Recurrent Hospitalizations for Fluid Overload in Diabetes with Kidney Failure Treated with Dialysis

The prevalence of end-stage kidney disease (ESKD) increased by nearly 50% worldwide between 2003 and 2016, so that there were nearly 2.5 million people receiving kidney replacement therapy in 2016 [1]. Patients with ESKD are frequently hospitalized and readmitted, compared with the general population [2]. Early readmissions within 30 days occurred in 10–25% of patients treated with peritoneal dialysis and hemodialysis [3‒5]. These unplanned readmissions are associated with increased risk of death [6, 7] and higher healthcare costs [2, 3, 6, 7]. Several studies have explored all-cause readmissions after hospitalizations for all-causes or heart failure [7‒14], but few have specifically examined potentially avoidable, unplanned readmissions [8], or fluid overload readmissions in patients with ESKD [4]. Fluid overload in ESKD has been attributed to congestive heart failure, missed dialysis sessions, nonadherence to fluid restriction, and inappropriate dry weight prescriptions [15, 16]. Locally, fluid overload-related hospitalizations increased between 2017 and 2022; this trend may be related to a corresponding increase in the proportion with stage G5 chronic kidney disease (CKD) [17]. Fluid overload is a common cause of hospitalization and accounted for 12–44% of readmissions in patients treated with dialysis [3, 4]. Moreover, the recurrence of fluid overload after an index hospitalization for the same condition can be considered an ambulatory-sensitive, potentially avoidable cause for unplanned readmission [8]. Hence, readmissions for fluid overload deserve greater attention, focused efforts, and dedicated resources to implement preventative interventions. It is thus important to understand the predictors of fluid overload readmissions. These predictors may inform strategies to identify individuals likely to have recurrent fluid overload, and modifiable risk factors can become targets for interventions to reduce avoidable, unplanned readmissions [4].

Diabetes is the most common cause of ESKD locally [18] and among developed countries worldwide [19]. Additionally, individuals with diabetes and kidney disease have a high prevalence of atherosclerotic cardiovascular disease and heart failure [20, 21]. These patients are likely to form the majority of those with hospitalizations and readmissions for fluid overload, so implementing preventative interventions in these patients may have the greatest impact on improving clinical outcomes and healthcare utilization. Hence, we aimed to identify the risk factors for fluid overload readmissions in individuals with diabetes and dialysis-dependent ESKD.

We conducted a retrospective cohort study of all hospitalizations for fluid overload among adults with diabetes and ESKD treated with dialysis at the SingHealth hospitals between 2018 and 2021. The SingHealth cluster of healthcare institutions is the large public healthcare cluster in Singapore that integrates primary and tertiary care and includes four acute hospitals in the east and central part of Singapore [22]. We utilized the comprehensive SingHealth Diabetes Registry that incorporated electronic medical records across the different levels of care in SingHealth [22]. Hospitalizations for fluid overload were identified from discharge codes for fluid overload, heart failure, congestive heart failure, pulmonary edema, and generalized edema (online suppl. Table S1; for all online suppl. material, see https://doi.org/10.1159/000542446). Patients were excluded if they died during the hospitalization or were lost to follow-up. Demographics data (age, gender, ethnicity), comorbidities (hypertension, cardiovascular disease, atrial fibrillation, history of cancer, systolic and diastolic blood pressure [BP], body mass index [BMI], Charlson Comorbidity Index [CCI]), biochemistry (hemoglobin), medications, and healthcare visits were retrieved until the last visit before 31 December 2022. Cardiovascular disease was present if the patient had heart failure, ischemic heart disease (IHD), myocardial infarction, stroke, or peripheral vascular disease (PVD). Hypertension was defined as systolic BP >140 mm Hg, diastolic BP >90 mm Hg at hospitalization, or received antihypertensive treatment. The Charlson Comorbidity Index (CCI) was derived using the Quan et al. [23] algorithm. The LACE index was calculated according to the length of stay (LOS), acuity of admission, CCI, and number of emergency department visits in the past 6 months to pinpoint those at elevated risk for early readmission [24] and is one of the most commonly used models in patients with and without heart disease [25, 26]. Healthcare visits were defined as visits to ambulatory clinics or emergency departments and hospitalizations. Prescriptions for sodium-glucose cotransporter-2 (SGLT2) inhibitors, glucagon-like peptide-1 (GLP1) receptor agonists, renin-angiotensin system (RAS) blockers such as angiotensin-converting enzyme inhibitors and angiotensin-receptor blockers, diuretics, mineralocorticoid receptor blockers (MRB), and statins within 6 months before the hospitalization and at discharge were retrieved from electronic prescription records. Other than clinical factors related to disease severity and patterns of healthcare utilization, this study also considered smoking status, history of alcoholism, and rental public housing as proxies for socioeconomic health determinants, given their potential impact on hospital readmissions [27, 28]. The primary outcomes of interest were hospital readmissions for fluid overload within 30 days and 90 days of discharge from the index hospitalization.

