Impact of Dialysis Dose and Membrane on Infection-Related Hospitalization and Death: Results of the HEMO Study

Study Design

The design and methods of the HEMO Study have been reported previously (9). In brief, the HEMO Study was a multicenter, prospective, randomized, 2 × 2 factorial clinical trial that evaluated the effect of dialysis dose and flux on the morbidity and mortality of hemodialysis patients. The study was approved by the Institutional Review Board at each of 15 clinical centers associated with 72 participating dialysis units, and all patients gave written informed consent.

Baseline

The subjects were enrolled in the baseline phase between March 1995 and October 2000 and randomized between May 1995 and February 2001. Eligibility requirements for baseline enrollment included age between 18 to 80 yr, receiving in-center hemodialysis thrice weekly, and on hemodialysis > 3 mo. Demographic and clinical information was collected at baseline. Patients were excluded during baseline if (1) their residual urea clearance in a 24 to 46-hr urine collection was > 1.5 ml/min per 35 L of urea volume, (2) their serum albumin (nephelometry) was < 2.6 g/dl, (3) they failed to achieve the high target dialysis dose in ≤ 4.5 h on two of three consecutive monitored dialysis sessions, (4) they had serious comorbid medical conditions, including active malignancy or infection, unstable angina, or end-stage cardiac, pulmonary, or hepatic disease, or (5) they were scheduled for a living donor kidney transplant.

Interventions

Patients meeting the inclusion and exclusion criteria were randomized to the study interventions in a 2 × 2 factorial design with equal allocation. Each patient was randomized to receive either a standard (eKt/V 1.45) or high (eKt/V 1.05) dialysis dose, and to dialyze with either a low-flux (32 β₂-microglobulin clearance < 10 ml/min) or high-flux (32 β₂-microglobulin clearance > 20 ml/min) membrane. Unmodified cellulose dialyzers were excluded. The target dialysis dose was achieved by manipulating dialyzer clearance, dialysis blood flow, and treatment time (minimum ≥ 2.5 h). Adherence to dose intervention was monitored by monthly urea kinetic modeling. β₂-microglobulin clearance was measured every 2 mo for high-flux dialyzers and every 6 mo for low-flux membranes. Dialyzer reuse was permitted, but the number of reuses was limited in the high-flux arm to meet the target β₂M clearance. The mean achieved equilibrated Kt/V (eKt/V) was 1.53 ± 0.09 in the high-dose group and 1.16 ± 0.08 in the standard-dose group; the single pool Kt/V (spKt/V) was 1.71 ± 0.11 and 1.32 ± 0.09, respectively. The mean achieved β₂-microglobulin clearances were 3.4 ± 7.2 ml/min and 33.8 ± 11.4 ml/min in the low-flux and high-flux groups, respectively. Other than the study interventions, the dialysis and medical management of the patients conformed to current standards of care. Data collection ended in December 2001. The mean patient followup for mortality was 2.84 yr.

Definition of Clinical Outcomes

The primary study outcome was all-cause mortality. Three main secondary outcomes were evaluated: first cardiac hospitalization or all-cause death, first infection-related hospitalization or all-cause death, and first declining albumin event (> 15% decrease from baseline) or all-cause death. A fourth main secondary outcome was the rate of non–access-related hospitalizations. Other outcomes included cause-specific mortality and composites of death and first hospitalizations for cardiac disease or infection.

Determination of Clinical Outcomes

Each hospitalization of a study subject was reported by the clinical center, following review of the discharge summary, pertinent medical records, and diagnostic tests. For each death, the clinical center provided the death certificate, USRDS death notification form, and a brief narrative summary prepared by the local principal investigator. All infection-related hospitalizations or deaths were subclassified as “bacteremia or sepsis,” “deep tissue infection,” or a combination of both. In addition, each infectious outcome was categorized as being access-related or non–access-related. Finally, the clinical center assigned specific medical diagnosis codes from a predefined list for each hospitalization (up to four diagnoses were permitted).

An Outcome Review Committee, comprising investigators blinded to the patient’s randomization group, performed independent audits of all clinical center reports of deaths and first cardiac or infection-related hospitalization. All mortality reviews required independent audits of the clinical center’s death report by two members of the committee. Agreement on the category of the primary cause of death was required. When the reviewers disagreed, the cause of death was adjudicated during a conference call of the Outcome Review Committee.

