18-year change in serum intact fibroblast growth factor 23 from midlife to late-life and risk of mortality: the ARIC Study

Abstract

Objective:

Fibroblast growth factor 23 (FGF23) concentration increases in response to declining kidney function to preserve normal phosphate concentrations. However, the etiological association of change in FGF23 concentration with mortality has not been examined in the general population.

Design and Methods:

We analyzed 5,458 participants of the Atherosclerosis Risk in Communities Study who had intact FGF23 and estimated glomerular filtration rate (eGFR) assessed during midlife (visit 3, 1993–1995, mean age 58 years) and late-life (visit 5, 2011–2013, 76 years) to examine the association of FGF23 change over 18 years from mid-life to late-life with the subsequent risk of mortality in late-life using Cox regression models.

Results:

The median 18-year change in intact FGF23 was +17.3 pg/mL. During a median follow-up of 7.2 years following visit 5, 1,176 participants died. In multivariable Cox models, elevated mortality was seen in the highest quartile of FGF23 change (ΔFGF23, ≥31.3 pg/mL) (adjusted hazard ratio [aHR], 1.61 [95%CI, 1.36 to 1.90], or 1.37 [1.15 to 1.64] after additionally adjusting for eGFR change, compared with the lowest quartile [≤6.4 pg/mL]). When both FGF23 change and FGF23 in late-life were simultaneously entered into the Cox model, FGF23 in late-life, but not FGF23 change, was an independent predictor of mortality; however, we observed a high correlation between FGF23 change from midlife to late-life and FGF23 in late-life (r=0.77).

Conclusions:

Serum intact FGF23 change from midlife to late-life was associated with subsequent risk of mortality independent of decline in kidney function. Our findings further support the implications of FGF23 beyond its association with kidney function.

Introduction

Fibroblast growth factor 23 (FGF23) is an endocrine hormone that regulates bone mineral metabolism.¹ FGF23 promotes phosphaturia and suppresses the synthesis of vitamin D, resulting in the net negative phosphate balance. Chronic kidney disease (CKD) is the most frequent underlying condition for an elevated concentration of FGF23, and the blood concentration of FGF23 is increased from its early stages to maintain normal phosphate concentraions.^{2, 3}

Several studies showed that an elevated concentration of FGF23 at baseline was associated with risk of various adverse outcomes,^4–7 although one study of community-dwelling older adults reported non-significant association of intact FGF23 with mortality once accounting for kidney function.⁸ In addition, a recent report demonstrated that a rapid increase in FGF23 concentrations over ~5 years, compared to stable FGF23 trajectory, was associated with a higher risk of mortality in individuals with moderate to severe CKD.⁹ However, it remains unclear whether longitudinal changes of FGF23 are prognostic beyond changes in kidney function in the general population.

We hypothesized that FGF23 change would be associated with mortality independent of change in kidney function. To test this hypothesis, we examined the association of change in intact FGF23 concentrations over 18 years from midlife to late-life with the subsequent risk of mortality in a community-based cohort, the Atherosclerosis Risk in Communities (ARIC) Study.

Materials and Methods

Study population

Details of the ARIC Study have been reported previously.¹⁰ In brief, the ARIC Study is a community-based cohort study that enrolled 15,792 participants from four US communities (Forsyth County, NC; Jackson, MS; suburban Minneapolis, MN; and Washington County, MD) between 1987 and 1989 (visit 1). For the present study, we included participants who attended the study visits that occurred in 1993–1995 (visit 3) and 2011–2013 (visit 5) and had data on serum intact FGF23 concentrations and estimated glomerular filtration rates (eGFRs) at both visits. Among 6,538 visit 5 participants, 5,817 participants met these criteria (Figure S1). We then excluded those who self-identified race other than Black or White (n=17), as well as those with prevalent end-stage kidney disease (n=14), and missing covariates (n=328), leaving 5,458 participants in the analytic sample. The study was conducted in compliance with the Declaration of Helsinki. The institutional review board at each study site approved the study (#H.34.99.07.02.A1 at Johns Hopkins University), and written informed consent was obtained from all participants.