This study abided by the Declaration of Helsinki. Our institution’s Centralized Institutional Review Board (CIRB 2023/2216) determined that the study did not require ethical deliberation for using de-identified data generated during routine clinical care. The study was reported according to the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) checklist (online suppl. Table S2).

Statistical analysis was performed using performed using R Statistical Software (v4.2.1; R Core Team 2022). Categorical variables were presented as proportions, and continuous variables were summarized as mean ± standard deviation or median with interquartile range (25th percentile, 75th percentile) as appropriate. The Prentice, Williams and Peterson Total Time (PWP-TT) model was used to assess covariable effects on readmissions following discharge from the index admission [29], accounting for intraindividual correlations [30]. In recognition of the complex readmission patterns that may occur (online suppl. Fig. S1), the PWP-TT model stratifies by prior readmission count that allows baseline risk to increase with each readmission. For clinical relevance, data were truncated at 30 and 90 days after discharge from the index admission. Detailed information on the data structure is provided in online supplementary Table S3. Covariables were chosen a priori as clinically important to fluid overload, heart failure, and all-cause hospitalizations in diabetes [4, 8, 31]. The multivariable model was adjusted for age, sex, race, prior hospitalization for fluid overload, prior emergency visits, history of alcoholism, current smoker status, rental public housing status, CCI, BMI, heart failure, IHD, myocardial infarction, stroke, PVD, atrial fibrillation, cancer, hypertension, diabetes duration, admission via the emergency department, medications at discharge (thiazide, loop diuretic, MRB, RAS blocker, statin), hemoglobin level at discharge, hospitalization LOS, and LACE score. Multicollinearity was assessed using generalized variance inflation factors (GVIF), adjusted for the degrees of freedom of each variable. The adjusted GVIF, calculated as GVIF^(1/(2*df)), was used with values <2.236 (equivalent to VIF <5) considered acceptable. We performed a complete case analysis as the base analysis. To identify whether the effects of potential risk factors on readmissions varied depending on the presence of known heart failure, we evaluated for interactions between heart failure and potential predictors. We further examined the magnitude of the associations between the predictors in subgroup analysis according to the presence and absence of heart failure. To address missing data in BMI, diabetes duration, and hemoglobin levels, we performed multiple imputation using the Multivariate Imputation by Chained Equations algorithm [32]. We performed a sensitivity analysis on 100 imputed datasets generated using predictive mean matching as the imputation method. Predictive mean matching is a semi-parametric approach that imputes missing values by selecting observed values with the closest predicted values, thus preserving the distribution of the observed data. The results were combined using Rubin’s rules for pooling to produce overall estimates and confidence intervals. Additionally, sensitivity analyses were performed with the addition of SGLT2 inhibitors and GLP1 receptor agonists in the multivariable model. All tests were two-tailed, and statistical significance was defined as p < 0.05 unless otherwise stated.

We identified 3,234 hospitalizations for fluid overload among 1,821 patients, after excluding 502 hospitalizations without follow-up visits (Fig. 1). The demographic and clinical characteristics of the hospitalizations are detailed in Table 1. The study cohort predominantly consisted of patients with type 2 diabetes (99.4%). Most were older adults aged ≥65 years (58.2%) and male (61.1%). Cardiovascular disease and hypertension were prevalent. The majority was admitted via the emergency department, and almost all had a LACE score of 10 or more, signifying a cohort at high risk for readmission [24].

Of the 3,234 hospitalizations during the study period, 585 (18.1%) resulted in readmission for fluid overload within 30 days, with each hospitalization considered an independent event. Table 1 presents the characteristics of hospitalizations with and without 30-day readmission for fluid overload. Hospitalizations with 30-day readmission showed a higher proportion of males and active smokers, more instances of prior hospitalization due to fluid overload, and more cases with two or more emergency department visits in the past 6 months. These hospitalizations also more frequently involved admission via emergency for the prior hospitalization, presence of cardiovascular disease (including heart failure, IHD, myocardial infarction, and PVD), and atrial fibrillation. This group also had higher CCI, higher LACE index, and shorter LOS. At discharge during the fluid overload hospitalization, a smaller proportion of these hospitalizations with readmissions involved prescriptions for RAS blockers and statins, while a larger proportion received MRB.

