGF23 measurements. Compared with participants who survived, those who died were older, were more likely to be smokers, had higher prevalence of diabetes and prior cardiovascular disease, had lower eGFR, and had higher urine albumin-to-creatinine ratio (UACR) and FGF23 levels at baseline (Table 1).

Figure 1.:
Sampling strategy for the case-cohort study population. The randomly selected subcohort included 1135 CRIC Study participants, of whom 203 died during the longitudinal follow-up period and 932 survived. All 390 remaining deaths from outside the subcohort were also sampled, providing a total study population of 1525 participants represented by the blue-shaded boxes.
Table 1. - Baseline characteristics of the subcohort and added cases from outside the subcohort
Baseline characteristics Random Subcohort Added Cases, n=390 Entire Population, n=3939
Survived, n=932 Died, n=203
Age, yr 57.0±11.0 61.6±9.0 62.3±9.2 58.2±11.0
Women, % 45.8 31.0 38.7 45.2
Black, % 40.8 50.3 48.2 42.1
Hispanic, % 12.5 9.4 11.0 12.6
Current smoking, % 9.2 16.8 21.5 13.1
BMI, kg/m2 32.0±7.9 33.3±8.5 32.0±7.4 32.1±7.8
SBP, mm Hg 125.8±21.0 132.8±22.1 134.8±24.5 128.5±22.2
Hypertension, % 84.7 93.1 92.3 86.1
Diabetes, % 44.4 68.0 60.8 48.4
Heart failure, % 4.8 21.2 20.3 9.7
Stroke, % 7.1 17.7 16.4 10.0
Peripheral vascular disease, % 2.7 18.2 13.9 6.7
Coronary artery disease, % 15.5 43.8 37.7 21.9
eGFR, ml/min per 1.73 m2 46.3±14.7 38.2±12.1 39.0±12.8 44.9±16.8
UACR, mg/g 36.1 (7.7–336.1) 154.2 (21.6–797.7) 145.6 (19.3–938.8) 52.0 (8.7–457.5)
Hemoglobin, g/dl 12.7±1.8 12.2±1.6 12.4±1.8 12.6±1.8
Serum albumin, g/dl 4.0±0.4 3.8±0.5 3.9±0.5 3.9±0.5
Calcium, mg/dl 9.2±0.5 9.1±0.5 9.1±0.6 9.2±0.5
Phosphate, mg/dl 3.7±0.6 3.9±0.7 3.8±0.8 3.7±0.7
PTH, pg/ml 51.0 (34.0–80.0) 73.0 (44.3–121.0) 68.0 (41.1–116.0) 54.0 (35.0–89.1)
FGF23, RU/ml 129.4 (90.4–194.7) 194.2 (129.0–324.9) 198.5 (121.1–344.6) 145.5 (95.8–239.2)
Results are reported as proportions, means±SD, or medians (interquartile ranges). BMI, body mass index; SBP, systolic BP; RU, reference unit.

Distribution of Longitudinal FGF23 and Covariates

Across five annual measurements (the CRIC baseline through the year 4 visit), eGFR declined modestly, whereas systolic BP and levels of mineral metabolites, including median FGF23, rose slightly (Table 2). In contrast, the mean and SD of FGF23 rose more rapidly, indicating a gradual right skew in the distribution over time. Figure 2 confirms the relative stability of the majority of the FGF23 distribution but a smaller subset of severely elevated outliers.

Table 2. - Repeated measures of eGFR, systolic BP, calcium, phosphate, PTH, and FGF23 in the case-cohort sample
Year eGFR SBP Calcium Phosphate PTH FGF23
N Mean (SD) N Mean (SD) N Mean (SD) N Mean (SD) N Median (IQR) Mean (SD) N Median (IQR) Mean (SD)
Y0 1525 43.4±14.4 1525 129.0±22.5 1515 9.2±0.5 1505 3.7±0.7 1504 57.3 (36.1–92.2) 78.2±76.1 1525 152.6 (98.8–249.0) 242.5±548.6
Y1 1473 40.8±15.7 1478 128.1±22.5 1478 9.3±0.6 1349 3.9±0.8 1337 66.0 (42.7–108.4) 94.5±97.3 1372 156.0 (95.1–308.4) 410.5±2400.2
Y2 1280 40.7±16.6 1280 129.3±23.4 1282 9.3±0.6 1211 4.1±0.8 1090 61.6 (38.7–112.4) 95.3±104.7 1208 154.1 (93.0–309.7) 360.7±957.9
Y3 1109 40.2±16.6 1116 126.5±21.1 1108 9.4±0.5 1060 4.1±0.8 821 62.9 (37.0–109.7) 96.1±102.9 1061 163.1 (99.8–305.5) 384.3±1453.5
Y4 935 41.7±17.5 939 128.3±22.3 935 9.4±0.5 858 4.1±0.9 505 63.0 (40.2–108.3) 93.6±94.1 903 159.5 (97.9–306.9) 561.8±3481.2
Values are means±SD or medians (IQRs). SBP, systolic BP; IQR, interquartile range; Y, year.

Figure 2.:
Longitudinal FGF23 levels are stable in the majority of the CRIC Study case-cohort participants, but there is a small subset of severely elevated outliers. Insets show the right tails of the distributions beginning at 375 reference units (RU) per 1 ml. For clarity, values >3000 RU/ml are not shown (five observations at baseline, 11 observations at year 1, 13 observations at year 2, nine observations at year 3, and 13 observations at year 4 post-baseline).

Baseline and Time-Varying FGF23 and Risk of Death

During a median follow-up of 7.06 years, 593 participants died. In fully adjusted analyses, higher baseline and time-varying FGF23 levels were independently associated with increased risk of death (Table 3): hazard ratio per 1-SD increase in natural log–transformed fibroblast growth factor 23 (lnFGF23), 1.46; 95% confidence interval (95% CI), 1.27 to 1.69 for baseline FGF23 and hazard ratio per 1-SD increase in lnFGF23, 1.84; 95% CI, 1.67 to 2.03 for time-varying FGF23. Results from analyses of ESRD-censored death were qualitatively similar (Table 3).

