Millayh CHAPTER. Edna St. Vincent

Transcription

1 Where you used to be, there is a hole in the world, which I find myself constantly walking around in the daytime, and falling into night. Edna St. Vincent Millayh Letters D mortality 8 Introduction Overall hospitalizion D mortality {admissions by diagnosis, vintage, & modality} {survival} {mortality res} Cause-speciIc hospitalizion {admission res for cardiovascular disease, infection, cardiovascualr proceures, & vascular access insertions} Cause-speciIc mortality {mortality res, by vintage, for cardiovascular & other events} {relive of mortality after cardiovascular & other events} 8 Major infections D cardiovascular events {event res & relive s for sepsis, pneumonia, & cardiovascular events} Hospitalizion D mortality, by modality {admissions for principal procedures & diagnoses} {five-year survival} {cause-specific mortality} Complicions D pregnancy {hospitalizion & trement res for cancer} {pregnancy, birth, & complicion res} Fractures D mortality {fracture res by vintage & demographic characteristics} {relive of mortality after fractures} Standardized mortality rios Standardized hopitalizion rios 8 Summary Outcomes: Hopitalizion CHAPTER

2 hapter Six has been expanded this year, as we upde da on hospitalizion, mortality, and survival, and present new informion on the relion between fractures and mortality, on cancers in women treed with dialysis, and on pregnancy res and complicions. Figure. shows th all-cause hospitalizion res for prevalent dialysis pients fell cent between 99 and. Admissions for vascular access declined. cent, but those reled to cardiovascular disease and infections grew 7. and. cent, respectively. Cardiovascular admission res in children increased. cent over the iod, a particularly striking increase. Hospitalizion res by dialysis pient vintage continue to fall for pients with less than two on the therapy, and to increase for those of longer vintage. Compared to those of hemodialysis pients, res are higher in pients on itoneal dialysis, particularly for infections. This difference appears reled to time on the therapy; while res are similar for the two modalities in the first two, they increase for itoneal dialysis pients as vintage increases. In further analyses we will investige how pients who fail on hemodialysis and move on to itoneal therapy may influence these res. Res of mortality due to major cardiovascular and infectious events have improved in both the dialysis and transplant populions, though with notable exceptions. Mortality due to malignancy, for example, has risen over the past decade, as has mortality due to a cardiac arrest or septic event. (Outcomes after cardiac events are addressed further in Percent change in admissions pient, 99 to - Age Gender Race Primary diagnosis -cause Cardiovascular Infection Vascular access Male Female White Black N Am Asian DM HTN GN CK {.} Percent change in adjusted hospitalizion res for prevalent dialysis pients, 99, by demographic characteristics & primary diagnosis iod prevalent dialysis pients; res for all pients are adjusted for age, gender, race, & primary diagnosis; res by age are adjusted for gender, race, & primary diagnosis; res by gender are adjusted for age, race, & primary diagnosis; res by race are adjusted for age, gender, & primary diagnosis; res by primary diagnosis are adjusted for age, gender & race. Direct comparison of adjusted res is approprie only within each graph, not between graphs. Dialysis pients,, used as reference cohort. Introducion 8 Outcomes: hospitalizion & mortalityh

3 Chapter Nine.) Considerable progress has been achieved in the trement of ischemic heart disease in the general populion; there has been little investigion, however, into its diagnosis and trement in ESRD pients, including the use of electrophysiologic studies, pacemakers, and defibrillors. The character of infectious hospitalizions in the dialysis populions has changed over time, with hospitalizions for pulmonary infections now twice as likely in hemodialysis pients as in those on itoneal dialysis. The re of admission for itonitis in itoneal dialysis pients has declined, while in hemodialysis pients hospitalizions for vascular access infections have doubled in the past ten. The competing effects of growing infection res and better identificion and trement of cardiovascular disease appear to be associed with net improvement in five-year survival. Analyses by pient vintage, however, show th time on a therapy continues to have a growing associion with mortality, as mortality res decline for pients of younger vintage, and rise for those on a modality five or more. More detailed assessments by cause of deh and vintage are needed to determine whether a shift has occurred th may help explain these changing pterns. We give expanded tention here to infectious complicions and mortality. Particularly striking are da showing th the mortality re six months after a septicemia diagnosis is seven times higher than in the reference populion, and remains. times higher after four. While sepsis events may identify pients for early deh, an alternive hypothesis reles these events to an increasing inflammory burden and cardiovascular disease. We introduce da this year on newly identified cancers in women receiving dialysis therapy, and on pregnancy res and complicions. Res of pregnancies in dialysis pients have remained relively stable since 99, while those in transplant pients have fallen dramically. Reasons for this are unknown, but may be explained in part by women choosing to delay pregnancy until more than three after their transplants; after this point, we can no longer track these pients. Live birth res are significantly lower in dialysis pients, though babies born to these pients have noticeably lower res of fetal distress and poor fetal growth. Further studies are needed to investige potential explanions for changes in these res and for the often dramic differences between the modalities. Mortality comparisons on a provider level have long been the subject of review by policy makers and health plans. The Standardized Mortality Rio (SMR), the primary method used in these comparisons, has been modified to incorpore adjustments for age, gender, race, primary diagnosis, time on dialysis, and comorbidity initiion, but the SMR s year-to-year variability, which could reflect either real changes or random variions, has been a major concern. The USRDS Coordining Center recently published a new Bayesian adjusted mortality re methodology, addressing more completely the inherent variability of small versus large providers to determine which units are true outliers, and which are within the normal degree of variion. To illustre differences between the traditional and new methods we have included two new spreads in this chapter, presenting da by provider and census division, and looking mortality and hospitalizion rios; transplantion rios are presented in Chapter Seven. We provide informion here, for example, on how frequently each of the two methods would classify a unit in the same group, demonstring the difficulties in simply interpreting a rio as good or suboptimal. The two methods generally provide similar results, but we believe the Bayesian method is suior for identifying true outliers, and provides more confidence in assessing providers. This will now be the sole method used by the USRDS to report these rios, and we will continue to examine alternive methods of assessing outcomes {.} Adjusted mortality res in the provider level. Areas of disagreement should be viewed with cau- pients continue to prevalent dialysis tion, as it may be difficult, fall for those with less than five on the particularly in smaller providers, to distinguish a increased in pients with a vintage therapy, but since 99 have real finding from of five or greer. {. &.} Admissions for cardiac arrest have random variion. those for bacteremia/septicemia. {. } increased in the dialysis populion, as have Pregnancy res for women with ESRD, as well as antepartum and post-partum complicion res, show differential pterns between dialysis and transplantion. {.7 8} In calculions of mortality and hospitalizion rios, the Bayesian method, developed by the USRDS, takes into account the inherent variability of small providers and produces more stable res. Chapter highlights H USRDS Annual Da Report 9

