Dialyzer Reuse with Peracetic Acid Does Not Impact Patient Mortality

Abstract

Summary

Background and objectives

Numerous studies have shown the overall benefits of dialysis filter reuse, including superior biocompatibility and decreased nonbiodegradable medical waste generation, without increased risk of mortality. A recent study reported that dialyzer reprocessing was associated with decreased patient survival; however, it did not control for sources of potential confounding. We sought to determine the effect of dialyzer reprocessing with peracetic acid on patient mortality using contemporary outcomes data and rigorous analytical techniques.

Design, setting, participants, & measurements

We conducted a series of analyses of hemodialysis patients examining the effects of reuse on mortality using three techniques to control for potential confounding: instrumental variables, propensity-score matching, and time-dependent survival analysis.

Results

In the instrumental variables analysis, patients at high reuse centers had 16.2 versus 15.9 deaths/100 patient-years in nonreuse centers. In the propensity-score matched analysis, patients with reuse had a lower death rate per 100 patient-years than those without reuse (15.2 versus 15.5). The risk ratios for the time-dependent survival analyses were 0.993 (per percent of sessions with reuse) and 0.995 (per unit of last reuse), respectively. Over the study period, 13.8 million dialyzers were saved, representing 10,000 metric tons of medical waste.

Conclusions

Despite the large sample size, powered to detect miniscule effects, neither the instrumental variables nor propensity-matched analyses were statistically significant. The time-dependent survival analysis showed a protective effect of reuse. These data are consistent with the preponderance of evidence showing reuse limits medical waste generation without negatively affecting clinical outcomes.

Introduction

Since the introduction of the concept in the early 1960s, the potential risks and benefits of reusing dialysis filters have been actively debated in the medical literature. The earliest studies reported increased mortality associated with reuse, but suggested that mortality was associated with the specific reprocessing agent and not reuse per se. Use of high-flux dialyzers became much more common in the late 1980s and early 1990s (1). Research from this period did not find an association between reuse and mortality risk (Table 1). The most recent study published was a single arm pre-/postanalysis of facilities switching from dialyzer reuse to single-use. Lacson et al. (2) examined changes in mortality and inflammatory markers in 23 newly acquired dialysis centers who switched from peracetic acid–based dialyzer reuse to single-use filters and found a remarkable drop in mortality (hazard ratio [HR] = 0.74) and mean C-reactive protein (26.6 to 20.2 mg/L). However, there was significant channeling bias in this study (only a subset of centers made the switch), and the study did not control for several potential sources of confounding.

Table 1.

Summary of published reuse mortality analyses

Citation	Study Design	Year Outcomes Collected	Sample Size	Primary Outcome	Results
Collins et al. 1998 (7)	Retrospective analysis of Medicare data	1989 to 1993	34,348	Mortality risk	Inconsistent relationship between mortality and reuse
Ebben et al. 2000 (9)	Period-prevalent model using five cohorts	1991 to 1995	84,279	Mortality risk	No difference in mortality between reuse and non reuse units
Fan et al. 2005 (6)	Prospective, observational using Medicare patients	2000 to 2002	74,831	Mortality risk	No difference between reuse and single-use in an intent-to-treat approach (HR 0.98; 95% CI: 0.94 to 1.02) and in an as-treated analysis (HR 0.97; 95% CI: 0.93 to 1.01)
Feldman et al. 1999 (11)	Nonconcurrent cohort	1986 to 1987	1491	Mortality risk	Higher rate of death in analyses unadjusted for confounders (RR 1.25, 95% CI 0.97 to 1.61), adjusted for demographics and renal diagnosis (RR 1.16, 95% CI 0.96 to 1.41), and analyses additionally adjusted for comorbidities (RR 1.25, CI, 1.03 to 1.52) in reuse facilities
Feldman et al. 1996 (12)	Nonconcurrent cohort	1986 to 1987	27,938	Mortality risk	Combination of peracetic and acetic acid (RR 1.10, 95% CI: 1.02 to 1.18)
					No difference with formaldehyde or glutaraldehyde compared with non-reuse
Held et al. 1994 (13)	Historic prospective study of Medicare patients	1989 to 1990	66,000	Mortality risk	Peracetic acid (RR 1.13, P < 0.001)
					Flutaraldehyde (RR 1.17, P = 0.01)
					Formalin — no increase in mortality
Lacson et al. 2011 (2)	Prospective crossover	2007	Reuse 1259; single-use 1354	Morality risk and C-reactive protein	Decrease in mortality (HR = 0.74) and mean C-reactive protein (26.6 to 20.2 mg/L) in single-use compared with reuse
Lowrie et al. 2004 (14)	Single arm, crossover	2001	Reuse 52,985; single-use 18,137	Mortality risk	A survival advantage for single-use was observed after a 90-day lag period (RR 0.93, 95% CI: 0.88 to 0.99).
Port et al. 2001 (8)	1- and 2-year follow-up analysis of USRDS data	1994 to 1995	12,791	Mortality risk	No difference between reuse and single-use (RR 0.95, 95% CI: 0.86 to 1.08)
					Within reuse, low flux synthetic membrane versus high flux (RR 1.24, 95% CI: 1.02 to 1.52)
					Without bleach versus with bleach (RR 1.24, 95% CI: 1.01 to 1.48)

