The Dialysis Outcomes and Practice Patterns Study (DOPPS): Design, Data Elements, and Methodology Ronald L. Pisoni, PhD, MS, Brenda W. Gillespie, PhD, David M. Dickinson, MS, Kenneth Chen, MS, Michael H. Kutner, PhD, and Robert A. Wolfe, PhD ● The Dialysis Outcomes and Practice Patterns Study (DOPPS) is a prospective, observational study designed to elucidate aspects of hemodialysis practice that are associated with the best outcomes for hemodialysis patients. In DOPPS I, 308 hemodialysis units from 7 countries participated, including 145 facilities from the United States (1996-2001), 62 facilities from Japan (1999-2001), and 101 facilities from France, Germany, Italy, Spain, and the United Kingdom (all 1998-2000). DOPPS II (2002-2004) has included 320 hemodialysis units and more than 12,400 hemodialysis patients from the 7 DOPPS I countries as well as Australia, Belgium, Canada, New Zealand, and Sweden. Dialysis units are chosen via a stratified random selection procedure to provide proportional sampling by region and type of facility within each country. In DOPPS I and II, longitudinal data have been collected from both a prevalent (cross-sectional) patient sample and an incident patient sample. Data have also been collected on numerous facility practice patterns. Most DOPPS analyses incorporate both facility- and patient-level data in regression-based analyses to investigate predictors of survival, hospitalization, quality of life, vascular access type, and other outcomes. DOPPS longitudinal data also help identify trends in subject characteristics, practice indicators, medication use, and outcomes. The DOPPS remains a unique source of data on hemodialysis patients and facilities. It continues to refine its methods of data collection and analysis with the goal of improving hemodialysis practice and end-stage renal disease patient lives worldwide. Am J Kidney Dis 44(S2):S7-S15. © 2004 by the National Kidney Foundation, Inc. INDEX WORDS: Dialysis Outcomes and Practice Patterns Study (DOPPS); hemodialysis outcomes; study design; comorbidities; mortality.
T
HE DIALYSIS OUTCOMES and Practice Patterns Study (DOPPS) is a prospective, observational study designed to elucidate aspects of hemodialysis (HD) practice that are associated with the best outcomes for hemodialysis patients. DOPPS is international in its scope, representing approximately 70% of the world’s hemodialysis patient population. This international participation in the DOPPS greatly enhances the study’s ability to investigate hemodialysis practices because of the large variation in approaches for hemodialysis therapy that are seen between countries. The DOPPS has produced a substantial body of clinically useful information, as documented by the broad range of hemodialysis practice areas discussed in over 40 DOPPS papers either published or currently in press. This article provides an overview of the unique DOPPS methodology and study design as an update to an earlier description of DOPPS methodology by Young et al.1 This information will be useful for understanding the design and structure of the DOPPS, which provides important evidence for developing both national and international best practice guidelines and helps improve hemodialysis patient care worldwide. DOPPS I AND II
There have been 2 phases of DOPPS data collection since study initiation in 1996; a third phase is
scheduled to begin in January 2005. In the first phase of DOPPS (DOPPS I), 308 hemodialysis units from 7 countries participated, including 145 facilities from the United States (1996-2001), 62 facilities from Japan (1999-2001), and 101 facilities from 5 European countries (France, Germany, Italy, Spain, and the United Kingdom, 1998-2000). The study design, sampling, and methodology used in DOPPS I have been well described before,1 and therefore this article focuses on DOPPS II, which has collected data from 2002 to 2004 using a
From the University Renal Research and Education Association, Ann Arbor, MI; Department of Biostatistics, School of Public Health, University of Michigan, Ann Arbor, MI; Amgen, Thousand Oaks, CA; and Department of Biostatistics, Rollins School of Public Health, Emory University, Atlanta, GA. The Dialysis Outcomes and Practice Patterns Study is supported by research grants from Amgen and Kirin without restrictions on publications. The NKF gratefully acknowledges the support of Amgen, founding and principal sponsor of K/DOQI. The publication of this supplement was supported by the DOPPS. Address reprint requests to Ronald L. Pisoni, PhD, MS, University Renal Research and Education Association, 315 W. Huron Street, Suite 260, Ann Arbor, MI 48103. E-mail:
[email protected] © 2004 by the National Kidney Foundation, Inc. 0272-6386/04/4405-0102$30.00/0 doi:10.1053/j.ajkd.2004.08.005
American Journal of Kidney Diseases, Vol 44, No 5, Suppl 2 (November), 2004: pp S7-S15
S7
S8
PISONI ET AL Table 1. DOPPS II Data Collection Instruments Questionnaire
Data Collection Schedule
Cumulative Hemodialysis Census
Completed at unit’s entry into DOPPS; maintained throughout study participation.