4 {.} Adjusted hospital admissions, by primary diagnosis & modality: prevalent pients... Admissions pient year Diabetes Hytension Glomerulonephritis {.} Adjusted hospital admissions, by vintage & modality: prevalent pients.... year.... pient Admissions < -< ospital admissions by primary diagnosis have remained relively steady since 99 (Figure.). Pients whose ESRD is caused by diabetes are admitted most frequently. times year in while res are lowest for pients with glomerulonephritis. pients have the lowest admission res. pients with diabetes are admitted slightly more often than their hemodialysis counterparts. By vintage a pient s time with ESRD admission res in the hemodialysis populion are highest for pients who have had ESRD less than two, and have been relively steady in this populion since the early 99s (Figure.). Res for pients with five or more on the therapy, in contrast, have been rising, from.8 admissions pient year in 99 to. in. This growth of nearly 9 cent has occurred in itoneal dialysis pients of older vintage as well, as res for younger vintage pients on the modality have fallen slightly. Compared to those of the previous iod, five-year survival probabilities for the populion incident in increased across modalities cent overall, and cent for pients on itoneal dialysis (Figures. ). By primary diagnosis, the greest improvement has occurred in the diabetic populions, with probabilities rising and 8 cent for hemodialysis and itoneal dialysis, respectively. Pients with diabetes continue, however, to have the lowest probabilities of survival, with only 7 cent of those on hemodialysis, and cent of itoneal dialysis pients, expected to survive five after initiion. We show in Chapter Three th incident pients carry a greer degree of comorbidity than ever before. Despite this, however, their survival res across modalities and primary diagnoses are improving. This improvement may be even greer if disease burden is taken more directly into account. Among prevalent dialysis pients, overall mortality res have fallen cent since their 988 peak, in reaching 8 dehs, pient (Figure.). This slight fall, however, masks significant differences by pient vintage. Since 98, res for pients on dialysis less than two have fallen cent. Res for pients on the modality five or more, in contrast, have increased cent since their lowest point in 99, to a level of 8 dehs, pient. These da show the need for more tention to conditions th develop over time, such as lipid disorders and diabetic, cardiovascular, and infectious complicions. The poor long-term survival of ESRD pients is illustred by comparisons to the general U.S. populion (Table.a). The expected remaining lifetimes of white dialysis pients are only onefourth to one-sixth those of the general populion; differences are particularly high among women aged. And despite slightly higher survival res, expected remaining lifetimes of black dialysis pients are only one-third to one-fifth those of the general populion. By modality, expected lifetimes for transplant pients are double those of dialysis pients among black males, and more than three times as high for white females ages and older. Expected remaining lifetimes in transplant pients are still, however, only 9 cent as long as those in the general populion. {Figure.} iod prevalent ESRD pients; adjusted for age, gender, & race. ESRD pients,, used as reference cohort. {Figure.} iod prevalent ESRD pients; adjusted for age, gender, race, & primary diagnosis. ESRD pients,, Overall hopitalizion D mortality Outcomes: hospitalizion & mortalityh

5 {.} Adjusted survival: incident pients probability Survival pients, by modality Dialysis HD PD HD pts, by primary diagnosis Diabetes HTN GN PD pts, by primary diagnosis DM HTN GN Months after initiion used as reference cohort. {Figures. } incident dialysis pients & pients receiving a first transplant in the calendar year. probabilities are adjusted for age, gender, & race; overall probabilities are also adjusted for primary diagnosis. ESRD pients, 99, used as reference cohort. Modality determined on first ESRD service de; excludes pients transplanted or dying during the first 9 days. {Figure.} iod prevalent dialysis pients; res adjusted for age, gender, race, & primary diagnosis. Dialysis pients,, used as reference cohort. {Table.a} U.S. da: from Table A in the United Stes life tables (Arias E). Available da provided only for whites & blacks. ESRD da: prevalent dialysis & transplant pients,. {.} Adjusted survival: incident pients {.} Adj. mortality res, by vintage: dialysis pts. pients, by modality HD pts, by primary diagnosis PD pts, by primary diagnosis probability Survival.8... Dialysis HD PD Diabetes HTN GN Diabetes HTN GN Dehs, pient 7 + -< < Months after initiion {.a} Expected remaining lifetimes () of the general U.S. populion, & of prevalent dialysis & transplant pients, by race & gender General U.S. populion, races White Black Age M F M F M F ESRD: Dialysis, races White Black Nive American Asian Age M F M F M F M F M F , H USRDS Annual Da Report