RR, relative risk; HR, hazard ratio; CI, confidence interval; USRDS, United States Renal Data System.

Determining whether reuse is associated with increased morbidity and mortality has important patient and environmental implications. When financial implications are also present, as would be the case for dialysis providers and dialyzer manufacturers, steps must be taken to rigorously control sources of potential bias. Accordingly, we sought to address the methodological shortcomings of previous work by applying multiple statistical techniques to control for confounding and to fully explore the association between reuse and mortality, if any.

Materials and Methods

We conducted a series of analyses to determine the association between dialyzer reuse and patient mortality in a large cohort of in-center hemodialysis (HD) patients: instrumental variables, propensity-score matching, and time-dependent survival analysis. All analyses were conducted in SAS 9.2 (Cary, NC).

Instrumental Variables Analysis at Single-Use Versus Reuse Centers

In many instances, the decision to reuse dialyzers is made by the medical director or facility administrator and is applied to all patients in the center. Centers without reuse facilities are perforce single-use centers. In either case, the center acts an “instrument,” controlling potential channeling bias through restriction. The robustness of this method has been shown in numerous studies (3–5).

To define the instrumental variable, we selected centers in which 100% of dialysis sessions were conducted using single-use filters during the study period. The comparator group defined high reuse as centers with ≥95% of patients using a reused filter over the study period. Of the 1140 centers with ≥20 prevalent HD patients receiving services at this large dialysis organization (LDO) during this period, 183 (16.1%) qualified as single-use centers and 301 (26.4%) as reuse centers. Any death that occurred within 30 days of the last treatment with the organization was counted. Prevalent patients (>120 days) as of January 1, 2009 were followed for 1 year, and days at risk were counted as the time from the beginning of the period through (1) the last day of the period, (2) the last date of dialysis at the LDO, or (3) the date of death.

We calculated the Cox proportional hazard comparing mortality in the single-use versus high reuse centers using the PHREG procedure (SAS 9.2). A crude HR and an HR adjusted for patient age, vintage, race (African-American, Caucasian, Hispanic, Asian/Pacific Islander, Native American, other), gender, primary cause of ESRD (diabetic or nondiabetic), dialysis access (arteriovenous [AV] fistula or other), dialysis adequacy (Kt/V), and Charlson comorbidity index was calculated between the two groups. All covariates, defined at the start of the observation period, were tested to determine whether they met the proportional hazards assumption, and all two-level interaction (effect modification) terms were analyzed.

