Dialysis Unit and Patient Characteristics Associated with Reuse Practices and Mortality:

Transcription

1 J Am Soc Nephrol 9: , 1998 Dialysis Unit and Patient Characteristics Associated with Reuse Practices and Mortality: ALLAN J. COLLINS, JENNIE Z. MA, EDWARD G. CONSTANTINI, and SUSAN E. EVERSON Hennepin County Medical Center, University of Minnesota, Minneapolis, Minnesota. Abstract. The diverse patient and dialysis unit characteristics in the United States pose challenges for assessing the safety and efficacy of reuse practices. A 10% random sample of periodprevalent hemodialysis patients from units practicing conventional dialysis (<25% of patients with high-efficiency/highflux dialysis) were analyzed. The data included 13,926 patient observations in and 20,422 in Centers for Disease Control and Prevention and Health Care Financing Administration facility survey Medicare data were analyzed with a Cox regression model, evaluating the risk of reuse compared with no reuse and adjusting for comorbidity, unit characteristics, and profit status. In , freestanding and hospital-based units that did not reuse dialyzers were not significantly different from each other in mortality rates. In , however, no-reuse, freestanding, for-profit units had higher risks (relative risk [RR] = 1.23, P = 0.003) compared with no-reuse, hospital-based, nonprofit units. Noreuse, hospital-based, for-profit units, in contrast, were associated with a bower mortality risk (RR = 0.70, P = ). An isolated higher risk associated with peracetic acid manual reuse in freestanding units ( ) was identified in for-profit units only. In the period, an increased mortality risk was noted in hospital-based, nonprofit units practicing formaldehyde automatic reuse, and in freestanding, for-profit units using glutarabdehyde, which accounted for <5% of all units. All other interactions of reuse germicide and technique were not different from no-reuse. The varying mortality rates idenrifled in both no-reuse and reuse units using conventional dialysis suggest that other factors, such as dialysis therapy and anemia correction (both known predictors of patient survival), have a greater influence on U.S. mortality than reuse germicides and techniques. Dialyzer reuse was first described by Shabdon et a!. in 1964 (1), a time when the scarcity of dialysis materials made the development of reuse techniques a necessity. Since then the practice of reuse has accelerated, particularly after the introduction of hollow-fiber dialyzers in the mid-1970s and again after the economic pressures of dialysis reimbursement began in 1983 (2). Standards for reusing dialyzers were first published by the Association for the Advancement of Medical Instrumentation in I 986 (3) and were adopted by the Health Care Financing Administration (HCFA) in 1988 (HCFA Federal Register: Medicare Program: Protocol for the Reuse of Dialysis Bloodlines, 1987). Since the introduction of reuse in the United States, many researchers have evaluated the safety and efficacy of this now widespread practice. Several studies have been conducted to evaluate the impact of reuse on the biocompatibility of diabyzer membranes (4-8). Others have focused on clinical aspects of reuse, evaluating the relationship between reuse and intratreatment complications (9-13). Adverse patient events relative to reuse and diabyzer membranes have also been investigated and Received January 5, Accepted June 1, Correspondence to Dr. Allan I. Collins, 825 South Eighth Street, Suite 816, Minneapolis, MN / $03.00/0 Journal of the American Society of Nephrology Copyright (1) 1998 by the American Society of Nephrology are compiled in annual surveillance information reports by the Centers for Disease Control and Prevention (CDC) (14-17). The dialysis community has been particularly concerned with the association of reuse and patient outcomes. Recently, Held et a!. (18) reported on the link between reuse practices and patient survival. Using HCFA facility survey data, ESRD Program Management and Medical Information System (PM- MIS) data, and the CDC annual survey, they investigated the association of reuse and mortality in a point-prevalent population from 1989 and 1990, finding that patients in freestanding dialysis units that disinfected diabyzers with peracetic acid, hydrogen peroxide, acetic acid mixture, or glutaraldehyde experienced higher mortality rates than patients in units performing no reuse or formaldehyde reuse. The authors suggested a number of possible hypotheses for this result but did not address confounding areas of concern, such as comorbidity and dialysis unit profit status. The authors also limited their study to patients in freestanding dialysis units, excluding those in hospital-based units. Using the same methods and focusing on the same time period, Luehmann and Cosentino (19) reported mortality rates in hospital-based units and, in contrast, found no effect or reduced mortality associated with reuse germicides. A third national study by Feldman et al. (20), based on incident cohort patients who started hemodialysis in 1986 and 1987, supported Held et a!. s findings in freestanding dialysis facilities, but, like the Luehmann and Cosentino study, showed no increased risk in hospital-based units.

