P

i

p

c

M

t

o

J

d

g

t

t

c

f

R

Mppvtotow

Saawat

MMaPkaI

dg

DJS

1

Interventions to Reduce Mortality among Patients Treated inIntensive Care Units

Peter J. Pronovost,*† Michael L. Rinke,‡ Katherine Emery,‡ Cheryl Dennison,§ Charles Blackledge,‡

and Sean M. Berenholtz*

t

f

w

y

C

c

m

c

i

a

r

t

i

©

urpose: Using sensitivity analysis to estimate the

mpact, in terms of patient lives, of the failure to use

roven therapies known to reduce mortality in criti-

ally ill intensive care unit patients.

aterials and Methods: We identified high-impact in-

erventions published in the last 5 years in the Journal

f the American Medical Association or New England

ournal of Medicine, extracted the absolute risk re-

uction associated with each intervention and

leaned the national incidence of each condition and

he percent of the population not receiving the cited

herapy from the literature. From this information, we

alculated national estimates of the excess deaths

rom failure to use these therapies.

esults: With consistent and appropriate implemen-

tteicstaotmUk

S

rJNtpdwpijv

D

nedoi:10.1016/j.jcrc.2004.07.003

58 Journa

ation of the 5 cited evidence-based interventions, we

ound a total of 167,819 lives could be saved per year,

ith a range of 137,670 to 197,965 lives saved per

ear.

onclusions: Mistakes of omission are common in the

ritical care setting and lead to significant preventable

ortality. There is a significant gap between the dis-

overy of effective interventions and their use in clin-

cal practice. By viewing the delivery of healthcare as

science and increasing funding for health services

esearch, we may be able to increase the use of effec-

ive therapies and, as a result, reduce patient mortal-

ty.

2004 Elsevier Inc. All rights reserved.

OUNTING EVIDENCE suggests that effec-tively constructed and implemented thera-

ies can improve care and increase survival ofatients in health care settings.1,2 Despite a largeolume of rigorously tested and verified interven-ions that improve patient outcomes, patients cannly count on receiving these interventions half ofhe time.3,4 The impact of these errors of omissionn patient outcomes is poorly understood and notell documented.Intensive care units (ICUs), caring in the United

tates for 4.4 million people a year while costingpproximately 1% of the gross domestic product,re not exempt from this trend, varying widelyith regard to practice, organizational structure

nd morbidity rates.5-10 There are several interven-ions, identified from well-conducted randomized

From the *Departments of Anesthesia and Critical Careedicine, and Surgery, Johns Hopkins University School ofedicine, Baltimore, MD, and †Department of Health Policy

nd Management, School of Nursing, Bloomberg School ofublic Health ‡Johns Hopkins School of Medicine, Johns Hop-ins Medicine Center for Innovations in Quality Patient Care;nd §School of Nursing, Johns Hopkins Medicine Center fornnovations in Quality Patient Care.

P. J. P. is supported by a grant from the VHA Foundation toevelop measures of quality sepsis care. S. B. is supported by arant from VHA Foundation for Sepsis Measures.Address reprint requests to Peter J. Pronovost, MD, PhD,

epartment of Anesthesiology & Critical Care Medicine, Theohns Hopkins University School of Medicine, 901 S. Bond St,uite 318, Baltimore, MD 21231; E-mail: ppronovo@jhmi.edu© 2004 Elsevier Inc. All rights reserved.0883-9441/04/1903-0006$30.00/0

rials or systematic reviews that can reduce mor-ality in critically ill patients.11-15 Although thextent to which these therapies are routinely useds unknown, ICU mortality rates have not de-reased to the extent predicted by recent clinicaltudies.16,17 Systematic implementation of ICU in-erventions, either independently or concurrently ifppropriate, could result in a significant reductionf the current 12% mortality rate among ICU pa-ients.18 The specific aim of this paper is to esti-

ate the number of preventable deaths in thenited States from the failure to use interventionsnown to reduce mortality in critically ill patients.

