QTc Interval Prolongation Is Independently Associated with FGF23 and Predicts Mortality in Predialysis Chronic Kidney Disease

Premature death is more likely than progression to dialysis dependence in the approximately one in ten people worldwide who suffer from chronic kidney disease (CKD) [1]. Cardiovascular disorders including coronary artery disease, heart failure, and especially sudden death cause about half of these premature deaths and occur at much higher rates than in the general population [3]. This cardiovascular morbidity is only partly explained by higher prevalence rates of traditional risk factors including diabetes and hypertension, and CKD-associated biomarkers such as FGF23 and pathologic left ventricular hypertrophy (LVH) are important contributors [2]. FGF23 is a phosphaturic hormone that rises to prevent or ameliorate hyperphosphatemia in CKD. It is tied to cardiac fibrosis, atrial fibrillation, LVH, heart failure, and both vascular and nonvascular death in patients with dialysis-dependent and predialysis CKD [3‒5]. Pathologic LVH – which occurs in up to half of people with advanced CKD and in response to FGF23 elevation, hypertension, anemia, and vascular stiffness – predicts overall, vascular, and sudden death in patients across the spectrum of CKD [2].

Acquired QTc prolongation also occurs at higher rates in patients with advancing CKD than in the general population. It is predicted by pathologic LVH, atrial fibrillation, coronary artery disease, diabetes, heart failure, electrolyte disturbances, use of QT prolonging medications, and hemodialysis-related fluxes in electrolyte concentrations [5‒8]. However, while it predicts overall and sudden death in patients with dialysis-dependent CKD, the associations and mortality impact of QTc prolongation in patients with predialysis CKD (who significantly outnumber those that require dialysis) remain understudied [5‒7].

Given that FGF23 induces LVH, LVH is associated with QTc prolongation, and QTc prolongation predicts death in patients with dialysis-dependent CKD, we surmised the existence of a pathway leading from FGF23 elevation to premature death in patients with predialysis CKD that entails LVH and QTc prolongation. To study this, we first investigated whether FGF23 was prospectively associated with QTc prolongation via LVH and then whether QTc prolongation predicted death in the CANadian study Assessing Inflammatory Markers to PREdict EVEnts in Nephrology Trial (CAN AIM to PREVENT), a prospective observational cohort study of 2,254 patients followed in three predialysis CKD clinics in Toronto, Canada.

The CAN AIM to PREVENT was an investigator-initiated prospective observational cohort study of 2,254 patients followed in three predialysis CKD clinics in Toronto, Canada, from 2010 to 2015. The primary objective of the study was to identify patterns of inflammation markers that predict progression to dialysis. It was registered at http://www.clinicaltrials.gov (#NCT01974713). Patients were invited by their usual nephrologist during regular clinic visits to participate in this study where they would be followed every 6 months for up to 3 years or until they died, required renal replacement therapy, were prescribed an erythropoietin-stimulating agent, withdrew consent, or were transferred to another institution. Inclusion criteria were receipt of care by a nephrologist (or awaiting nephrology consultation with an estimated glomerular filtration rate [GFR] less than 60 mL/min/m²) in one of the participating clinics and age over 18 years. Exclusion criteria were active renal replacement therapy (including functioning organ transplant), life expectancy of less than 12 months, use of or contraindications to an erythropoietin-stimulating agent, and the expectation that renal replacement therapy would be needed within 6 months.

After obtaining signed informed consent, baseline data from patients and charts were collected. Ethnicity was self-identified as Asian or Pacific Islander, black, white, or other. All medications were recorded, and measurements of vital signs, routine laboratory values, and serum C-terminal FGF23 levels were conducted at baseline and at each 6-month clinic visit. FGF23 was measured using an enzyme-linked immunosorbent assay targeting the carboxy-terminal fragment using reagents from Immutopics (FGF23) (San Clemente, CA, USA). Electrocardiograms (ECGs) were conducted at baseline and annually and performed at the CKD clinics. Two-dimensional surface echocardiograms were performed at regional cardiology clinics and conducted at baseline and at the end of the study.