Table 3. - Baseline and time-varying FGF23 levels and risk of death
Deaths, n Unadjusted Model 1 Model 2 Model 3 Model 4
Baseline FGF23 a
 Including all deaths
  593 1.56 (1.38−1.78) 1.72 (1.51−1.95) 1.60 (1.39−1.85) 1.50 (1.30−1.72) 1.46 (1.27−1.69)
 ESRD-censored deaths
  398 1.72 (1.54−1.93) 1.80 (1.57−2.05) 1.76 (1.51−2.04) 1.62 (1.38−1.89) 1.58 (1.34−1.85)
Time-varying FGF23 b
 Including all deaths
  593 1.74 (1.60−1.89) 1.82 (1.67−1.97) 1.81 (1.65−2.00) 1.75 (1.59−1.92) 1.84 (1.67−2.03)
 ESRD-censored deaths
  398 1.94 (1.77−2.12) 2.01 (1.82−2.21) 2.06 (1.85−2.30) 1.99 (1.78−2.22) 2.05 (1.84−2.29)
All results are reported per 1-SD increase in lnFGF23 from the CRIC Study baseline visit. Model 1: stratified by center and adjusted for age, sex, race, and ethnicity. Model 2: model 1 plus eGFR, UACR, serum albumin, and hemoglobin. Model 3: model 2 plus diabetes, smoking, systolic BP, body mass index, history of coronary artery disease, history of heart failure, history of stroke, and history of peripheral vascular disease. Model 4: model 3 plus calcium, phosphate, and PTH.
aAdjusted for covariates at the CRIC Study baseline visit.
bAdjusted for time-varying covariates except sex, race, ethnicity, and UACR from the CRIC Study baseline visit.

FGF23 Trajectories and Risk of Death

Because analyzing repeated measures of FGF23 using a traditional time-varying approach generates mean estimates of effect for the entire population, we performed trajectory analyses using latent class model,18–23 which is an emerging method that enables agnostic detection of multiple subpopulations with distinct longitudinal patterns of FGF23 exposure. Trajectory analyses may be more relevant to practitioners who discern dynamic trends over time in laboratory test results in their individual patients rather than average changes in all of their patients. Using all five annual FGF23 measurements (Figure 3A), we identified three distinct FGF23 trajectories: stable (n=724), slowly rising (n=486), and rapidly rising (n=99) (Figure 3E); no subpopulations with falling or fluctuating trajectories emerged. Participants in the stable trajectory group had the lowest baseline FGF23 levels and experienced minimal change over time (slope of lnFGF23 per year =0.03; 95% CI, 0.02 to 0.04). Participants in the slowly rising group experienced slow but significant increases in FGF23 above their intermediate baseline levels (slope=0.14; 95% CI, 0.12 to 0.16). Participants in the rapidly rising group experienced a steeper rise in FGF23 above their high baseline values (slope =0.46; 95% CI, 0.38 to 0.54). The FGF23 slopes of each group were significantly different (P value for group × time interaction <0.001). Among all baseline characteristics, lower eGFR was the strongest independent predictor of a rising FGF23 trajectory.

Figure 3.:
Trajectory modeling identified three distinct FGF23 trajectory groups. (A) The primary analysis derived FGF23 trajectory groups using the maximum of five time points, set the onset of survival time (time 0) at the fifth time point (year 4 visit after baseline), and adjusted for concurrent covariates. (B and C) Secondary analyses repeated the weighted Cox modeling strategy after deriving FGF23 trajectory groups using four and three time points and adjusted for covariates ascertained at the analysis-specific time 0, which corresponded to the timing of the last possible FGF23 test for that specific analysis. (D) A sensitivity analysis introduced a 1-year lag, starting survival time at the year 5 visit after baseline, with adjustment for concurrent covariates. (E) Primary FGF23 trajectory analysis. Observed medians and interquartile ranges of FGF23 across the five time points (solid lines) and predicted FGF23 trajectories (dashed lines) with 95% CIs (shaded areas). The predicted FGF23 values approximate the observed values. The posterior predicted probabilities for each individual of being a member of a given trajectory group ranged from 0.91 to 0.96 across the trajectory groups. RU, reference unit.

Participants’ characteristics at the onset of the survival period (time 0), which coincided with the dates of their final FGF23 tests (Figure 3A), are presented in Table 4 according to FGF23 trajectory groups. In all univariate and multivariable-adjusted analyses of the five time point trajectories, membership in a rising FGF23 trajectory group was associated with significantly increased risk of death (Table 5). Compared with the stable FGF23 trajectory group, the full multivariable-adjusted hazard ratio of death was 4.49 (95% CI, 3.17 to 6.35) for the slowly rising group and 15.23 (95% CI, 8.24 to 28.14) for the rapidly rising group. The estimates were independent of multiple covariates, including final FGF23, baseline and final eGFR, or concomitant eGFR trajectories. When we repeated the trajectory analyses using four or three annual FGF23 measurements (Figure 3, B and C), the magnitude of risk associated with membership in a rising FGF23 trajectory group decreased but remained significant and qualitatively similar (Table 5).