6 Adjusted cause-specific hospital admissions: prevalent pients {.7} by age year pient Admissions..... {.8} by gender year pient Admissions hospitalizions Infection (other than int. device) hospitalizions Infection (other than int. device) {.9} by race/ethnicity year pient Admissions hospitalizions Infection (other than int. device) White Black Cardiovascular disease (9-) - - Infection due to internal device 99 Cardiovascular disease (9-) Male Female Infection due to internal device 99 Cardiovascular disease (9-) Infection due to internal device Nive American Asian Hispanic 99 ialysis pients age 7 and older have the highest hospital admission res overall, for cardiovascular disease, and for infections not due to internal devices (Figure.7). Admissions for infections th are due to internal devices, however, are most common in pients age. Hospital admission res continue to be higher in female dialysis pients than in males cent higher for all hospitalizions, and 9 cent higher for admissions reled to cardiovascular disease (Figure.8). Overall res for admissions which include a cardiovascular procedure are also greer in women (Figure.). Res for admissions including a stent/angioplasty or a bypass procedure, however, are and cent higher, respectively, in men than in women. And while women have greer admission res for inpient vascular access insertions overall and for cheters and grafts, admission res for fistula creion were cent higher in men (Figure.). While overall admission res by race and ethnicity differ little for most pient groups, they are consistently lowest for dialysis pients of Asian descent in,. admissions pient year, compared to.9. in Hispanic pients and those of other races (Figure.9). Admissions for cardiovascular disease in the Asian populion have, however, been rising slowly, and in were cent higher than in 99. Overall res for admissions with a cardiovascular procedure are greest in black dialysis pients, and lowest in those of other races (Figure.). Black pients also have, however, the lowest res of admissions with a bypass procedure, and among the lowest res of admissions with a stent or angioplasty, while the highest res consistently occur in white pients. Since 99, the re of admissions including a stent or angioplasty has grown 8 cent for white pients, and 89 cent for blacks. As more vascular access procedures are done on an outpient bases, res of inpient vascular access insertions overall, and of cheter and graft placements, have decreased for most pient groups (Figures. and.). Guidelines of the NKF s Kidney Disease Outcomes Quality Initiive (K/DOQI) advoce the increased use of fistulas; admissions for fistula creion have been relively steady over time, decreasing slightly in whites and in blacks. { figures} dialysis pients,, used as reference cohort. Res by age are adjusted for gender, race, & primary diagnosis; res by gender are adjusted for age, race, & primary diagnosis; & res by race/ethnicity are adjusted for age, gender, & primary diagnosis. CMS began collecting Hispanic ethnicity da on the Medical Evidence form in April 99. Because the model-based adjustment method uses da from the current & previous two, res for Hispanic pients are shown only for 998 & ler. {Figures.7 9} iod prevalent dialysis pients age & older. At the end of 998 a new ICD-9-CM code was added for infections due to internal devices in itoneal dialysis pients; da prior to this de are omitted. {Figures. } iod prevalent dialysis pients age & older. Res reflect all admissions with a cardiovascular procedure (excluding vascular access procedures), not just those for the purpose of a cardiovascular procedure. Cegories are not mutually exclusive, so a hospitalizion th includes more than one type of cardiovascular procedure will be counted under each cegory. {Figures. } iod prevalent hemodialysis pients age & older. Part B physician/supplier claims for vascular access insertions in an inpient setting. Cause-pecific hopitalizion Outcomes: hospitalizion & mortalityh

7 Adjusted admissions with cardiovascular procedures: prevalent pients {.} by age Adjusted inpient VA insertions: prevalent pients {.} by age cardiovascular procedures Bypass 7 vascular access insertions Cheters Admissions, pient Stent/angioplasty Valve procedures Insertions, pient Fistulas Grafts {.} by gender {.} by gender cardiovascular procedures Bypass 7 vascular access insertions Cheters Admissions, pient Stent/angioplasty Male Female 7 Valve procedures Insertions, pient Fistulas Male Female Grafts {.} by race/ethnicity {.} by race/ethnicity cardiovascular procedures Bypass 7 vascular access insertions Cheters Admissions, pient Stent/angioplasty White Black Hispanic Valve procedures Insertions, pient Fistulas White Black Hispanic Grafts H USRDS Annual Da Report

8 {.} Adjusted mortality, by vintage: AMI {.7} Adjusted mortality, by vintage: ASHD {.8} Adj. mortality, by vintage: cardiomyophy, pient Dehs Dialysis Dialysis Dialysis < < , pient Dehs , pient Dehs. < {.9} Adj. mortality, by vintage: card. arrhythmia {.} Adj. mortality, by vintage: cardiac arrest {.} Adj. mortality, by vintage: cerebrovascular 8 Dialysis Dialysis Dialysis Dehs, pient 8.. < + Dehs, pient < + Dehs, pient 9 < {.} Adj. mortality, by vintage: bacteremia/sept. {.} Adj. mortality, by vintage: pulmonary infection {.} Adjusted mortality, by vintage: malignancy Dialysis Dialysis Dialysis Dehs, pient 8 < + Dehs, pient < + Dehs, pient < Cause-pecific mortality Outcomes: hospitalizion & mortalityh

