The association between changes in echocardiography and risk of heart failure hospitalizations and death in adults with chronic kidney disease

Adults with chronic kidney disease (CKD) are at increased risk for developing heart failure (HF). However, longitudinal cardiac remodeling in CKD has not been well-characterized and its association with HF outcomes remains unknown. We evaluated the association between change in echocardiographic parameters between baseline and year 4 with the subsequent risk of HF hospitalization and death using Cox proportional hazard models in a landmark analysis of a prospective multicenter CKD cohort. Among 2673 participants, mean ± SD age was 61 ± 11 years, with 45% women, and 56% non-white. A total of 472 hospitalizations for HF and 776 deaths occurred during a median (interquartile range) follow-up duration of 8.0 (6.3–9.1) years. Patients hospitalized for HF experienced larger preceding absolute increases in left ventricular (LV) volumes and decreases in LV ejection fraction. Adverse changes in LV ejection fraction, LV cavity volume, LV mass index, and LV geometry were independently associated with an increased risk of HF hospitalization and death. Among adults with CKD, deleterious cardiac remodeling occurs over a relatively short timeframe and adverse remodeling is associated with increased risk of HF-related morbidity and mortality.


Echocardiographic parameters.
Baseline and follow-up transthoracic echocardiograms were performed at years 1 and 4 after enrollment using a standardized protocol and transferred to an echocardiography core laboratory at the University of Pennsylvania (Philadelphia, PA) for systematic analysis. The median (interquartile range [IQR]) time between the year 1 and year 4 echocardiogram was 3.0 (2.9-3.1) years. Images were evaluated by a single physician according to contemporaneous guidelines from the American Society of Echocardiography (ASE) 15 . All sonographers performing and physicians interpreting studies in the echocardiography core laboratory were blinded to the participant's other clinical characteristics.
LV end-diastolic volume (LVEDV) and end-systolic volume (LVESV) were calculated using the biplane method of discs 16 and indexed to body surface area using the Mosteller formula (LVEDVI and LVESVI, respectively) 17 . LVEF was calculated as (LVEDV-LVESV)/LVEDV × 100 and further stratified into reduced (< 40%), mid-range (40-49%), and preserved LVEF (≥ 50%). LV mass (LVM) was calculated using the area-length method 16 and indexed to body surface area (LVM index [LMVI]). LV geometry was categorized as normal, concentric remodeling, eccentric hypertrophy, and concentric hypertrophy based on relative wall thickness and LVMI as defined by the ASE 16 . Prevalence of missing echocardiographic variables is provided in the supplement (Supplement, Table 1).
Covariates. Demographic information, medical comorbidities, and medication use was ascertained via self-report questionnaires administered during annual in-person visits and updated during bi-annual telephone visits. Systolic and diastolic blood pressure, heart rate, height, weight, and urine/blood specimens were collected during annual in-person visits using standardized methods. An average of three seated blood pressure measurements was recorded. All laboratory testing was performed in a biomarker core laboratory, and eGFR was calculated using the 2009 Chronic Kidney Disease Epidemiology (CKD-EPI) Collaboration formula 18 . All covariates used in the analysis were obtained at year 4, except for B-type natriuretic peptide, which was only obtained at the time of enrollment.
Follow-up and outcomes. The primary outcomes of hospitalization for HF and death were collected from the year 4 visit through November 30, 2018. CRIC study personnel screened participants or proxies for hospitalizations and death on a bi-annual basis via telephone interviews. All identified hospitalizations with International Classification of Disease 9th Edition discharge diagnosis codes related to HF (i.e., 398.91, 402.01, 402.11, 402.91, 425.xx, 428.xx, 429.xx, 514.xx, 518.4) were reviewed by two physicians for evidence of documented symptoms, physical exam findings, chest radiographs, echocardiograms, and invasive hemodynamic monitoring if available. A HF hospitalization was considered to have occurred when both reviewers agreed it was "probable" or "definite" based on the modified Framingham clinical criteria 19 . Deaths were also identified through death certificates or obituaries, review of hospital records, and from the Social Security Master File, as available. Participants were censored at the end of follow-up, study withdrawal, or death. Statistical analysis. All analyses were conducted using SAS, version 9.4 (Cary, NC, USA) and R, version 4.0.2 (https:// www.r-proje ct. org/). Baseline characteristics were collected at year 4 (i.e., second echocardiogram) for this landmark analysis and were stratified by the presence or absence of HF hospitalization. Categorical variables are presented as frequencies with percentages and continuous variables as means with standard deviations. We describe the changes from year 1 to year 4 as both absolute and relative differences for continuous vari- www.nature.com/scientificreports/ ables, and category changes for categorical variables. Differences between baseline characteristics and changes in echocardiographic parameters were compared using analysis of variance or Kruskal-Wallis tests for continuous variables and Chi-square tests for categorical variables. We first used Kaplan-Meier survival curves to assess differences in hospitalization for HF and all-cause death among participants stratified by the level and direction of 3-year changes in LVEF, LVMI, LVESVI index, and LVEDVI. Cut points for Kaplan-Meier curves were defined using 10% of the population mean value at year 1; for example, if the mean LVEF at year 1 was 50%, we categorized individuals as having "No Change" if the difference in LVEF from year 1 to 4 was between -5% and 5%, an "Increase" if the difference in LVEF was > 5%, or a "Decrease" if the difference in LVEF was < − 5%. Next, we evaluated the independent associations between 3-year absolute changes in echocardiographic measurements and the outcomes of interest using multivariable Cox proportional hazard models controlling for potential confounders at year 4, including demographics (age, sex, race/ethnicity), medical history (presence of HF, HF hospitalizations before the year 4 study visit, atrial fibrillation, acute myocardial infarction, coronary revascularization, peripheral vascular disease, stroke, tobacco use, alcohol use), vital signs (body mass index, systolic blood pressure), and laboratory values (eGFR by CKD-EPI equation, low-density lipoprotein cholesterol, hemoglobin, and glycosylated hemoglobin), and echocardiographic parameter value at baseline (i.e., year 1), and medical therapy for HF (beta blockers, loop diuretics, and potassium sparing diuretics). We included a random effect for CRIC clinical center to account for within-center correlation and used a multiple imputation approach across 50 imputed datasets to account for missing data in echocardiographic and laboratory variables. Finally, an interaction analysis was performed between change in echocardiographic parameters and markers of kidney function (eGFR and urine protein-to-creatinine ratio) for the outcomes of HF hospitalization and death.