2 J Am Soc Nephrol 9: , 1998 Reuse Practices and Mortality: These varying results from hospital-based and freestanding dialysis units raise the question of whether the higher mortality rates identified in the studies are germicide-specific or are due to unidentified factors, such as differences in dialysis unit characteristics, rates of comorbidity, dialysis therapy, nutritionab factors, or correction of anemia. We obtained the HCFA and CDC data to replicate the original Held et al. study, extending the analysis into to further investigate these possibilities. We expanded the analysis to include both freestanding and hospital-based dialysis facilities, controlling for potential heterogeneity in the patient population by adjusting for patient-specific comorbid conditions and for characteristics specific to individual dialysis units. We also advanced the previous studies by including profit status of dialysis units and by analyzing no-reuse units independently. Materials and Methods Data Sources and the Study Population The data sets used in this study were obtained from the HCFA ESRD database, HCFA annual facility surveys, and the annual CDC National Surveillance of Dialysis-Associated Diseases surveys. Patient-specific data were obtained from the HCFA database for ESRD patients (PMMIS), which includes information on patients entering ESRD from 1978 to the present. The PMMIS contains patient demographic characteristics, data from the Medical Evidence Form 2728, Medicare entitlement data, death data from HCFA Form 2746, hospitalization data, and modality change notifications. Patient comorbid conditions were derived from the HCFA institutional Part A claim data. The claims on medical services were billed on Uniform Bill 82 (UB82) forms, by institutional providers from outpatient dialysis services, hospitals, skilled nursing facilities, home health agencies, and hospices. These claims have fields for International Classification of Diseases (ICD-9) diagnostic and procedure codes, indicating medical conditions associated with each service. The detailed comorbidity profiling method from the ICD-9 codes is described in the next section. Dialysis unit-specific information, including unit ownership, age, and size, was determined from HCFA annual facility survey data from Dialysis unit-specific characteristics related to dialyzers and reuse-including germicide type, reprocessing technique (manual or automated), dialysate composition (acetate or bicarbonate), and water treatment method-were obtained from the CDC surveillance data. All information was entered and processed on a Digital Equipment Corp. (DEC) Alpha 2100 computer with an E mhz processor and maintained under an OpenVMS operating system. To make the data set more manageable and to reduce the computational requirements, we chose a 10% random sample of patients from , including patients from both freestanding and hospital-based units. This sample was created by extracting from health insurance claim numbers (social security numbers) all numbers with a ninth digit equal to four, i.e., XXX-XX-XXX4, since the last four digits of the social security number are randomly assigned by the Social Security Administration (21). The random sampling method in the study is similar to that used by HCFA to construct the 5% Public Use claims data for actuarial analyses on Medicare patient expenditures (22). We then further restricted the study population to hemodialysis patients, identified through hemodialysis revenue codes 82X from the dialysis unit claims, and used the 90-day restriction for a period-prevalent cohort defined by the U.S. Renal Data System (23), secondary to the Medicare eligibility restrictions for patients under age 65. For each individual year, we created a period-prevalent cohort, which included all patients alive when the year began, along with those who entered during that year (23). The date of entry into the study was defined for prevalent patients as January 1 of each year, and for incident patients as 90 days after the onset of ESRD. Patients were followed to modality change (hemodialysis to peritoneal dialysis), renal transplantation, death, or December 3 1 of the same year, whichever occurred first. Links between patients and dialysis units were determined from PMMIS quarterly dialysis records (derived from HCFA outpatient UB82 claims) and from CDC surveillance reuse data. A unit designation for each patient was assigned based on hospital-based or freestanding profit status from the HCFA facility survey data, and the reuse status of dialysis units was assigned from CDC surveys. Hospital-based, freestanding, and profit status was assigned for units at the beginning of each cohort year for patients prevalent on January 1, and assigned at the start of hemodialysis for the incident patients entering during the year. Reuse germicide and other CDC data were assigned only at the start of each cohort year. To replicate the Held et a!. study, we further narrowed our study population to include only those patients using conventional dialyzers. CDC survey data provided the percentage of patients in each unit using high-efficiency and high-flux dialyzers. To define conventional dialyzer use, we performed a distributional analysis and categorized a unit as conventional when no more than 25% of the patients were on high-efficiency or high-flux therapy. Comorbidity Profiling Because dialysis patient survival is significantly influenced by comorbidity (24-27), we included comorbid conditions in our mortality analysis. A comorbidity profiling methodology was developed based on the ICD-9 diagnosis and procedure codes in the Part A inpatient and institutional claims. These claims provide one principle diagnosis, four secondary diagnoses, and up to three procedure codes representing conditions identified during the patient s hospitalization. Additional information on the prior medical conditions of each patient is provided by ICD-9 codes indicating prior medical conditions and exposure to communicable diseases (V-codes). We longitudinally traced the codes and dates, and classified the codes into seven major diagnostic groups for pre-existing conditions: atherosclerotic heart disease (ASHD); congestive heart failure (CHF); other cardiac diseases, including valvular heart disease, pacemaker, and arrhythmia; peripheral vascular disease; cerebrovascular accident/transient ischemic attack; non-skin