Socioeconomic Influences on the Outcomes of Dialysis-Requiring Acute Kidney Injury in Brazil

Introduction Although research suggests that socioeconomic deprivation is linked to a higher incidence of acute kidney injury (AKI) and worse outcomes in high-income countries, there is limited knowledge about these epidemiologic factors in developing countries. In addition, the impact of medical institution administration (private versus public) on AKI outcomes remains to be determined. Methods We studied 15,186 pediatric and adult patients with dialysis-requiring AKI (AKI-D) admitted to private and public hospitals in Rio de Janeiro, Brazil. According to Brazil's demographic census, socioeconomic indicators were derived from patient zip codes. Propensity score matching analysis and a mixed-effect Cox regression were used to assess the impact of socioeconomic indicators and hospital governance on patient survival. Results Crude mortality rates were higher in private hospitals than in public hospitals (71.8% vs. 59.5%, P < 0.001) and were associated with significant differences in age (75 years, interquartile range [IQR]: 61–83 vs. 53 years, IQR: 31–66), baseline renal function (prevalence of chronic kidney disease [CKD]: 33.2% vs. 23%, P < 0.001), comorbidities (Charlson score: 2.03 ± 0.87 vs. 1.72 ± 0.75, P < 0.001), and severity of presentation (mechanical ventilation: 76.5% vs. 58% and vasopressors: 72.8% vs. 50.5%, P < 0.001). After adjustments and propensity score matching, we found no effect of different hospital administrations or socioeconomic factors on mortality. Baseline characteristics and the severity of presentation primarily influenced AKI-D prognosis. Conclusions Despite significant racial and socioeconomic differences in hospital governance, these indicators had no independent influence on mortality. Future epidemiologic studies should investigate these relevant assumptions to allow healthcare systems to manage this severe syndrome promptly.

A KI remains associated with poor short-term and long-term outcomes and rising treatment costs. 1 Therefore, obtaining high-quality data on the epidemiology of AKI worldwide, not only in its many clinical aspects but also in its socioeconomic determinants, is critical to assisting physicians and healthcare policymakers in consistently and sustainably managing this critical syndrome. 2,3or many disease conditions, the influence of socioeconomic differences on health outcomes has been documented.For example, studies from different countries with different healthcare delivery models have consistently found an association between CKD and socioeconomic deprivation. 4][7][8][9][10] Information is even scarcer in developing countries, where health information systems may be deficient or inaccurate. 3Moreover, AKI epidemiology may differ within the same country because of differences in sanitation, disease distribution, and access to quality healthcare. 11,12ne factor contributing to disease outcomes is the disparity in quality of care provided by the public and private sectors when they coexist in 2-tiered healthcare systems.Unfortunately, it is difficult to distinguish which factors are related to the type of provider and which are influenced by demographic, environmental, and educational or economic factors. 13,14his study investigates the epidemiologic features and associated outcomes of patients with AKI-D in a large metropolitan area of Brazil, focusing on the performance of healthcare institutions that are either market-oriented private or government-funded sector.We also analyzed the impact of other socioeconomic factors (education, income, longevity, human development, and inequality) on AKI outcomes.

Study Protocol
We performed a retrospective analysis of the NefroWeb database, an 11 year (2002-2012) prospective, multicenter study on clinical characteristics and outcomes of AKI-D in the metropolitan area of Rio de Janeiro, Brazil. 15The study included 170 private and public hospitals and medical institutions.Personally sponsored or company-sponsored private health plans or insurance (market oriented) were the primary source of funding in private governance hospitals.In public governance hospitals, the primary source of financing was the National Healthcare System (Sistema Unico de Saúde), and it was provided free of charge at the point of consumption.After nephrology consultation and AKI-D diagnosis, data were entered into the clinical module of NefroWeb, a customized Microsoft SQL Server database, by the attending nephrologist at the time of renal replacement therapy (RRT) initiation. 15ecause the study was noninterventional, the institutional research ethics board approved the study protocol and waived the need for informed consent.This paper follows the STROBE guidelines for cohort studies (Supplementary File S1, Appendix)

Study Population
All adult or pediatric patients, admitted to one of the 170 hospitals and medical facilities, and required dialysis during hospitalization were eligible for inclusion in this cohort.Out of the 25,624 patients who were dialyzed, we excluded 7289 patients diagnosed with end-stage kidney disease (either on previous chronic outpatient RRT or patients with CKD who were commencing dialysis treatment at the time of hospital admission).After chart review, we excluded 3149 patients with incomplete reports, missing values, or missing zip codes (which were required to compute socioeconomic data), leaving the final cohort of 15,186 patients (Figure 1).