MATERIALS AND METHODS

tudy Design

To achieve our aims, we conducted a literatureeview. We identified interventions published in theournal of the American Medical Association or theew England Journal of Medicineestimated how of-

en people in the United States received these thera-ies, and calculated national estimates of the excesseaths from failure to use these therapies. Our goalas not to identify every intervention known to im-rove mortality in critically ill patients but rather todentify significant interventions published in majorournals for which there is consensus that the inter-entions ought to be used.16,19

ata Sources

We required four data sources to estimate theumber of preventable deaths. First, we identified

ffective interventions. Second, we estimated the

l of Critical Care, Vol 19, No 3 (September), 2004: pp 158-164

npewpnc

I

tiNtdlrimmc2PcswFgs

E

tlki2pl

ET

sucApsap

E

tntowdmip1bnd

wsss

Uctaapf

I

w4rtmvdN3mtc1

A

INTERVENTIONS TO REDUCE MORTALITY 159

umber of patients at risk for that disease in the USopulation. Third, using the results of the study, westimated the absolute risk reduction associatedith the use of the intervention. Fourth, we ap-roximated the percent of the exposed populationot receiving the identified therapy and then cal-ulated the number of preventable deaths.

dentification of Interventions

To identify interventions known to reduce mor-ality in critically ill patients published in highmpact journals, we searched 2 major journals:ew England Journal of Medicine and Journal of

he American Medical Association. Two indepen-ent observers performed a web-based, systematiciterature review of these journals. We includedandomized trials or systematic reviews, publishedn the last 5 years, of interventions that reduceortality in critically ill patients that were imple-ented in an ICU (search terms: ICU, mortality,

ritically ill and January 1998 through August003, JAMA limited to Caring for the Critically Illatient, N ENGL J MED limited to Original Arti-les). Two ICU physicians (P. J. P and S. M. B.)elected articles that met inclusion criteria and forhich there was general agreement about their use.or each selected article, we abstracted data re-arding study population, intervention, and the ab-olute effect of the intervention on mortality.

stimates of the Number of Patients at Risk

For each disease that the above interventionsargeted, incidence data were abstracted from pub-ished data and a pharmaceutical company’s mar-eting research.20-24 When needed, we multipliedncident data by the current US population of80,000,000 to estimate numbers at risk.25 Theopulation numbers chosen were well cited in theiterature.

stimate of Population Percentage not Receivingherapy

When available, data were gleaned from publishedtudies.20,26,27 When not available, we estimated these of these therapies based on performance in aohort of 23 ICUs through the Volunteer Hospitals ofmerica (VHA) working to improve ICU care, or onharmaceutical company information.21,28 If no dataource were available, we estimated, consistent with

broad range of other interventions, that half the3

atients received the therapy. s

stimate of Preventable Deaths

For each intervention identified, we calculatedhe number needed to treat (NNT), that is theumber of patients that must receive the interven-ion to prevent one death. The NNT is equal to onever the absolute risk. Annual preventable deathsere estimated by the following calculation: Inci-ence of specific diseases in the United Statesultiplied by the percentage of patients not receiv-

ng therapy divided by the NNT. Although the timeeriod for the studies varied, they were all less thanyear. Because some of our estimates of the num-

er of patients at risk and the percentage of patientsot receiving the therapy are imprecise, we con-ucted sensitivity analysis for these variables.

RESULTS

Our literature review identified 89 abstracts ande included 5 studies (4 randomized trials and 1

ystematic review) in our analysis that met inclu-ion criteria and are widely accepted.16,20,29 Weummarize these studies in Table 1.

Table 2 lists the number of patients at risk in thenited States, the percentage of patients not re-

eiving a given intervention, the number of pa-ients needed to treat in order to prevent 1 death,nd the preventable deaths in the United Statesnnually. Our assumptions to estimate the relevantopulations and the number of preventable deathsor each intervention are discussed below.