malignancies, including melanoma; and chronic obstructive pulmonary disease. The codes are defined in Table 1. For each patient in each year of survival, the codes for these conditions were extracted from claims prior to the entry date for the study year, then used as pre-existing risk factors. Statistical Analyses Unless otherwise indicated, we have expressed descriptive statistics as a percentage for categorical variables, and as the mean ± SD for continuous variables. Differences were tested by and a tests, and considered significant at a P value <0.05. We used the Cox proportional hazards model to evaluate the relative impact of risk factors on patient death rates (28). Because of our concern about the nonproportionality of some covariates, and to replicate the Held et a!. study, we stratified the Cox regression on age (0 to 19, 20 to 44, 45 to 64, 65 to 74, and 75 years and older), gender, race (black, white, and other), and four categories of primaryrenal diagnoses (all-cause diabetes, hypertension and renal vascular diseases, gbomerulonephritis of all forms, and other). The Cox regression analyses were carried out in a stepwise manner:

3 21 10 Journal of the American Society of Nephrology J Am Soc Nephrol 9: , 1998 Table 1. Coding categories for major risk factors from Part A claims Risk ICD-9 Diagnostic Codes ICD-9 V-Codes ICD-9 Procedures Cancer VlO; V76 Diabetes mellitus 250 ASHD V45.81; V8l.0; V45.82; V XX CHF 425; 428, 402.Xl CVA/TIA PVD ; 447; ; Cardiac (other) ; ; 429; COPD ; 496; 510; V8l V42.l; V42.2; V43.3; V45.0; V X; 37.8X; a lcd International Classification of Diseases; ASHD, atherosclerotic heart disease; CHF, congestive heart failure; CVATI IA, cerebrovascular accident/transient ischemic attack; PVD, peripheral vascular disease; COPD, chronic obstructive pulmonary disease. patient-specific and dialysis unit-specific characteristics were added into the model first, patient comorbidity second, and profit status last. In addition to the stratification variables, patient-specific characteristics consisted of patient age (as a continuous variable within each stratum), duration of ESRD exposure before study entry, and the comorbid conditions described above. Dialysis unit-specific characteristics included reuse germicide type (formaldehyde, peracetic acid, and glutaraldehyde), dialysis unit size and age in log scale, profit status (for-profit versus nonprofit), water treatment methods (reverse osmosis versus other), dialysate composition (acetate versus bicarbonate), and dialyzer reprocessing techniques (manual versus automated). To evaluate the impact of reuse germicides thoroughly, two types of regression analyses were performed. First, for each unit designation (hospital-based or freestanding) and each time period, the germicide types were used as independent covariates adjusting for other patientand dialysis unit-specific characteristics. Patients in no-reuse units (within each type of unit designation and time period) served as the baseline group. Similar regression analyses were then repeated on all patients from both hospital-based and freestanding units to examine the independent effect of germicides for each time period ( or ). To further identify whether reuse germicides are associated with a specific type of dialysis unit characteristic, a regression analysis with a four-way interaction among germicides, unit designation, profit status, and dialyzer reprocessing techniques was performed separately for each time period ( and ). Patients in hospital-based, nonprofit, no-reuse units formed the baseline group within each time period. All results from the Cox regression analyses are reported as relative risks (RR) with 95% confidence intervals (CI). All analyses were performed using SAS version for Windows and 6.09 for OpenVMS. Five consecutive period-prevalent cohorts from were combined and divided into two study populations: and A patient followed for more than 1 year was included in several period-prevalent cohorts, each treated as a separate observation. To avoid potential prevalent cohort biasing, prior ESRD exposure was added (18,29). The period of , as defined by Held et a!., was analyzed separately to confirm the association of germicide and patient outcome in freestanding dialysis units. To investigate the impact of reuse in more recent years, we applied the same analytical techniques to data from the latter period of Using these cohorts and methods, we investigated two sets of hypotheses: those proposed by previous investigators or implicit in their studies (Held et a!., Feldman et a!.), and those based on the enhanced data set we constructed for this analysis. Hypotheses from earlier studies included: (1) Freestanding units have less complex patients than hospital-based units and should therefore be analyzed separately. (2) Germicide-associated outcomes are general phenomena in freestanding units. Our hypotheses included the following: (1) Hospital-based and freestanding units have equally complex patients. (2) Outcomes in units that do not reuse dialyzers are not consistent across all units. (3) Profit status is associated with patient outcomes. (4) Reuse-associated outcomes are not general phenomena but are specific to unit practices and dialysis unit characteristics. Results This large-scale observational study encompasses five preyalent years, diverse reuse practices, and numerous patient and dialysis unit characteristics, correlating each variable to patient survival. The results of our complex analyses are presented in the following order: fixed patient characteristics (including comorbidity), dialysis unit characteristics (hospital-based versus freestanding designation; for-profit versus nonprofit), independent associations between germicides and profit status, analysis of baseline no-reuse units, and, finally, the interactions of dialysis unit characteristics and reuse practices. Patient Demographics Our study included 34,348 prevalent patient observations in 23,908 individuals: 19,207 in freestanding units (13,212 mdividuals) and 15,141 in hospital-based units (10,696 individuals), with 13,926 in the period (10,357 individuals) and 20,422 in the period (13,551 individuals). This 10% random sample is representative both of the U.S. hemodialysis population as a whole (30), as noted in Table 3, and of the population studied by Held et a!. (18) Table 2 shows the distribution of patient observations by germicide type for the five prevalent years. In freestanding units, formaldehyde practice declined during the 5 years of the study, glutaraldehyde was consistently used with fewer than 5% of the patients, and peracetic acid use increased during the first 3 years and then became stable. Units practicing no reuse accounted for fewer than one-fifth of freestanding unit patients, in sharp contrast to hospital-based units, in which no-reuse facilities accounted for 72 to 78% of the patients. Of the