Table 4. - Clinical characteristics of the FGF23 trajectory groups at the onset of the survival time (time 0) in the primary five-time point trajectory analysis
Characteristics Overall, n=1309 FGF23 Trajectory Group
Stable, lnFGF23 per 1 yr =0.03, n=724 Slowly Rising, lnFGF23 per 1 yr =0.14, n=486 Rapidly Rising, lnFGF23 per 1 yr =0.46, n=99
Age, yr 62.6±10.8 62.4±10.6 63.2±11.0 61.5±10.9
Women, % 43.4 40.7 46.3 48.5
Black, % 43.2 41.2 46.3 43.4
Hispanic, % 11.9 8.4 15.2 21.2
Current smoking, % 11.4 8.7 14.1 17.8
BMI, kg/m2 32.2±8.3 31.7±7.9 32.8±8.8 32.1±8.8
SBP, mm Hg 127.3±23.3 126.2±22.1 128.6±24.5 129.3±25.3
Diabetes, % 53.4 47.5 60.1 63.4
Heart failure, % 17.9 11.4 24.0 35.4
Stroke, % 12.5 9.6 15.6 19.2
Peripheral vascular disease, % 12.1 7.9 15.8 24.8
Coronary artery disease, % 32.6 27.2 38.6 43.2
eGFR, ml/min per 1.73 m2 40.0±17.4 45.8±16.6 33.2±15.4 31.4±16.7
UACR, mg/g a 52.6 (9.3–457.5) 18.3 (5.8–133.8) 185.1 (27.9–887.1) 809.6 (105.7–2025.1)
Hemoglobin, g/dl 12.4±1.8 12.8±1.8 12.0±1.8 11.6±1.8
Serum albumin, g/dl 3.9±0.4 3.9±0.4 3.8±0.4 3.8±0.5
Calcium, mg/dl 9.3±0.6 9.3±0.5 9.3±0.6 9.3±0.6
Phosphate, mg/dl 4.2±0.9 4.0±0.8 4.3±1.0 4.7±1.1
PTH, pg/ml 99.2 (51.4–178.1) 75.5 (42.1–143.9) 126.4 (72.3–208.5) 153.2 (85.5–227.7)
Baseline FGF23, RU/ml 145.0 (96.0–231.8) 105.2 (80.2–139.9) 214.3 (159.8–304.0) 471.9 (334.8–749.1)
Final FGF23, RU/ml 259.0 (120.7–1211.9) 134.1 (88.1–241.0) 584.5 (294.5–1612.6) 2124.5 (1027.4–7186.4)
eGFR slope, ml/min per 1.73 m2 per year 0.0±2.2 0.7±2.0 −0.7±2.1 −1.9±1.7
Serum phosphate slope, mg/dl per year −0.02 (−0.05 to 0.03) −0.03 (−0.06 to 0.00) 0.00 (−0.03 to 0.05) 0.07 (0.02 to 0.12)
Results are reported as proportions, means±SD, or medians (interquartile ranges). BMI, body mass index; SBP, systolic BP; RU, reference unit.
aUACR is from the CRIC Study baseline visit.

Table 5. - FGF23 trajectories and risk of death
Analysis Total, N Deaths, N Unadjusted Model 1 Model 2 Model 3 Model 4 Model 5
Up to five annual time points, median duration of subsequent follow-up time 3.4 yr in 1309 total participants at risk
 Stable, lnFGF23 per 1 yr =0.03 724 102 Reference Reference Reference Reference Reference Reference
 Slowly rising, lnFGF23 per 1 yr =0.14 486 221 4.96 (3.83–6.43) 5.27 (3.98–6.97) 4.87 (3.55–6.68) 4.86 (3.51–6.73) 4.54 (3.20–6.44) 4.49 (3.17–6.35)
 Rapidly rising, lnFGF23 per 1 yr =0.46 99 67 13.06 (8.50–20.07) 18.16 (11.37–29.01) 18.70 (11.22–31.16) 17.98 (10.56–30.62) 14.32 (7.73–26.55) 15.23 (8.24–28.14)
Up to four annual time points, median duration of subsequent follow-up time 4.3 yr in 1376 total participants at risk
 Stable, lnFGF23 per 1 yr =0.06 725 113 Reference Reference Reference Reference Reference Reference
 Slowly rising, lnFGF23 per 1 yr =0.16 528 257 4.79 (3.75–6.10) 5.03 (3.87–6.53) 4.70 (3.50–6.31) 4.19 (3.10–5.67) 3.94 (2.84–5.47) 4.06 (2.93–5.64)
 Rapidly rising, lnFGF23 per 1 yr =0.34 123 85 11.78 (8.00–17.35) 14.59 (9.46–22.52) 13.71 (8.63–21.79) 13.66 (8.59–21.73) 12.17 (7.10–20.87) 12.63 (7.44–21.47)
Up to three annual time points, median duration of subsequent follow-up time 5.1 yr in 1412 total participants at risk
 Stable, lnFGF23 per 1 yr =0.05 780 137 Reference Reference Reference Reference Reference Reference
 Slowly rising, lnFGF23 per 1 yr =0.20 528 280 4.85 (3.86–6.08) 5.15 (4.03–6.60) 4.80 (3.64–6.33) 4.08 (3.09–5.40) 3.11 (2.30–4.22) 3.28 (2.41–4.46)
 Rapidly rising, lnFGF23 per 1 yr =0.49 104 73 10.88 (7.21–16.43) 13.28 (8.39–21.01) 12.59 (7.66–20.70) 12.04 (7.38–19.63) 6.21 (3.56–10.85) 6.54 (3.75–11.43)
Covariate adjustment is for covariates obtained at the analysis-specific time 0 except for baseline UACR and baseline eGFR, which are obtained at the CRIC Study baseline visit. Model 1: stratified by center and adjusted for age, sex, race, and ethnicity. Model 2: model 1 plus eGFR, baseline UACR, serum albumin, and hemoglobin. Model 3: model 2 plus diabetes, smoking, systolic BP, body mass index, history of coronary artery disease, history of heart failure, history of stroke, and history of peripheral vascular disease. Model 4: model 3 plus calcium, phosphate, PTH, lnFGF23, and baseline eGFR. Model 5: model 4 substituting eGFR trajectory groups for baseline and updated eGFR at time 0.