CU Physician Staffing

Using Medicare data and published literature,e estimated the number of ICU admissions to be,400,000 annually.5 The range of patients noteceiving therapy (63%-90%) was based on a na-ional health care reporting database containing 40illion patients and a nationally representative sur-

ey of hospitals, as well as from a survey con-ucted for the Leapfrog Group.20,26 To estimate theNT for ICU physician staffing, we applied the0% relative risk reduction to the average ICUortality of 12% to provide NNT of 25.11,18 The

otal number of annual preventable deaths wasalculated to be 134,640 with a range of 110,880 to58,400.

ctivated Protein C

We estimated that 190,000 patients had severe

epsis with an APACHE II score greater than 24,

bco

ato

P

B

A

V

A

P

B

A

V

A

160 PRONOVOST ET AL

ased on a correspondence with a major pharma-eutical company.21 The exclusion criteria used inur estimate are stricter than those cited in the

Table 1. Intervent

Citation Intervention

ronovost, et alJAMA 288:2151-2162, 2002

High-intensity v low-intensity ICU staassociated with reduced hospital aICU mortality and length of stay.11

ernard, et alN Engl J Med344:699-709, 2001

Treatment with human activated Prosignificantly reduced mortality inpatients with severe sepsis.12

nnane, et alJAMA 288:862-871,2002

A 7-day treatment with low doses ofhydrocortisone and fludrocortisonesignificantly decreased risk of deatpatients with septic shock and relaadrenal insufficiency who were trein the ICU without increasing adveevents.13

an den Berghe, et alN Engl J Med 345:1359-1367, 2001

Intensive insulin therapy to maintainglucose at or below 110 mg/dL redmorbidity and mortality among criill patients in the surgical ICU.14

RDS NetworkN Engl J Med 342:1301-1308, 2000

In patients with acute lung injury andacute respiratory distress syndromlower tidal volume ventilation resuin decreased mortality and numberventilator days.15

*Study provided a relative risk reduction of 30% that was a

Table 2. Calculatio

Citation Incidence

P

ronovost, et alJAMA 288:2151-2162,2002

4,400,000 (5)63-

ernard, et alN Engl J Med 344:699-709, 2001

190,000 (16,20)8

nnane, et alJAMA 288:862-871,2002

101,112 (22)†

15

an den Berghe, et alN Engl J Med 345:1359-1367, 2001

484,000 (5,32)†

70-

RDS Network,N Engl J Med 342:1301-1308, 2000

89,000† (24)50-

*Population presented as 76.6% of septic shock patients, w†Population presented as 11% of the total ICU admissions p

Data presented in incidence per 100,000 person-years was multipli

rticle, and instead followed recommendations byhe Food and Drug Administration.30 Therefore,ur incidence number may be conservative, espe-

Reduce Mortality

Time Periodfor Mortality

Mortality RateWithout

Intervention

MortalityRate With

Intervention

AbsoluteRisk

Reduction

Hospitaldischarge

12* 8 4

28 day 30.8 24.7 6.1

28 day 63 53 10

1 year 8 4.6 3.4

28 day 39.8 31 8.8

o an average ICU mortality of 12%.11,18

reventable Deaths

otg

y

NumberNeeded to

Treat toPrevent One

Death

Range of PreventableDeaths in the United

States Annually

0,26)25 134,640

110,880-158,400

21)16.4 10,311

9,847-10,774

,29)10 5,056

1,517-8,595

8,31)29.4 12,347

11,523-13,170

7,31)11.4 5,465

3,903-7,026

Total: 167,819Range: 137,670-197,965

3% of the total ICU admissions per year (22,32)(5,32)

25

ions to

ffingnd

tein C

h intiveatedrse

blooducedtically

e,ltedof

pplied t

ns of P

ercent NReceivinTherap

76.590% (2

895-93% (

50-85% (3

7580% (2

7090% (2

ho areer year

ed by the current United States population of 280,000,000.