4 J Am Soc Nephrol 9: , 1998 Reuse Practices and Mortality: Table 2. Patient observations distribution by germicide type Observation Freestanding. Units No. of patient observations No reuse (%) Formaldehyde (%) Peracetic acid (%) Glutarabdehyde (%) Hospital-Based Units No. of patient observations No reuse (%) Formaldehyde (%) Peracetic acid (%) Glutarabdehyde (%) Table 3. Patient characteristics by site or location (percentage of n)a Characteristic Freestanding Hospital-Based Freestanding Hospital-Based Patient observations (n) , Mean age ± SD 59.9 ± 153b 57.8 ± ± ± 16.4 Male 49#{149}7C White 575b b 57.7 Black Diabetes melbitus 268b Hypertension 310b b Gbomerubonephritis Other renal diagnosis ASHD 56b b I 1.9 CHF Othercardiac ls.4c PVD CVA/TIA c 6.9 Cancer 25b b 3.7 COPD a p values were calculated based on comparisons of variables between freestanding and hospital-based units within each prevalent period. Abbreviations bp < as in Table 1. C p < hospital-based units that did reuse dialyzers, approximately half used peracetic acid. The number of patient observations in conventional units increased from and then dedined in By defining patient complexity, Tables 3 and 4 address the first hypothesis that freestanding units have less complex patients. Table 3 compares patient characteristics in freestanding and hospital-based units. Freestanding units accounted for the greatest number of patients in both time periods, particularly in These units had significantly older patients, slightly fewer males, significantly fewer whites, and significantly greater percentages of diabetic and hypertensive patients. With the exception of ASHD and cancer, the percentage of patients having each type of comorbid condition appears comparable in freestanding and hospital-based units. Table 4 examines patient distribution and complexity by unit profit status: for-profit units had significantly older patients, fewer whites, slightly lower percentages of patients with ASHD and cancer, and significantly greater percentages of patients with diabetes and CHF. Table 5 shows the distribution of unit characteristics by germicide, reuse technique, unit designation, and profit status. As expected, the distribution of no-reuse units was similar to the patient distribution noted in Table 2, whereas the percentage of freestanding, for-profit units practicing no-

5 21 12 Journal of the American Society of Nephrology J Am Soc Nephrol 9: , 1998 Table 4. Patient characteristics by unit profit status (percentage of n)a Characteristic For-Profit Nonprofit For-Profit Nonprofit Patient observations (n) , Mean age ± SD 59.3 ± 154b 58.5 ± ± 153b ± 16.6 Male White 551b b 63.0 Black Diabetes mellitus 270b b 27.5 Hypertension Gbomerubonephritis Other renal diagnosis ASHD 56b b 12.1 CHF 191b b 20.9 Othercardiac PVD CVAITIA Cancer 25b b 3.7 COPD a p values were calculated based on the comparison of variables between for-profit and nonprofit units. Abbreviations as in Table 1. bp < o.os. Table 5. Distribution of unit characteristics by germicide type and technique (% of units) Characteristic Hospital-Based Units Freestanding Units Hospital-Based Units Freestanding Units Nonprofit For-profit Nonprofit For-profit Nonprofit For-profit Nonprofit For-profit No. of unit observations Non-reuse Formaldehyde - Auto Formaldehyde - Manual Glutaraldehyde - Auto Glutaraldehyde Manual Peracetic acid - Auto Peracetic acid - Manual reuse was less than that of freestanding, nonprofit units. Manuab reprocessing techniques were used more frequently in freestanding, for-profit units. Independent Impact of Patient Characteristics, Profit Status, and Germicide Table 6 shows the relative risks of death associated with patient- and dialysis unit-specific characteristics, separately for each type of unit designation and each time period. Patients in nonprofit, no-reuse units served as the baseline in each of the four separate analyses. The results show, first of all, that older patients within each stratum with shorter exposure on ESRD, and with CHF and other cardiac diseases, consistently have significantly higher mortality. Patients in smaller dialysis units with longer history of certification are also consistently associated with a higher risk of death. The impact of other patientand dialysis unit-specific characteristics on mortality differed according to unit designations and time periods. Once profit status was addressed as a covariate, there was no significant association of germicides with mortality in for either type of unit designation, whereas patients in freestanding units with reuse in had a significantly lower risk. Formaldehyde and peracetic acid reuse was associated with significantly lower risks in freestanding units. Glutaraldehyde reuse had no significant impact on patient survival for both types of units and time periods. The results also show that profit status was significantly associated with patient mortality, with differential effects based on types of units. Freestanding, for-profit units, for instance, had significantly higher risks in both time periods ( : RR =