Exposure

The primary exposure was change in serum intact FGF23 concentrations from visit 3 to visit 5, as calculated by the subtraction of FGF23 at visit 3 from FGF23 at visit 5. Between 2018 and 2019, the Advanced Research and Diagnostic Laboratory at the University of Minnesota measured serum intact FGF23 concentrations in previously frozen stored serum samples collected at visit 3 and visit 5 using a 2-step ELISA (FGF23 ELISA Kit; Kainos Laboratories, Inc, Tokyo, Japan). All stored serum samples were never thawed prior to the study, except for 2376 of 5458 (43.5%) samples collected at visit 3, which had undergone a freeze-thaw cycle. No calibration was made, since intact FGF23 concentrations were stable to a freeze-thaw process in our visit 3 serum samples.¹¹ Inter-assay coefficients of variations were 8.1% and 5.4% at mean concentrations of 23.7 and 87.2 pg/mL.

Outcome

The outcome of interest was all-cause mortality because mortality is the most important outcome and FGF23 has been associated with various types of adverse events.^4–7 Death records were obtained through the active surveillance as well as the linkage to the National Death Index. The follow-up period was from visit 5 until December 31, 2019.

Covariates

All covariates were based on data collected at visit 5. Age, sex, race, and smoking status (never vs. ever) were self-reported. Systolic blood pressure was based on the average of the last two of three readings measured after 5 minutes rest in a sitting position. The information on medications was based on self-report and medication containers participants brought to the research clinics. Diabetes was defined as taking antidiabetic drugs, self-reported diagnosis of diabetes by a physician, a fasting blood glucose ≥126 mg/dL, or a non-fasting blood glucose ≥200 mg/dL. History of heart failure was defined according to the algorithm recommended by ARIC Study, using combinations of self-reported diagnosis, diagnostic code, biomarker, use of medication, and adjudicated cases of heart failure. History of coronary heart disease, and stroke were based on self-reported questionnaire at visit 1 (1987–1989) and adjudicated incident cases between visit 1 and visit 5. Plasma concentrations of total cholesterol and high-density lipoprotein cholesterol were measured using enzyme methods according to the standard protocol. Serum calcium and phosphate concentrations were measured using colorimetric methods. eGFR was calculated using the Chronic Kidney Disease Epidemiology Collaboration 2012 equation, incorporating serum creatinine and cystatin C, age, sex, and race.¹² Serum creatinine was measured using an enzymatic method (Roche Diagnostics, Indianapolis, IN) and standardized to isotope-dilution mass spectrometry traceable method. Serum cystatin C was measured using the Gentian cystatin C immunoassay (Gentian AS, Moss, Norway).

Statistical analysis

Characteristics of the study population were compared by quartiles of FGF23 change between visit 3 and visit 5 using ANOVA (analysis of variance) or chi-square test. Mortality rates and their 95% confidence intervals (95%CIs) were estimated using Poisson regression, and we modeled FGF23 change as restricted cubic splines with knots at the values corresponding to the 25th, 50th, and 75th percentiles. We estimated the hazard ratios (HRs) of death using Cox regression models for quartiles of FGF23 change, with the lowest quartile as the reference. We considered the following three models: Model 1 adjusted for age, sex, and race; Model 2 further accounting for diabetes, ever smoke, systolic blood pressure, medication use (antihypertensive, vitamin D, calcium, and iron), prevalent heart failure, coronary heart disease, stroke, total cholesterol, high-density lipoprotein cholesterol, calcium, and phosphate; and Model 3 additionally adjusted for eGFR change. We considered Model 3 as the primary model.