Medical Questionnaire
Collected at time of study entry for randomly selected study sample patients.
Interval Summary
Collected at 4-month intervals once a randomly selected patient has entered DOPPS.
Patient Questionnaire
Completed by randomly selected study patients at study entry and annually thereafter.
Termination Form
Completed for study sample patients who depart from the study.
Unit Practices Survey and Medical Directors Survey Vascular Access Surgery Questionnaire
Completed shortly after unit’s entry into DOPPS and annually thereafter.
Completed by surgeon or physician having major involvement in creating vascular accesses for hemodialysis patients in the dialysis unit participating in DOPPS.
Research Items
Complete census of all chronic maintenance hemodialysis patients treated in the dialysis unit during the study. Collects data regarding mortality, demographics, diabetes as cause of ESRD, and reasons for patient entering into and departing from facility, including transfer to/from peritoneal dialysis or kidney transplantation. Baseline medical information. Captures demographics, date of starting ESRD, primary cause of ESRD, ESRD modality history, medical insurance coverage, socioeconomic and psychosocial status, placement on kidney transplant waiting list, selected aspects of the family history of kidney disease, smoking status, and medical history prior to study enrollment, including 72 comorbidity-related factors, height, weight, recent laboratory data prior to study enrollment, hemodialysis prescription, delivered dialysis dose, residual renal function, selected measures of pre-ESRD care, vascular access use, nutrition therapy, and medication use. Longitudinal data, including most recent laboratory data, hemodialysis prescription, delivered dialysis dose, causes and dates for any hospitalizations or outpatient events occurring during prior 4-month period, any vascular access-related events during prior 4-month interval, medication use, and placement on kidney transplant waiting list. Assessment of kidney disease quality of life, measures of self-reported depression and patient satisfaction, and other aspects of medical care. Indicates reason for study departure. If patient has died, date and cause(s) of death. For kidney transplant patients, date and whether cadaveric or living related transplant. Detailed facility practice information. UPS completed by nurse manager (or designee); MDS completed by unit’s medical director. Data regarding surgeon’s training and experience in vascular access placement; perceptions, preferences, and types of surgical/monitoring techniques used in vascular access surgical practice.
Abbreviations: ESRD, end-stage renal disease; UPS, Unit Practices Survey; MDS, Medical Directors Survey.
DOPPS METHODS
S9
Fig 1. DOPPS framework for hypothesis.
protocol modified from that used in DOPPS I. DOPPS II has included 320 hemodialysis units and more than 12,400 hemodialysis patients from the 7 DOPPS I countries as well as Australia, Belgium, Canada, New Zealand, and Sweden. In both DOPPS I and II, the study design is structured around 5 important elements: (1) within each country, random selection of dialysis units stratified by type of facility and geographic region, with facility sampling proportional to size within each stratum; (2) collection of demographic data, diabetes as cause of end-stage renal disease, and mortality data for all chronic maintenance HD patients in each study dialysis unit (cumulative hemodialysis census); (3) collection of additional detailed patient data from a random selection of 20 to 40 patients within each dialysis unit at study entry (medical questionnaire) and at 4-month intervals throughout the study (interval summary); (4) collection of kidney disease quality of life information and other data regarding a patient’s medical care as indicated in a questionnaire completed by the patient at study entry and annually thereafter (patient questionnaire); and (5) collection of detailed facility practice information, assessed from patient data and from questionnaires completed annually by the dialysis unit’s medical director (medical directors survey) and by the unit’s nurse manager or designee (unit practices survey). These elements form the main study structure on which DOPPS data analyses are based and form the basis of the study instruments shown in Table 1. These data allow DOPPS to examine a wide spectrum of relationships between many types of outcomes (eg, mortality, hospitalization, patient quality of life, vascular access, and so on) and patient- or facility-level characteristics, while
at the same time allowing for numerous adjustments for potentially confounding factors (Fig 1). DOPPS II FACILITY SAMPLING
Table 2 lists the number of dialysis units in each country and the total number of patients at these facilities. It also gives the number of geographic regions in each country and the types of facilities (eg, hospital versus free standing) used for stratification. The location of DOPPS II dialysis units is shown in Fig 2, showing the wide geographic distribution of DOPPS sites across the 12 participating countries. One inclusion criterion incorporated into the DOPPS study design is that a dialysis unit must treat ⱖ25 hemodialysis patients within the unit to be eligible for study participation (ⱖ20 patients at last registry reporting was used for contacting facilities). This minimum facility size was chosen to provide a sufficient patient sample size to obtain accurate estimates of facility practices and outcomes. It is estimated that in most DOPPS II countries, this study inclusion criterion resulted in exclusion of ⱕ5% of all in-center hemodialysis patients. In most countries, ⬎95% of hemodialysis patients receive hemodialysis therapy within a dialysis unit; however, in some countries such as Australia and New Zealand, 15% to 29% of hemodialysis patients receive home hemodialysis therapy, and these patients are not eligible for participation in DOPPS. Therefore, the DOPPS study design represents the great majority of hemodialysis patients within the 12 countries participating in DOPPS II; however, outcomes for some hemodialysis patients (eg, home hemodialysis patients) are not reflected in the results from the study. Patients treated by peritoneal dialysis are not included in the DOPPS.