Extending the analysis to 90 days, readmissions for fluid overload occurred in 967 out of 3,234 hospitalizations (29.9%). In addition to the characteristics observed in the 30-day analysis, hospitalizations with 90-day readmission included a higher proportion of cases with hypertension and patients residing in public rental housing. At discharge during the prior fluid overload hospitalization, a smaller proportion of these hospitalizations involved prescriptions for RAS blockers and statins, while a larger proportion received MRB, similar to the pattern observed in the 30-day analysis.

To accommodate the PWP-TT model’s data structure, which defines risk intervals from discharge dates, we excluded hospitalizations with discharge dates extending beyond 2021. Our final analysis included 3,190 hospitalizations from 1,801 patients. Over a median follow-up period of 20.4 months, 641 out of 1,801 patients (35.6%) experienced a total of 1,389 readmissions for fluid overload, with the number of readmissions per patient ranging up to 21 (online suppl. Fig. S2). This comprised 641 first readmissions and 748 subsequent readmissions, illustrating the recurrent nature of fluid overload events.

For clinical relevance, we truncated the data at two time points: 30 and 90 days after discharge from the index admission. Within 30 days, there were 208 readmissions from 1,952 hospitalizations among 1,801 patients. Within 90 days, there were 425 readmissions from 2,192 hospitalizations among 1,801 patients. Online supplementary Table S4 shows the number of patients and time to readmission according to the number of readmissions. Among those with one and two readmissions within 30 days, the median time between the previous discharge and readmission were 9 (5, 18) and 4.5 (2, 7) days, respectively. Among those with one, two and three readmissions within 90 days, the median time between the previous discharge and readmission were 24 (8.75, 47), 14 (4, 28), and 19 (9.5, 32.5) days, respectively. Due to missing covariate data, the multivariable regression analysis for factors associated with readmissions included 1,528 hospitalizations from 1,425 patients (164 readmissions) for the 30-day outcome, and 1,697 hospitalizations from 1,444 patients (329 readmissions) for the 90-day outcome, in the complete case analysis (Table 2). Patients with IHD, PVD, and lower hemoglobin levels showed an increased risk of readmission within both 30 and 90 days. Heart failure was associated with higher readmission risk within 90 days. In contrast, patients on peritoneal dialysis (compared to hemodialysis) and those prescribed statins had a lower risk of readmission for fluid overload within 90 days. There was no significant multicollinearity in the models (online suppl. Table S5).

Online supplementary Table S6 showed that there were significant interactions between heart failure and several factors: ethnicity (specifically for Malay and Indian patients compared to Chinese patients), PVD, and statin at discharge in the 30-day readmission risk model. In the 90-day readmission risk model, significant interactions were observed between heart failure and atrial fibrillation. The subgroup analyses according to the presence and absence of cardiovascular disease (online suppl. Table S7 and S8, respectively) confirmed that the stratum-specific adjusted odds ratios were different. Among patients with heart failure, several factors were associated with an increased risk of fluid overload readmission. For 30-day readmissions, Indian ethnicity (compared to Chinese), the presence of PVD, and the absence of statin prescription at discharge were significant risk factors. When examining 90-day readmissions, the risk was higher for heart failure patients with PVD, those without stroke and atrial fibrillation, and those not prescribed statins at discharge. For patients without heart failure, MRB prescription and lower hemoglobin levels were associated with higher risk of fluid overload readmission within both 30 and 90 days. Additionally, the presence of PVD was linked to increased readmission risk within 90 days.

Online supplementary Table S9 shows the characteristics of the multiple imputation cohort, and Online supplementary Table S10 shows the sensitivity analysis. In the analysis using the multiple imputation cohort, 30-day readmission risk was associated with IHD, PVD, and lower hemoglobin levels, as in the complete case analysis. Additionally, MRB prescription at discharge and shorter LOS were identified as risk factors. For 90-day readmissions, the imputation cohort analysis aligned with the complete case analysis, showing increased risk associated with PVD, statin prescription, and lower hemoglobin levels. However, unlike the complete case analysis, peritoneal dialysis (compared to hemodialysis), heart failure, and IHD did not show significant associations with readmission risk in the imputation cohort.

Online supplementary Table S11 shows sensitivity analyses with the inclusion of SGLT2 inhibitors and GLP1 receptor agonists to the base model. After adjustment for the demographic, clinical, and psychosocial variables, neither SGLT2 inhibitors nor GLP1 receptor agonists were associated with readmissions for fluid overload.