Sensitivity Analyses

Despite widening of the 95% CIs owing to fewer events, the results of the five-time point trajectory analysis were qualitatively unchanged when we introduced a 1-year lag after the fifth FGF23 test and the onset of the survival observation period (Figure 3D, Supplemental Table 1). The results were also unchanged when we further adjusted the full five time point models for intercurrent cardiovascular events, recent hospitalizations, number of antihypertensive medications, baseline markers of inflammation, time-updated use of phosphate binders, active vitamin D and calciferols, and levels of 25-hydroxyvitamin D and 1,25-dihydroxyvitamin D at the year 2 visit (Supplemental Table 1). A complete case analysis consisting of individuals without missing data yielded qualitatively similar results (data not shown). In contrast to FGF23, trajectories of eGFR (stable, slowly declining, or rapidly declining), systolic BP (low stable, moderate stable, or elevated rising), and serum phosphate (stable or rising) were not consistently associated with increased risk of death in adjusted analyses, and the effect estimates were lower than for the rising FGF23 trajectory groups (Figure 4, Supplemental Tables 2–4).

Figure 4.:
Comparison of risks of mortality according to trajectories of FGF23, eGFR, systolic BP (SBP), and phosphate demonstrates specificity of the association between FGF23 trajectories and risk of mortality. Fully adjusted hazard ratios comparing the higher-risk with the lowest-risk trajectory groups for each exposure are shown. Trajectory groups were derived using the maximum of five time points, and the onset of survival time (time 0) in the Cox models was set at the fifth time point (year 4 visit after baseline). R, reference.


In a large population of individuals with moderate to severe CKD, median FGF23 levels did not increase substantially across five annual assessments. However, the unexpected longitudinal stability of FGF23 in the overall population was driven by the majority of participants whose unchanging FGF23 levels obscured rising levels experienced by the minority. Individuals in whom FGF23 levels rose rapidly were at >15-fold higher risk of death compared with the majority of participants with stable FGF23 levels. Participants with slowly rising FGF23 levels also were at approximately fourfold higher risk of death than those with stable levels. These results were independent of numerous covariates, including final FGF23 levels and baseline and follow-up kidney function, and unmatched by parallel analyses of eGFR, serum phosphate, or systolic BP trajectories.

The one previous study that measured FGF23 at baseline and 1 year later in 439 individuals with CKD found only a modestly stronger association of time-averaged versus baseline FGF23 with risk of death, leading the authors to question the value of repeated FGF23 measurements.24 Similarly, we observed modest increases in the point estimates for FGF23-associated risk when transitioning from analyses of single baseline FGF23 measurements to analyses of time-varying FGF23. However, our results indicate that traditional time-varying analytic approaches may underestimate effects of FGF23 on mortality in small subpopulations that trajectory analyses can uncover. Whether a similar degree of risk stratification could be obtained using repeated FGF23 measured more frequently in clinical practice (for example, over months instead of years) will require further study.

The high risk of mortality associated with membership in a rising FGF23 trajectory group contrasts the modest risks associated with membership in the higher-risk trajectory groups of eGFR, serum phosphate, and systolic BP. Although we are not able to exclude the possibility of residual confounding by eGFR, the discrepant results for FGF23 versus eGFR trajectories and the independence of the FGF23 effects from eGFR modeled multiple ways indicate that elevated FGF23 is likely not acting only as a biomarker of more severe CKD. The minimal effects of a rising serum phosphate trajectory suggest that the mechanisms underlying the link between rising FGF23 and death likely extend beyond phosphate homeostasis. The nonsignificant effects of systolic BP trajectories on mortality risk and the lack of confounding of the FGF23 effects by systolic BP or number of antihypertensive medications suggest that hypertension is not the primary driving force underlying the link between rising FGF23 and mortality. Instead, we speculate that chronic high-level exposure to FGF23, represented by high and rapidly rising FGF23 levels, exerts systemic toxicity as supported by experimental data implicating FGF23 excess in the pathogenesis of left ventricular hypertrophy, anemia, inflammation, and immune dysregulation.9–12,16,17 Alternatively, elevated and rising FGF23 levels could contribute to risk of mortality indirectly through adverse effects of inflammation and iron deficiency that are known to increase both intact and C-terminal FGF23 levels in CKD.25

Strengths of this study include its large and well characterized CKD cohort with minimal loss to follow-up, the efficient case-cohort design, use of traditional and emerging analytic methods to assess repeated FGF23 measurements, detailed covariate data that enabled comprehensive multivariable adjustment, and multiple sensitivity analyses that support the robustness of the primary results. The study also has limitations. Because our study design required participants to have a baseline and at least one additional FGF23 measurement, our source population was not identical to the entire CRIC Study population, which included participants who died or were lost to follow-up during the first year after enrollment. We used multiple imputation to account for missing data, and we did not have information on repeated measurements of markers of inflammation, iron stores, or intact FGF23 levels. Although we identified baseline eGFR as the strongest determinant of a rising FGF23 trajectory, additional basic and clinical research is needed to identify mechanisms underlying why specific individuals manifest particular trajectories and whether sources other than bone contribute to the extreme elevations in circulating FGF23 among individuals with rising trajectories. Because the CRIC Study did not collect cause-specific mortality data, understanding the extent to which our results were driven by cardiovascular, infectious, or other causes of death must await future studies. Likewise, additional studies in the full CRIC Study population are needed to confirm our findings and conduct formal risk prediction analyses.

Our results support novel research and clinical applications for serial FGF23 testing. Prior clinical trials that aimed to reduce FGF23 levels in patients with CKD yielded conflicting and often disappointing results.26–28 Our data suggest that more than one half of the participants in those trials probably had flat FGF23 trajectories. If so, interventions that are truly capable of lowering or at least preventing a rise in FGF23 levels might fail to show efficacy in trials that enroll an abundance of patients on stable FGF23 trajectories. Future trials could leverage serial FGF23 testing to enrich recruitment of high-risk patients with rising FGF23 trajectories who might be most amenable to and in need of reduction. Serial FGF23 testing could also help clinicians identify their highest-risk patients with CKD who require the closest clinical follow-up and most aggressive CKD management. By assessing risk in their patients on the basis of synthesizing visit to visit changes in laboratory results, clinicians are already attuned to the conceptual framework of trajectory analyses, which could offer a pathway for serial FGF23 testing to transition from research assay to clinical practice.