9 {.} Adj. RR of mortality after AMI: dialysis {.} Adj. RR of mortality after heart failure: dial. {.7} Adj. RR of mortality after CVA/TIA: dialysis Adjusted relive of mortality 8 8 Months after AMI Adjusted relive of mortality 8 8 Months after heart failure Adjusted relive of mortality 8 8 Months after CVA/TIA {.8} Adj. RR of mortality after PVD: dialysis {.9} Adj. RR of mortality after pneumonia: dial. {.} Adj. RR of mortality after VA event: dialysis Adjusted relive of mortality 8 8 Months after PVD Adjusted relive of mortality 8 8 Months after pneumonia Adjusted relive of mortality Months after VA event ince the early 99s, adjusted cause-specific mortality res have generally decreased in both the dialysis and transplant populions. Res of mortality due to acute myocardial infarction, for instance, have fallen cent since 99 for all dialysis pients, and cent for those on the modality less than three (Figure.). The change has been slightly smaller for older vintage pients, for whom res have dropped 8 cent. Similar changes have occurred for herosclerotic heart disease, cardiomyophy, and cardiac arrhythmia (Figures.7 9). While res for transplant pients tend to be less stable, they have decreased since 99 for all causes examined here. Res in the dialysis populion have grown, however, for mortality due to cardiac arrest, bacteremia/septicemia, and malignancy (Figures.,., and.). This is particularly true for pients of older vintage, in whom res since 99 have increased,, and cent, respectively. The highest of mortality in ESRD pients with cardiovascular, respirory, and vascular access events occurs in the interval immediely following the event (Figures. ). Pients who suffer an AMI or pneumonia are the most vulnerable, with a of deh times higher than th in pients without the event. Mortality s decrease dramically as time after an event increases. The of deh months following an AMI, for instance, falls from the six-month level by a factor of three (from. to.), and remains relively stable, though four it is still 8 cent higher than in pients without the event. In pients suffering from episodes of pneumonia, the of deh decreases by almost half from six months to months, and four is cent higher than in the reference group. The s of mortality in pients with heart failure, CVA/TIA, or PVD are similar immediely following the event and over the fouryear study iod. The relive of deh six months for pients with heart failure, for example, is, compared to. and. for CVA/TIA and PVD. At the end of four, s are cent higher than in pients without these conditions. The lowest of deh overall occurs in pients with a vascular access event, but the six months is still cent higher than in the reference group. It declines thereafter, however, to only cent higher months, and after this point there is essentially no difference when compared to the reference group. {Figures. } iod prevalent dialysis & transplant pients; adjusted for age, gender, race, & primary diagnosis. Period prevalent ESRD pients,, used as reference cohort. {Figures. } incident dialysis pients with 9-day rule, Medicare as primary payor, & Part A & B claims, 99 combined; adjusted for age, gender, race, & primary diagnosis. Reference cohort: pients without corresponding diseases in the first year after ESRD initiion + 9 days. H USRDS Annual Da Report

10 {.} Adjusted hospital admissions & days, by modality: prevalent pients.... Admissions pient year Hospital days pient year hile the number of admissions pient year has remained steady since the early 99s, the number of hospital days has fallen cent for hemodialysis pients, cent for those with a transplant, and 9 cent for those on itoneal dialysis (Figure.). Dialysis pients have nearly the same number of yearly admissions, but itoneal dialysis pients spend more days in the hospital. Across modalities, causes other than cardiovascular disease and infection account for the greest number of hospital admissions and days (Figure.). Admissions for infection in hemodialysis pients have increased 8 cent since 99, while they are only cent different in itoneal dialysis and transplant pients. {.} Adjusted cause-specific hospital admissions & days, by modality: prevalent pients. Admissions: Admissions pient year Hospital days Cardiovascular disease Infection year 8 pient days Hospital {.} Adjusted admissions for principal procedures & diagnoses, by modality: prevalent pients Admissions, pt 8 Pulmonary infection Infection Cardiovascular procedures Heart cheterizions : itonitis Peritoneal dialysis : vascular access infection Peritoneal dialysis Hopitalizion D mortality, by modality Outcomes: hospitalizion & mortalityh

11 Admissions for pulmonary infections in hemodialysis pients have grown 9 cent since 99, while those for vascular access infections have more than doubled (Figure.). Survival probabilities for pients incident in increased slightly compared to those of the previous iod, with cent of hemodialysis pients, and cent of those on itoneal dialysis, expected to survive five after the beginning of trement (Figure.). Since 98, all-cause mortality res have decreased 9 cent for hemodialysis pients, and cent for pients on itoneal dialysis (Figure.). As with the overall res shown in Figure., however, these modere changes conceal dramic relionships between mortality res and vintage. For hemodialysis pients on the modality less than two, res have fallen cent since 98; for itoneal dialysis pients, res have declined cent. Res for pients on their modality five or more, in contrast, have increased. Though equivalent in 99, mortality res for older vintage hemodialysis pients are now cent higher than for those on the modality less than two, and have risen cent since their lowest point in 99. Res in the itoneal dialysis populion were essentially equivalent in 98, but res for older vintage pients have since increased 8 cent. Cause-specific mortality res remain relively stable, with res of mortality due to cardiovascular disease and infection higher in the itoneal dialysis populion, and res for other causes slightly lower (Figure.). {.} Adjusted five-year survival, by modality: incident pients probability Survival, pient Dehs < -< + (adjusted for vintage) Months after initiion {.} Adjusted all-cause mortality, by vintage: prevalent pients {.} Adjusted cause-specific mortality, by modality: prevalent pients {Figures. } iod prevalent pients; res adjusted for age, gender, race, & primary diagnosis. ESRD pients,, used as reference cohort. Cardiovascular procedure cegory excludes vascular access procedures. Heart cheterizions for transplant pients excluded prior to 99 because of few events. {Figure.} incident dialysis pients; adjusted for age, gender, race, & primary diagnosis. ESRD pients, 99, used as reference cohort. Modality determined on first ESRD service de; excludes pients transplanted or dying during the first 9 days. {Figures. } iod prevalent dialysis pients; res adjusted for age, gender, race, & primary diagnosis. Dialysis pients,, used as reference cohort. The Deh Notificion form was revised in September 99 to include more detailed cegories for cause of deh; prior to this time cardiovascular dehs were often classified as being of other causes. Because of this, da for cardiovascular & other dehs prior to 99 have been omitted here. Dehs, pient 7 7 Cardiovascular Cardiovascular Infection Infection H USRDS Annual Da Report 7