S10
PISONI ET AL Table 2. Summary of Hemodialysis Units in Sampling Frame, by DOPPS Country
Country
Total Facilities
Total HD Patients
Total Geographic Regions
Australia/ New Zealand
Australia: 167* New Zealand: 13*
Belgium
50 units (n ⱖ 25 patients)
Canada
189 units
Australia: 4,635* New Zealand: 648* (year 2000) 3,081 (in units, n ⱖ 25 patients) 10,059 (year 2000)
France
316
20,106
Germany
1,065
54,385 (year 2001)
Italy Japan
897† 3,358‡ (year 2000)
32,281† 197,278 (year 2000)
Spain
341 (year 2001)
15,508 (year 2001)
8
Sweden
62
2,332
5
United Kingdom
147
United States
3,886 (year 2000)
13,107§ (year 2002) 244,950储 (year 2000)
Facility Types
5
Center-hospital, satellite
2
Center-hospital, satellite
8
Hospital, satellite (not self-care), satellite (self-care) General, private, university, association Clinics (medical centers), nonprofit free standing, private practice Public, private Private clinic, private hospital, prefecture or city hospital, other hospital, university Center (hospital) units satellite (clinic) units Hospital, satellite, and university Center, Satellite
22 9
19 47
11 10
Hospital, for-profit free standing, not-for-profit free standing, whether unit was created since 1995
*12/31/00 ANZDATA Registry 2001 Report. †Data obtained directly from International Federation of Renal Registries (data for 1999). ‡12/31/00, JSDT Registry Report, Nakai et al, J Japanese Soc Dialysis Therapy 35:1155–1184, 2002. §2003, Renal services for dialysis—a position paper by the National Kidney Research Fund, Peters J in association with the Project Advisory Group, http://www.nkrf.org.uk. 储2003, US Renal Data System, USRDS 2003 Annual Data Report: Atlas of End-Stage Renal Disease in the United States, National Institutes of Health, National Institute of Diabetes and Digestive and Kidney Diseases, Bethesda, MD, 2003.
DOPPS II: PREVALENT AND INCIDENT HD PATIENT SAMPLES
In DOPPS II, detailed patient data have been collected from 2 different hemodialysis patient samples: (1) a prevalent (cross-sectional) patient sample and (2) an incident patient sample collected over time. The prevalent patient sample was obtained by randomly selecting 20 to 40 patients from among all patients receiving longterm, maintenance hemodialysis therapy within each selected dialysis unit at the time the dialysis unit initiated DOPPS II data collection. A greater number of sample patients (up to 40) was se-
lected from larger dialysis units within each country as defined by a target size algorithm. In DOPPS II, to reduce the data collection burden, sampled patients departing from the study were not replaced, whereas in DOPPS I, patients departing from the study typically were replaced every 4 months by randomly selecting from patients who had entered the dialysis unit since the time of the prior random selection. Enrollment of the incident patient sample was begun at each site once the facility had completed 6 weeks of study participation. Up to 15 patients who had initiated chronic hemodialysis
DOPPS METHODS
S11
Fig 2.
