The dialysis facility levels and sizes are associated with outcomes of incident hemodialysis patients

The outcomes of patients with incident kidney failure who start hemodialysis are influenced by several factors. Whether hemodialysis facility characteristics are associated with patient outcomes is unclear. We included adults diagnosed as having kidney failure requiring hemodialysis during January 1, 2001 to December 31, 2013 from the Taiwan National Health Insurance Research Database to perform this retrospective cohort study. The exposures included different sizes and levels of hemodialysis facilities. The outcomes were all-cause mortality, cardiovascular death, infection-related death, hospitalization, and kidney transplantation. During 2001–2013, we identified 74,406 patients and divided them in to three groups according to the facilities where they receive hemodialysis: medical center (n = 8263), non-center hospital (n = 40,008), and clinic (n = 26,135). The multivariable Cox model demonstrated that a larger facility size was associated with a low mortality risk (hazard ratio [HR] 0.991, 95% confidence interval [95% CI] 0.984–0.998; every 20 beds per facility). Compared with medical centers, patients in non-center hospitals and clinics had higher mortality risks (HR 1.13, 95% CI 1.09–1.17 and HR 1.11, 95% CI 1.06–1.15, respectively). Patients in medical centers and non-center hospitals had higher risk of hospitalization (subdistribution HR [SHR] 1.11, 95% CI 1.10–1.12 and SHR 1.22, 95% CI 1.21–1.23, respectively). Patients in medical centers had the highest rate of kidney transplantation among the three groups. In patients with incident kidney failure, a larger hemodialysis facility size was associated with lower mortality. Overall, medical center patients had a lower mortality rate and higher transplantation rate, whereas clinic patients had a lower hospitalization risk.


Scientific Reports
| (2021) 11:20560 | https://doi.org/10.1038/s41598-021-00177-x www.nature.com/scientificreports/ tion as a teaching hospital, and the completeness of subspecialty services. The dialysis centers are operated by different level of healthcare facilities, and we stratified them into 3 levels, namely medical centers, non-center hospitals, and clinics, based on the Institution Accreditation subdatabase in the NHIRD. Each hemodialysis session receives same reimbursement regardless of the facility levels. The accreditation of hemodialysis service further regulates the physician-patient and staff-patient ratio. The dialysis facility size was presented as HD bed numbers, which was extracted from the Institution Bed subdatabase in the NHIRD.
Outcomes. The outcomes were all-cause mortality, cardiovascular death, infection-related death, major cardiac and cerebrovascular events (MACCEs), all-cause admission, infection-related hospitalization, and kidney transplantation during the 2-year follow-up and at the end of the follow-up. Death was indicated by withdrawal from the NHI program 21 . The cause of death was determined using the main diagnosis in discharge records for inpatient hospital deaths, the primary diagnosis of the last emergency room visit, or the cause of hospitalization within 7 days of death for out-of-hospital deaths. MACCEs were acute myocardial infarction (AMI), ischemic stroke, and cardiovascular death. AMI and ischemic stroke were defined by the principal diagnosis on admission. The date of kidney transplantation is detected by using the NHI reimbursement codes of the inpatient claims data. Kidney transplantation status was confirmed by the presence of ICD-9-CM code V42.0 in the outpatient diagnoses with a corresponding immunosuppressant prescription after the kidney transplantation date. All patients were followed from the index date to the event occurrence date, date of withdrawal from the NHI program, the day of kidney transplantation, or December 31, 2013, whichever came first.