12 igure.7 shows unadjusted res of hospitalizion with a diagnosis of septicemia, while Figure.8 shows an otherwise similar analysis in which adjustment has been made for age, gender, race, and primary diagnosis. Septicemia res are consistently higher in hemodialysis pients than in itoneal dialysis pients, and in these lter pients appear to have largely stabilized after 997. In contrast, res of septicemia in hemodialysis pients continue to climb, from.8 pient in 99 to. in. This apparent doubling of res over a time interval of ten remains unchanged when adjustment is made for age, gender, race, and primary diagnosis. Mortality res after admission with septicemia are high. Figure.9 shows adjusted mortality res of. in the initial six months after septicemia, declining to.7 pient between and 8 months ler; Figure. shows th these res are 7. and. times, respectively, those of pients not exiencing an episode of septicemia. Figure. presents unadjusted res of pneumonia events, while Figure. shows an otherwise similar analysis in which adjustment has been made for age, gender, race, and primary diagnosis. Pneumonia res are consistently higher in hemodialysis pients than in itoneal dialysis pients. While res in the lter populion were relively constant between 99 and, res in hemodialysis pients have climbed gradually, from.8 pient in 99 to. in. Mortality res after pneumonia are high. Figure. shows Overall first-year hospital admission res for septicemia, by modality {.7} Raw res Mortality after first bacteremia/septicemia event {.9} Adjusted mortality res Admissions pient Dehs pient 8 With bacteremia/septicemia Without bacteremia/septicemia 8 8 Months after event {.8} Adjusted res {.} Adjusted relive of deh Admissions pt deh Adjusted relive of Months after event Major infections D cardiovascular events 8 Outcomes: hospitalizion & mortalityh

13 Overall first-year pneumonia event res, by modality {.} Raw res Mortality after first pneumonia event {.} Adjusted mortality res Admissions pient Dehs pient 8 7 With pneumonia Without pneumonia 8 8 Months after event {.} Adjusted res {.} Adjusted relive of deh Admissions pient deh Adjusted relive of 8 8 Months after event adjusted mortality res of 78. pient in the initial six months after pneumonia, declining to. between and 8 months ler; Figure. shows th these res are. and.8 times, respectively, those of pients not exiencing an episode of pneumonia. Pneumonia is a classic inflammory ste, and a considerable body of research suggests th inflammion is a factor for cardiovascular disease. Cardiovascular event res are high after a diagnosis of pneumonia. Figure. shows adjusted cardiovascular event res of.8 pient in the initial six months after pneumonia, declining to.7 between and 8 months ler; Figure. shows th these res are. and. times, respectively, those of pients not exiencing an episode of pneumonia. {Figures.7 8} incident dialysis pients with 9-day rule; adjusted res adjusted for age, gender, race, & primary diagnosis. Pients with Medicare as a secondary payor or enrolled in an HMO on day 9 are excluded, as are pients with septicemia claims overlapping the start de of the followup iod. {Figures.9 } incident dialysis pients, 99, with 9-day rule & with Medicare as primary payor; adjusted res adjusted for age, gender, race, primary diagnosis, & vintage. Pients without sepsis in the first year + 9 days after initiion are used as the reference cohort. {Figures. } incident dialysis pients with 9-day rule; adjusted res adjusted for age, gender, race, & primary diagnosis. Pients with Medicare as a secondary payor or enrolled in an HMO on day 9 are excluded, as are pients with pneumonia claims overlapping the start de of the followup iod. {Figures. } incident dialysis pients, 99, with 9-day rule & with Medicare Parts A & B as primary payor; adjusted res adjusted for age, gender, race, primary diagnosis, & vintage. Pients without pneumonia during the first year + 9 days after initiion are used as the reference cohort. Cardiovascular event res after first pneumonia event {.} Adjusted cardiovascular event res pient CV events CV event Adjusted relive of With pneumonia Without pneumonia 8 8 Months after event {.} Adjusted relive of cardiovascular event 8 8 Months after event H USRDS Annual Da Report 9