6 J Am Soc Nephrol 9: , 1998 Reuse Practices and Mortality: Table 6. Relative risks (95% CI) for patient- and unit-specific characteristic? Category and Characteristic Freestanding Hospital-Based Patient-specific characteristics patient age (per year) 1.03 ( 1.02 to 1.04) 1.04 ( 1.03 to 1.05) 1.04 ( 1.03 to I.06) 1.05 ( 1.03 to 1.06) ESRD time (per year) 0.98 (0.96 to 1.00) 0.97 (0.96 to 0.99) 0.97 (0.95 to 0.99) 0.98 (0.96 to 1.00) ASHD 1.02 (0.84 to 1.24) 0.96 (0.85 to 1.09) 0.77 (0.61 to 0.96) 1.03 (0.89 to 1.18) CHF 1.42 (1.25 to 1.61) 1.54 (1.39 to 1.70) 1.38 (1.18 to 1.60) 1.29 (1.14 to 1.46) othercardiac l.31(l.l3tol.50) l.24(l.lltol.38) l.32(l.l3tol.54) b.14(l.o0tol.30) PVD 1.05 (0.93 to 1.17) 0.97 (0.88 to 1.07) 1.09 (0.95 to 1.24) 1.17 (1.05 to 1.32) CVAITIA 1.16 (0.93 to 1.45) 1.22 (1.05 to 1.41) 1.30 (1.01 to 1.68) 1.27 (1.07 to 1.51) cancer 1.42 (1.10 to 1.82) 1.06 (0.85 to 1.31) 1.53 (1.18 to 1.99) 1.15 (0.92 to 1.44) COPD 1.27 (1.01 to 1.60) 1.40 (1.22 to 1.60) 1.29 (1.01 to 1.64) 1.12 (0.94 to 1.33) Unit-specific characteristics unitage(logscabe) l.13(l.o6tol.21) 1.12(l.O6tol.18) l.15(l.o6tol.25) l.09(1.o3tol.16) unit size (log scale) 0.92 (0.85 to 1.00) 0.96 (0.89 to 1.03) 0.91 (0.84 to 0.99) 0.93 (0.87 to 0.99) RO water treatment 1.07 (0.96 to 1.19) 0.81 (0.68 to 0.97) 1.13 (1.00 to 1.28) 0.92 (0.76 to 1.15) >50% patients on 1.03 (0.91 to 1.17) 0.97 (0.83 to 1.14) 1.25 (1.08 to 1.46) 0.96 (0.80 to 1.15) HCO3 automatic reuse techn (0.76 to 1.00) 1.06 (0.95 to 1.18) 0.99 (0.69 to 1.43) 1.12 (0.81 to 1.54) glutaraldehyde reuse 1.15 (0.82 to 1.62) 1.03 (0.80 to 1.33) (0.72 to 1.73) 1.09 (0.74 to 1.59) formaldehyde reuse 0.83 (0.69 to 0.98) 0.82 (0.72 to 0.93) 1.00 (0.69 to 1.45) (0.86 to 1.55) peracetic acid reuse 1.02 (0.85 to 1.22) 0.83 (0.72 to 0.96) 1.09 (0.74 to 1.60) 0.94 (0.67 to 1.32) For-profit/nonprofit 1.38 (1.16 to 1.63) 1.16 (1.02 to 1.32) 0.86 (0.74 to 0.99) 0.70 (0.62 to 0.80) a The results are from four independent Cox regression analyses with the baseline as no-reuse units within the cohort years and unit characteristics. CI, confidence interval; ESRD, end-stage renal disease; RO, reverse osmosis; techn., technique. Other abbreviations as in Table ; : RR = 1. 16), whereas hospital-based, forprofit units had significantly bower risks in both time periods. When patients in both hospital-based and freestanding units were analyzed together, the impact of patient demographics and comorbid conditions was very similar to those in the separate analyses presented in Table 6. However, peracetic acid reuse in in all units was associated with a 15% increased risk compared with no reuse (RR = I.15; 95% CI, 1.01 to 1.31), but had no significant association in the latter time period of (RR = 1.03; 95% CI, 0.92 to 1.15). The impact of formaldehyde or glutarabdehyde was not significant in both time periods. The overall analyses across hospitalbased and freestanding units also showed that profit status was no longer a significant risk factor of death (RR = 1.06; 95% CI, 1.96 to 1.16 in ; RR = 0.94; CI, 0.87 to 1.02 in ). The significantly higher risk of death for profit status in freestanding units and significantly lower risk in hospital-based units appear to be canceled out in the pooled analyses. This result demonstrates the impact of combining results and the need for more careful evaluation of the interactions of these risk factors. Besides the patient- and dialysis unit-specific characteristics listed in Table 6, other identified risk factors or other changes in medical care may affect mortality patterns over the study time period, such as the introduction of erythropoietin during late 1989, or the improvement of health care standards and changes in dialysis therapy. To adjust for these unidentifiable factors and to sort out the real impact of germicides, additional analyses were performed by including the prevalent year as a risk factor in the survival analyses, separately for each time period and each type of unit designation. The results showed that the impact of germicides is similar to that shown in Table 6. Significant progression over the time span only exists in the freestanding dialysis units in , with a relative risk of0.93 (95% CI, 0.88 to 0.98) for each yearly period. That is, after adjusting all of the patient- and dialysis unit-specific characteristics in the study, there is still a 7% risk reduction each year during for patients in freestanding dialysis units. Outcomes Associated with No Reuse To test the hypothesis that no-reuse outcomes are consistent across all units, we analyzed no-reuse units in the different time periods, thereby defining the baseline population before evaluating any additional reuse effects. Figure 1 shows the relative risks from the Cox proportional hazards model, comparing no-reuse mortality by hospital-based/freestanding designation and profit status. On the basis of these characteristics, although the no-reuse units had a mortality pattern in 1989-

7 21 14 Journal of the American Society of Nephrology I Am Soc Nephrol 9: , Non-Profit. Profit Hospital-Based Freestanding Hospital-Based Freestanding Figure 1. Comparison of no-reuse units (relative risk with 95% confidence interval). No-reuse units in showed no significant difference when compared with hospital-based, nonprofit units. The pattern of mortality risk in for-profit units in was similar to that of , but was highly significant that was not significant, it was significant in No-reuse, for-profit units that were hospital-based (RR = 0.70; P = ) differed markedly from those that were freestanding (RR = I.229; P = ). These changing patterns were taken into consideration in the subsequent analyses, particularly when assessing the interactions of dialysis unit characteristics and the impact of reuse practices. Interactions Between Germicides, Profit Status, Unit Status, and Reuse Techniques Our analyses up to this point evaluated each variable independently and showed clear differences between the two time periods. The complex results led us to hypothesize that independent risk factor analyses did not adequately represent the combination of unit characteristics that exist in the true clinical setting, and that such characteristics are more closely depicted by risk factor analyses that consider interactions of the independent variables. Therefore, we analyzed interactions among freestanding/hospital-based designation, germicide type, profit status, and dialyzer reprocessing technique, separating the data by time period. The results from Cox regression analyses with these interactions are shown in Table 7. Patients in nonprofit hospital-based dialysis units with no-reuse served as the baseline in each Linie period. In I 989- I 990, it appears there was a significant association of mortality in only three specific types of dialysis units (shown in boldface in Table 7); no other interactions in this time period differed from the baseline group. Formaldehyde manual reuse performed in nonprofit, freestanding units had an associated risk of 0.66 (P < ), whereas the risk associated with peracetic acid manual reuse in these units was 0.47 (P = ). For-profit, freestanding units practicing peracetic acid manual reuse, in contrast, were associated