All first infection-related or cardiac hospitalization reports were audited by one member of the Outcome Review Committee. This review process required agreement between the Clinical Center’s principal investigator and the Outcome Review Committee member as to the classification of the cause of hospitalization. When agreement could not be reached between the Outcome Review Committee member and the Clinical Center investigator, the case was adjudicated by the Outcome Review Committee. To evaluate the accuracy of the review process, the level of agreement between the diagnosis reported by the Clinical Center and the Outcome Committee was examined. When the clinical center determined that a hospitalization was due to infection, the Outcome Committee agreed 96% of the time. The Outcome Committee also concurred with the clinical center classification of the first hospitalization as “bacteremia or sepsis,” “deep tissue infection,” and “access-related hospitalization” in 92%, 95%, and 88% of cases. When the clinical center did not report an infectious etiology for the hospitalization, the Outcome Committee disagreed in only 4% of the time.

Subsequent infection-related and cardiac hospitalizations were not routinely audited by the Outcome Committee. To further assess the validity of the classification in those instances, a random subset of subsequent hospitalizations was adjudicated by the Outcome Committee. The sensitivity and specificity of the clinical center’s diagnosis of infectious hospitalization was 92% and 99%, respectively, when the Outcome Committee’s diagnosis was taken as the gold standard. On this basis, we concluded that the Clinical Center classification could be accepted with a high degree of confidence in those instances that were not audited by the Outcome Committee.

Each month, the clinical center completed a form for each study patient specifying what type of vascular access (fistula, graft, or catheter) was being used for dialysis. For the purpose of this analysis, we assumed that the type of access specified in the form preceding a death or hospitalization was the one in use at the time of the infectious event.

Statistical Analyses

The effects of the dose and flux interventions were investigated for two classes of infectious outcomes. The first class consisted of outcomes defined by a single event for each patient. These included (1) all-cause mortality, (2) the main secondary composite outcome including the first infection-related hospitalization or death from any cause, (3) infection-related death, (4) the first infection-related hospitalization, and (5) the composite outcome including the first infection-related hospitalization or infection-related death. The time from randomization to the second outcome can be interpreted as the duration of survival free of infection-related hospitalization, while the third, fourth, and fifth outcomes specifically address only infection-related events. Each of these outcomes was defined on the basis of the classifications of the HEMO Study Outcome committee, as described above.

The effects of the randomized interventions on each of these outcomes was evaluated using Cox regression (10), stratified by clinical center and controlling for the seven pre-specified baseline covariates: age, gender, race, years on dialysis, diabetes, Index of Coexisting Disease (ICED) score computed excluding diabetes (11,12⇓), and serum albumin. The interaction of baseline albumin with follow-up time was also included as a covariate to account for a reduction in the association of baseline albumin with mortality rate over time. Kidney transplants were censored for each outcome. In addition, the time of transfer to a nonparticipating dialysis unit or alternative dialysis modality was treated as a censoring event for the second, fourth, and fifth outcomes, which included hospitalization events; non–infection-related deaths were censored for the third and fifth outcomes, which included infection-related deaths; and all deaths were censored for the fourth outcome. The analyses with censored deaths must be viewed in a competing risk framework, in which the censored deaths are regarded as competing risk.

The second class of outcomes included multiple events for patients having more than one occurrence of the designated infection-related hospitalization. These outcomes were (1) all infection-related hospitalizations, (2) all infection-related hospitalizations that were access-related, (3) all infection-related hospitalizations that were non–access-related, (4) all infection-related hospitalizations presenting as deep-tissue infections, and (5) all infection-related hospitalizations presenting as bacteremia or sepsis. The effects of the randomized treatment interventions on each of these outcomes were evaluated using overdispersed Poisson regression analysis (13) while controlling for clinical center and the same seven baseline covariates described above, with deaths, transplants, and transfers censored.

Event rates for both classes of outcomes defined above were defined as the ratio of the total number of events to the total patient years of follow-up for that outcome, using the same rules for censoring as in the Cox and Poisson regressions described above.

The relationships of access type with infection-related death and with the composite outcome including the infection-related hospitalization or infection-related death were evaluated by Cox regressions relating the indicated outcomes to the access type designated on the monthly kinetic modeling form as a time-dependent covariate. These Cox regression models included the seven pre-specified baseline factors and randomized treatment group as covariates.

The relationship of the conditional probability that a patient’s infection-related hospitalization resulted in death to the randomized treatment groups and the seven pre-specified baseline factors was investigated using multiple logistic regression. Deaths occurring within 3 d of discharge were attributed to the infection-related hospitalization in this analysis.

All analyses of the effects of the interventions were intent-to-treat. Kaplan-Meier survival curves (14) were constructed for the composite outcome including first infection-related hospitalization or all-cause death. Statistical differences between survival curves were determined by the log rank test. All reported P-values are two-sided, without adjustment for multiple comparisons.