caT82TAar

S

ctIIwcsswcstwtM2ttrwTtw8

G

Idalhartapps

spudceI

id1

A

dtT5luagob5

mdmedoOpftiaH

mgtpotti

INTERVENTIONS TO REDUCE MORTALITY 161

ially when ongoing studies looking at broaderpplications of protein C therapy are concluded.16

he range of the population not receiving therapy,5%-93%, was obtained from 2002 and projected003 prescriptions of activated protein C (APC).21

he NNT, 16.4, was obtained from the randomizedPC trial.12 The total number of annual prevent-

ble deaths was calculated to be 10,311 with aange of 9,847 to 10,774.

teroids in Sepsis

We estimated 132,000 cases of septic shock areared for in the ICU annually based on the applica-ion of a 3% septic shock incidence in an multicentertalian ICU study to the total number of AmericanCU patients.5,22 To estimate the population at risk,e conservatively multiplied this number by the per-

entage of corticotrophin non-responders in thetudy, 76.6%, creating a population of 101,112 septichock nonresponders. This is a cautious estimatehen the projected 1.5% per year increase in sepsis

ases due to the growing elderly population is con-idered.23 Because there is no national database forhe percentage of patients not receiving this therapy,e estimated that 50% of patients were receiving this

herapy, based on recent New England Journal ofedicine findings.3 In addition, our experience with

4 hospitals working with the VHA, Inc to improvehe case of patients with septic shock demonstratedhat only 30% of septic shock patients received ste-oids. Given these imprecise estimates, a large rangeas used for these numbers in the sensitivity analysis.he NNT, 10, was obtained from the randomized

rial.13 The total number of annual preventable deathsas calculated to be 5,056 with a range of 1,517 to,595.

lucose Control

We estimated that 484,000 patients were in theCU with a length of stay (LOS) greater than 7ays. This data were obtained from the total ICUdmission data as stated above, and a large, year-ong LOS study, which found that 11% of patientsad a LOS greater than 7 days.5,31 The entire ICUdmission population was used, based on a recenteview article and a recent critical care and infec-ious disease consensus conference suggesting thepplication of tight glucose control to all ICUatients, even though the original article’s studyopulation was only surgical ICU patients.16,19 De-

pite this, we feel our estimate to be conservative, t

ince the study found a decrease in mortality inatients with a LOS of greater than 5 days, and wesed patients with a LOS of greater than 7ays.16,32 The percentage of the population not re-eiving tight glucose control therapy, 70%-80%, wasstimated from an unpublished, nationwide study ofCUs conducted by the VHA in 23 ICUs. 28

The NNT, 29.4, was obtained from the random-zed trial.14 The total number of annual preventableeaths was calculated to be 12,347 with a range of1,523 to 13,170.

cute Respiratory Distress Syndrome

We estimated 89,000 cases of acute respiratoryistress syndrome (ARDS) annually based on a Na-ional Institutes of Health ARDS network database.24

he percent of the population not receiving therapy,0%-90%, was taken from an abstract on the use ofow-tidal volume therapy and the ICU data of onerban, academic medical center.29 A wide range waslso used for this number in the sensitivity analysisiven the imprecise estimation. The NNT, 11.4, wasbtained from the randomized trial.15 The total num-er of annual preventable deaths was calculated to be,465 with a range of 3,903 to 7,026.

DISCUSSION

The Institute of Medicine report “To Err is Hu-an” suggested between 44,000 and 98,000 people

ie each year in the United States because ofedical errors.33 These estimates are based in gen-

ral on mistakes of commission, things caregiverso to patients, and did not include mistakes ofmission– the failure to use therapies we ought to.ur study suggests that mistakes of omission mayose a much greater threat to patient safety. Weound that consistent and appropriate implementa-ion of 5 evidence-based interventions in criticallyll patients can prevent 137,670 to 197,965 deaths,

far greater number of patients than “To Err isuman” estimated for all of US healthcare.While these numbers are clearly a rough esti-ate and subject to debate, they highlight the stag-

ering number of preventable deaths from failingo implement evidence-based interventions in ICUatient care. While we used conservative estimatesf incidence and intervention use, the impact ofhese interventions may not be additive. In addi-ion, our study included only 5 high profile ICUnterventions and neglected the multiple other in-

erventions that may also reduce ICU mortality and

ms

baitrstitHlttrnaiuoVHo

A(upWwpsb

Faccsdvoad

WaAmciotecotptrvegtstflocdantcie

ksma1rstspTo

i

162 PRONOVOST ET AL

ortality within other high risk specialties such asurgery, cardiovascular disease, and dialysis.