Concise Methods

Source Population

The CRIC Study is a multicenter, prospective observational study of risk factors for cardiovascular disease, CKD progression, and death.29 Individuals ages 21–74 years old with an eGFR of 20–70 ml/min per 1.73 m2 were enrolled between June 2003 and August 2008 (n=3939). Because CKD is more common in minorities, black and Hispanic participants were oversampled. Exclusion criteria included inability to consent, institutionalization, enrollment in other studies, pregnancy, New York Heart Association class 3 or 4 heart failure, HIV, cirrhosis, myeloma, polycystic kidney disease, renal cancer, recent chemotherapy or immunosuppressive therapy, organ transplantation, or prior treatment with dialysis for 1 month. The protocol was approved by institutional review boards at each study site, and participants provided written informed consent.

After enrollment, all participants underwent annual in-person visits and biannual updates by telephone. Demographic and clinical data were ascertained by interview, questionnaire, and physical examination, and blood and urine specimens were collected. Serum creatinine was measured in real time, and aliquots of blood and urine were stored at −80°C for future use. Adjudication committees validated clinical events.13 During the first 5 years of follow-up, >90% of living participants were retained in the study.

Study Design

We performed a case-cohort study nested in the CRIC Study. A case-cohort study design maximizes power to detect precise estimates of association between exposures and outcomes while reducing cost.30,31 Among participants who had baseline FGF23 levels measured at enrollment6 and stored plasma samples from at least one additional annual visit, we selected a random subcohort of 1135 participants, 203 of whom died during prospective follow-up. We sampled all remaining 390 deaths from outside the subcohort to reach a total analytic sample of 1525 individuals, including 593 who died. To minimize bias that could be introduced by differential rates of CKD progression, the case-cohort sampling captured all deaths during follow-up without censoring for onset of ESRD.

Exposure, Outcome, and Covariates

The primary exposure was serial levels of FGF23, which the central CRIC laboratory measured in duplicate using a second generation C-terminal assay (Immutopics, San Clemente, CA) in stored frozen plasma samples collected at the baseline and subsequent annual visits. The mean intra-assay coefficient of variation for paired assays was <6.5%. The laboratory also measured calcium, phosphate, and parathyroid hormone (PTH) levels annually. The primary outcome was all-cause mortality. Participants were followed until the occurrence of death, voluntary withdrawal from the study, loss to follow-up, or mid-2013, when the database was locked.

Statistical Analyses

We used descriptive statistics to characterize the study population and longitudinal changes in systolic BP, eGFR, calcium, phosphate, PTH, and FGF23. We analyzed FGF23 and mortality using several complementary strategies. To benchmark the risk of death associated with a single measurement of FGF23, we used weighted Cox proportional hazards models modified for the case-cohort design32 to analyze the hazard ratio of time to death per 1-SD increments of lnFGF23 measured at baseline. By comparison, we analyzed time to death for time-varying lnFGF23 per the same 1-SD increments in baseline lnFGF23 using weighted Cox models for the case-cohort design. To test the robustness of the primary analyses that included all deaths, we performed secondary analyses, in which we censored at onset of ESRD. For analyses of baseline and time-varying FGF23, follow-up time began at the CRIC Study baseline visit.

We used multivariable-weighted Cox models to hierarchically adjust for potential confounding. Model 1 stratified by center and adjusted for demographic factors (age, sex, race, and ethnicity). Model 2 further adjusted for CKD-specific risk factors (eGFR, UACR, serum albumin, and hemoglobin). Model 3 further adjusted for traditional cardiovascular risk factors (diabetes, smoking, body mass index, and systolic BP as well as individually, history of coronary artery disease, heart failure, stroke, and peripheral vascular disease). Model 4 further adjusted for calcium, phosphate, and PTH levels. In the models that examined baseline FGF23 as the exposure, we adjusted for baseline covariates. In the models that examined FGF23 as a time-varying exposure, we adjusted for time-varying covariates, for time-invariant covariates including sex, race, and ethnicity, and for UACR, which was only available at baseline.

Trajectory Analyses

Our primary trajectory analysis (Figure 3A) included participants who survived beyond their fifth annual study visit (the CRIC baseline through the year 4 visits) and considered all available FGF23 results that were obtained before onset of ESRD, because ESRD increases FGF23 exponentially.33 We used SAS Proc Traj to fit the longitudinal FGF23 data as a discrete mixture of two or more trajectories via maximum likelihood.18–20 For analyzing trajectories of FGF23, this procedure relies on a semiparametric group–based modeling strategy, which incorporates hierarchical modeling and latent growth curve modeling. An underlying assumption is that the population is made up of multiple trajectory groups. Therefore, the model simultaneously estimates probabilities for multiple trajectories. We tested models with different numbers of trajectory groups and different forms of potential trajectories (linear, quadratic, or cubic) for the best model fit, which we assessed with the Bayesian Information Criterion. Given continuous outcomes, we used a censored normal model, in which annual visits served as the time variable. On the basis of the best model fit, we identified three FGF23 trajectory groups, which we named according to their visual appearance and clinically meaningful categories: stable, slowly rising, and rapidly rising. To further enhance model fit, we evaluated different functional forms for the three groups. In the final model, we modeled the stable FGF23 trajectory group in quadratic terms and the slowly rising and elevated rising groups in linear terms. Next, we calculated the posterior predicted probabilities for each individual of being a member of a given trajectory group. We assigned the participants to the trajectory group for which they had the highest posterior predicted probability. Our final model classified the participants into trajectory groups with good discrimination: the mean probability of final group membership was 0.94 (ranging from 0.91 to 0.96) across the trajectory groups. We compared the slopes of the FGF23 trajectory groups using the “TRAJTEST” macro,18 and we tested for heterogeneity.