Additionally, we tested models adjusting for FGF23 at visit 5 in addition to covariates in Model 3, since a clinically relevant question may be whether a physician should take into account information in the past or can rely on laboratory data obtained “today”, although FGF23 change incorporates both past and present FGF23 concentrations by way of its calculation (Supplementary Figure 2).¹³

To examine whether the associations were consistent within a priori determined demographic and clinical subgroups. We performed subgroup analyses by age (< vs. ≥75 years), sex (male vs. female), race (White vs. Black), ever smoke (no vs. yes), antihypertensive medication use (no vs. yes), diabetes (no vs. yes), prevalent cardiovascular disease (no vs. yes), FGF23 at visit 5 (<55.0 vs. ≥55.0 pg/mL), eGFR at visit 5 (≥ vs. <60 ml/min/1.73m²), and phosphate at visit 5 (<3.51 vs. ≥3.51 mg/dL). For subgroup analyses, the HRs were presented for the highest quartile of FGF23 change or lowest quartile of eGFR change, compared to the rest of the quartiles as the reference to increase the effective sample size. The statistical interaction was assessed with the use of likelihood ratio tests.

For sensitivity analysis, we first examined the mortality risk in the combined categories of FGF23 change and eGFR or phosphate change to quantify the joint associations of FGF23 change and eGFR (or phosphate) change with mortality, because eGFR and phosphate are the two major regulators of FGF23 concentration. Second, due to the cohort attrition between visit 3 and visit 5, participant characteristics might be different between those who did and did not attend visit 5. To account for this attrition bias, we repeated the analysis with the use of the inverse probability of weighting.¹⁴ All statistical analyses were performed using Stata version 15. A two-sided P-value of less than 0.05 was considered statistically significant.

Results

Baseline characteristics

The mean age of participants at visit 5 was 76 [5.2] years, 58% were female, and 21% were identified as Black. The median time between visit 3 (1993–1995) and visit 5 (2011–2013) was 17.7 years (interquartile interval [IQI], 17.1 and 18.4 years). The distribution of FGF23 change was mostly normal, with the median change of 17.3 pg/mL (IQI, 6.5 and 31.2 pg/mL) (Figure 1A). The correlation between visit 3 and visit 5 FGF23 concentrations was weak (r=0.36, Figure 1B), although there was a high correlation between FGF23 at visit 5 and FGF23 change (r=0.77). There was a modest negative correlation between FGF23 change and eGFR change (r=−0.42, Figure 1C)

Figure 1: Intact FGF23 change from midlife (visit 3, 1993–1995) to late-life (visit 5, 2011–2013).

(A) Histograms of intact FGF23 change, (B) Scatter plots for intact FGF23 at visit 3 and intact FGF23 at visit 5, (C) Scatter plots for intact FGF23 change and eGFR change. r indicates Pearson correlation coefficient. Outliers as defined by values that deviated from the mean greater than the 3-fold of stand deviation were not included in the plots. FGF23 indicates fibroblast growth factor 23. eGFR indicates estimated glomerular filtration rate.

When baseline characteristics were compared across the quartiles of FGF23 change, participants who had a greater change in FGF23 concentrations were more likely to be female, White, and on antihypertensive medication, and have diabetes and prevalent cardiovascular disease (Table 1). Participants with higher quartiles of FGF23 change tended to have a higher FGF23 concentration at visit 5, but lower FGF23 concentration at visit 3. Participants with a greater FGF23 change also tended to have a lower eGFR at visit 5. On the other hand, there was no clear association between eGFR at visit 3 and quartiles of FGF23 change (Table 1).

Table 1:

Characteristics of ARIC Study participants in late-life (visit 5, 2011–2013). Data are presented as n (%) or as mean ±S.D.