The main outcomes assessed in this report were associations between FGF23, LVH, and QTc intervals and between prolonged QTc intervals and all-cause mortality. The QTc interval (corrected for the heart rate using the Bazett formula) and the presence of atrial fibrillation were recorded from each automated ECG report. The QTc interval was analyzed as a continuous variable and after categorization as normal (350–449 ms [350–459 if female]), borderline (450–499 ms [460–499 if female]), or prolonged (500 ms or greater). QTc intervals with concurrent atrial fibrillation were excluded when analyzing associations with FGF23 and included in survival analyses where atrial fibrillation was added as a covariate. LVH was defined as a left ventricular mass index (LVMi) greater than 131 g/m² (>100 g/m² if female) [9]. LVMi was defined using M-mode echocardiographic parameters and body surface area using the following formula: {0.8 × (1.04 × [left ventricular end diastolic dimension + septal thickness + posterior wall thickness]³ – left ventricular end diastolic dimension³) + 0.6}/body surface area [10]. The GFR was estimated using the 2009 CKD EPI equation and analyzed as a continuous or categorical variable with the following category designations: 60–46, 45–31, 30–16, and ≤15 mL/min/1.73 m². Any medications reported to affect QTc intervals were identified and categorized according to the Arizona Center for Education and Research on Therapeutics Registry into those that probably risk QTc prolongation, conditionally risk QTc prolongation, or possess known risk for torsades de points [11]. The number of medications taken in each category were counted (irrespective of dose) at each study visit. Time to death was defined as the number of days from the last ECG to death, and survival time was defined as the number of days from the last ECG to the following ECG or final assessment.

All data were analyzed in Stata (StataCorp. 2013. Stata Statistical Software: Release 13. College Station, TX: StataCorp LP). Descriptive statistics were reported using means, medians, and frequencies as appropriate. χ² analysis was used to assess the independence of QTc categories in relation to stage of CKD and time. The natural log of FGF23 was empirically determined to exhibit a linear trend with QTc intervals and used for all analyses (online suppl. Fig. 1; for all online suppl. material, see https://doi.org/10.1159/000535133). After confirming negative results for each fixed effect in multicollinearity testing (tolerance >0.4), mixed-effects general linear/logistic regression was fitted examining QTc, borderline QTc, and prolonged QTc in relation to the fixed effects of FGF23; LVH; age; sex; GFR; time; serum potassium, calcium, and phosphate; number of QTc-associated medications in each of the 3 categories; left ventricular ejection fraction; history of diabetes, coronary artery disease, or heart failure; and to the patient identification number as the random effect (models 1–3). To account for possible longer time period-dependent associations between FGF23 levels and QTc intervals, the models were repeated replacing FGF23 with previous FGF23 (i.e., that was measured 6 months before the QTc interval measurement) (models 4–6). To account for the possible effects of changes in LVH status over time, the initial 3 models were repeated using time-averaged (i.e., [final – baseline]/2) instead of baseline LVMI to define LVH (models 7–9). Mediation analysis was then performed using the Baron and Kenny approach provided in the structural equation modelling package by Stata. This allowed comparisons of direct, indirect (mediated by LVH), and total (direct + indirect) effects of FGF23 on prolonged and borderline QTc intervals and was repeated using previous FGF23 and then time-averaged LVH.

Unadjusted rates of all-cause deaths in patients with prolonged (vs. borderline or normal) QTc intervals and borderline (vs. normal) QTc intervals were compared using Kaplan-Meier survival curves and log-rank testing. Cox proportional survival hazards regression was performed adjusting for age, sex, albumin, phosphate, estimated GFR, the presence of atrial fibrillation, and LVH. These variables were chosen because they are reported to predict mortality and yet did not predict prolonged QTc in the current study.