14 {.7} Fracture res, by age {.8} Fracture res, by gender {.9} Fracture res, by modality, pient Fractures Vertebral Rib Long bone Hip , pient Fractures Male Female Vertebral Rib Long bone Hip , pient Fractures Vertebral Rib Long bone Hip {.} Fracture res, by race {.} Fracture res, by ethnicity {. Fracture res, by primary diagnosis, pient Fractures White Vertebral Rib Long bone Hip, pient Fractures Hispanic Non-Hispanic Vertebral Rib Long bone, pient Fractures Diabetes Glomerulonephritis Hytension Vertebral Rib Long bone Hip ip fractures in the dialysis populion as a whole occur a re of only.8, pient ; the re of long bone fractures, in contrast, has grown to. (Figure.7). Fractures occur most frequently, as expected, in the oldest pients. The re of long bone fractures in these pients grew 7 cent between the and iods, to, pient. Long bone fractures occur in women a re 7 cent higher than in men, and in hemodialysis pients a re cent higher than in those on itoneal dialysis; they are also more than twice as common in whites as in pients of other races (Figures.8 ). By diagnosis, pients whose ESRD is caused by diabetes are most likely, and those with glomerulonephritis least likely, to have a long bone fracture (Figure.). At six months, the relive s of deh in pients suffering vertebral, rib, or long bone fractures are.,, and., respectively (Figure.). While decreasing 8 months, they still remain 8 cent higher than in pients without similar fractures. Compared to diabetic pients, pients with hytension have higher initial s of deh six months after a fracture. versus.9 after a rib fracture, for example, and.7 versus.9 after a long bone fracture (Figure.). As time after the event increases, however, the of deh shows no clear ptern between diabetics and pients with hytension. Males have a higher of deh than females in the first year following a fracture (Figure.). The of deh six months after a vertebral fracture, for example, is.9 for men, and. for women; for a rib fracture, the s are.9 and., respectively. Risks of deh after a fracture vary by race (Figure.). At months, for example, black pients with a vertebral fracture have a mortality of, compared to.7 in whites; for rib fractures, however, s are even between the races. {Figures.7 } incident dialysis pients, age & older. Included pients from used as reference cohort. Hip fracture includes fracture of pelvis (ICD-9-CM code 88.), while fracture of neck of femur (8.xx) is counted as a long bone rher than hip fracture. See Appendix A for full list of codes. {Figures. } incident ESRD pients with 9-day rule, Medicare as primary payor, & Parts A & B claims, 99 combined. Reference cohort: pients without corresponding fractures in the first year after ESRD initiion + 9 days. x {Figure.7} adjusted for gender, race, & primary diagnosis; res for all also adjusted for age. Res by age are not directly comparable to those for all dialysis pients. {Figure.8} adjusted for age, race, & primary diagnosis. {Figure.9} adjusted for age, gender, race, & diabetic stus. {Figure.} adjusted for age, gender, & primary diagnosis. {Figure.} adjusted for age, gender, race, & primary diagnosis; hip fractures not shown by ethnicity because of small cohort size. {Figure.} adjusted for age, gender, & race. {Figure.} adjusted for age, gender, race, & primary diagnosis. {Figure.} adjusted for age, gender, & race. {Figure.} adjusted for age, race, & primary diagnosis. {Figure.} adjusted for age, gender, & primary diagnosis. Fractures D mortality Outcomes: hospitalizion & mortalityh

15 {.} Adjusted relive of mortality after fracture, overall {.} Adjusted relive of mortality after fracture, by primary diagnosis Vertebral Vertebral Diabetes Hytension deh Adjusted relive of Rib Long bone deh Adjusted relive of Rib Long bone 8 8 Months after fracture 8 8 Months after fracture {.} Adjusted relive of mortality after fracture, by gender deh Adjusted relive of Vertebral Rib Long bone Male Female {.} Adjusted relive of mortality after fracture, by race deh Adjusted relive of Vertebral Rib Long bone White Black 8 8 Months after fracture 8 8 Months after fracture H USRDS Annual Da Report

16 {.7} New cancer hospitalizion res: female dialysis pients omen who develop cancer or become pregnant while on ESRD therapy clearly face enormous healthcare challenges. We present here new da on hospitalizion and trement for breast, cervical, uterine, and ovarian cancers in dialysis pients, and on pregnancies in dialysis and transplant pients. Res of hospitalizion with a new cancer diagnosis in women on dialysis declined slightly during the 99s 7 cent for breast cancer, to. hospitalizions, pient, and cent for cervical, uterine, and ovarian cancers, to. (Figure.7). In , cent of pients with any of these cancers received chemotherapy; cent of those with breast cancer received radiion, compared to cent of those with cervical, uterine, or ovarian cancer; and and cent, respectively, received radiion and/ or chemotherapy (Figure.). Among women age, pregnancy res are consistently highest for those without diabetes, regardless of modality (Figure.). Changes in these res over time, however, are quite different between the modalities. Res for dialysis pients have been relively stable since 99. For transplant pients, in contrast, res have declined steadily for non-diabetics, from pregnancies, pient in 99 to in, and for diabetics, from to 8. The use of immunosuppressive medicions may help explain this change. In addition, recent literure has suggested th women wait several after a kidney transplant to become pregnant, and, since Medicare covers only the first three post-transplant, many of these ler pregnancies are not captured in the dabase. Complicions during pregnancy have followed no clear ptern over time (Figure.). In nearly one-third of dialysis pients, and one-fifth of those with a transplant, suffered a hemorrhage, while and cent, respectively, went into early labor. Pregnancy outcomes vary quite dramically by modality (Figure.). Among dialysis pients, only 7 cent of preg- yrs, pt / hospitalizions cancer First yrs, pt / hospitalizions cancer First Breast Cervical, uterine, & ovarian {.8} New cancer hospitalizion res, by diabetic stus: female dialysis pts yrs, pt / hospitalizions cancer First Breast Diabetic Non-diabetic Breast Breast White Cervical, uterine, & ovarian {.9} New cancer hospitalizion res, by race: female dialysis pients pients of Percent Radiion Chemotherapy Either Cervical, uterine, & ovarian {.} Trement res for new cancers: female dialysis pients Cervical, uterine, & ovarian Complicions D pregnancy in women with ESRD Outcomes: hospitalizion & mortalityh