These results suggest there is an enormous gapetween therapies that are known to be effectivend their use in clinical practice. Although alarm-ng, these results are entirely predictable based onhe dispersion of the United States government’sesearch dollars. Ninety-nine percent of US re-earch dollars, largely through the National Insti-utes of Health and Industry, are spent understand-ng disease biology and identifying effectiveherapies with 1%, largely through the Agency forealthcare Research and Quality and Foundations,

eft to learn how to ensure that patients receivehese therapies safely.34 With only half of the pa-ients receiving the interventions they ought to, thisesearch paradigm needs to be reconsidered. Weeed to view the delivery of healthcare as a sciences well as an art and rigorously learn how toncrease the extent to which effective therapies aresed. Focused and funded efforts to improved usef effective therapies have benefited Medicare andA patients, patients in the VHA and Institute forealthcare Improvement, as well as patients inther fields and specialties.35-43

Indeed, our research team, with funding fromgency for Healthcare Research and Quality

AHRQ), is currently implanting the interventionssed to reduce blood stream infections and im-rove ventilator care in the 108 ICUs in Michigan.e can no longer assume that our learning endsith discovering effective therapies; to improveatients’ outcomes, we need a more balanced re-earch portfolio that values the end as well as theeginning of the translation superhighway.44

There are several limitations to our study.irst, we selected only 5 interventions for ournalysis. We sought to be conservative and in-lude interventions that are widely accepted byaregivers. Had we included a larger number oftudies, the estimated number of preventableeaths would increase. Second, to estimate pre-entable deaths, we relied upon various sourcesf information, with varying degrees of randomnd systematic error. Nevertheless, we con-

ucted sensitivity analyses of our assumptions. i

REFERENC

gM

e recognize that there is some controversybout the intervention in the control group in theRDS trial potentially causing us to overesti-ate the treatment benefit. Even with the most

onservative assumptions, the results are signif-cant. Indeed, the lower estimate of the numberf preventable deaths (137,670) from not usinghese 5 therapies in ICU care exceeds the totalstimate of preventable deaths from mistakes ofommission. Third, we assumed that the effectf each intervention was independent and addi-ive. This may not be so and would decrease theotential number of lives saved. Fourth, we es-imated the effect of the intervention based onesults from one study. Nevertheless, the inter-entions are supported with a large body ofvidence and consensus conference practiceuidelines.19 Fifth, our literature search was nei-her complete nor based on highly systematicearch methods such as journal impact or cita-ion counts. We sought to identify interventionsrom the most highly read US journals. We be-ieve more rigorous research methods wouldnly have added to the number of lives thatould be saved with broad applications of evi-enced based medicine. Finally, the resultschieved in the randomized trial (efficacy) mayot be replicated when broadly applied (effec-iveness) due to such factors as study exclusionriteria and study bias. Nonetheless, the random-zed trials provide us with the best availablestimate of the efficacy of the interventions.

Mistakes of omission, the failure to use therapiesnown to be effective, are common and present aignificant opportunity to improve care. We esti-ate the failure to use 5 commonly accepted ther-

pies in critical care results in between 137,670 to97,965 preventable deaths per year.The currentesearch paradigm that focuses mainly on under-tanding disease biology and identifying effectiveherapies needs to be reconsidered. Greater re-earch and effort is needed to help ensure thatatients receive the therapies they ought to receive.his will likely come from evaluating how therganization, finance and delivery of health care

nfluence physician behavior.