The baseline data of the 2,254 CAN AIM to PREVENT participants are presented in Table 1. Their mean age was 68.7 years, and the estimated GFR was 41.4 mL/min/1.73 m². More than 98% self-identified their ethnicity as Asian or Pacific Islander, black, or white. They attended a mean of four study visits over a median 990 days (range: 3–1,568; interquartile range: 959–1,159 days). Five hundred and thirty-four (23.7%) participants did not complete the study – 204 (9.1%) withdrew consent or were unable to continue attending study visits, 138 (6.1%) died, 102 (4.5%) were lost to follow-up, 71 (3.1%) required renal replacement therapy, 14 (0.6%) transferred to another institution, and 5 (0.2%) initiated erythropoietin-stimulating agent therapy (Fig. 1).

A total of 12,785 FGF23 measurements (mean 5.7 per participant) were performed (Fig. 1). Echocardiograms were conducted in 2,085 participants at baseline and 1,616 at study end, with 1,543 receiving both. Baseline and time-averaged LVH was diagnosed in 10.8 and 8.6%, respectively. Of the 7,317 ECGs (mean 3.2 per participant), atrial fibrillation was identified in 288 (3.8%), and the QTc interval (mean 414+/−41.8 ms) was normal, borderline, and prolonged in 86.9, 10.9, and 2.2% of the remaining tracings, respectively. Borderline and prolonged QTc intervals were more frequent with advancing CKD and time, and significant relationships were demonstrated using χ² tests of independence in cross-tabulations between QTc intervals and CKD stage (χ² [6, N = 47], p < 0.001) or time (χ² [6, N = 90)] p < 0.001) (online suppl. Tables 1, 2).

In models 1–3, each unit increase in the natural log of FGF23 predicted continuous (3.54 ms [1.91, 5.18, p < 0.001]) and prolonged (odds ratio [OR] = 1.76 [1.15, 2.70, p = 0.009]) but not borderline (OR = 1.21 [1.00, 1.47, p = 0.053]) QTc intervals, while LVH predicted continuous (7.86 ms [3.26, 12.56, p = 0.001]) and borderline (OR = 1.81 [1.10, 2.96, p = 0.019]) but not prolonged (OR = 2.52 [0.99, 6.40, p = 0.051]) QTc intervals (Table 2; Fig. 2). In models 4–6, each unit increase in the natural log of the previous FGF23 predicted continuous (4.55 ms [2.76, 6.33, p < 0.001]), borderline (OR = 1.40 [1.09, 1.81, p = 0.010]), and prolonged (OR = 1.94 [1.17, 3.21, p = 0.010]) QTc intervals. LVH was not associated with continuous (4.62 ms [−0.59, 9.83, p = 0.082]), borderline (OR = 1.53 [0.80, 2.91, p = 0.196]), or prolonged (OR = 1.62 [0.56, 4.70, p = 0.379]) QTc intervals (Table 3; Fig. 3).

In models 7–9, each unit increase in the natural log of FGF23 predicted continuous (3.40 ms [1.66, 5.13, p < 0.001]) and prolonged (OR = 1.80 [1.11, 2.90, p = 0.016]) but not borderline (OR = 1.23 [1.00, 1.51, p = 0.059]) QTc intervals, while time-averaged LVH predicted continuous (7.86 ms [2.49, 13.22, p = 0.004]) and borderline (OR = 1.90 [1.06, 3.40, p = 0.030]) but not prolonged (OR = 1.89 [0.61, 5.81, p = 0.267]) QTc intervals (Table 4; Fig. 4). In mediation analysis, LVH indirectly mediated 11% and 13% of the total effects of FGF23 on borderline and prolonged QTc intervals. It indirectly mediated 6% and 5% of the total effects of previous FGF23, while time-averaged LVH indirectly mediated 8% and 9% of the total effects of FGF23 on borderline and prolonged QTc intervals, respectively. Compared with FGF23, previous FGF23 had larger direct (and similar indirect) effects on borderline and prolonged QTc intervals (Fig. 5).

One hundred thirty-eight (6.1%) participants died after a median period following study enrollment of 531 days (range: 13–1,175; interquartile range: 256–838). This included 18 (13.6%) of 132 with at least one prolonged QTc interval, 34 (5.4%) of 625 with at least one borderline prolonged QTc interval, and 86 (5.8%) of 1,477 participants with only normal QTc intervals. Participants with prolonged QTc intervals (vs. normal or borderline QTc intervals) had higher unadjusted (log rank p < 0.0001) and adjusted (hazard ratio = 2.06 [1.08, 3.92, p = 0.028]) risks for death (Fig. 6; Table 5). Mortality rates did not differ between patients with borderline versus normal QTc intervals (hazard ratio = 1.504 [0.99, 2.27, p = 0.057]).