Data Collection
The study protocol was previously described in detail. 15Briefly, we collected patient demographic data (age, sex, and race) and clinical variables (primary clinical or surgical diagnosis, renal diagnosis, and comorbidities [the Charlson comorbidity score was derived from the database's extensive list of comorbidities and clinical diagnoses available in the database], precipitant cause of AKI-D, and severity of presentation and organ failures), AKI-D phenotype (e.g., de novo AKI vs. acute-on-chronic kidney disease [ACKD] and hospital-acquired vs. community-acquired AKI), dialysis indication, and modality at the time of dialysis initiation.By conducting detailed medical histories and laboratory studies at enrollment, we identified and excluded patients with CKD who were beginning dialysis at the time of hospital admission.This allowed us to differentiate them from patients with de novo AKI or underlying nondialytic ACKD.Categorization of the patients with ACKD was done on the day of nephrology evaluation based on available information, such as previous medical records, admission serum creatinine, past laboratory results (when available), medical history, relatives' or caregivers' interviews, kidney sonography, and clinical judgment.As a result, the patients were categorized as having either de novo AKI or ACKD.Dialysis modalities included continuous automated peritoneal dialysis and hemodialysis with standard machinery in the following methods: conventional intermittent hemodialysis, prolonged intermittent RRT (PIRRT), or PIRRT in continuous mode (meaning continuous hemodialysis provided with standard dialysis equipment and PIRRT set-up parameters). 16After nephrology RRT indication, data were entered into the clinical module of NefroWeb by the attending nephrologists (for more information on the NefroWeb database, data collection process, definitions used [e.g., ACKD, sepsis, cardiorenal syndrome, and liver failure definitions] and dialysis modalities; see Supplementary File S1, sections 1 and 2).At hospital discharge, the primary outcomes were death, complete or partial recovery, and dialysis dependence.Complete recovery was defined as serum creatinine returning to the same level or a lower level than the enrollment value, whereas partial recovery was defined as serum creatinine being lower than the peak but higher than the enrollment value.In addition, hospitals were classified based on their governance (private vs. public), bed capacity (small: <50 beds; medium: 50-150 beds; and large: >150 beds), and specialty (general, pediatric, cardiology, emergency/ trauma, obstetrics, oncology, and infectious diseases).Finally, each record was reviewed by a pediatric nephrologist (TLCY) or adult nephrologist (CLRG or JHRS) for implausible values, event ascertainment, adjudication, and data cleaning.

Socioeconomic Indicators
We recovered the following variables based on the geographic region of individual patient residence derived from postal zip codes: global municipal human development index (HDIM), further divided into 3 categories (education, income, and longevity), Gini index for socioeconomic disparities, and mean monthly per capita income.Data were presented as continuous variables and categorized in terciles or dichotomized based on mean values in the state of Rio de Janeiro.The United Nations Development Program in Brazil, the Institute for Applied Economic Research, and the João Pinheiro Foundation adapted the country's overall human development index to calculate these indicators for municipalities and states, which are available on the website http://atlasbrasil.org.br(data from the 2010 census).The HDIM is interpreted in the same way as the overall human development index, but its components aim to adapt the global human development index methodology to Brazilian municipalities at municipal and state levels.It has a scale of 0 to 1 and allows for the comparison of Brazilian cities over time.