In this retrospective cohort study of hospitalizations for fluid overload among adults with diabetic kidney disease on dialysis, the 30-day and 90-day readmission rates for fluid overload were 18.1% and 29.9%, respectively. Heart failure, IHD, PVD, hemodialysis (compared to peritoneal dialysis), lower hemoglobin level, and absence of statin at discharge were independently associated with increased risk of readmissions among individuals with diabetes and ESKD treated with dialysis. Atherosclerotic cardiovascular disease and heart failure are common among individuals with diabetes and kidney disease [20, 21], and the prevalence of cardiovascular disease increases incrementally with reduced kidney function [33]. The American Heart Association recognized that multidirectional interactions between metabolic risk factors (such as diabetes), CKD, and the cardiovascular system in the cardiovascular-kidney-metabolic syndrome result in multiple organ system dysfunction [34], with subsequent decompensation that manifests as fluid overload. Indeed, fluid overload measured by bioimpedance spectroscopy was associated with structural and functional cardiac abnormalities [35]. Anemia was associated with increased hospitalizations and readmissions among patients with CKD and dialysis [9, 36]. Plasma volume expansion in early fluid overload may cause hemodilution and “apparent” anemia [37]. Thus, a higher hemoglobin at discharge may indicate better fluid volume control that lowers the risk of rapid recurrence of symptomatic fluid overload. Anemia can also exacerbate cardiac ischemia and myocardial dysfunction with resultant decompensated fluid overload [36]. A systematic review of 18 studies on heart failure (368 patients) found that red blood cell volume was negatively associated with left ventricular ejection fraction (LVEF) [38]. Thus, guidelines for heart failure and CKD recommend anemia management with intravenous iron and erythropoiesis stimulating agents [21, 39].

Our patients on peritoneal dialysis had a significantly lower risk of 90-day readmission due to fluid overload compared to those on hemodialysis. However, the difference in 30-day readmission was not statistically significant. Perl et al. demonstrated that the risk of 30-day readmission was higher for patients on home-based peritoneal dialysis [40], while Xu et al. [5] did not find any difference in all-cause readmission between dialysis modalities. We postulated that daily fluid removal and possibly better-preserved residual kidney function in peritoneal dialysis may reduce the risk of fluid overload recurrence compared to our practice of in-center intermittent hemodialysis performed over 4 h, three times per week. The larger fluid volume accrued between hemodialysis sessions, compared to daily peritoneal dialysis sessions, can lead to elevated BP and left ventricular hypertrophy, while the faster ultrafiltration rate increases the risk of myocardial stunning, intradialytic hypotension, and cramps that limit fluid volume removal, thereby predisposing individuals treated with hemodialysis to fluid overload and/or heart failure. Peritoneal dialysis was effective in sodium removal [41, 42], improving heart failure symptoms and reducing hospitalization LOS among individuals with heart failure [43], including those with refractory heart failure, without significantly affecting kidney function [44].

The prescription of statins at discharge was associated with lower risk of 90-day readmissions for fluid overload. Although several trials [45‒47] failed to demonstrate the cardiovascular benefits of statins in patients on dialysis, subsequent post hoc analyses have shown that statins significantly reduced cardiovascular events in dialysis patients with high atherosclerotic cardiovascular disease risk [48‒50]. Notably, our study population comprised patients with diabetic kidney disease with ESKD and, thus, a high-risk group for atherosclerotic cardiovascular disease. The potential benefit of statins in the dialysis population remains controversial and warrants further research to identify patients who would benefit from these treatments. The prescription of RAS blockers at discharge was not associated with 30-day and 90-day readmissions for fluid overload in this study. The evidence for RAS blockers to reduce heart failure and cardiovascular disease in ESKD has been limited by trial exclusion [51]. An observational study found no association between the type of antihypertensive medications and fluid overload-related readmissions in hemodialysis [52]. Furthermore, systematic reviews did not find an association between RAS blockers and mortality or cardiovascular events in hemodialysis [53, 54].