Statistical analysis.
This was a retrospective cohort study and the distribution of patient's characteristics among different facility levels might be substantially different. To mitigate the possible selection bias, we applied an inverse probability of treatment weighting (IPTW) of propensity scores (PSs) when comparing the outcomes among different dialysis facility levels. To obtain optimal balance between dialysis facility levels, the PSs were estimated using the generalized boosted model (GBM), which was based on 50,000 regression trees 22 . Explanatory variables included in the PS estimation were all the covariates listed in Table 1, except that the follow-up duration was replaced with the index date. Dependent variables in the PS estimation were the facility levels (1 = medical center, 2 = non-medical center hospital, and 3 = clinics). The balance among different dialysis facility levels before and after GBM-IPTW was investigated using standardized differences 23 ; a threshold of > 0.1 indicated an imbalance. The association between covariates and all-cause mortality risk was studied using the multivariable Cox proportional hazards model in the original cohort before propensity score weighting. The event rate of outcomes and survival analyses were estimated in the cohort after propensity score weighting. The risk of time to event outcomes among different dialysis facility levels was compared using Cox proportional hazard model. The facility level was the only explanatory variable in the aforementioned survival analyses. A 2-sided P value of < 0.05 was considered statistically significant. Statistical analyses were performed using SAS version 9.4 (SAS Institute, Cary, NC) 24 .
Ethics approval and consent to participate. The NHIRD contains no identifiable personal information; the need for informed consent was waived by the Chang Gung Medical Foundation Institutional Review Board because of this study's retrospective noninterventional design and because patient data confidentiality and privacy were maintained. The study was approved by the Chang Gung Medical Foundation Institutional Review Board.

Results
Patient selection. Between January 01, 2001 and December 31, 2013, 158,910 patients with KFRT were screened. After the exclusion of patients aged < 20 years, who received dialysis before the index date, whose initial modality was PD, and who had malignancies, ventilator dependency, or bed confinement status, 92,638 adult ambulatory HD patients were included. During the first 90 days following the index date, patients who were lost to follow-up, received HD less than twice per week, or did not receive HD predominantly at a certain facility level were also excluded. A total of 74,406 patients were eligible for analysis, and they were divided into three groups based on the facility level: medical centers, non-center hospitals, and clinics ( Fig. 1).
Patient characteristics and trends of epidemiology. The HD center sizes varies in different levels of facility (mean HD bed numbers in medical center: 88.7 ± 46.7, non-center hospital: 49.6 ± 35.7 and the clinic: 28.5 ± 11.3) (data not shown). Before propensity score weighting, patients undergoing dialysis at medical centers were younger, resided in urban areas, had higher incomes, had fewer comorbidities and prior cardiovascular events, and had higher numbers of nephrology clinic visit before HD. After propensity score weighting, all covariates in the three groups had maximum absolute standardized differences of < 0.1, indicating the groups were well-balanced (Table 1). In addition, the distribution of the facility levels was balanced across the study period after propensity score weighting ( Supplementary Fig. 1). Figure 2A depicts the time from HD initiation to death. Of the 92,638 PD patients, 49,176 (53.1%) died during the follow-up. Our study demonstrated that the mortality peak occurs within 1 year following HD initiation, and the mortality rate decreased steadily with time. From 2001 to 2013, the proportion of patients who received HD at clinics gradually increased (Fig. 2B). HD patients included older age, male sex, low monthly income, low urbanization area residence, diabetes mellitus, coronary arterial disease, chronic obstructive pulmonary disease, liver cirrhosis, prior cardiac and cerebrovascular events, and use of nonnative fistula access. Conversely, hypertension, dyslipidemia, hyperuricemia, gout, large HD station number, and receive HD at medical center than other levels of facility were associated with low risks of all-cause mortality ( Table 2).

Comparison of outcomes between facility levels.
After propensity score weighting, the time-to-event analysis showed that patients who attended medical centers had the lowest risks of all-cause mortality, cardiovascular death, and infection-related death, whereas patients who attended clinics had the lowest risk of all-cause and infection-related hospitalization. Medical center patients were more likely to receive kidney transplantation Table 1. Baseline characteristics of patients according to the dialysis facility level before and after IPTW. IPTW inverse probability of treatment weighting, MASD maximum absolute standardized difference, AV arteriovenous. a Data were presented as frequency (percentage) or mean ± standard deviation; b Data were presented as percentage or mean ± standard deviation. No. of nephrologist outpatient visits in the previous year 11.7 ± 9.2 8.9 ± 8.9 9.8 ± 9.1 9.7 ± 8.9 9.5 ± 9.0 9.5 ± 9.0 0.02  Table 3. The propensity score weighting-adjusted rate of kidney transplantation among different facility levels is illustrated in Fig. 3.