Statistical Analysis
Analyses were performed using the R statistical environment, version 4.2.2, with a significance level set at P < 0.05.Categorical variables, summarized as numbers and percentages, were compared using the c2 test.Continuous variables were expressed as mean and standard deviation (SD) or median and IQR and compared using the Student's t-test.A shared frailty gamma Cox regression with random effects for individual hospitals of enrollment was employed to compare survival according to hospital governance.The model was based on variables such as age, sex, HDIM, per capita income (indices stratified by the mean value of the population of the state of Rio de Janeiro), Gini index (stratified in tertiles), AKI phenotype (community-acquired vs. hospital-acquired, de novo vs. ACKD, septic vs. nonseptic AKI, and oliguric vs. nonoliguric AKI), number of failing organs, and Charlson comorbidity score.The results were presented as hazard ratios with a 95% confidence interval.The Schoenfeld residuals analysis was employed to test the assumption of the Cox proportional hazards model.In addition, a separate analysis was conducted that included interaction terms of socioeconomic factors, type of hospital, outcomes, and stratification of hospitals according to the number of beds.
In addition to comparing the entire cohort (14,149 patients in the private group vs. 1037 patients in the public group), a propensity score matching approach was used to create a balanced cohort.A logistic regression model that included demographic and clinical characteristics was employed (Supplementary File S1, section 3, Supplementary Table S1).Propensitymatched cohorts were created using a 1:1 ratio greedy-nearest-neighbor matching on the propensity score and a caliper of 0.1 SD of the propensity score logit with no replacement.The standardized mean differences (SMD) were used to compare covariate balances before and after matching, with successful balancing considered when the SMD was <0.10.To investigate differences between matched groups and possible relationships to outcomes, socioeconomic indicators were not included in the propensity score.
Thirty-day Kaplan-Meier survival curves were constructed for both unmatched and matched cohorts to compare public and private hospitals and HDIM and income.A log-rank test was used to determine the significance of survival differences between curves.In addition to considering individual hospitals as random effects in the mixed model and acknowledging the heterogeneity of medical institutions involved (i.e., the predominance of medium and small private medical facilities), a sensitivity analysis was conducted using only large hospitals under private and public governance.Pearson correlation was employed to compare the associations between HDIM, the Gini index, and per capita income in different hospital governance populations.

Characteristics of Private and Public Hospitals and Patient Profiles
We examined data from 15,186 patients with AKI-D collected over an 11-year period (Figure 1).Of the 170 medical institutions involved, 82% (n ¼ 139) were under private administration and 18% (n ¼ 31) were under public administration.Small-sized and medium-sized general hospitals and medical facilities predominated in the private sector, whereas medium-sized and larger medical institutions predominated in the public sector.Public hospitals had a more diverse specialty distribution, including a higher proportion of emergency or trauma units (Supplementary File S1, section 4, Supplementary Table S2).The distribution of studied hospitals in the cohort within Rio de Janeiro was predominantly concentrated in high-income districts, deviating from the expected pattern of aligning with the most populated communities (Supplementary File S1, section 4, Supplementary Figure S1).
The cohort's demographic and clinical characteristics are summarized in Table 1 as follows: 14,149 patients were admitted to private hospitals, and 1037 patients were admitted to public hospitals.The median age in private hospitals was 75 years (IQR, 61-83) compared with 53 years (IQR, 31-66) in public hospitals.Private hospitals had a higher proportion of patients who self-declared as of White ethnicity than public hospitals (84.3% vs. 61.0%,P < 0.001).Most patients presented with de novoacquired and hospital-acquired AKI-D.However, community-acquired AKI-D was more common in public hospitals (45.7% vs. 27.0%,P < 0.001), whereas ACKD disease was more common in private hospitals (33.2% vs. 23%, P < 0.001).
Comorbidities were more prevalent in private hospitals, except for malignancy (Table 1).The Charlson comorbidity score (2.03 AE 0.87 vs. 1.72 AE 0.75, P < 0.001) was higher in private hospitals.Medical reasons for admission and intensive care unit admission predominated in private hospitals (P < 0.001).Incidence of sepsis at admission, major surgery, nephrotoxicity, and glomerular diseases as AKI-D causes were similar.Private hospitals had a higher rate of sepsis after admission, cardiorenal syndrome type I and II, and hepatorenal syndrome.Hypovolemia, obstructive uropathy, and preeclampsia were more common in public hospitals (Figure 2).Mechanical ventilation and vasopressor support were used more in private hospitals (P < 0.001).Isolated kidney failure was more common in public hospitals (27.5% vs. 16.8%,P < 0.001).
Most patients were treated with PIRRT in private hospitals, whereas intermittent hemodialysis was the most common initial RRT modality in public hospitals (Table 1).However, the use of PIRRT in continuous mode was greater in public hospitals than in private hospitals.Finally, the length of stay in private hospitals was significantly longer (21 days, IQR: 11-39) than in public hospitals (17 days, IQR: 8-32).
Each group received 1037 cases after propensity score matching (Table 1).The matching process resulted in a well-balanced cohort.Most baseline-matched covariates showed a decrease of <0.1 in SMD between private and public hospitals, except for socioeconomic covariates, which were purposefully excluded from the matching process.Baseline demographics, AKI-D phenotype, comorbidities, AKI-D precipitating causes (except obstetric conditions) and RRT modalities were comparable.