ES

1. Chassin MR: Practice guidelines: Best hope for qualitymprovement in the 1990s. J Occup Med 32:1199-1206, 1990

2. Audet AM, Greenfield S, Field M: Medical practice

uidelines: Current activities and future directions. Ann Interned 30:709-714, 1990

3. McGlynn EA, Asch SM, Adams J, et al: The quality of

hJ

hA

tu

wC

oI

tc

I1

to

sA

ss

to2

s3

Vtd

m(

oN

tpC

sa

AI

P

S

tC

mi2

a1

2S

cmdJ

V2

aC

fmc

t

I4

m

hi

cWW2

to

Ce7

cf2

in2

st

INTERVENTIONS TO REDUCE MORTALITY 163

ealth care delivered to adults in the United States. N EnglMed 348:2635-2645, 2003

4. Institute of Medicine: Crossing the quality chasm: a newealth system for the 21st century. Washington, DC, Nationalcademies Press, 2001

5. Young MP, Birkmeyer JD: Potential reduction in mor-ality rates using an intensivist model to manage intensive carenits. Eff Clin Pract 6:284-289, 2000

6. Halpern NA, Bettes L, Greenstein R: Federal and nation-ide intensive care units and healthcare costs: 1986-1992. Critare Med 22:2001-2007, 1994

7. Knaus WA, Draper EA, Wagner DP, et al: An evaluationf outcome from intensive care in major medical centers. Annntern Med 104:410-418, 1986

8. Knaus WA, Wagner DP, Zimmerman JE, et al: Varia-ions in hospital mortality and length-of-stay from intensiveare. Ann Intern Med 118:753-761, 1993

9. Zimmerman JE, Shortell SM, Rousseau DM, et al:mproving Intensive Care. Crit Care Med 21:1443-1451,99310. Pronovost PJ, Jenckes MW, Dorman T, et al: Organiza-

ional characteristics of intensive care units related to outcomesf abdominal aortic surgery. JAMA 281:1310-1317, 199911. Pronovost PJ, Angus DC, Dorman T, et al: Physician

taffing patterns and clinical outcomes in critically ill patients:systematic review. JAMA 288:2151-2162, 200212. Bernard GR, Vincent JL, Laterre PF, et al: Efficacy and

afety of recombinant human activated protein C for severepsis. N Engl J Med 344:699-709, 2001

13. Annane D, Sebille V, Charpentier C, et al: Effect ofreatment with low doses of hydrocortisone and fludrocortisonen mortality in patients with septic shock. JAMA 288:862-871,00214. Van den Berghe G, Wouters P, Weekers F, et al: Inten-

ive insulin therapy in critically ill patients. N Engl J Med45:1359-1367, 200115. The Acute Respiratory Distress Syndrome Network:

entilation with lower tidal volumes as compared with tradi-ional volumes for acute lung injury and the acute respiratoryistress syndrome. N Engl J Med 342:1301-1308, 200016. Vincent JL, Abraham E, Annane D, et al: Reducingortality in sepsis: new directions. Crit Care 6:S1-S18, 2002

suppl)17. Martin GS, Mannino DM, Eaton S, et al: The epidemi-

logy of sepsis in the United States from 1979 through 2000.Engl J Med 348:1546-1554, 200318. Zimmerman JE, Wagner DP, Draper EA, et al: Evalua-

ion of acute physiology and chronic health evaluation IIIredictions of hospital mortality in an independent database.rit Care Med 26:1317-26, 199819. Dellinger RP, Carlet JM, Masur H, et al: Surviving

epsis campaign guidelines for management of severe sepsisnd septic shock. Crit Care Med 32:858-887, 2004

20. The Leapfrog Group: ICU Physician Staffing Fact Sheet.vailable from: URL: http://www.leapfroggroup.org/FactSheets/

CU_FactSheet.pdf. Accessed September 12, 200321. Email correspondence with Thane Wettig, Elli Lilly

harmaceuticals, March 18, 200322. Salvo I, de Cain W, Musicco M, et al: The Italian