Discussion
Our study revealed that advanced age, low socioeconomic status, higher comorbidities, vascular types other than fistula at the initial access, non-medical center based facilities and smaller facility size were associated with worse short-and long-term outcomes of adult incident HD patients in Taiwan. Several risk factors for mortality, including DM, CAD, COPD, liver cirrhosis, and prior cardiac events, were consistently reported in previous large registries or cohorts [3][4][5][6]13,[25][26][27] . The overall comorbidity burden, collectively presented as the Charlson Comorbidity Index, was associated with mortality risk. By contrast, hypertension, dyslipidemia, and hyperuricemia/gout were associated with low mortality in our study. This could be explained by reverse epidemiology finding in the dialysis population in which relatively high blood pressure, plasma cholesterol, and uric acid levels usually also represent a favorable nutritional status rather than risk factors [28][29][30] .
The level and station number of the HD facility were associated with mortality. Large HD bed numbers usually represent a large dialysis service group with more resources in staff training, continued education, and www.nature.com/scientificreports/ interdisciplinary care. Taiwan's NHI is a single-payer healthcare program with mandatory enrollment that covers > 99.8% of Taiwan's 23 million residents. With this single-payer system, the overall reimbursement is positively correlated with service volume. However, the fixed cost per service may decrease when the service volume increases to a significant level. The reduction of fixed costs in large service groups may make them resource abundant. In 2013, Yan et al. analyzed data from the USRDS to investigate the relationship among HD facility size, ethnicity, and mortality. By using 26-30 HD beds as a reference, an HD station number below this level was associated with a high mortality risk 4 . Our study found that medical center patients had a higher survival benefit than non-center hospital and clinic patients; however, medical center and non-center hospital patients also had a higher hospitalization rate than the clinic patients. This result has not been previously reported and may be explained through several factors. Because the accessibility of health care services for patients who received HD at medical centers is high, they may receive a diagnosis of cardiovascular disease or a confirmation of infection at an early stage. HD patients were at a high risk of MACCEs wherever they receive dialysis therapy, and hence, a dialysis facility where emergency response staff are well-trained, fully available, and can provide services in a timely manner may influence patient survival. For example, Anderson et al. reported that patients with cardiac arrest who received in-hospital resuscitation were more likely to survive than those who received out-of-hospital resuscitation 31 . As for AMI, shortening door-to-balloon time to < 60 min can improve survival compared with a door-to-balloon time of 60-90 min 32 . Such a short door-to-balloon time is undoubtedly more likely to be achieved in medical center patients. The same reason also applies to infection. Liu   www.nature.com/scientificreports/ of appropriate antibiotics on sepsis recognition, the relative mortality risk increased by 9% 33 . These findings indicate that in a medical center, emergent situations could be managed faster than at other non-center facilities owing to the lack of need for transportation. Despite the extensive control of possible confounding factors, some unmeasured factors were present. From our daily practice, we have observed that many frail patients who chose to receive HD at non-center hospitals or clinics in terms of proximity. Frail elderly patients and non-frail ambulatory patients may have similar numbers of comorbidities, but frailty is an independent risk factor for mortality 34,35 . The extent of frailty could not be Table 3. Outcome of the patients according to the dialysis facility level during the follow up after IPTW. HR hazard ratio, SHR subdistribution hazard ratio, CI confidence interval, MACCE major adverse cardiac and cerebrovascular events. Data were presented as percentage. *P value < 0.05; a Including acute myocardial infarction, acute ischemic stroke and cardiovascular death. www.nature.com/scientificreports/ adjusted in most claims databases, including Taiwan's NHIRD, and this may contribute to more adverse events in facilities where patients' frailty is higher. Another inadequately adjusted factor could be socioeconomic status. Almost all medical centers in Taiwan are located in highly urbanized areas, and patients who received regular HD at medical centers usually have a higher income than others. The differences in income could not be adequately controlled because the upper limit of incomes in Taiwan's NHIRD is below the