Results
Among 2,673 participants with serial echocardiograms, mean age was 61.7 ± 10.6 years, 45% were women, and 56% self-identified as non-white (Table 1). There was a high burden of cardiovascular and non-cardiovascular disease, with 12.4% of participants diagnosed with HF. Participants hospitalized for HF were more likely to be older, non-white, and had a higher burden of medical comorbidities. The average eGFR was lower among those hospitalized for HF, while glycosylated hemoglobin and brain natriuretic peptide were higher (Table 1).
Year 1 and year 4 echocardiographic parameters, stratified by subsequent HF hospitalization status, are presented in Table 2. Patients hospitalized for HF after year 4 had more severe preceding abnormalities in cardiac structure and function. Between year 1 and year 4, average LV volumes increased, while mean LVEF declined across the cohort, but the magnitude of change was larger in those with a subsequent HF hospitalization (Table 3, Fig. 1). Mean LVMI increased in patients subsequently hospitalized for HF and declined in those without a hospitalization for HF, although the differences were not statistically significant. Eccentric hypertrophy became more common over the study period, especially among those who experienced a subsequent HF hospitalization. Participants with more severe baseline cardiac abnormalities (i.e., lower LVEF and higher LVESV/LVEDV) experienced greater improvements in LVEF and decreases in LV volumes between year 1 and year 4 ( Fig. 2).
After adjustment for clinical confounders, baseline echocardiographic parameters, and medical therapy for HF, adverse changes in echocardiographic measures were generally associated with an increased risk of HF hospitalizations and, to a lesser extent, death (Table 4). Decreases in LVEF and increase in LVMI, LVESVI, LVEDVI, and left atrial area were independently associated with higher risks of both outcomes. Normalization of LV geometry was associated with a lower risk of both HF hospitalization and death, while an improvement in LVEF category was associated with a lower risk of HF hospitalization. There was no interaction between the change in echocardiographic parameters and baseline eGFR and urine protein-to-creatinine ratio for the outcomes of HF hospitalization and death (Supplement, Table 2).