Propensity-Score Matched Patient-Level Analysis

To obtain a more comprehensive sample, we conducted a propensity-score matched patient-level analysis of likelihood of death by single-use versus high reuse across all of the LDO clinics. The propensity score for each patient was defined as the predicted probability of receiving dialysis with single-use filters, given the included set of predictors, and was calculated for each prevalent (>120 days) in-center HD patient as of January 1, 2009. An initial logistic regression was run with a dichotomous dependent variable (no reuse versus any reuse). The covariates were entered in a stepwise fashion, with a probability of 0.2 both to enter and remove from the propensity score equation.

To select the propensity-score matched population, a “greedy” algorithm was applied to the propensity-score-ranked data using the caliper support matching method and selection without replacement. Single-use patients and potential matches among reuse patients were sorted by their propensity score. Single-use patients were assigned a random number from the uniform probability function, and the algorithm was allowed to run iteratively over the available reuse patients selecting control individuals matched on five digits of the propensity score (i.e., a caliper of 0.0001). If more than one potential control matched at the five-digit level, all potential matches were sorted by the uniform random variable assigned and the first match was selected. After all possible five-digit propensity-score matches had been made, the procedure was repeated for all unmatched individuals using a matching caliper of 10⁻⁴ through 10⁻¹ as described (3).

To examine the effectiveness of the propensity match, we conducted t tests (for continuous variables) or χ² tests (for categorical variables) to evaluate whether statistically significant differences remained between the groups after match.

A logistic model (LOGISTIC procedure in SAS 9.2) was used to assess the association of reuse and mortality over the observation period. Differences between the propensity-score matched groups that were found to be significant were included in adjusted models of this potential association.

Time-Dependent Survival Analysis

Whereas the previous analyses can help to define an association between reuse and mortality, they define reuse dichotomously, which does not allow for the patient's changing exposure to reuse as filters are used again and again and then replaced. We conducted two time-dependent analyses to test two competing hypotheses: (1) reuse has a cumulative effect (the greater percent of sessions using a reused filter, the greater the risk) and (2) reuse has an acute effect (a filter that is used more often will have a proximal effect on health outcomes).

To test these hypotheses, we conducted survival analyses in which each day was defined as a new exposure period. The cumulative effect analysis defined exposure as the percentage of sessions over the selected period in which the patient had used a previously used filter. Thus, a patient who received a new filter every 10 sessions would have an exposure level that varied over time but did not exceed 90%. The proximal effect analysis used the most recently used filter as the exposure level. Thus, a first use filter received a value of 0 and a filter used 27 times received a value of 27. Measured in this way, exposure level could change considerably from day to day, e.g., when a new filter was used. For this reason, exposure was tested with no lag and with a lag of 7 days.

These models were adjusted for patient characteristics at the start of the observation period, including age, vintage, race, gender, primary cause of ESRD (diabetes, hypertension, polycystic disease, and other), and primary insurance type (commercial, Medicare, Medicaid, Veterans Affairs, other, none). Comorbidities were considered as classes of comorbidities (cardiovascular disease, chronic obstructive pulmonary disease, diabetes, liver disease, gastrointestinal bleeding, cancer) and Charlson comobidity index. The model with the better fit to the data (as determined by lower Akaike information criterion) was to be selected.

These analyses were limited to prevalent HD patients (>120 days) over 2 years (July 1, 2008 through June 30, 2010). Any death that occurred within 30 days of the last treatment with the LDO was counted. Days at risk were counted as described earlier. Data for the 3 months preceding the observation period were used to establish the baseline exposure level in the cumulative analysis. Analyses were conducted in SAS 9.2, using the PHREG procedure.

The reuse process is standard across all facilities. Dialyzers are reprocessed within 2 hours of use or are stored in a designated reuse refrigerator up to 36 hours to retard bacterial growth until reuse is begun. Association for the Advancement of Medical Instrumentation quality water is used to dilute Renalin (a stabilized mixture of hydrogen peroxide, peroxyacetic acid, and acetic acid) to create a 1% solution, which was used as the sterilant. Dialyzers must be disinfected for a period of ≥30 minutes but ≤24 hours before use. The minimum and maximum allowable storage times for disinfected dialyzers are 11 hours and 14 days.