In this study, we looked for associations between FGF23, LVH, and QTc prolongation and between QTc prolongation and overall deaths in 2,254 patients followed prospectively for up to 3 years in predialysis CKD clinics in Toronto, Canada. We found that FGF23 but not LVH was independently associated with QTc prolongation of ≥500 ms and that the adjusted mortality risk increased when QTc intervals were prolonged to this degree.

FGF23 rises early in CKD in response to subtle increases in serum phosphate concentrations and prevents or delays hyperphosphatemia by activating complexes of FGF23 receptors and transmembrane klotho co-receptors in proximal tubules that lead to urinary excretion of phosphate [12]. It is markedly elevated in hyperphosphatemic states such as advanced CKD where it is linked to adverse cardiovascular outcomes mediated by the cardiac receptor FGFR4 [13]. Klotho-independent FGF23/FGFR4 signaling activates nuclear factor of activated T-cells (NFAT)-dependent mechanisms that lead to LVH in human and animal models of CKD [13]. Since LVH also predicts QTc interval prolongation, we hypothesized it might mediate an indirect effect between FGF23 and QTc prolongation in CKD [15]. Instead, FGF23 had an LVH-independent association with QTc prolongation ≥500 ms, and only 13% of this effect was mediated by LVH. Furthermore, FGF23’s direct effect on QTc prolongation increased (and the indirect effect mediated by LVH remained unchanged) when values 6 months prior to the QTc assessment were used instead. This suggests there may be a time-dependent association between FGF23 and QTc prolongation that is LVH-independent. A more contemporaneous association between FGF23 and QTc intervals is also possible as FGF23/FGFR4 signaling increases calcium influx through L-type Ca2+ channels, thus raising transient calcium concentration and sarcoplasmic reticulum calcium content [16]. This delays after depolarizations and can lead to ventricular ectopy and arrhythmias. In their series of ex vivo and in vivo experiments, Graves et al. [16] demonstrated QTc prolongation and increased ventricular ectopy following an infusion of FGF23 to levels found in CKD in young, healthy mice. These effects were attenuated following antibody-mediated FGFR4 blockade, lending support for a direct effect of FGF23/FGFR4 signaling on dynamic QTc prolongation. While more research is needed to understand the time course and mechanisms of FGF23’s association with QTc interval prolongation, the available evidence suggests that FGF23 can underpin LVH and QTc≥500 ms separately in patients with predialysis CKD. This might have implications in other disease states – such as hypertension, stroke, and heart failure – where FGF23 elevation, LVH, and QTc prolongation intersect [17‒20].

While QTc prolongation predicts death in multiple populations including patients receiving dialysis, evidence for increased QTc mortality in patients with predialysis CKD are more limited [5]. Deo et al. [23] found evidence for increased death in Chronic Renal Insufficiency Cohort participants with borderline QTc intervals. However, this finding was no longer apparent when LVMi and left ventricular ejection fraction were added to their survival model. Malik et al. [24] found evidence for worse survival in National Health and Nutrition Examination Survey participants with both renal function impairment and borderline prolonged QTc intervals (vs. those with both normal renal function and normal QTc intervals). They also showed improved mortality risk stratification when QTc intervals were added to survival models of patients with renal impairment. However, no significant interaction was observed between eGFRs and prolonged QTc intervals, and it is uncertain how well their cohort represented patients with established predialysis CKD as the degree of renal impairment was mild (mean eGFR: ∼51 mL/min/1.73 m²); based on a single eGFR; and at 571 participants, was relatively small. The current study confirms that QTc prolongation can predict death in a large multiethnic cohort of patients with established predialysis CKD and contains design aspects that future studies concerned with QTc-related mortality or risk stratification might consider. For example, by conducting ECGs at baseline and annually, the current study confirmed the dynamic nature of QTc intervals [6]. It identified more cases of QTc prolongation – and allowed for more frequent, shorter, and more plausible observation periods to evaluate QTc mortality risk – than other studies employing single baseline QTc values and longer observation periods [25]. Furthermore, by categorizing the QTc interval as normal, borderline, or prolonged, we found that the increased mortality risk was restricted to patients with QTc intervals ≥500 ms. An increased risk with prolonged versus borderline QTc intervals has previously been described, suggesting that including patients with borderline intervals into the high-risk group might lead to an underestimation of QTc mortality risk [27].