17 {.} Pregnancy res, by modality & DM stus {.} Complicions in pregnancy, by modality {.} Pregnancy outcomes, by modality, pient Re 8 Dialysis Diabetic Non-diabetic pregnancies of Percent Dialysis Infection Hemorrhage Preeclampsia Early labor pregnancies of Percent Dialysis 8 Unspecified outcome Live birth Early terminion {.} Early terminions & live births, by modality early terminions of Percent pregnancies of Percent Early terminion: dialysis Dialysis Fetal distress Poor fetal growth Unspecified terminion Spontaneous abortion Induced abortion Ectopic/molar live births of Percent Live birth: dialysis 8 Unspecified birth Caesarean section Vaginal delivery {.} Neonal health, by modality {.} Postpartum complicions, by modality 8 pregnancies of Percent 9 9 Dialysis Infection Hemorrhage nancies since 99 have resulted in a live birth, while live births have occurred in more than half, and up to cent, of pregnancies among women with a transplant. Regardless of modality, the most common cause of an early terminion is spontaneous abortion; induced abortions have been slightly more common among transplant pients than among those on dialysis (Figure.). In live births, the use of Caesarean sections has declined among transplant pients; for women on dialysis, because of the small number of pients, types of live births have varied more over time. Though pregnancies among women on dialysis are less likely to result in a live birth, the children born to these women tend to be somewh healthier than those born to transplant pients, exiencing less fetal distress and better fetal growth (Figure.). {Figures.7 9} incident female adult dialysis pients without breast, cervical, uterine, or ovarian cancer in the first year of dialysis. Res include first new cancer admissions (in the second year) with cancer as a principal or secondary diagnosis code. Included pients from used as reference cohort. {Figure.7} adjusted for age, race, & diabetic stus. {Figure.8} adjusted for age & race. {Figure.9} adjusted for age & primary diagnosis. {Figure.} incident female adult dialysis pients with a new cancer hospitalizion. {Figures. } prevalent pients initiing therapy least 9 days before December of prior year, age on January of prevalent year, alive on December of th year, & with Medicare as primary payor & Medicare Part B coverage during the year. Trement modality defined on December of previous year, & assumed to be fixed during prevalent year. A complicion event or pregnancy outcome is tributed to the year of the first claim indicing pregnancy, not the year in which the event or outcome occurred (although these may be identical). H USRDS Annual Da Report

18 n the ESRD community, mortality and morbidity res have long been used as a quality assurance tool to improve pient outcomes. Mortality or any form of morbidity are assessed on the simplest level by comparing the actual number of events such as dehs or hospitalizion to the expected number of events. Provider-level mortality and morbidity rios have been used by the USRDS and others for many, with varying degrees of adjustment based on populion complexity and provider characteristics. Wh has been clear from the outset is the intrinsic variability in these estimes, which is highly reled to the size of the populion being assessed. Event re estimes have a greer degree of stistical instability for small providers than they do for larger ones. This is crudely exemplified in the lower graph of Figure.7, which presents standardized mortality rios (SMRs) calculed using the traditional method, along with their 9 cent confidence intervals, for a random sample of providers. This plot shows th the stabilities of estimed SMRs using this method differ widely, and are closely reled to sample size. The lower graph of Figure.9 provides a different illustrion of the impact of provider size on the variability of SMR estimes the smaller the provider, the wilder the SMR estime. The Bayesian hierarchical model, however, provides an alternive method of SMR estimion, stabilizing the estimes and making comparisons more approprie. To distinguish two methods, we here use the term BMR to designe the SMR estimed using the Bayesian method, and continue to use SMR to designe estimes using the traditional method. The top graphs in Figures.7 {.7} Standardized mortality rios & 9 cent confidence intervals (random sample of providers) BMR (with 9% CI) BMR 9 Small providers (< pient ) Large providers (+ pient ) and.9 show the dramic improvement in the variion of estimes using the Bayesian method, with the length of the confidence intervals flter and almost all BMRs in a constant band. For details on this method, we refer readers to our pa in Health Services and Outcomes Research Methodology (September issue), accessible on the web 7/contents. BMRs and SMRs in this section are adjusted for pient age, gender, race, primary diagnosis, and ESRD vintage. In Figure.8 we examine the agreement of results produced by these two methods. For providers with SMRs in the lowest quintile, for example, the BMR places cent in the same quintile, and the remaining units in higher quintiles. For providers with SMRs in the cent quintile, the BMR places cent in the same quintile, 7 cent in the lowest quintile, and cent in the middle quintile. Comparisons within this middle quintile give a 7 cent agreement in results of the two methods; cent are placed one quintile higher, and cent one quintile lower. For providers whose SMRs are in the 8 cent quintile, the BMR places 77 cent in the same quintile, 8 cent one quintile higher, and cent one lower. And in the highest SMR quintile, the two methods produce an 8 cent agreement; the BMR places 8 cent of units in the next lowest quintile. It is important for policy makers, providers, and health plans to recognize th, compared to the traditional SMR method, the Bayesian method needs a relively smaller sample size to produce an accure estime. If all providers were large enough, the two methods would produce the same results. Because this is not the case, however, we have pursued an alternive analytical approach, and have determined th more accure results are obtained through the Bayesian model. Overall, Figures.7 7 show th almost all provider groups by quintile, unit affiliion, and geographic region have similar median BMR and SMR values, but th the ranges are tighter with the BMR estimes. This holds true as well for hospital-based units, in which mortality rios calculed by both methods are slightly higher than in other groups. Pients receiving trement these providers may SMR {.8} Agreement of BMR & SMR, by quintile SMR (with 9% CI) Provider BMR: % units in each SMR quintile SMR quintiles (%) Traditional D Bayesian methods for estiming standardized mortality rios Outcomes: hospitalizion & mortalityh