Results

Instrumental Variables Analysis

Table 2 shows characteristics of patients at single-use and high reuse centers. The patients at high reuse centers were significantly more likely to be Hispanic or non-African American, have diabetes as primary cause of ESRD, use an AV fistula as primary vascular access, have higher dialysis adequacy as measured by Kt/V, and have a higher Charlson comobidity index. However, with a sample size in excess of 27,000 patients, clinically nonmeaningful differences can be statistically significant. Of these variables, only the differences in race and percent of patients with diabetes as primary cause of ESRD would be considered clinically meaningful. There was also high consistency in the types of dialyzers used. More than 79% of patients at single-use and 85% at high reuse centers used Polyflux dialyzers. At single-use centers, 76.4% of patients used Polyflux 170H and 210H dialyzers, 2.7% used another Polyflux dialyzer, 16.0% used Optiflux NR, and 4.9% used other dialyzers. At high reuse centers, 81% of patients used Polyflux 17R, 21R, and 24R dialyzers, 3.6% used another Polyflux dialyzer, and 14.4% used other dialyzers.

Table 2.

Patient characteristics by reuse policy (centers with 0 and ≥95% reuse)

	At Center with No Reuse (0%) (10,182 patients in 183 centers)	At Center with High Reuse (≥95%) (17,223 patients in 301 centers)	P
Age (years), mean, SD	62.8 (±14.9)	62.8 (±14.8)	<0.001
Vintage (years), mean, SD	4.34 (±2.73)	4.31 (±2.76)	<0.001
Race (%)			<0.001
African-American	51.4	26.7
Caucasian	35.0	34.1
Hispanic	8.2	27.9
Asian/Pacific Islander	1.2	5.5
Native American	0.9	3.0
other	3.3	2.7
Male (%)	56.1	55.3	<0.001
Diabetes cause of ESRD (%)	42.7	50.8	<0.001
AV fistula (%)	56.6	58.5	0.003
Kt/V, mean, SD	1.66 (±0.33)	1.71 (±0.33)	<0.001
Charlson comobidity index, mean, SD	5.95 (±2.25)	6.04 (±2.16)	0.001
Dialyzer type
Polyflux 170H	41.4%	2.4%	NA
Polyflux 210H	35.0%	1.2%
Polyflux 17R	—	8.0%
Polyflux 21R	—	39.5%
Polyflux 24R	—	34.1%
Optiflux NR	16.0%	0.4%
all others	7.6%	14.4%

NA, not applicable; AV, arteriovenous.

Table 3 shows differences in unadjusted and adjusted mortality among patients in single-use and high reuse centers. Patients at high reuse centers had a slightly higher unadjusted death rate per 100 patient-years (16.2 versus 15.9), but the HR of 1.02 (95% confidence interval [CI]: 0.95 to 1.09) was not statistically significant. After adjusting for race and percentage of patients with diabetes as the cause of ESRD, the HR was slightly higher but still not significant: 1.04 (95% CI: 0.97 to 1.12). All covariates met the proportional hazards assumption, and no interaction or effect modification terms were found to be significant. Adjusted survival curves for the two groups of patients can be seen in Figure 1.

Table 3.

Unadjusted and adjusted mortality by single-use and high reuse centers

	At Center with No Reuse (0%) (10,182 patients in 183 centers)	At Center with High Reuse (≥95%) (17,223 patients in 301 centers)
Deaths	1357	2362
Mortality (%)	13.33	13.71
Deaths/100 patient-years (95% CI)	15.9 (15.0, 16.7)	16.2 (15.5, 16.8)
Crude HR (95% RL)	Ref	1.02 (0.95, 1.09)
Adjusted HR (95% RL)a	Ref	1.04 (0.97, 1.12)

CI, confidence interval; RL, reference limit; Ref, reference.

^aAdjusted for race and percentage of patients with diabetes as cause of ESRD.

Figure 1.