average monthly regular earnings of all employees (US$760 vs US$1400), which makes the scale of income less sophisticated 36,37 . For example, a patient living in urban area, earning 3,000 USD per month and receiving HD in the medical center will be categorized into similar income group with another patient living in countryside, earning 1000 USD per month and receiving HD in the clinic. Despite a low mortality rate, all-cause and infection-related hospitalizations were the highest in the medical center group than in the other two groups. Approximately half of the hospitalizations in the three groups were infection related. Although all healthcare facilities require accreditation for infection control, the hospitals may be more difficult to keep clean because the complexity of patient flow compared with a simple, primary clinic. The larger bed numbers in the hospital-based dialysis makes the possibility of patient-patient or staff-patient transmission of infectious disease much higher. Meanwhile, the possibility of antibiotic resistance is higher in hospitals than in primary care settings. These factors may explain the result that patients who received HD at a hospital level had a high infection risk 38,39 . Conversely, an increased hospitalization rate with reduced mortality imply that medical center patients may receive more meticulous care with regard to their health problems. However, the balance among hospitalization, life quality, and mortality remained unanswered in our study.
The transplantation rate is significantly higher for medical center patients than in non-center hospital and clinic patients. Studies have focused on different factors related to kidney transplantation referral; however, the association between kidney transplantation rate and dialysis facility level had never been reported. In a study investigating factors associated with transplantation referral rates in the United States, Patzer et al. found that the referral rate was high in young patients with less comorbidities who received dialysis in a facility that was nonprofit or had a high social worker-patient ratio 40 . In Taiwan, most kidney transplantation centers are also medical centers. Medical center patients may receive more information about transplantation, be advised to register for transplantation, and eventually receive a kidney allograft.
This study has limitations. First, because of a lack of laboratory data and frailty scores, we were unable to adjust for these factors. However, a laboratory or severity score at single time point could not correctly predict mid-or long-term outcomes. Second, this database did not have information regarding dialysis adequacy. Survival was worse in patients whose single pool Kt/V was < 0.8, as shown in the National Cooperative Dialysis Study 41 . Nevertheless, in the HEMO study, achieving Kt/V of > 1.7 did not confer better survival than did a Kt/V of 1.3 42 . The calculation of Kt/V is subject to some bias such as under-and over-estimation in overweight and underweight patients, respectively. The use of Kt/V to guide dialysis prescription has also encountered criticisms. Despite a lack of aforementioned lab data, we use propensity score weighting to balance the patient related factors among three groups. The propensity score weighting has a greater statistical power than matching, and it also has a better control of confounding than traditional multivariable adjustment. Third, because of the complex data cleaning, validation and de-identifying process with resultant long lag time in NHIRD, we only obtain data from 2001 to 2013. However, the reimbursement and contents of hemodialysis service has not substantially changed in the last 10 years in Taiwan, we believed that our result would not be affected too much without the latest data.
In conclusion, the study demonstrated that larger HD facility sizes and medical center-based HD facilities were associated with lower all-cause mortality after adjustment for multiple patient factors. Compared with clinic patients, patient in medical center tended to have lower mortality rates but higher hospitalization rates. Patients in medical centers received transplantation more often than did clinic patients. Under the single-payer health care systems in Taiwan, the relocation of resources may help minimize the discrepancy of health-related outcomes in HD patients who are treated at facilities of different sizes or levels.

Data availability
The data underlying this study are from the National Health Insurance Research Database (NHIRD), which have been transferred to the Health and Welfare Data Science Center (HWDC). Interested researchers can obtain the data through formal application to the HWDC, Department of Statistics, Ministry of Health and Welfare, Taiwan (http:// dep. mohw. gov. tw/ DOS/ np-2497-113. html). The authors have no privilege over others in obtaining the data from the NHIRD.