Characteristics	Overall	Quartile of 18-year change in FGF23 from midlife to late-life, pg/mL
		Q1: ≤6.4	Q2: 6.5 to 17.3	Q3: 17.4 to 31.2	Q4: ≥31.3
n	5458	1364	1365	1364	1365
Age, years	76 ±5.2	75 ± 5.1	75 ± 4.9	76 ±5.2	76 ±5.3
Females	3172 (58)	709 (52)	778 (57)	817 (60)	868 (64)
Black	1159 (21)	333 (24)	272 (20)	273 (20)	281 (21)
Ever smoke	3156 (58)	817 (60)	789 (58)	766 (56)	784 (57)
Systolic BP mmHg	130 ± 18	131± 18	130± 18	130±18	130± 19
Diabetes	1784 (33)	435 (32)	388 (28)	422 (31)	539 (40)
Medication use
Antihypertensive	4095 (75)	931 (68)	953 (70)	1053 (77)	1158 (85)
Vitamin D	1501 (28)	302 (22)	339 (25)	397 (29)	463 (34)
Calcium	1437 (26)	297 (22)	333 (24)	384 (28)	423 (31)
Iron	232 (4.3)	62 (4.5)	45 (3.3)	46 (3.4)	79 (5.8)
Past medical history
Prevalent HF	692 (13)	157 (12)	137 (10)	168 (12)	230 (17)
Prevalent CHD	809 (15)	173 (13)	190 (14)	178 (13)	268 (20)
Prevalent stroke	208 (3.8)	53 (3.9)	33 (2.4)	48 (3.5)	74 (5.4)
Laboratory tests
Total cholesterol, mg/dL	181± 42	181± 42	183± 41	182± 41	179± 44
HDL cholesterol, mg/dL	52± 14	52± 14	53 ± 14	52 ± 13	51±13
Calcium, mg/dL	9.4 ± 0.4	9.3± 0.4	9.3± 0.4	9.4± 0.4	9.5± 0.4
Phosphate, mg/dL	3.5 ± 0.4	3.4± 0.4	3.4± 0.4	3.5 ± 0.4	3.6± 0.5
FGF23 at visit 3, pg/mL	40± 19	49 ± 30	37± 12	36 ± 12	38 ± 14
FGF23 at visit 5, pg/mL	61±26	44±16	49±12	59± 12	89±30
eGFR at visit 3, ml/min/1.73m2	87± 13	86± 13	88± 13	88± 13	86± 14
eGFR at visit 5, ml/min/1.73m2	65±18	72±16	70± 16	65± 16	55± 18

eGFR and FGF23 at visit 3 were based on data in 1993–1995. Between-group difference based on ANOVA or chi-square test was significant (p<0.05) for all variables except for ever smoke (p=0.3), systolic blood pressure (p=0.3), and total cholesterol (p=0.07). To covert to SI units, multiply total cholesterol and HDL cholesterol by 0.026, calcium by 0.250, and phosphate by 0.323. Abbreviations: ARIC, Atherosclerosis Risk in Communities; FGF23, fibroblast growth factor-23; HDL, high-density lipoprotein; eGFR, estimated glomerular filtration rate.

Associations of changes in intact FGF23 and eGFR with all-cause mortality

During a median follow-up of 7.2 years after visit 5, 1,177 participants had died (crude mortality rate, 32.9 per 1,000 person-years [95%CI, 31.1 to 34.9]). Figure 2 shows the age-, sex-, and race-adjusted mortality rates across the range of FGF23 change from midlife to late-life. Although there were continuous associations of FGF23 change with mortality, the risk gradient appeared steepest in the highest quartile of FGF23 change (≥31.3 pg/mL), intermediate in the third quartile, and modest or flat in the lowest two quartiles of FGF23 change. The higher risks of mortality associated with FGF23 change remained significant in its highest vs. lowest quartile after adjusting for clinical factors including eGFR change (HR, 1.37 [95%CI, 1.15 to 1.64]) (Model 3 in Table 2).

Figure 2: All-cause mortality rates for change in intact FGF23 from midlife (visit 3, 1993–1995) to late-life (visit 5, 2011–2013): ARIC Study, 1993–2019.

The model was adjusted for age, sex, and race. Solid lines indicate point estimates. Dashed lines indicate upper and lower bound of 95% confidence intervals. Histograms show the distribution of FGF23 change. Abbreviations: FGF23, fibroblast growth factor 23; ARIC, Atherosclerosis Risk in Communities.