Survival by dialyzer type. Propensity-score-matched cohorts, adjusted for race and cause of ESRD.

It is important to note that 392 of 17,223 patients at high reuse centers received only single-use filters. These patients are included in the results for high reuse centers to avoid selection bias. Removal of these patients from analysis did not affect the results or conclusion.

Propensity-Score Matched Patient-Level Analysis

A 1:1 propensity-score matched sample of single-use and reuse patients was created from all available prevalent patients in the organization, resulting in 13,801 single-use patients and an equal number of reuse controls (Table 4). The propensity-score match eliminated differences on all variables with the exception of race. Patients who reused filters were significantly less likely to be African American (37.7% versus 52.2%) and more likely to be Hispanic (17.7% versus 8.8%). Because black patients are known to have significantly lower postdialysis mortality (4), the higher proportion of African-American patients in the single-use group is likely to bias the analysis in favor of single-use. Catheter rates were similar among single-use and high reuse patients (18.0% and 19.2%, respectively), and this small difference was eliminated by the propensity match (20.1% in each propensity matched group).

Table 4.

Patient characteristics: propensity-score-matched sample

	Single-Use (13,801 patients)	Reuse (13,801 patients)	P
Age (years), mean, SD	61.7 (±14.8)	61.6 (±15.3)	<0.001
Vintage (years), mean, SD	4.36 (±2.73)	4.32 (±2.74)	<0.001
Race			<0.001
African-American	52.2%	37.7%
Caucasian	32.9%	36.3%
Hispanic	8.8%	17.7%
Asian/Pacific Islander	2.3%	4.0%
Native American	0.6%	1.5%
other	3.2%	2.8%
Male (%)	55.8%	56.7%	<0.001
Diabetes cause of ESRD (%)	40.1%	39.6%	<0.001
AV fistula (%)	55.3%	55.6%	<0.001
Kt/V, mean, SD	1.66 (±0.34)	1.66 (±0.32)	<0.001
Charlson comobidity index, mean, SD	5.80 (±2.24)	5.77 (±5.74)	<0.001

As shown in Table 5, patients who reuse filters had a lower crude mortality (15.2 deaths per 100 patient-years; 95% CI: 14.5 to 15.9) than patients dialyzing with single-use filters (15.5 deaths per 100 patient-years; 95% CI: 14.8 to 16.2). However, this difference was not significantly different. A logistic model fitted to adjust for differences in racial composition yielded a risk ratio for death within the observation year at 0.98 (0.91, 1.05) for reuse versus single-use. A model with adjustments for race, age, vintage, and the interaction of these factors yielded a risk ratio of 1.00 (95% CI: 0.93 to 1.07) for reuse versus single-use. Thus, none of the effects in these analyses reached statistical significance.

Table 5.

Mortality: propensity-score-matched sample

	Single-Use (13,801 patients)	Reuse (13,801 patients)
Deaths	1785	1789
Mortality (%)	12.93%	12.96%
Deaths/100 patient-years (95% CI)	15.5 (14.8 to 16.2)	15.2 (14.5 to 15.9)
Crude RR (95% RL)	Ref	1.00 (0.94 to 1.08)
Adjusted RR (95% RL)a	Ref	0.98 (0.91 to 1.05)
Adjusted RR (95% RL)b	Ref	1.00 (0.93 to 1.07)

RR, relative risk; RL, reference limit.

^aAdjusted for race.

^bAdjusted for race, age, vintage, and the interaction of these factors.

Time-Dependent Survival Analysis

One time-dependent survival analysis of all prevalent HD patients used percent of sessions with a reused filter as the exposure, thus measuring the cumulative effects of reuse over time. Over the 2-year period, the percentage of sessions with reuse was marginally related to improved survival, with an adjusted odds ratio of 0.993 (0.992, 0.995) for mortality with each increasing percentage point. However, the range of percentages was narrow, with a mean of 91.8% and an interquartile range (25% to 75%) of 89.4% to 96.9%.