Discussion
In this prospective study of 2673 adults with mild-to-moderate CKD at baseline, we observed significant cardiac remodeling over a 3-year time period with reductions in LVEF and increases in LV volumes, particularly LVESV. Participants with more severe baseline LV cavity dilation and systolic dysfunction experienced the greatest improvements (i.e., normalization) in echocardiographic parameters during the 3-year period. Participants subsequently hospitalized for HF experienced more adverse cardiac remodeling (i.e., reduction in LVEF and increase in LV volume) over the preceding years. There was a trend toward increased LVMI in subjects hospitalized for HF and they were more likely to develop eccentric LV hypertrophy than those not hospitalized for HF. In fully adjusted models that accounted for baseline clinical characteristics and echocardiographic parameters, adverse changes in LV volumes, LVEF, and LVMI were all independently associated with increased risk of HF hospitalizations and death. These findings suggest serial echocardiograms over a timeframe of several years can detect cardiac remodeling in individuals with CKD, and that even modest changes in cardiac structure and function can be prognostically meaningful for outcomes of HF-related morbidity and mortality.
To our knowledge, this is the first study to prospectively report longitudinal echocardiographic changes in a large contemporary cohort with CKD. Serial echocardiographic measurements among the subset of participants  20,21 . Compared to our larger cohort, the 417 patients who progressed to ESKD had modest but significant declines in LVMI and experienced no change in LVEDVI, but other echocardiographic parameters were consistent with our findings. In a smaller study of 98 patients with stage 5 CKD, LVMI and LV geometry did not change appreciably over a 2-year study period 22 . The mean baseline eGFR of participants who developed ESKD during the CRIC study was predictably lower than the overall cohort (24.6 vs. 41.7 ml/min/1.73 m 2 ). The discovery of distinct patterns of cardiac remodeling at different eGFR ranges within the same cohort suggests that the relationship between eGFR and cardiac remodeling may change as CKD progresses. Alternatively, other clinical circumstances associated with advanced CKD may influence cardiac remodeling in a manner that opposes the effect of CKD (i.e., effect modification). For example, an elevated glycosylated hemoglobin is independently associated with abnormal cardiac remodeling 23 , www.nature.com/scientificreports/ but in advanced CKD, reduced insulin clearance contributes to lower glycosylated hemoglobin levels 24 . Data from cross-sectional studies, which are more vulnerable to this type of effect modification, are heterogenous but have generally shown that markers of deleterious cardiac remodeling worsen as eGFR declines 4,6,12 . Serial echocardiograms across a longer time horizon could detect non-linear changes in cardiac structure and function and provide a more comprehensive understanding of cardiac remodeling across a wide spectrum of eGFR.
In clinical practice, reversal of abnormal echocardiographic findings is typically interpreted as a favorable prognostic finding. Given this assumption, some of our findings may at first seem counterintuitive. In our Kaplan-Meier analysis, participants with seemingly favorable cardiac remodeling (i.e., increases in LVEF and decreases in LVESVI or LVEDVI) consistently experienced higher crude rates of HF hospitalization and death than subjects with no change in their echocardiogram. However, after adjustment for baseline echocardiographic parameters, an improvement in LVEF category and normalization of LV geometry had strong favorable associations with subsequent HF hospitalization and death. Review of participant-level data reveals that subjects with lower baseline LVEF and larger LV volumes experienced more substantial improvement in echocardiographic measures. Measurement error may partially explain this finding, but a physiologic mechanism is also possible. A potential explanation is that adults with more abnormal baseline echocardiography were aggressively treated with guideline directed medical therapy for HF which had beneficial effects on cardiac remodeling. Regardless of mechanism, the ultimate consequence of this trend was that subjects with more favorable echocardiographic changes also had more abnormal echocardiograms and therefore appeared to have worse outcomes in analysis that did not account for baseline echocardiography. Our findings highlight the critical importance of interpreting changes in echocardiography within the clinical context of baseline values.
Our study had several limitations. The CRIC study excluded subjects with NYHA functional class III/IV HF, which likely reduced participation among adults with prevalent HF and reduced systolic function and lowered the event rate for HF hospitalizations and potentially death. However, this decision allowed us to describe the natural history of echocardiographic findings in the setting of mild-to-moderate CKD patients with a low rate of prevalent HF which is unique. The protocol required echocardiograms at only two time points separated by a median of 3 years, potentially limiting our ability to detect non-linear associations between CKD progression   www.nature.com/scientificreports/ and cardiac remodeling. Low rates of moderate to severe valvular heart disease at baseline precluded analyzing these findings. The approximately 20% of CRIC participants who did not have a year 4 echocardiogram were sicker than our analytical cohort (Supplement, Table 3) and approximately 40% died prior to year 4. Therefore, our results may not be fully representative of critically ill patients or those near the end of life. Contemporary measures of diastolic function such as tissue Doppler imaging and estimates of pulmonary artery systolic pressure were not collected and limited our ability to analyze diastology. However, we did include left atrial area which is an important marker of chronically elevated left-sided filling pressures. Finally, adjudication of HF hospitalizations did not require a change in HF-related therapy, which may have reduced the diagnostic specificity of the operational definition for this endpoint.
In conclusion, we found that deleterious cardiac remodeling occurs over a relatively short time horizon in the setting of mild-to-moderate CKD and that adults subsequently hospitalized for HF experienced larger preceding  www.nature.com/scientificreports/ absolute decreases in LV systolic function and increases in LV volumes. Furthermore, adverse changes in LVEF, LV volumes, LVMI, and LV geometry were all independently associated with increased risks of subsequent HF hospitalization and death. Effective blood pressure control and guideline directed medical therapy for HF, when indicated, may prevent deleterious cardiac remodeling and improve outcomes in CKD populations. Echocardiography can help identify patients at high risk for HF-related morbidity, and the potential benefit of routine echocardiography in CKD populations should be investigated in prospective trials. Future studies should also explore potential non-linear cardiac remodeling in CKD and further investigate the prognostic significance of (pseudo)normalization of echocardiographic parameters.

Data availability
The datasets generated during and/or analyzed during the current study are available from the corresponding author on reasonable request.