Table 2:

Hazard ratios of all-cause mortality by quartiles of change in intact FGF23 concentrations from midlife (visit 3, 1993–1995) to late-life (visit 5, 2011–2013): ARIC Study, 1993–2019

18-year intact FGF23 change, pg/mL	HR for all-cause mortality (95%CI)
	Model 1	Model 2	Model 3
Q1: ≤6.4	1 [Reference]	1 [Reference]	1 [Reference]
Q2: 6.5 to 17.3	1.02 (0.86 to 1.22)	1.09 (0.91 to 1.30)	1.06 (0.88 to 1.27)
Q3: 17.4 to 31.2	1.20 (1.01 to 1.42)	1.24 (1.04 to 1.47)	1.13 (0.95 to 1.35)
Q4: ≥31.3	1.71 (1.45 to 2.00)	1.61 (1.36 to 1.90)	1.37 (1.15 to 1.64)

Model 1 was adjusted for age, sex, and race. Model 2 was adjusted for age, sex, race, diabetes, ever smoke, systolic blood pressure, medication use (antihypertensive, vitamin D, calcium, iron), prevalent heart failure, prevalent coronary heart disease, prevalent stroke, total cholesterol, high-density lipoprotein cholesterol, calcium, and phosphate. Model 3 was adjusted for eGFR change in addition to covariates in model 2.

When we simultaneously modeled FGF23 change and FGF23 at visit 5, FGF23 at visit 5, but not FGF23 change, was significantly associated with mortality (e.g., HRs, 1.48 [95%CI, 1.17 to 1.87] for the highest quartile of FGF23 at visit 5 and 1.10 [95%CI, 0.88 to 1.39] for the highest quartile for FGF23 change) (Supplemental Table 1). The HRs for FGF23 change were mostly unchanged with the further adjustment for eGFR at visit 5 (Supplemental Table 1). In subgroup analyses, the associations of FGF23 change with mortality were consistent regardless of age, sex, race, ever smoke, antihypertensive medication, diabetes, prevalent cardiovascular disease, FGF23, eGFR, and phosphate without significant interactions (Figure 3).

Figure 3: Subgroup analysis.

The hazard ratios are shown for the highest quartile of intact FGF23 change (Q4) compared to the other quartiles (Q1–3) as reference. The model was adjusted for age, sex, race, diabetes, ever smoke, systolic blood pressure, medication use (antihypertensive, vitamin D, calcium, iron), prevalent heart failure, prevalent coronary heart disease, prevalent stroke, total cholesterol, high-density lipoprotein cholesterol, calcium, and phosphate. Circles indicate point estimates, and horizontal lines indicate 95% confidence intervals. Abbreviations: FGF23, fibroblast growth factor 23; CVD, cardiovascular disease; eGFR, estimated glomerular filtration rate.

When the risk of mortality was assessed in the cross-categories of changes in FGF23 and eGFR, the HRs were higher in greater FGF23 increase and greater eGFR decline (Table 3). The risk was highest for the highest quartile of FGF23 change plus the lowest quartile of eGFR change (i.e., the largest decline) (HR, 2.16 [95%CI, 1.66 to 2.82]), compared to the lowest quartile of FGF23 change plus the highest quartile of eGFR change. There was no evidence of interaction between FGF23 change and eGFR change regarding their associations with mortality (p-for-interaction, 0.67). When the HRs were assessed in the cross-categories of FGF23 change and phosphate change, an increased risk was consistently observed in the highest quartile of FGF23 change, but not necessarily in the highest quartile of phosphate change (Supplementary Table 2). Finally, the associations were mostly unchanged when accounting for the attrition bias between visit 3 and visit 5 with the use of the inverse probability of weighting (Supplementary Table 3).