Another time-dependent survival analysis used the number of times a filter was reused at each session as the time-dependent variable, measuring the proximal or acute effects of dialyzers that are reused more frequently. The adjusted odds ratio per increased unit of last filter reuse was 0.995 (95% CI: 0.994 to 0.996), showing a marginal protective effect of increased reuse on mortality over the observation period. When the reuse number with a 7-day lag was substituted for the most recent number, the odds ratio was unchanged. The final adjusted models included Charlson score (rather than individual comorbidities) because of better model fit.

Medical Waste

We added the total number of reuses to determine the number of dialyzers saved over this 2-year period. Among prevalent HD patients, this amounted to >13.8 million filters. At 1.6 pounds per average high-flux filter (5), this equates to >22 million pounds (10,000 metric tons) of medical waste.

Discussion

Our analyses showed no meaningful association between reuse and mortality. Both the instrumental variables analysis and the propensity-score match failed to reject the null hypothesis, despite being powered to detect miniscule effects. Between two time-dependent survival analyses—one in which reuse could have a cumulative effect over time and one that looked at acute impact—only the latter showed a small but statistically significant effect in favor of reuse. This result could be interpreted to mean that greater reuse was associated with slightly lower mortality or that first use was associated with a slightly higher mortality risk, which decreased over time. We prefer to take a more conservative approach and interpret the findings to indicate a lack of clinically meaningful association between reuse and mortality risk. The savings of 10,000 metric tons of waste over 2 years is an indisputable advantage of reuse.

These data supplement and support the larger historical body of literature, in which studies that adequately address confounding show no adverse effect of reuse on clinical outcomes (6–9). These results also add to the literature, inasmuch as early research was done before the wide use of high-flux dialyzers. As shown in this analysis, reuse of high-flux dialyzers is not associated with increased mortality. In the past, questions have been raised regarding the impact that infection and exposure to germicides and reuse agents may have on mortality. These concerns have not been born out in the majority of published studies. Instead, many potential benefits of reuse have been noted: delivery of a higher dose of dialysis, increased biocompatibility, and avoidance of first-use syndrome. From an environmental perspective, the practice of reuse decreases the generation of biomedical waste in the form of discarded single-use dialyzers.

Dialyzer reuse has existed for nearly 50 years in the United States. As a result, there are specific guidelines issued by Association for the Advancement of Medical Instrumentation on how reuse should be conducted. Guidance from the two dominant regulatory agencies, the Food and Drug Administration and the Centers for Medicare and Medicaid Services, exist to ensure appropriate reuse as it relates to dialyzer and dialysis unit certification. It is important to reinforce the need for monitoring with reused dialyzers, because reprocessing of high-flux dialyzers with Renalin may result in degradation of water permeability over time (10).

We believe it is important to remind the greater dialysis community of the importance of using appropriate statistical techniques when conducting observational research. Without properly controlling for confounding, type 1 errors can easily occur, and effects can be reported where no real associations exist. It is critical that appropriate techniques, such as those used here, or others such as marginal structural modeling, be used to minimize the occurrence of such errors.

Conclusions

We examined the association between dialyzer reuse and patient mortality in a large cohort of in-center HD patients using three analytical approaches to control for potential confounding. With one exception, there was no association between dialyzer reuse and mortality. The one small statistically significant effect showed decreased mortality associated with the number of times an individual dialyzer was reused.

Limitations

Without the benefit of a randomized controlled trial, observational research must be used. As with any retrospective study, the inability to prospectively assign exposure leads to the possibility of confounding by indication. We have attempted to minimize this through multiple statistical techniques, but one cannot rule out the possibility of residual confounding. Additionally, there is the possibility of temporal confounding related to improvements in the overall population affecting the outcomes of interest but not flowing through the variable of interest. We attempted to address that by limiting the time period being studied to 2 years. Finally, the patients in this study were followed for 1 year, which does not preclude cumulative effects given a longer follow-up and period of exposure.

Disclosures

The authors are employees of DaVita Inc.