Socioeconomic Disparities in Private and Public Hospital Admissions
The socioeconomic indicators differed significantly between patients in private and public hospitals (Table 2 and Figure 3).This pattern persisted after the matching process.The HDIM was significantly higher in private hospitals than in public hospitals, and this difference was consistent across all HDIM index subdivisions.A lower Gini index distinguished patients in public hospitals from private hospitals.Income was significantly higher in private hospitals (Table 2).Regardless of hospital governance, there was a significant relationship between socioeconomic indicators and income level as follows: higher incomes were associated with higher HDIM (R ¼ 0.94, P < 0.001) and greater inequality (measured by the Gini index) (R ¼ 0.51, P < 0.001), in both private and public hospitals (Figure 4).All socioeconomic indicators were lower among non-White patients than among White patients, but this difference was only significantly different after stratification in private hospital patients (Supplementary File S1, section 5, Supplementary Tables S3 and S4, Supplementary Figure S2).

Mortality and Discharge Outcomes
Within 30 days after RRT initiation, the crude mortality rate was 70.9% (10,771 patients), with the private group accounting for 71.8% (10,154 patients) and the public group accounting for 59.5% (617 patients, P < 0.001).In contrast, mortality rates in the balanced matched cohort were comparable (60.8% vs. 59.5% in private vs. public hospitals, P ¼ 0.597).The median length of hospital stay was higher in private hospitals (21 days, IQR: 11-39) compared to public hospitals (18 days, IQR: 8-32.75,P < 0.001).The Kaplan-Meier curves for overall mortality in the unmatched and matched cohorts are shown in Figure 5a and 5b.Worse socioeconomic indicators (HDIM, income, and Gini index) in the univariate mixed-effects Cox proportional model of the unmatched cohort were all associated with higher mortality (Table 3).This difference remained significant only for higher income in the multivariate model.However, no socioeconomic variable was significantly associated with mortality in the multivariate model (Figure 5c-f).
Multivariate analysis in the unmatched and matched cohorts showed that age, oliguric AKI, septic AKI, and the number of organ failures were all independently associated with mortality.In the unmatched cohort, male gender, community-acquired AKI-D, and ACKD were associated with better survival (Table 3).
Among the 4415 survivors (3995 in private hospitals and 420 in public hospitals), 1034 (23.4%) were ) than in public hospitals (16%), whereas discharge with partial recovery was higher in public hospitals (49% vs. 30.3%).The kidney status at discharge was unknown in 17.8% of the patients.The discharge outcomes did not differ based on socioeconomic indicators (Supplementary File S1, section 7, Supplementary Table S7).

Sensitivity Analysis Results
A sensitivity analysis of larger hospitals (20 private vs. 21 public, 6993 patients) revealed no significant differences in patient characteristics and outcomes compared to the entire cohort (Supplementary File S1, section 6, Supplementary Table S4 and S5).Comparing the original mixed-effect Cox model with the model including additional interaction terms and hospital stratification by the number of beds showed no significant improvement in performance (Supplementary File S1, section 6, Supplementary Table S6).