Table 3:

Cross-category of changes in intact FGF23 and eGFR from midlife (visit 3, 1993–1995) to late-life (visit 5, 2011–2013), and subsequent risk of all-cause of mortality: ARIC Study, 1993–2019

18-year change in intact FGF23 (pg/mL)	18-year change in eGFR (ml/min/1.73m2)
	Q4: ≥−11.3		Q3: −19.9 to −11.4		Q2: −30.4 to −20.0		Q1: ≤−30.5
	Deaths/ Participants, n	HR (95% CI)	Deaths/ Participants, n	HR (95% CI)	Deaths/ Participants, n	HR (95% CI)	Deaths/ Participants, n	HR (95% CI)
Q1: ≤6.4	77/567	1 [Reference]	68/409	0.99 (0.72–1.38)	56/254	1.38 (0.98–1.96)	36/134	1.51 (1.01–2.25)
Q2: 6.5 to 17.3	57/406	1.01 (0.71–1.42)	72/423	1.15 (0.83–1.59)	69/343	1.31 (0.94–1.82)	48/193	1.67 (1.16–2.41)
Q3: 17.4 to 31.2	39/267	1.07 (0.73–1.57)	63/328	1.38 (0.99–1.93)	83/401	1.28 (0.93–1.75)	106/368	1.78 (1.32–2.41)
Q4: ≥31.3	27/125	1.67 (1.07–2.58)	41/204	1.41 (0.96–2.07)	96/367	1.58 (1.17–2.15)	238/669	2.16 (1.66–2.82)

The model was adjusted for age, sex, race, diabetes, ever smoke, systolic blood pressure, medication use (antihypertensive, vitamin D, calcium, iron), prevalent heart failure, prevalent coronary heart disease, prevalent stroke, total cholesterol, high-density lipoprotein cholesterol, calcium, and phosphate.

Discussion

In a community-based cohort, serum intact FGF23 change over 18 years from mid-life to late-life was significantly associated with the subsequent risk of mortality. The observed association was independent of eGFR change and consistent across demographic and clinical subgroups. When both FGF23 change and FGF23 in late-life were simultaneously entered into the Cox model, FGF23 in late-life, but not FGF23 change, was an independent predictor of mortality. However, it was of note that FGF23 change from midlife to late-life was highly correlated with FGF23 concentrations in late-life.

Previous cross-sectional studies have shown an inverse association between FGF23 concentrations and eGFR,^{2, 3} which is generally consistent with a modest correlation between FGF23 change and eGFR change in our study (r=−0.42). Nonetheless, we noted that a non-negligible proportion of participants had a small increase in FGF23 despite a large decline in eGFR (or a large increase in FGF23 despite a small decline in eGFR) (Table 3), suggesting that the increase in FGF23 from midlife to late-life may not be simply explained by decreased kidney function. Future studies should investigate non-kidney factors that are associated with FGF23 change, although physiological changes during aging from mid-life to late-life may likely have some role.¹⁵

The association of FGF23 change with mortality risk was independent of eGFR change, although the association was certainly attenuated after accounting for kidney function. In addition, our cross-category analysis demonstrated that greater FGF23 increase and eGFR decline were jointly associated with high risk of mortality without clear evidence of interaction. Such findings extend previous studies relating single-measured intact FGF23 concentrations to the risk of mortality in older adults,^{7, 8} and are consistent with a recent report of 1,135 patients with moderate to severe CKD demonstrating a higher risk of mortality associated with a rapid increase in C-terminal FGF23 concentrations over 5 years.⁹

There are a few unique aspects of our study. First, to our knowledge, this is the first study to show an independent association of FGF23 change with subsequent risk of mortality in the general population. Second, we explored changes in FGF23 over almost two decades. Third, we rigorously assessed the prospective associations in the Cox models that simultaneously incorporated both FGF23 change and FGF23 at visit 5. Finally, FGF23 change even within its normal ranges was robustly associated with mortality. Although future confirmation studies are needed, if our findings are replicated, it will be worth exploring exact mechanisms linking increasing FGF23 concentrations even within normal ranges to mortality. Nonetheless, taken together, these findings further support the broad etiological implications of FGF23 beyond its relevance with kidney function.