When individual hospitalizations were considered in the univariable analysis, a recent hospitalization for fluid overload in the past 6 months was strongly associated with 30- and 90-day readmission for fluid overload. Similarly, recent hospitalization for pulmonary edema was the strongest predictor for 30-day readmission for pulmonary edema, with a 2.4-fold increased risk compared to those without pulmonary edema [4]. However, prior hospitalization for fluid overload was not significantly associated with readmissions in the multiple failure models after accounting for other comorbidities. The post-discharge recurrence of severe symptomatic fluid overload requiring readmissions has been attributed to challenges in ensuring patient adherence to salt and fluid intake restrictions, fluid status and dry weight assessment, and inadequate care coordination between dialysis facilities and hospitals after discharge [14]. Early post-discharge assessment of fluid volume status in the dialysis facility, including the use of adjunctive techniques such as bioimpedance spectroscopy [55‒57] and lung ultrasonography assessment of interstitial lung water [58‒61], may empower dialysis healthcare staff to assess fluid status, improve dry weight adjustment, and thus mitigate the risk of fluid overload and subsequent hospitalization. Although these adjunctive techniques are not widely available to many dialysis providers currently, many patients with ESKD receive incenter hemodialysis [1], so the frequent contact with hemodialysis care providers provides opportunities to address precipitants of fluid overload and prevent its recurrence, and/or identify mild fluid overload early and institute lifestyle and treatment modifications to reverse the progression to severe symptomatic fluid overload requiring hospitalizations. Notably, more frequent provider visits after discharge reduced the risk of 30-day readmission [3, 13], although interventions to improve communication and coordination between hospitals and dialysis centers did not significantly reduce 30-day readmissions [62]. Our institution routinely generates a detailed memo at discharge that aims to inform the community dialysis provider about changes to their medications and dry weight to improve transitions in care after discharge. Additional strategies to reduce readmissions for fluid overload can be targeted at individuals with elevated readmission risks based on our findings.

Our study has several limitations. This observational study cannot determine causality but generates hypotheses that should be tested by future randomized controlled trials. The absence of data on dialysis vintage, dry weight, interdialytic weight gain, N-terminal (NT) pro-b-type natriuretic (BNP), echocardiography parameters such as LVEF, and functional status limited the ability to assess these as possible predictors for readmissions. However, BNP or NT-proBNP are often elevated in diabetes and kidney disease [63] and therefore less likely to be performed in patients with ESKD. The prevalence of preexisting heart failure may be under-recognized in our cohort of patients with kidney failure [21], as we observed that the prevalence (30.6%) was relatively low considering that 60% had IHD. SGLT2 inhibitors and GLP1 receptor agonists were not included in the base model due to the low prevalence of use. Instead, they were included in the sensitivity analysis and were not significantly associated with readmission risk. While these medications reduced cardiovascular events (including heart failure hospitalizations) in diabetes and CKD [64], there are little data on their impact in ESKD. Data on the volume status at discharge were not available. However, the healthcare teams usually aim to achieve euvolemia and reassessment of dry weight during the routine clinical care of individuals hospitalized for fluid overload. While we excluded patients who died and those without follow-up visits, we did not have information on death after discharge and was thus unable account for mortality as a competing risk. As the study cohort comprised patients with ESKD secondary to diabetic kidney disease and were largely older adults with multimorbidity, our results may not be generalizable to other ESKD populations with different cardiovascular and comorbidity risk profiles. Despite these limitations, this study included a comprehensive range of conditions such as congestive heart failure, pulmonary edema, and generalized edema that indicated severe fluid overload so that we did not neglect those with unrecognized heart failure, regardless of ejection fraction [21, 31, 63]. We also evaluated clinical and socioeconomic health determinants that may affect the risk of readmissions for fluid overload among individuals with diabetes and EKSD on dialysis. Additionally, this study included all eligible hospitalizations for each patient and captured the full trajectory of recurrent events.

In conclusion, heart failure, IHD, PVD, hemodialysis (compared to peritoneal dialysis), lower hemoglobin level, and absence of statin at discharge were associated with increased risk of readmissions among individuals with diabetes and ESKD treated with dialysis. The findings can stratify patients according to their risks of readmission and inform strategies that aim to reduce readmissions for fluid overload among individuals with diabetic kidney disease receiving dialysis.

The authors thank Ms. Xin Xiaohui and Ms. Nur Nasyitah from the Health Service Research Unit, Singapore General Hospital, for their support for the project.

This study abided by the Declaration of Helsinki, and the Centralized Institutional Review Board (2023/2216) determined that the study did not require further ethical deliberation for use of de-identified data. Written informed consent from participants was not required for this study in accordance with local/national guidelines.

All authors declare no relevant conflicts of interest.

This study was not supported by any sponsor or funder.

Chee Chin Phang and Cynthia Lim conceptualized and designed the study; Hanis Abdul Kadir, Chee Chin Phang, and Cynthia Lim analyzed and interpreted data; Chee Chin Phang drafted the manuscript; and Li Choo Ng, Peiyun Liu, Sheryl Gan, Lina HuiLin Choong, Chieh Suai Tan, and Yong Mong Bee provided critical input to the manuscript.

Data are not publicly available due to institutional data policies. Data sharing will be subjected to data sharing agreements. Further inquiries can be directed to the corresponding author.