19 have higher s caused by factors other than age, gender, race, primary diagnosis, and ESRD vintage. In the West North Central census division the BMR median is larger (closer to ) than the SMR median; providers in this area are much smaller than those in other areas of the country. Theoretically, the rank of BMR is optimal under mean squareerror loss. But even optimal procedures may form poorly if the da do not provider enough informion. For very small providers, the Bayesian method may over-shrink the estimes to. Caution is needed to report the results, and it is necessary to evalue how much informion is contained in the da (Liu et al). Based on these results, the USRDS will now be conducting all analyses of mortality rios along with hospitalizion and transplantion rios using this new Bayesian method. On the following spread we present similar comparisons of hospitalizion rios, using the traditional and Bayesian methods. { figures} adjusted for age, gender, race, primary diagnosis, & vintage. Informion on the U.S. Census divisions is available maps/cp_mapproducts; a map of the divisions is presented on page 7. x {Figure.7} prevalent dialysis pients,, in a random sample of dialysis providers. {Figures.8 7} prevalent dialysis pients,, in all dialysis providers. {.9} Variion of provider-level BMR & SMR, by provider size (in pient ) {.7} Distribution of provider-level mortality rios, by unit affiliion. BMR BMR SMR BMR (log plot).. Mortality rio.. SMR NC HB Unit affiliion (see box below for codes) SMR (log plot).. {.7} Distribution of provider-level mortality rios, by U.S. Census division. BMR SMR. Provider size Mortality rio.. {.7} Distribution of provider-level mortality rios, overall & by BMR quintile Overall By BMR quintile BMR SMR.. PAC MTN WNC WSC ENC ESC SA MA NE U.S. Census division (see box below for codes) Mortality rio BMR SMR BMR quintiles (%) Chain Fresenius Chain Gambro Chain DaVita Chain Renal Care Group Chain Dialysis Clinics, Inc. Chain N l Nephrology Assoc. NC Non-chain units HB Hospital-based units PAC Pacific MTN Mountain WNC West North Central WSC West South Central ENC East North Central ESC East South Central SA South Atlantic MA Middle Atlantic NE New England H USRDS Annual Da Report

20 n this spread we assess morbidity using the standardized hospitalizion rio (SHR), which is adjusted for age, gender, race, primary cause of ESRD, and vintage. As noted with the standardized mortality rio (pages ), there is considerable variion in the SHR based on provider size. To address this, along with the inherent variion in predicted hospitalizion res, we have applied the Bayesian model here as well. In Figure.7 we compare the ranking of providers, using the traditional SHR method and the newer Bayesian model. Providers within the lowest SHR quintile have an only 8 cent agreement with the BHR model, while the remaining 7 cent are placed into higher BHR quintiles. The two methods appear to give comparable results overall, but they differ the extremes. On a provider level it appears th the chains and independents have similar BMR and SHRs, while hospital-based providers have higher rios (Figure.7). We look here as well cardiovascular and infectious hospitalizions. Based on the modeling of the intrinsic variability of small versus larger providers, the range is smaller here in the BHR compared to the SHR (Figures.77 and.8). Chain-owned and independent providers appear to have comparable rios, while rios for hospital-based providers are higher (Figures.78 and.8). Cardiovascular hospitalizion rios are higher in the West South Central and East North Central census divisions, while infectious hospitalizion rios are higher in the Middle Atlantic and New England areas (Figures.79 and.8). Used in the calculion of mortality and hospitalizion rios, the Bayesian model offers a more stable way to compare yearly formance while taking into account normal variion. This method can be also used to assess transplant centers for acute graft and pient events and for longer-term outcomes. Whereas the traditional SMR/SHR calculion may show wide swings in results based on smaller provider variability, the Bayesian model may over-adjust for random variion, thereby shrinking the results toward.. Just as it is unlikely th a provider would truly have a zero mortality or hospitalizion rio, it is also of concern th most units may have rios close to one. In these cases a more direct review of a provider is required to ensure safety. It is also important not to penalize providers whose increasing variion is due to small size. Another important competing event is th of kidney transplantion. Mortality and hospitalizion rios are vulnerable to this event, which removes healthier pients. The USRDS will continue to investige these issues in order to provide multiple methods of assessing providers and determining their true formance. { figures} prevalent dialysis pients,, in all dialysis providers; adjusted for age, gender, race, primary diagnosis, & vintage. Informion on U.S. Census divisions is available {.7} Agreement of BHR & SHR, by quintile {.7} Distribution of provider-level hosp. rios, overall & by BHR quintile. Overall BHR quintile BHR: % units in each SHR quintile SHR quintiles (%) Hospitalizion rio..... BHR SHR BHR SHR BHR quintile (%) {.7} Distribution of provider-level hospitalizion rios, by unit affiliion {.7} Distribution of hospitalizion rios, by U.S. Census division Hospitalizion rio..... BHR SHR Hospitalizion rio... BHR SHR. NC HB. PAC MTN WNC WSC ENC ESC SA MA NE Unit affiliion (see box right for codes) U.S. Census division (see box right for codes) Traditional D Bayesian methods for estiming standardized hospitalizion rios Outcomes: hospitalizion & mortalityh