DISCUSSION
Our study aims to investigate the impact of socioeconomic indicators on outcomes among patients with AKI.This topic has only been explored in high-income countries using large administrative databases.In addition, our study is the first to compare the outcomes of patients with AKI-D in private versus public hospital settings.Although previous studies have examined the incidence of AKI in different hospital administrations as a complication after various medical procedures, 17 none have specifically investigated the epidemiology of AKI in these settings.Age, sepsis, oliguria, and the number of failing organs were associated with poor survival, whereas male versus female gender, community-acquired versus hospital-acquired AKI, and ACKD versus de novo AKI had better outcomes.9][20][21][22][23][24][25] Interestingly, in the context of the present study, some of these covariates were unequally distributed between patients assisted in private versus public hospitals.Previous research in Brazilian intensive care units compared sepsis outcomes in patients admitted to public versus private hospitals, with contradictory results.One early multicenter study using a convenience sample of patients with sepsis found that those admitted to public hospitals fared worse than those treated in private hospitals.In contrast, a more extensive and possibly less biased multicenter study on a large stratified pseudorandom sample of healthcare institutions found no difference in mortality between inpatients with sepsis admitted to public or private intensive care units. 13,26In the present study, which only included patients with AKI-D, crude mortality was higher in private hospitals.However, after adjusting for confounding, mortality rates were not significantly different, particularly in the propensity-matched cohort, indicating that, at least in the particular environment under study, other   The socioeconomic indicators varied depending on the hospital administration.Higher socioeconomic indicators were associated with higher mortality in the unadjusted analysis, a pattern that was also observed in other population-level analyses. 5,8However, after adjustments, there was no significant impact of HDIM, its components (education, income, and longevity), and the Gini coefficient on survival.Higher income was associated with poorer survival, but other unmeasured factors may have influenced this, because high-income patients were primarily treated in private hospitals.The median age was also significantly higher among private hospital patients.An essential aspect of aging that was not specifically quantified is the increasing prevalence of frailty, which is associated with worse outcomes.Our data suggest that the higher mortality in higher-income patients predominantly treated in private hospitals may reflect the unmeasured frailty component of the aging process. 8It has already been demonstrated that crude mortality rates following AKI are higher in high-income areas, indicating an older patient population. 5Notably, this relationship was not observed in the multivariate model of the balanced matched cohort, where the median age was considerably lower.Finally, our findings differ from those in high-income countries, where the burden of comorbidity and the severity of AKI at presentation may contribute to the impact of social deprivation on AKI-associated mortality. 5,8,10espite significant differences in socioeconomic indicators related to ethnicity and a significant difference in the distribution of these groups in the 2 hospital systems, we found no independent influence of race on survival, contrary to findings in high-income countries. 27More significant economic deprivation and a higher burden of comorbidities in the non-White population are characteristics of social discrepancies in Brazil, 28 but they did not translate into a higher mortality risk in our study.Sepsis at admission was the most common condition associated with AKI-D development in private and public hospitals, according to several cohort studies of AKI. 6,19,20,23However, markers of poorer outcomes, such as sepsis after admission, cardiorenal syndrome type I, and hepatorenal syndrome, were more common in private hospitals, whereas factors associated with better survival odds, such as obstructive uropathy and preeclampsia, were more common in public hospitals.This difference partly explains the lower survival observed in private hospitals.
We found no association between socioeconomic indicators and renal recovery among survivors.This finding differs from an epidemiologic study conducted in Taiwan, which identified higher regional economic status as an independent factor for renal recovery. 6Our study revealed a high prevalence of patients discharged with partial renal recovery (33.03%) or receiving RRT (23.3%).Interestingly, more patients in private hospitals were discharged on RRT compared to public hospitals (23.7% vs. 21.2%).Our previous publications with the same cohort showed a significant association between preexisting CKD and these adverse kidney outcomes. 15The higher median age of the cohort may partly explain these findings.However, we acknowledge that the lack of long-term follow-up of the patients may have affected our analysis of kidney recovery outcomes.
Our findings must be interpreted in the context of the study's limitations.First, the distribution of private and public hospitals in this cohort does not reflect the actual situation of the Brazilian healthcare system.Approximately 90% of inpatients in this cohort were admitted to private hospitals, even though only 20% to 25% of the Brazilian population has private health insurance. 27Furthermore, it was not our intention to produce a population-level analysis regarding the incidence and prevalence of AKI-D.As stated, the institutions involved do not represent the whole scope of Rio de Janeiro's medical facilities, as can be gathered by the higher prevalence of private hospitals in the cohort. 15It is also important to note that the hospitals studied in the manuscript were unequally distributed in the city of Rio de Janeiro, showing higher concentrations in neighborhoods with higher human development index and not according to population distribution.This uneven distribution of hospitals may be indicative of existing inequities in the accessibility and availability of healthcare services within the city, which could impact the generalizability of our findings.Second, significant differences were observed in the phenotypes of hospitals, with small or medium medical institutions in the private sector outnumbering large hospitals in the public sector.Therefore, we conducted a sensitivity analysis limited to larger hospitals from both types of governance, which reproduced the results of the entire cohort.Third, given the various health determinants and potential disparities in care quality across our county, these findings may not be extrapolated to all regions of the vast Brazilian territory. 28Fourth, socioeconomic data were retrieved from census information using the residence code rather than individual-level measures of socioeconomic status, which may have resulted in some misclassification. 10However, this strategy is mainly used in studies that assess the socioeconomic status and AKI incidence and outcomes. 5,8,10Finally, this was a shortterm follow-up study, from admission to discharge, and we were not able to follow-up with these patients in the long-term to account for outcomes among survivors.Patients were discharged either with complete or partial recovery or with dialysis dependence.We acknowledge that some AKI survivors discharged on RRT may eventually become dialysis independent during the follow-up period, and unfortunately, we were not able to capture this important information. 29nclusion AKI-D has short-term and long-term adverse consequences.Despite differences in socioeconomic characteristics, the present study found no impact of different hospital governance systems on overall mortality.Poor AKI outcomes were primarily driven by the patient's baseline characteristics and the severity of the presentation.Future large-scale epidemiologic studies should further investigate these relevant assumptions, allowing healthcare systems to manage this severe syndrome promptly and accurately.