with a risk of 1.36 (P = ). These results more specifically illustrate that the association of reuse practices with mortality varies dramatically with different dialysis unit settings. In the latter time period of , the significant impact of reuse-associated mortality was isolated to two types of dialysis units (shown in boldface in Table 7), with all other types of units showing no significant difference from the baseline. Hospital-based, nonprofit, formaldehyde-automatic reuse units were associated with a risk of 1.35 (P = ), and freestanding, for-profit units using glutaraldehyde (all techniques combined) were associated with a risk of 1.34 (P = ). Our analyses of these interactions show that only isolated cells were associated with significantly higher or lower risks. No other interactions in either time period differed significantly from the baseline group of hospital-based, nonprofit, no-reuse units. Discussion Although the diverse dialysis unit and patient characteristics in the United States pose significant challenges to any national study

8 J Am Soc Nephrol 9: , 1998 Reuse Practices and Mortality: Table 7. Interaction between germicides and other unit-specific characteristics Hospital-B ased Units Freestand ing Units Characteristic Nonprofit For-Profit Nonprofit For-Profit RR 95%CI RR 95%CI RR 95%CI RR 95% CI F-Auto (0.82 to 1.29) 1.15 (0.66 to 2.00) 0.85 (0.51 to 1.40) 0.93 (0.74 to 1.16) F-Manual 1.08 (0.62 to 1.89) 0.53 (0.16 to 1.71) 0.66 (0.48 to 0.92) 0.99 (0.87 to 1.12) G-Abl 1.12 (0.82 to 1.53) 1.00 (0.55 to 1.83) 0.82 (0.44 to 1.53) 1.30 (0.92 to 1.25) P-Auto (0.90 to 1.35) 0.87 (0.56 to 1.34) 0.75 (0.52 to 1.09) 1.06 (0.90 to 1.25) P-Manual 1.35 (0.72 to 2.54) 0.92 (0.29 to 2.88) 0.47 (0.30 to 0.82) 1.36 (1.17 to 1.59) F-Auto 1.35 (1.05 to 1.74) 0.69 (0.38 to 1.25) 1.09 (0.79 to 1.50) 1.03 (0.87 to 1.22) F-Manual 1.17 (0.71 to 1.94) 0.84 (0.55 to 1.29) 0.94 (0.69 to 1.29) 0.96 (0.87 to 1.07) 0-All 1.16 (0.85 to 1.56) 0.91 (0.61 to 1.38) 1.03 (0.64 to 1.64) 1.34 (1.04 to 1.73) P-Auto 1.04 (0.87 to 1.23) 0.89 (0.65 to 1.22) 0.93 (0.72 to 1.21) 1.06 (0.93 to 1.20) P-Manual 0.53 (0.13 to 2.14) 0.27 (0.07 to 1.10) 0.77 (0.55 to 1.07) 1.02 (0.87 to 1.18) a Boldface indicates significant results. Baseline group: no-reuse, nonprofit, hospital-based units. F, formaldehyde; 0, glutaraldehyde; P. peracetic acid. on reuse safety and efficacy, the ability to analyze these characteristics was previously limited by the detail provided in the federal Medicare ESRD data. Additional sources of data are now available from the CDC, which provides information about dialysis unit practices, including water treatment, reuse germicide, reuse technique, and the use of high-efficiency and high-flux dialyzers. Dialysis unit characteristics are also available from HCFA facility surveys, which identify a unit s profit status and its freestanding or hospital-based designation. The merging of these data sets by Held et al., Feldman et al., and ourselves provided a unique opportunity to analyze dialysis unit and patient factors associated with reuse practices and mortality. To replicate earlier studies, we included only those units using conventional diabyzers. We extracted comorbid conditions from HCFA outpatient institutional and Part A inpatient claims spanning the years The strengths of the HCFA national data set remain its large sample size and patient-specific and dialysis unit-specific information, while its weaknesses continue to center on the lack of data regarding patient dialysis therapy and nutrition. The expanded data set we constructed allowed us to evaluate several major issues in our study: the distribution of patientrelated factors, dialysis unit characteristics, and reuse germicides; the impact of each factor on patient mortality; and the interaction of these factors and their associated mortality risks. To enhance our understanding of dialysis unit and patient characteristics, we addressed these areas through three major hypotheses: (1) Hospital-based and freestanding units have equally complex patients and should therefore be analyzed together. (2) Outcomes in no-reuse units are not consistent across dialysis unit characteristics, suggesting that factors other than reuse may influence mortality. (3) Dialysis unit profit status is associated with patient outcomes. Patient-related factors, including age, gender, race, and renal diagnosis, have previously been shown to influence patient survival (18,31). Comorbid conditions, however, also clearly affect patient survival, a fact that bed us to develop a detailed comorbidity profiling system. Our analysis of patient characteristics and the results of the comorbidity profiling show that conventional dialysis hospital-based units did not, in fact, have more complex patients, but instead had a younger population and fewer diabetic and hypertensive patients. The majority of the remaining comorbid conditions were not different between hospital-based and freestanding units but clearly were strong predictors of mortality. The absolute impact of a comorbid condition varied by time period and unit characteristic. It therefore appears that hospitalbased units do not dialyze sicker patients. Dialysis unit characteristics include not only unit designation (freestanding or hospital-based), but also unit profit status, which we obtained from the ownership classification field in the HCFA facility survey. Although Held et al. provided statistics on the percentage of freestanding units that were for-profit, this factor was not considered in the survival analysis of Held et al. or Feldman et al. (18,20). Our analyses showed that the distribution of patient factors in for-profit units was significantly different than in nonprofit units, with for-profit units treating older patients, more minorities, and a greater percentage of diabetic patients. For-profit units, which account for the majority of freestanding units, treated patients who were as complicated or more so than those in nonprofit units (which account for the majority of hospital-based units). Our study also found a significant increased mortality risk associated with for-profit units, a risk not previously reported. This association as an overall effect suggests that profit status can and should be addressed as a factor that may influence treatment practices and patient outcomes, particularly in specific settings (described below).