Given that the QTc interval was not associated with renal function in any of the models tested, its increasing prevalence in progressive CKD likely reflects an accumulation of predictors. In contrast to FGF23, the frequency of hypokalemia (the only other modifiable covariate to correlate with QTc≥500 ms in all models tested) would be expected to decline as CKD progresses. We thus hypothesize that hypokalemia (which is strongly linked to QTc interval prolongation in numerous other studies) might play a more prominent role in QTc interval prolongation in settings such as the immediate post-hemodialysis period and peritoneal dialysis where it is more common [28‒30].

This study was strengthened by its prospective cohort design, its socialized health care setting, the number of covariates examined, and the frequencies of outcome measurements. The number and multiethnic makeup of enrolled patients improves the generalizability of the study results. However, certain limitations should be mentioned. First, QTc interval measurements were restricted to those from the automated reports, and individual ECG tracings were not manually reviewed. This approach was taken for practical reasons and is not expected to have systematically biased the results. Second, information on QRS intervals was not extracted from the automated reports, and tracings with intraventricular conduction delays were not excluded. This would cause the QTc interval to be overestimated in those instances, and the current study findings are thus extended to patients with concurrent QTc and QRS prolongation. Third, the FGF23 assay used was unable to differentiate between full-length FGF23 and its carboxy-terminal fragments. While the clinical consequences of conflating these may be minimal, future studies could determine whether associations between FGF23 and QTc are modified by type of FGF23 assay [13]. Fourth, baseline histories of coronary artery disease and heart failure were extracted from chart reviews rather than adjudicated. While this risked misclassification, it is unlikely to have systematically biased the study findings. Finally, as the causes of death could not always be ascertained, we resorted to analyzing overall rather than sudden death as a surrogate for QTc-related premature death. However, numerous studies have demonstrated concordance between all-cause mortality and sudden death risks with QTc prolongation [5].

In this prospective observational cohort of multiethnic patients with predialysis CKD, FGF23 was associated with LVH-independent QTc prolongation to at least 500 ms, and when present, this predicted a significant increase in mortality. Future studies delineating the mechanisms and interdependence behind these observations may be warranted.

Ethics Review Board approval was granted to conduct the study, and the protocol was registered at http://www.clinicaltrials.gov (#NCT01974713). As a post hoc analysis of data acquired in the CAN AIM to PREVENT, no additional Ethics Review Board approval was sought for the current study.

Written informed consent was obtained from participants (or their parent/legal guardian/next of kin) to participate in the CAN AIM to PREVENT. Additional written informed consent was not obtained for this post hoc analysis. The study was conducted ethically and in accordance with the World Medical Association Declaration of Helsinki.

The authors have no conflicts of interest to declare.

No external financial support was applied for or obtained for this study. None of the authors received any funding of relevance to this study.

Concept and design: Tabo Sikaneta, Natalie Ho, Antonio Bellasi, Bhavanesh Makanjee, and Jason Roberts; acquisition, analysis, or interpretation of data: Tabo Sikaneta, Sara Mahdavi, Hulya Taskapan, George Wu, Bharat Nathoo, and Paul Tam; statistical analysis: Tabo Sikaneta and Anton Svendrovski. Natalie Ho and Antonio Bellasi contributed equally to this work. Each author contributed important intellectual content during manuscript drafting or revision.

Natalie Ho and Antonio Bellasi contributed equally to this work.

The research data that support the findings of this study are not publicly available. Further inquiries can be directed to the corresponding author.