When we included both FGF23 change and FGF23 at visit 5 simultaneously in a Cox model, FGF23 at visit 5, but not FGF23 change itself, remained significant. However, this should not be interpreted as the lack of prognostic value of changes in FGF23 since FGF23 change is a key determinant of FGF23 levels at visit 5 as represented by its strong correlation with FGF23 at visit 5 (r=0.77). Nonetheless, if the goal is prognostication, our results suggest that a single-point assessment of FGF23 in late-life should be prioritized over its change from midlife.

Our study has several clinical implications. Our study provides a logical basis to target FGF23 in interventions to improve health outcomes. Previous studies have suggested that interventions to phosphate regulation (e.g., low phosphate diet, phosphate binders) may lower FGF23 concentrations in patients with advanced CKD.^16–19 Thus, whether low phosphate intake can lower FGF23 concentrations in the general population may be a relevant future research topic, given that the average dietary phosphate intake among the US adults was higher than currently recommended.^{20, 21}

In addition, FGF23 is known to interact not only with phosphate metabolism but also with iron metabolism and erythropoiesis.²² Thus, the cross-talk between FGF23 and these conditions warrant future investigations. Finally, while our study is supportive of the utility of FGF23 to assess mortality risk in older adults, further data, such as the cost-effectiveness, should be evaluated to consider its routine measurements in clinical practice.²³

Our study findings should be interpreted in light of several limitations. First, we measured FGF23 concentrations at two time points approximately 18 years apart, which were based on the serum sample availability. Thus, whether FGF23 trajectory based on measurements with more than three time points provides additional prognostic information should be examined in future studies. Nonetheless, the mean ages at visit 3 (58 years) and visit 5 (76 years) allowed us to uniquely examine the implications of FGF23 change from midlife to late-life, an age period relevant to changes in bone-mineral metabolism and an increase in mortality. Second, our inclusion criteria required to have FGF23 measured at both visit 3 and 5. However, we confirmed that our findings were mostly unchanged when accounting for the attrition bias between visit 3 and visit 5 with the use of the inverse probability of weighting.¹⁴

Third, due to the nature of observational study, residual confounding is possible. Fourth, the study population consisted of Black and White older adults residing in the US communities, which may limit the generalizability of our findings to other age groups and races. Fifth, while our analysis was based on intact FGF23, a biologically active form of FGF23, whether our findings are similar when using C-terminal FGF23 should be explored. Sixth, degradation of intact FGF23 during the long-term sample storage is possible,²⁴ although previous studies have shown that intact FGF23 concentrations were stable to the storage in frozen samples up to 6 years at −80°C.^{25, 26} Seventh, analytical variation is likely to partly account for FGF23 change, particularly when the change was small (e.g., FGF23 change ≤17.3 pg/mL). Finally, although this study examined the risk of all-cause mortality because it was most clinically important outcome, future studies should explore how our findings are linked to other types of adverse events associated with FGF23, such as cardiovascular disease.

In conclusion, serum intact FGF23 change from midlife to late-life was associated with subsequent risk of mortality, independent of eGFR change. Our findings support the broad implications of FGF23 beyond its association with kidney disease.

Supplementary Material

01

Supplementary Figure 1: Cohort flow diagram.

02

Supplementary Figure 2: Directed acyclic graph to explain the association of intact FGF23 change from midlife to late-life with the risk of mortality.

03

Supplementary Table 1: Intact FGF23, eGFR, their changes from midlife (visit 3, 1993–1995) to late-life (visit 5, 2011–2013), and subsequent risk of all-cause of mortality: ARIC Study, 1993–2019

04

Supplementary Table 2: Cross-category of changes in intact FGF23 and phosphate from midlife (visit 3, 1993–1995) to late-life (visit 5, 2011–2013), and subsequent risk of all-cause of mortality: ARIC Study, 1993–2019.

05

Supplementary Table 3: Hazard ratios of all-cause mortality by quartiles of change in intact FGF23 and eGFR from midlife (visit 3, 1993–1995) to late-life (visit 5, 2011–2013) in a weighted sample accounting for the attribution bias: ARIC Study, 1993–2019.