DOPPS II target samples.
within the past 30 days were enrolled sequentially. This sampling has resulted in data collected from ⬎3,300 incident patients in DOPPS II, with the majority of these patients entering DOPPS within 2 days of their first-ever chronic dialysis treatment. These DOPPS II incident patients, along with 5,354 who entered in DOPPS I within 30 days of starting hemodialysis, allow for detailed investigations of the practices, characteristics, and outcomes for patients from the time of initiating hemodialysis therapy. DOPPS II PATIENT- AND FACILITY-LEVEL DATA
The data collected in DOPPS II provide detailed information on sampled patients, including demographics, more than 70 measures of comorbidity, socioeconomic status, insurance coverage, laboratory values, medication use, hospitalization and outpatient events, vascular access use and vascular access procedures, dialysis prescription, delivered dialysis dose, residual renal function, nutritional measures, aspects of medical care before end-stage renal disease, kidney trans-
plant wait-listing, patient quality of life, physician-diagnosed and patient reported measures of depression, date and cause(s) of death for patients who die during the study, and transfer to or from peritoneal dialysis or kidney transplantation. By achieving high consent rates in recruitment of patients into the study, this sample provides patient data representative of the practices and patient mix of each participating dialysis unit. Furthermore, the use of a representative sample of dialysis units in a country and a random sample of patients from each study dialysis unit allow for nationally representative results to be described for in-center hemodialysis therapy. In addition to the collection of detailed patient data, a major focus in DOPPS is the collection of information on many different aspects of facilitylevel hemodialysis practices. Table 3 highlights some of the major facility practice areas that are closely examined in DOPPS investigations.Anewly developed vascular access surgery questionnaire added a significant dimension to the study of vascular access in DOPPS II. This questionnaire gathers surgical practice data directly from the surgeons
S12
PISONI ET AL Table 3. Facility-Level Data Collection Items
Anemia and iron therapy Antihypertensive therapy Continuing education policies/practice Dialysate processing and composition Dialysis dose Dialysis machines Dialysis practices Dialyzer re-use Dialyzers Dietitian and nutrition practices
Facility characteristics Facility staffing practices Health care maintenance Hospital and outpatient practices Immunizations Information systems Initiation and discontinuation of dialysis Insurance policies Laboratory testing Local dialysis market
Mineral metabolism Nurse and technician practices Patient turnover Physician practices Pre-ESRD practices Quality assurance and improvement practices Scheduling practices Social service practices Vascular access Water treatment and surveillance
Abbreviation: ESRD, end-stage renal disease.
and other physicians creating vascular accesses at the DOPPS dialysis units regarding a wide range of surgical practices, perceptions, and preferences for placement of different types of vascular accesses for HD patients. DOPPS STATISTICAL ANALYSIS: METHODOLOGICAL APPROACHES
Types of Analyses To describe characteristics and relationships associated with hemodialysis practices and outcomes across the 12 countries in DOPPS, many different statistical analytical methods are needed. Table 4 indicates 3 general types of questions that can be addressed using DOPPS data and provides examples of the patient sample and analytical techniques that have been used for investigating each. For example, a simple descriptive characterization of the hemodialysis patient population (eg, mean age) or a hemodialysis practice (eg, percentage of patients who are female or diabetic, or percentage of patients outside the Kidney Disease Outcomes Quality Initiative guidelines for Kt/V or serum phosphorus)
can be performed using a prevalent crosssectional patient sample. Population Estimates The fraction of patients sampled varies widely across facilities, and survey sampling weights are used to compute descriptive statistics intended to represent an entire country or region. Facilities are also weighted by their stratum sampling weights to be representative of the total number of facilities in their stratum (region and facility type). For facility-level descriptions, facility weights alone are used; for patient level descriptions, both facility and patient weights are used. As SAS procedures (SAS Institute, Inc., Cary, NC) that handle survey weights, such as the survey means procedure, have become available, they have typically been used to calculate population estimates and standard errors. Multivariable Analyses, Time-Dependent Modeling, and Facility Clustering Techniques DOPPS analyses are performed at both the facility and patient level. Most analyses incorpo-
Table 4. Examples of Methods Type of Question
Sample Used
Analytic Technique
Characterization (eg, prevalence of patient conditions or facility practices, patient characteristics) Cross-sectional associations (no temporality) Prospective associations (baseline conditions predict future event)