9 21 16 Journal of the American Society of Nephrology J Am Soc Nephrol 9: , 1998 Our study also showed that hospital-based and freestanding units have dramatically different practices in terms of reuse. Approximately 75% of conventional hospital-based units did not reuse dialyzers, whereas the reverse was true for freestanding units, suggesting that comparisons of no-reuse and reuse outcomes may be influenced by dialysis unit designation. The large number of no-reuse units located in the hospital-based setting led us to carefully evaluate the no-reuse population to assess dialysis unit characteristics that were not confounded by reuse practices. In , for example, there was no significant difference among no-reuse, conventional diabyzer outcomes in hospital-based, freestanding, or for-profit units. This result changed, however, in the period, when freestanding, for-profit, no-reuse units had a 23% higher associated risk of death compared with hospital-based, nonprofit, no-reuse units. The opposite was true in for-profit units that were hospital-based, where the risks were significantly lower (RR = 0.70; P = ). Although the pattern of the effect for profit status was similar in , it was not significant compared with (Figure 1). These findings suggest that factors other than patient characteristics and reuse germicides can influence mortality. The differences in no-reuse outcomes may be explained by the financial structure and incentives of the varying types of units. The cost structures and overhead charges of a hospital-based dialysis unit are spread over many fiscal units, meaning that the unit is only one of many cost and revenue centers that can buffer losses in other hospital areas. Such units also refer patients directly to the hospitals themselves for services and are therefore a source of revenue. Freestanding units, in contrast, are vulnerable to more direct effects of cost structures and margins and lack additional sources of revenue such as hospital admissions. Nonprofit, freestanding units are under considerably less financial pressure and show no significant difference in outcomes compared with nonprofit, hospital-based, no-reuse units. These important no-reuse outcomes were integrated into our detailed interaction survival model, where a clearer picture emerged of their association with germicides, mortality, and dialysis unit characteristics (Table 7). In , only three of 20 interactive cells were associated with mortality differences, compared with no-reuse, hospital-based, nonprofit units. Nonprofit, freestanding units practicing either formaldehyde or peracetic acid reuse had a significantly lower risk of death, whereas peracetic acid manual reuse in for-profit units had a significantly higher risk in this time period. In , the findings were even narrower, with only two cells showing significant associations with mortality. Higher risks were noted in hospital-based, nonprofit units practicing formaldehyde automatic reuse and in for-profit, freestanding units using glutaraldehyde. These two of the 20 possible reuse cells represent only a small minority of the total patient population. These results clearly demonstrate that small cells (specific types of dialysis units) with strong associations can drive the overall analysis. The previously identified associations between reuse and increased mortality in , therefore, are the result of specific reuse practices and dialysis unit characteristics. The remaining data, in fact, suggest that noreuse and reuse outcomes do not differ. Our study was based on the period-prevalent hemodialysis patient population. There may be a potential bias associated with a prevalent study design due to preferential exclusion of patients who died early during their dialysis course. The potential bias, however, should theoretically change only the magnitude of an association, not the direction (29). A specific example is the study by Feldman et a!., which was incident-based ( ) and showed an effect of peracetic acid germicide on patient mortality that was similar to that in the Held et a!. point prevalent study, but did not show an effect of glutaraldehyde similar to our findings (18). Our analysis on the period-prevalent cohort did adjust for prior ESRD exposure, as did Held et al. ; thus, the potential selection bias should be reduced. Our findings, along with the no-reuse associations, suggest that factors other than reuse germicides should be evaluated to determine more specific associations with mortality rates. The most prominent of these additional factors rebates to dialysis therapy, nutrition, and anemia correction (32), which may explain why there was a 7% mortality risk reduction each year from in freestanding dialysis units. The earlier analyses included data from , a period in which dialysis therapy was first being identified as an important factor in patient survival (23). The U.S. Renal Data System has shown that dialysis therapy has been increasing from a mean Kt/V of 0.92 (before 1990) to 1.22 in December 1993 (23). Anemia correction also changed during this period, as recombinant human erythropoietin was introduced into Medicare reimbursement in August of Studies have evaluated the impact of hematocrit level on patient outcomes and have shown that hematocrit levels <30% have progressive associated increased mortality (32). These two important factors, both of which are more directly under the control of individual and chain dialysis units, may significantly affect patient outcomes, a fact of particular importance since our analyses of no-reuse units suggest that specific dialysis unit characteristics are associated with mortality. Conclusions Our analysis of the association between dialysis unit characteristics, reuse germicides, and mortality in and reinforces several points made in earlier studies but reaches additional conclusions as well. For the time period, we confirmed the association of peracetic acid manual reuse with increased mortality in for-profit, freestanding units, which accounted for 8.4% of all dialysis units in that time period, but we did not find this association in Reuse-associated mortality in this latter time period was limited to fewer than 3% of all dialysis units, those practicing glutaraldehyde reuse in freestanding, for-profit units, and formaldehyde automatic reuse in nonprofit, hospital-based units. It appears, therefore, that mortality is not consistently associated with reuse or no-reuse practices, and we hypothesize that factors such as dialysis therapy, nutrition, and anemia correction are better candidates to explain these varying mortality results.