DISCLOSURE
FR and JHRS are minor partners of KMR2, a spinoff company of Kidney Assistance Ltd that developed the NefroWeb product to manage logistics and billing related to dialysis services in hospitals and clinics in the metropolitan area of Rio de Janeiro.JHRS conceived NefroWeb's clinical module solely for epidemiologic research purposes, with no financial relationship that could appear to have influenced the submitted work.The 2 other authors have no affiliation to KMR2 and no conflicting interests to disclose.

Figure 2 .
Figure 2. Precipitant causes of severe dialysis-requiring acute kidney injury stratified by hospital governance.Causes are not mutually exclusive.

Figure 3 .
Figure 3. Socioeconomic indicators of patients with dialysis-requiring acute kidney injury among private versus public hospitals.(a) unmatched and (b) matched cohorts.HDIM, human development index (municipality).

Figure 4 .
Figure 4. Correlation between income and HDIM and Gini index according to hospital administration.HDIM: human development index (municipality).Income is presented in Brazilian currency (Reais, R$).HDIM, human development index (municipality).
CLINICAL RESEARCHdemographic and clinical features would appear to be the main drivers of poor outcomes in AKI-D.

Figure 5 .
Figure 5. Survival curves, according to Kaplan-Meier estimates according to hospital administration (a and b), HDIM (c and d), and income (e and f), with log-rank test for significance, in the unmatched and matched cohorts.HDIM: human development index (municipality).

Table 2 .
Socioeconomic indicators according to hospital governance in the unmatched and matched cohorts HDIM, human development index (municipality); PSM, propensity score matching; SMD, standardized mean difference.

Table 3 .
Univariate and multivariate mixed-effects Cox regression model on mortality in the matched and unmatched cohort AKI, acute kidney injury; ACKD, acute-on-chronic kidney disease; HR, hazard ratio; HDIM, human development index (municipality)