After deriving the FGF23 trajectory groups, we compared their risks of death using weighted Cox models modified for the case-cohort design, setting the onset of survival time (time 0) to coincide with the dates of participants’ fifth annual visits and final FGF23 tests. We used the same multivariable modeling strategy to adjust for potential confounding, such as in the continuous FGF23 analyses, except that we adjusted for updated covariates at time 0 (year 4 visit) and included further adjustments for baseline eGFR or eGFR trajectory groups and the final FGF23 test performed at time 0. In secondary trajectory analyses, we repeated the weighted Cox modeling strategy after deriving the FGF23 trajectory groups using four or three annual FGF23 tests. We adjusted for the same set of covariates as in the primary analyses but incorporated the values that were ascertained at the onset of the survival time for that specific analysis (at the year 3 visit for the four-time point analysis and at the year 2 visit for the three-time point analysis) (Figure 3, B and C).

Missing Data

To account for missing covariate data, we used multiple imputation, for which we used a multiple regression procedure in IVEware 2.0.34 We generated five imputed datasets, imputing values for missing data on the basis of the observed data, with the assumption of the data missing at random. Imputations were created through a sequence of multiple regressions and drawing values from the corresponding predictive distributions, varying the type of regression model by the type of variable being imputed.35 Then, we tested the association of membership in the three different FGF23 trajectory patterns with risk of death in weighted Cox models modified for the case-cohort design on each completed dataset separately. Finally, we combined the test results across the imputed datasets using the rules of Rubin.36

Sensitivity Analyses

To address potential residual confounding, we repeated the primary trajectory analysis after including a 1-year lag between the final FGF23 test and the onset of survival time, such that only individuals that survived to year 5 after baseline were considered (Figure 3D). Because cardiovascular events and hospitalizations could influence FGF23 trajectories and subsequent risk of death and would inform clinicians of their patients’ severity of illness and subsequent risk, we performed an additional sensitivity analysis, in which we adjusted the trajectory analyses for new adjudicated cardiovascular events and numbers of hospitalizations that occurred in the year preceding time 0. Because hypertension is a risk factor for mortality that is associated with higher FGF23 levels37 and because its severity cannot be fully ascertained by annual measurements, we performed an additional sensitivity analysis that further adjusted for participants’ numbers of antihypertensive medications in addition to their systolic BP. In additional analyses, we adjusted for baseline markers of inflammation; time-updated use of phosphate binders, active vitamin D, and calciferols; and levels of 25-hydroxyvitamin D and 1,25-dihydroxyvitamin D at the year 2 visit, and we conducted a complete case analysis. To test the specificity of the FGF23 trajectory analyses, we replicated our analytic strategy to examine the associations of longitudinal trajectories of eGFR, systolic BP, and serum phosphate with risk of mortality.

All analyses were performed with SAS 9.4 (SAS Institute, Cary, NC). All statistical tests were two sided, and P values <0.05 were considered significant.


T.I. received honoraria from Kyowa Hakko Kirin Co., Ltd. and grant support from Shire. I.d.B. served as a consultant or received honoraria from Amgen, Boehringer-Ingelheim, Janssen, and Ironwood and received equipment and supplies for research from MedTronic and Abbott. M.W. has served as a consultant or received honoraria from Akebia, Amag, Amgen, Ardelyx, Diasorin, Incyte, Keryx, Luitpold, Pfizer, Sanofi, and Ultragenyx and received grant support from Shire.

Published online ahead of print. Publication date available at www.jasn.org.

This article contains supplemental material online at http://jasn.asnjournals.org/lookup/suppl/doi:10.1681/ASN.2017070772/-/DCSupplemental.

This work was supported by grants R01DK102438 (to T.I.), R01DK110087 (to T.I.), K23DK095949 (to J.J.S.), R01DK099199 (to I.d.B.), R01DK081374 (to M.W.), R01DK076116 (to M.W.), R01DK094796 (to M.W.), K24DK093723 (to M.W.), and U01DK099930 (to M.W.) from the National Institutes of Health (NIH); a Department of Veterans Affairs Health Services Research and Development Service grant (to M.J.F.); and a Strategically Focused Research Network Center Grant on Health Disparities from the American Heart Association (to M.W.). Funding for the Chronic Renal Insufficiency Cohort Study was obtained under a cooperative agreement from the National Institute of Diabetes and Digestive and Kidney Diseases (grants U01DK060990, U01DK060984, U01DK061022, U01DK061021, U01DK061028, U01DK060980, U01DK060963, and U01DK060902). In addition, this work was supported in part by Perelman School of Medicine at the University of Pennsylvania Clinical and Translational Science award NIH/National Center for Advancing Translational Sciences (NCATS) UL1TR000003, Johns Hopkins University grant UL1 TR-000424, University of Maryland General Clinical Research Center grant M01 RR-16500, the Clinical and Translational Science Collaborative of Cleveland, grant UL1TR000439 from the NCATS component of the NIH and NIH roadmap for Medical Research, Michigan Institute for Clinical and Health Research grant UL1TR000433, University of Illinois at Chicago Clinical and Translational Science Award grant UL1RR029879, Tulane Center of Biomedical Research Excellence for Clinical and Translational Research in Cardiometabolic Diseases grant P20 GM109036, and Kaiser Permanente NIH/National Center for Research Resources University of California San Francisco Clinical and Translational Science Institute grant UL1 RR-024131.