10 J Am Soc Nephrol 9: , 1998 Reuse Practices and Mortality: Acknowledgments Research support for this study was provided in part by Hennepin Faculty Associates and the Minneapolis Medical Research Foundation (MMRF). Minntech Corp. provided an unrestricted grant to MMRF for this study. The investigators and staff involved in this study have no direct financial relationship to Minntech Corp., either by stock, options, consulting agreements, or other financial arrangements, except as noted above. We acknowledge the substantial efforts of the staff of Nephrology Analytical Services and their outstanding assistance in this study. In addition, we thank Dana D. Knopic for her administrative support and assistance in manuscript preparation. References 1. Shaldon 5, Silva H, Rosen 5: Technique of refrigerated coil preservation hemodialysis with femoral venous catheterization. Br Med J 2: , Tokars I, Alter M, Favero M, Moyer L, Bland L: National surveillance of hemodialysis-associated diseases in the United States, ASAJO J 39: 71-80, Association for the Advancement of Medical Instrumentation: Recommended Practice for Reuse of Hemodia!vzers, Arlington, VA, Diaz R, Washburn 5, Cauble L, Siskind M, Van Wyck D: The effect of dialyzer reprocessing on performance and beta 2-microglobulin removal using polysulfone membranes. Am J Kidney Dis 21: , Windus D, Atkinson R, Santoro 5: The effect of hemodialysis on platelet activation with new and reprocessed regenerated cellulose dialyzers. A,n J Kidney Dis 27: , Kerr P. Argiles A, Canaud B, Flavier J, Mion C: The effects of reprocessing high-flux polysulfone dialyzers with peroxyacetic acid on beta 2-microglobulin removal in hemodiafiltration. Am J Kidney Dis 19: , Pereira B, King A, Poutsiaka D, Strom I, Dinarelbo C: Comparison of first use and reuse of Cuprophan membranes on interbeukin-l receptor antagonist and interleukin-l beta production by blood mononuclear cells. Am J Kidney Dis 22: , Kaplan A, Halley 5, Lapkin R, Graeber C: Dialysate protein losses with bleach processed polysulphone dialyzers. Kidney mt 47: , Jackson B, Beck-Sague C, Bland L, Arduino M, Meyer L, Jarvis W: Outbreak of pyrogenic reactions and Gram-negative bacteremia in a hemodialysis center. Am J Nephrol 14: 85-89, Vanholder R, Vanhaecke E, Ringoir 5: Pseudomonas septicemia due to deficient disinfectant mixing during reuse. mt J Artif Organs 15: 19-24, Pegues D, Beck-Sague C, Woollen 5, Oreenspan B, Burns 5, Bland L, Arduino M, Favero M, Machow R, Jarvis W: Anaphylactoid reactions associated with reuse of hollow-fiber hemodialyzers and ACE inhibitors. Kidney mt 42: , Pegues D, Oettinger C, Bland L, Oliver I, Arduino M, Aguero 5, McAllister S, Gordon S. Favero M, Jarvis W: A prospective study of pyrogenic reactions in hemodialysis patients using bicarbonate dialysis fluids filtered to remove bacteria and endotoxin. J Am Soc Nephrol 3: , Cheung A, Dalpias D, Emmerson R, Leypoldt I: A prospective study on intradialytic symptoms associated with reuse of hemodialyzers. Am J Nephrol II: , Tokars I, Alter M, Favero M, Moyer L, Bland L: National surveillance of hemodialysis-associated diseases in the United States, ASAIO J 39: 71-80, Tokars II, Alter Mi, Favero MS, Moyer LA, Bland LA: National surveillance of dialysis associated diseases in the United States, ASAJO J 39: , Alter M, Favero M, Moyer L, Miller J, Bland L: National surveillance of dialysis-associated diseases in the United States, 1988.ASAIOTrans36: , Alter M, Favero M, Moyer L, Bland L: National surveillance of dialysis-associated diseases in the United States, ASAJO Trans 37: , Held PJ, Wolfe RA, Gaylin DS, Port FK, Levin NW, Turenne MN: Analysis of the association of dialyzer reuse practices and patient outcomes. Am J Kidney Dis 23: , Luehmann D, Cosentino L: Safety of dialyzer reuse with Renabin: The untold story. Dial Transplant 23: , Feldman H, Kinosian M, Bilker W, Simmons C, Holmes J, Pauly M, Escarce J: Effect of dialyzer reuse on survival of patients treated with hemodialysis. JAMA 276: , Social Security Administration: Social Security: Your Nu,nber [Pamphlet]. 4: Health Care Financing Administration: Data Users Reference Guide [Catalog]. Baltimore, Health Care Financing Administration, US Renal Data System: USRDS 1996 Annual Data Report, Bethesda, Keane W, Collins A: Influence of co-morbidity on mortality and morbidity in patients treated with hemodialysis. Am J Kidney Dis 24: , Shapiro F, Umen A: Risk factors in hemodialysis patient survival.asaioj6: , Gaylin DS, Held PJ, Port FK, Hunsicker LO, Wolfe RA, Kahan BD, Jones CA, Agodoa LY: The impact of comorbid and sociodemographic factors on access to renal transplantation. JAMA 269: , Held P1, Port FK, Gaylin DS, Wolfe RA, Levin NW, Blagg CR, Garcia JR. Agodoa LY: Evaluations of initial predictors of mortality among new ESRD patients: The USRDS Case Mix Study [Abstract]. J Am Soc Nephrol 2: 328, Cox D: Regression models and life-tables. J Roy Stat Soc Series B 34: , Brookmeyer R, Gail MH: Biases in prevalent cohorts. Biometrics 43: , U.S. Renal Data System: USRDS 1997 Annual Data Report, Bethesda, Collins A, Ma J, Umen A, Keshaviah F: Urea index and other predictors of hemodialysis patient survival. Am J Kidney Dis 23: , Madore F, Lowrie E, Brugnara C, Lew N, Lazarus M, Bridges K, Owen K: Anemia in hemodialysis patients: Variables affecting this outcome predictor. J Am Soc Nephrol 8: , 1997