EPSIS study: preliminary results on the incidence and evolu-

ion of SIRS, sepsis, severe sepsis and septic shock. Intensiveare Med 21:S244-2449, 1995 (suppl 2)23. Angus DC, Linde-Zwirble WT, Lidicker J, et al: Epide-iology of severe sepsis in the United States: Analysis of

ncidence, outcome, and associated costs of care. Crit Care Med9:1472-1474, 200124. Goss CH, Brower RG, Hudson LD, et al: Incidence of

cute lung injury in the United States. Crit Care Med 31:1607-611, 200325. United States Census Bureau: Your Gateway to Census

000. Available online at http://www.census.gov. Accessedeptember 12, 200326. Angus DC, Kelley MA, Schmitz RJ, et al: Caring for the

ritically ill patient. Current and projected workforce require-ents for care of the critically ill and patients with pulmonary

isease: Can we meet the requirements of an aging population?AMA 284:2762-2770, 2000

27. Rubenfeld GD, Caldwell D, Hudson L: Lung Protectiveentilation in Patients with acute lung injury. AJRCCM A295,00128. Pronovost PJ, Berenholtz, SM, Ngo K et al: Developing

nd pilot testing quality indicators in the intensive care unit. Jrit Care 18:145-155, 200329. National Quality Forum: A Comprehensive Framework

or Hospital Care Performance Evaluation: Executive Sum-ary. Available from: URL: http://www.qualityforum.org. Ac-

essed September 12, 200330. Siegel JP: Assessing the use of activated protein C in the

reatment of severe sepsis. N Engl J Med 347:1030-1034, 200231. Stricker K, Rothen HU, Takala J: Resource use in the

CU: short- vs. long-term patients. Acta Anaesthesiol Scand7:508-515, 200332. Montori VM, Bistrian BR, McMahon MM: Hyperglyce-ia in acutely ill patients. JAMA 288:2167-2169, 200233. Institute of Medicine: To err is human: Building a safer

ealth system. Kohn L, Corrigan J, Donaldson M, (eds): Wash-ngton, DC, National Academies Press, 2000

34. Gray BH, Gusmano MK, Collins SR: AHCPR and thehanging politics of health services research. Health Affairs.eb Exclusive. Available from: URL: http://healthaffairs.org/ebExclusives/Gray_Web_Excl_062503.htm. Accessed June

5, 200335. Pronovost PJ, Jenckes MW, Dorman T, et al: Organiza-

ional characteristics of intensive care units related to outcomesf abdominal aortic surgery. JAMA 281:1310-1317, 199936. Davis DA, Thomson MA, Oxman AD, Haynes R:

hanging physician performance. A systematic review of theffect of continuing medical education strategies. JAMA 274:00-705, 199537. Cretin S, Shortell SM, Keeler EB: An evaluation of

ollaborative interventions to improve chronic illness care;ramework for study and design. Evaluation review 28:28-51,00438. Feldman PH, Peng TR, Murtaugh CM, et al: A random-

zed intervention to improve heart failure outcomes in commu-ity-based home health care. Home Health Care Serv Q 23:1-3, 200439. Kresowik TF, Bratzler DW, Kresowik RA, et al: Multi-

tate improvement in process and outcomes of carotid endar-erectomy. J Vasc Surg 39:372-380, 2004

40. McClellan WM, Millman L, Presley R, et al: Improved dia-

beg

IC

fq

iJ

t

164 PRONOVOST ET AL

etes care by primary care physicians: results of a group-randomizedvaluation of the Medicare Health Care Quality Improvement Pro-ram (HCQIP). J Clin Epidemiol 56:1210-1217, 2003

41. McClellan WM, Frankenfield DL, Frederick PR, et al:mproving the care of ESRD patients: A success story. Healthare Financ Rev 24:89-100, 200342. Jha AK, Perlin JB, Kizer KW, et al: Effect of the trans-

ormation of the Veterans Affairs Health Care System on theuality of care. N Engl J Med 348:2218-2227, 200343. Ghali WA, Ash AS, Hall RE, et al: Statewide quality

mprovement initiatives and mortality after cardiac surgery.AMA 277:379-382, 1997

44. Lenfant C: Clinical research to clinical practice-lost inranslation? N Engl J Med 349:868-874, 2003