Prevalent cross-sectional sample (initial sample) or incidence sample
Descriptive statistics (mean, variance, proportion)
Cross-section of enrolled patients (prevalent or incident) Initial sample ⫹ replacement patients with follow-up
Linear, logistic, other regression models Cox regression (time to event with censoring); repeated measures models
DOPPS METHODS
rate aspects of both via regression-based analyses that simultaneously account for differences among patients and facilities. Associations between outcomes (eg, clinical, quality of life, and economic outcomes) and practice patterns are analyzed using Poisson, logistic, proportional hazards, log-linear, or linear regression models, as appropriate, that control for patient demographic and comorbidity differences. Mortality or time-to-event models are used to investigate the prognostic importance of baseline covariates, such as hemoglobin concentration at the time of study entry on the patient’s subsequent mortality risk. To investigate a patient’s laboratory values over time2-4 and relate these values to a desired outcome, time-dependent Cox survival models are applied to the longitudinal data typically collected every 4 months. DOPPS investigations have shown a substantially stronger relationship between a patient’s hemoglobin concentration and the risk of death when time-dependent models are used compared with models using only the patient’s study entry hemoglobin concentration. The time-dependent approach is more useful for considering mechanisms, whereas the baseline values approach is more useful for assessing prognosis. DOPPS longitudinal data also are useful for assessing longitudinal trends in patient characteristics, practice indicators, medication use, and outcomes. For these analyses, repeated measures models are used to account for serial measures over successive years or time intervals. The data collected in DOPPS II are often used to corroborate observed relationships with data collected in DOPPS I and vice versa. All analyses are carried out using SAS software (SAS Institute, Inc). Primary analyses of practice patterns are based on facility-level summaries of practices (eg, the percentage of facility patients using a catheter in a prevalent cross-sectional sample). Such facilitylevel measures assign all patients at a facility the same value, which avoids treatment by indication bias in some situations. For example, patientlevel results would be biased if poor prognosis were the “indication” for using a catheter. One issue requiring statistical attention is that patients within a single facility are likely to have more similar outcomes than patients from different facilities. Patient-level analyses account for the effects of facility-level clustering by using
S13
either random effects models (for linear or generalized linear models)5 or robust standard error estimates based on the sandwich estimator (for generalized linear models and for proportional hazards models).6 Standard Confounding Factors Controlled for in DOPPS Analyses Many DOPPS analyses control for patient age, race, gender, and 14 summary comorbid conditions. These 14 conditions represent an aggregation of 47 questions describing various aspects of the patients’ medical histories. Table 5 shows the questions that contribute to each standard summary comorbid condition typically used in most DOPPS analyses; these groupings were chosen based on clinical judgment. In addition to the 47 factors aggregated into the 14 summary comorbid conditions, data also are collected on 25 other aspects of patient comorbidity. Summary measures on these other 25 questions were not found to be predictive of patient survival independently of the current list of 14 summary factors. Missing Data Missing data are a relatively small but pervasive problem in the DOPPS data. Investigators have begun to implement data analyses incorporating multiple imputation. When used correctly, this technique produces estimates that are consistent, asymptotically efficient, and asymptotically normal when the data are missing at random.7 Data Validation Validating the integrity of data collected in DOPPS is an important activity occurring throughout the study. The first step in DOPPS data validation is to check the raw data files using quality control programs (eg, meet range checks, units of measure, and evaluation of nonresponse). Queries generated from these database quality checks are brought to the attention of the primary data collectors at dialysis units to verify or correct specific data elements and obtain missing data. Additionally, DOPPS findings have been validated by comparison of results with reports for registries or other representative data sources. A data validation study performed in DOPPS I showed a high level of quality regarding data
S14
PISONI ET AL
Table 5. Components of Standard Summary Comorbid Conditions Often Used in DOPPS Multivariable Analyses Summary Comorbid Condition
Component Questions*
Coronary heart disease