1. Go AS, Chertow GM, Fan D, McCulloch CE, Hsu CY: Chronic kidney disease and the risks of death, cardiovascular events, and hospitalization. N Engl J Med 351: 1296–1305, 200415385656
2. Isakova T, Wahl P, Vargas GS, Gutiérrez OM, Scialla J, Xie H, Appleby D, Nessel L, Bellovich K, Chen J, Hamm L, Gadegbeku C, Horwitz E, Townsend RR, Anderson CA, Lash JP, Hsu CY, Leonard MB, Wolf M: Fibroblast growth factor 23 is elevated before parathyroid hormone and phosphate in chronic kidney disease. Kidney Int 79: 1370–1378, 201121389978
3. Portale AA, Wolf M, Jüppner H, Messinger S, Kumar J, Wesseling-Perry K, Schwartz GJ, Furth SL, Warady BA, Salusky IB: Disordered FGF23 and mineral metabolism in children with CKD. Clin J Am Soc Nephrol 9: 344–353, 201424311704
4. Gutierrez O, Isakova T, Rhee E, Shah A, Holmes J, Collerone G, Jüppner H, Wolf M: Fibroblast growth factor-23 mitigates hyperphosphatemia but accentuates calcitriol deficiency in chronic kidney disease. J Am Soc Nephrol 16: 2205–2215, 200515917335
5. Gutiérrez OM, Mannstadt M, Isakova T, Rauh-Hain JA, Tamez H, Shah A, Smith K, Lee H, Thadhani R, Jüppner H, Wolf M: Fibroblast growth factor 23 and mortality among patients undergoing hemodialysis. N Engl J Med 359: 584–592, 200818687639
6. Isakova T, Xie H, Yang W, Xie D, Anderson AH, Scialla J, Wahl P, Gutiérrez OM, Steigerwalt S, He J, Schwartz S, Lo J, Ojo A, Sondheimer J, Hsu CY, Lash J, Leonard M, Kusek JW, Feldman HI, Wolf M; Chronic Renal Insufficiency Cohort (CRIC) Study Group: Fibroblast growth factor 23 and risks of mortality and end-stage renal disease in patients with chronic kidney disease. JAMA 305: 2432–2439, 201121673295
7. Kendrick J, Cheung AK, Kaufman JS, Greene T, Roberts WL, Smits G, Chonchol M; HOST Investigators: FGF-23 associates with death, cardiovascular events, and initiation of chronic dialysis. J Am Soc Nephrol 22: 1913–1922, 201121903574
8. Wolf M, Molnar MZ, Amaral AP, Czira ME, Rudas A, Ujszaszi A, Kiss I, Rosivall L, Kosa J, Lakatos P, Kovesdy CP, Mucsi I: Elevated fibroblast growth factor 23 is a risk factor for kidney transplant loss and mortality. J Am Soc Nephrol 22: 956–966, 201121436289
9. Faul C, Amaral AP, Oskouei B, Hu MC, Sloan A, Isakova T, Gutiérrez OM, Aguillon-Prada R, Lincoln J, Hare JM, Mundel P, Morales A, Scialla J, Fischer M, Soliman EZ, Chen J, Go AS, Rosas SE, Nessel L, Townsend RR, Feldman HI, St John Sutton M, Ojo A, Gadegbeku C, Di Marco GS, Reuter S, Kentrup D, Tiemann K, Brand M, Hill JA, Moe OW, Kuro-O M, Kusek JW, Keane MG, Wolf M: FGF23 induces left ventricular hypertrophy. J Clin Invest 121: 4393–4408, 201121985788
10. Grabner A, Amaral AP, Schramm K, Singh S, Sloan A, Yanucil C, Li J, Shehadeh LA, Hare JM, David V, Martin A, Fornoni A, Di Marco GS, Kentrup D, Reuter S, Mayer AB, Pavenstädt H, Stypmann J, Kuhn C, Hille S, Frey N, Leifheit-Nestler M, Richter B, Haffner D, Abraham R, Bange J, Sperl B, Ullrich A, Brand M, Wolf M, Faul C: Activation of cardiac fibroblast growth factor receptor 4 causes left ventricular hypertrophy. Cell Metab 22: 1020–1032, 201526437603
11. Rossaint J, Oehmichen J, Van Aken H, Reuter S, Pavenstädt HJ, Meersch M, Unruh M, Zarbock A: FGF23 signaling impairs neutrophil recruitment and host defense during CKD. J Clin Invest 126: 962–974, 201626878171
12. Singh S, Grabner A, Yanucil C, Schramm K, Czaya B, Krick S, Czaja MJ, Bartz R, Abraham R, Di Marco GS, Brand M, Wolf M, Faul C: Fibroblast growth factor 23 directly targets hepatocytes to promote inflammation in chronic kidney disease. Kidney Int 90: 985–996, 201627457912
13. Scialla JJ, Xie H, Rahman M, Anderson AH, Isakova T, Ojo A, Zhang X, Nessel L, Hamano T, Grunwald JE, Raj DS, Yang W, He J, Lash JP, Go AS, Kusek JW, Feldman H, Wolf M; Chronic Renal Insufficiency Cohort (CRIC) Study Investigators: Fibroblast growth factor-23 and cardiovascular events in CKD. J Am Soc Nephrol 25: 349–360, 201424158986
14. Mehta R, Cai X, Lee J, Scialla JJ, Bansal N, Sondheimer JH, Chen J, Hamm LL, Ricardo AC, Navaneethan SD, Deo R, Rahman M, Feldman HI, Go AS, Isakova T, Wolf M; Chronic Renal Insufficiency Cohort (CRIC) Study Investigators: Association of fibroblast growth factor 23 with atrial fibrillation in chronic kidney disease, from the Chronic Renal Insufficiency Cohort Study. JAMA Cardiol 1: 548–556, 201627434583
15. Chonchol M, Greene T, Zhang Y, Hoofnagle AN, Cheung AK: Low vitamin D and high fibroblast growth factor 23 serum levels associate with infectious and cardiac deaths in the HEMO study. J Am Soc Nephrol 27: 227–237, 201625971439
16. Coe LM, Madathil SV, Casu C, Lanske B, Rivella S, Sitara D: FGF-23 is a negative regulator of prenatal and postnatal erythropoiesis. J Biol Chem 289: 9795–9810, 201424509850