Prior diagnosis of CHD/CAD History of angina (at any time) Angina associated with exertion or dialysis within 12 months of enrollment Angina at rest within 12 months of enrollment Myocardial infarction ever Myocardial infarction within 3 months of enrollment Coronary bypass surgery ever Coronary angioplasty ever Cancer other than skin History of cancer other than skin cancer Other cardiovascular disease Cardiac arrest ever Atrial fibrillation Other arrhythmias Permanent pacemaker implanted Pericarditis Valvular heart disease by echocardiogram or heart catheterization Prosthetic heart valve (s/p valve replacement) Cerebrovascular disease Cerebrovascular accident (stroke) without major residual neurologic deficit Cerebrovascular accident (stroke) with major residual neurologic deficit Carotid endarterectomy Transient ischemic attacks Congestive heart failure Congestive heart failure Congestive heart failure requiring hospitalization within 12 months before enrollment Pulmonary edema Diabetes Diagnosis of diabetes (even if not primary cause of ESRD) Insulin therapy (any time before enrollment) Diabetes pills (any time before enrollment) Diabetic gastroparesis Gastrointestinal bleeding Gastrointestinal bleed within 12 months before enrollment HIV/AIDS HIV status AIDS diagnosis Hypertension Diagnosis of hypertension Lung disease Chronic obstructive pulmonary disorder Use of home oxygen Neurologic disease Seizure disorder Dementia Peripheral neuropathy (diabetic or other) Psychiatric disorder Depression within 12 months before enrollment Substance abuse within 12 months before enrollment Alcohol abuse within 12 months before enrollment Other psychiatric disorder Peripheral vascular disease Prior diagnosis of peripheral vascular disease Aortic aneurysm Arterial bypass surgery for peripheral vascular disease or surgical repair of aortic aneurysm Claudication (pain in extremities with exertion due to peripheral vascular disease) Rest pain of extremities because of peripheral vascular disease Amputation because of peripheral vascular disease Recurrent cellulitis, gangrene Recurrent cellulitis, skin infection, gangrene Abbreviations: CHD, coronary heart disease; CAD, coronary artery disease; ESRD, end-stage renal disease; HIV, human immunodeficiency virus; AIDS, acquired immunodeficiency syndrome. *Answering “yes” or “suspected” to any component results in the patient being coded as having the summary comorbid condition. Comorbid conditions are recorded at patient entry into the DOPPS sample.
collected by DOPPS. In this study, an external data collector reabstracted patient data from primary medical records regarding a specific set of
data elements. This study revealed ⬎90% identity between the original and reabstracted data (unpublished data).
DOPPS METHODS
S15
CONCLUSION
The design of the DOPPS has been critical to its success as a major international study. The DOPPS provides data from a representative sample of dialysis facilities in 12 countries with follow-up over several years. A wide variety of data elements are collected, incorporating both patient-specific and facility-specific characteristics. In addition, DOPPS data allow analyses to be performed for both prevalent and incident hemodialysis patient samples. The variety of facility practices shown in these diverse samples allows broad exploration of predictors of dialysis treatment and outcomes. Statistical adjustments for confounding or imbalance are required to isolate factors of interest and make “all else equal.” Because causality can only be inferred—not confirmed—from observational studies, thoughtful adjustments must be implemented. Additionally, it is crucial to evaluate data quality and appropriately address missing data for descriptive analyses. As indicated by the accompanying articles in this supplement, this large international study has provided numerous important findings that are relevant for practicing clinicians worldwide. ACKNOWLEDGMENT The authors gratefully acknowledge the dedication and hard work of the DOPPS Coordinating Center staff:
Jon Bodfish and Rameswari Metla in maintaining the DOPPS database; and the multitude of activities of Theresa Helm, Patrick Carlson, Thuy Bui, and Enrico Sassi in working with the sites participating in the DOPPS. We further acknowledge our gratitude for the great deal of dedication from the staff members and medical directors of over 300 participating dialysis units in 12 countries.
REFERENCES 1. Young EW, Goodkin DA, Mapes DL, et al: The Dialysis Outcomes and Practice Patterns Study (DOPPS): An international hemodialysis study. Kidney Int Suppl 74:S74S81, 2000 2. Lacson E Jr, Ofsthun N, Lazarus JM: Effect of variability in anemia management on hemoglobin outcomes in ESRD. Am J Kidney Dis 41:111-124, 2003 3. Berns JS, Elzein H, Lynn RI, et al: Hemoglobin variability in epoetin-treated hemodialysis patients. Kidney Int 64:1514-1521, 2003 4. Pisoni RL, Bragg-Gresham JL, Young EW, et al: Anemia management and outcomes from 12 countries in the Dialysis Outcomes and Practice Patterns Study (DOPPS). Am J Kidney Dis 44:94-111, 2004 5. SAS Institute, Inc. SAS/STAT User’s Guide. Version 8 (vol 2). Cary, NC, SAS Institute, 2000 6. Klein J, Moeschberger M: Survival Analysis Techniques for Censored and Truncated Data. New York, NY, Springer, 1997, pp 416-418 7. Allison PD: Missing Data. (Sage University Papers Series on Quantitative Applications in the Social Sciences, series no. 07-136). Thousand Oaks, CA, Sage, 2001, p 27