17. Mehta R, Cai X, Hodakowski A, Lee J, Leonard M, Ricardo A, Chen J, Hamm L, Sondheimer J, Dobre M, David V, Yang W, Go A, Kusek JW, Feldman H, Wolf M, Isakova T; CRIC Study Investigators: Fibroblast growth factor 23 and anemia in the Chronic Renal Insufficiency Cohort Study [published online ahead of print August 7, 2017]. Clin J Am Soc Nephrol doi:10.2215/CJN.0395041728784656
18. Jones B, Nagin D, Roeder K: A SAS procedure based on mixture models for estimating developmental trajectories. Sociol Methods Res 29: 374–393, 2001
19. Nagin DS, Odgers CL: Group-based trajectory modeling in clinical research. Annu Rev Clin Psychol 6: 109–138, 201020192788
20. Nagin DS, Odgers CL: Group-based trajectory modeling (nearly) two decades later. J Quant Criminol 26: 445–453, 201021132047
21. Gill TM, Gahbauer EA, Han L, Allore HG: Trajectories of disability in the last year of life. N Engl J Med 362: 1173–1180, 201020357280
22. Allen NB, Siddique J, Wilkins JT, Shay C, Lewis CE, Goff DC, Jacobs DR Jr ., Liu K, Lloyd-Jones D: Blood pressure trajectories in early adulthood and subclinical atherosclerosis in middle age. JAMA 311: 490–497, 201424496536
23. Song M, Hu FB, Wu K, Must A, Chan AT, Willett WC, Giovannucci EL: Trajectory of body shape in early and middle life and all cause and cause specific mortality: Results from two prospective US cohort studies. BMJ 353: i2195, 201627146280
24. Bouma-de Krijger A, Bots ML, Vervloet MG, Blankestijn PJ, Ter Wee PW, van Zuilen AD, Wetzels JF: Time-averaged level of fibroblast growth factor-23 and clinical events in chronic kidney disease. Nephrol Dial Transplant 29: 88–97, 201424215017
25. David V, Martin A, Isakova T, Spaulding C, Qi L, Ramirez V, Zumbrennen-Bullough KB, Sun CC, Lin HY, Babitt JL, Wolf M: Inflammation and functional iron deficiency regulate fibroblast growth factor 23 production. Kidney Int 89: 135–146, 201626535997
26. Oliveira RB, Cancela AL, Graciolli FG, Dos Reis LM, Draibe SA, Cuppari L, Carvalho AB, Jorgetti V, Canziani ME, Moysés RM: Early control of PTH and FGF23 in normophosphatemic CKD patients: A new target in CKD-MBD therapy? Clin J Am Soc Nephrol 5: 286–291, 201019965540
27. Isakova T, Barchi-Chung A, Enfield G, Smith K, Vargas G, Houston J, Xie H, Wahl P, Schiavenato E, Dosch A, Gutiérrez OM, Diego J, Lenz O, Contreras G, Mendez A, Weiner RB, Wolf M: Effects of dietary phosphate restriction and phosphate binders on FGF23 levels in CKD. Clin J Am Soc Nephrol 8: 1009–1018, 201323471131
28. Block GA, Wheeler DC, Persky MS, Kestenbaum B, Ketteler M, Spiegel DM, Allison MA, Asplin J, Smits G, Hoofnagle AN, Kooienga L, Thadhani R, Mannstadt M, Wolf M, Chertow GM: Effects of phosphate binders in moderate CKD. J Am Soc Nephrol 23: 1407–1415, 201222822075
29. Feldman HI, Appel LJ, Chertow GM, Cifelli D, Cizman B, Daugirdas J, Fink JC, Franklin-Becker ED, Go AS, Hamm LL, He J, Hostetter T, Hsu CY, Jamerson K, Joffe M, Kusek JW, Landis JR, Lash JP, Miller ER, Mohler ER 3rd, Muntner P, Ojo AO, Rahman M, Townsend RR, Wright JT; Chronic Renal Insufficiency Cohort (CRIC) Study Investigators: The Chronic Renal Insufficiency Cohort (CRIC) study: Design and methods. J Am Soc Nephrol 14[Suppl 2]: S148–S153, 200312819321
30. Wacholder S, Gail M, Pee D: Selecting an efficient design for assessing exposure-disease relationships in an assembled cohort. Biometrics 47: 63–76, 19912049514
31. Prentice RL: A case-cohort design for epidemiologic cohort studies and disease prevention trials. Biometrika 73: 1–11, 1986
32. Barlow WE, Ichikawa L, Rosner D, Izumi S: Analysis of case-cohort designs. J Clin Epidemiol 52: 1165–1172, 199910580779
33. Isakova T, Xie H, Barchi-Chung A, Vargas G, Sowden N, Houston J, Wahl P, Lundquist A, Epstein M, Smith K, Contreras G, Ortega L, Lenz O, Briones P, Egbert P, Ikizler TA, Jueppner H, Wolf M: Fibroblast growth factor 23 in patients undergoing peritoneal dialysis. Clin J Am Soc Nephrol 6: 2688–2695, 201121903990
34. Raghunathan TE, Solenberger PW, Van Hoewyk J: IVEware: Imputation and Variance Estimation Software, Ann Arbor, MI, University of Michigan, Institute for Social Research, Survey Research Center, 2000
35. Raghunathan TE, Lepkowski JM, Van Hoewyk J: A multivariate technique for multiply imputing missing values using a sequence of regression models. Surv Methodol 27: 85–95, 2001
36. Rubin DB: Multiple Imputation for Nonresponse in Surveys, New York, John Wiley and Sons, 2004
37. Gutiérrez OM, Wolf M, Taylor EN: Fibroblast growth factor 23, cardiovascular disease risk factors, and phosphorus intake in the health professionals follow-up study. Clin J Am Soc Nephrol 6: 2871–2878, 201122034506

CKD; FGF23; mortality risk

Copyright © 2018 by the American Society of Nephrology