RLAMAO

bbdsslc

wiaytcadasa

CdrtvepaSoIt

A

PS

AOb

©P

AC

IC

ORIGINAL ARTICLES: GENERAL THORACIC

CARDIOTHORACIC ANESTHESIOLOGY:The Annals of Thoracic Surgery CME Program is located online at http://cme.ctsnetjournals.org.To take the CME activity related to this article, you must have either an STS member or anindividual non-member subscription to the journal.

ecalibration of the Revised Cardiac Risk Index inung Resection Candidates

lessandro Brunelli, MD, Gonzalo Varela, MD, PhD, Michele Salati, MD,arcelo F. Jimenez, MD, Cecilia Pompili, MD, Nuria Novoa, MD, andrmando Sabbatini, MD

spedali Riuniti, Ancona, Italy, and University Hospital Salamanca, Spain

3RpimcaCs0

ulicc

GEN

ERA

LT

HO

R

Background. The revised cardiac risk index (RCRI) haseen proposed as a tool for cardiac risk stratificationefore lung resection. However, the RCRI was originallyeveloped from a generic surgical population including amall group of thoracic patients. The objective of thistudy was to recalibrate the RCRI in candidates for majorung resections to provide a more specific instrument forardiac risk stratification.

Methods. One thousand six hundred ninety-six patientsho underwent lobectomy (1,426) or pneumonectomy (270)

n two centers between the years of 2000 and 2008 werenalyzed. Stepwise logistic regression and bootstrap anal-ses were used to recalibrate the six variables comprisinghe RCRI. The outcome variable was occurrence of majorardiac complications (cardiac arrest, complete heart block,cute myocardial infarction, pulmonary edema, or cardiaceath during admission). Only those variables with a prob-bility of less than 0.1 in more than 50% of bootstrapamples were retained in the final model and proportion-

lly weighted according to their regression estimates.

cSrrisncolmm

P

Ttci

spedali Riuniti Ancona, Via Conca 1, Ancona 60020, Italy; e-mail:runellialex@gmail.com.

2010 by The Society of Thoracic Surgeonsublished by Elsevier Inc

Results. The incidence of major cardiac morbidity was.3% (57 patients). Four of the six variables present in theCRI were reliably associated with major cardiac com-lications: cerebrovascular disease (1.5 points), cardiac

schemia (1.5 points), renal disease (1 point), and pneu-onectomy (1.5 points). Patients were grouped into four

lasses according to their recalibrated RCRI, predictingn incremental risk of cardiac morbidity (p < 0.0001).ompared with the traditional RCRI, the recalibrated

core had a higher discrimination (c indexes, 0.72 versus.62; p � 0.004).

Conclusions. The recalibrated RCRI can be reliablysed as a first-line screening instrument during cardio-

ogic risk stratification for selecting those patients need-ng further cardiologic testing from those who can pro-eed with pulmonary evaluation without any furtherardiac tests.

(Ann Thorac Surg 2010;90:199–203)

© 2010 by The Society of Thoracic Surgeons

ardiac evaluation is a fundamental part of the pre-operative functional workup of lung resection can-

idates. Recent guidelines have indicated that cardiacisk scores may be used as a stratification tool to selecthose patients needing more accurate invasive or nonin-asive cardiologic tests before proceeding to pulmonaryvaluation to complete the operability decision-makingrocess. The recently published American Heart Associ-tion/American College of Cardiology [1] and Europeanociety of Cardiology/European Society of Anesthesiol-gy guidelines [2] indicate the Revised Cardiac Riskndex (RCRI) [3] as the preferable cardiologic risk scoreo assess cardiac risk in noncardiac surgery. Most re-

ccepted for publication March 4, 2010.

resented at the Forty-sixth Annual Meeting of The Society of Thoracicurgeons, Fort Lauderdale, FL, Jan 25–27, 2010.

ddress correspondence to Dr Brunelli, Division of Thoracic Surgery,

ently, the joint European Respiratory Society/Europeanociety of Thoracic Surgeons task force on fitness foradical treatment of lung cancer patients endorsed theseecommendations and proposed a cardiologic algorithmncorporating this scoring system as a preliminarycreening instrument [4]. However, the RCRI was origi-ally developed from a generic surgical population in-luding only a small group of thoracic patients. Thebjective of this study was to recalibrate the RCRI in a

arger and more modern population of candidates forajor lung resections to provide a more specific instru-ent for cardiac risk stratification.

atients and Methods

his is an observational study performed on prospec-ively collected data in two dedicated thoracic surgeryenters. The study was approved by the local hospital

nstitutional review boards, and patients gave their con-

0003-4975/$36.00doi:10.1016/j.athoracsur.2010.03.042

sc

tumi

pcGnt

p(ld(

teaaw

ddcaCge

msiwtmvo

mtcdop

ebmcaehfea

rhacs

hldcs

oitlipc

aswbotrfivbvmsmvbttc

a(ftcrco

tdsS

R

Tdp

200 BRUNELLI ET AL Ann Thorac SurgRECALIBRATED REVISED CARDIAC RISK INDEX 2010;90:199–203

GEN

ERA

LT

HO

RA

CIC

ent to collection and use of their data in the database forlinical and scientific purposes.

All 1,696 major lung resections (1,426 pulmonary lobec-omies and 270 pneumonectomies) performed from Jan-ary 2000 through December 2008 for benign, primaryalignant, or metastatic disease at the two centers were

ncluded in the study.The two centers share standardized perioperative

athways of care and definition of risk factors and out-omes. The principal investigators of the two centers (AB,V) meet at regular times to critically review and, ifeeded, update these data, which are then disseminated

o the respective teams.Operability exclusion criteria included a predicted

ostoperative forced expiratory volume in 1 secondppoFEV1) and predicted postoperative carbon monoxideung diffusion capacity (ppoDlco) less than 30% pre-icted in association with a peak oxygen consumption

peak V̇o2) less than 10 mL · kg�1 · min�1.All patients underwent a preoperative cardiac evalua-

ion by a cardiology specialist, including history, physicalvaluation, resting electrocardiogram, and, if indicated,dditional tests, such as echocardiography, stress test,nd more-invasive tests. Prophylaxis against arrhythmiaas not used in any of the patients in this series.All patients who then underwent surgery were

eemed in stable cardiac condition with optimized car-iologic treatment. If a patient was found to have a highardiac risk with unstable hemodynamic conditions, theppropriate treatment was instituted as per Americanollege of Cardiology/American Heart Associationuidelines [1, 5], and the patient was subsequently re-valuated for possible inclusion in the surgical plan.As a rule all the operations were performed through auscle-sparing lateral thoracotomy by qualified thoracic

urgeons. Patients were usually extubated in the operat-ng room and transferred to a dedicated thoracic wardith cardiologic monitoring. Patients were admitted to

he intensive care unit only in case of major cardiopul-onary complications requiring assisted mechanical

entilation, multiple inotropic support, or invasive cardi-logic monitoring.Postoperative treatment focused on as early as possibleobilization, physical and respiratory rehabilitation, an-

ibiotic and antithrombotic prophylaxis, and thoracotomyhest pain control. Chest pain was assessed at least twiceaily, and treatment was titrated to achieve a pain scoref less than 4 (on a scale from 0 to 10) in the first 72ostoperative hours.The RCRI is a four-class risk score comprising six

qually weighed factors: ischemic heart disease (definedy the presence of any of the following: history ofyocardial infarction, history of a positive exercise test,

urrent complaint of chest pain considered to be second-ry to myocardial ischemia, use of nitrate therapy, orlectrocardiogram with pathologic Q waves), congestiveeart failure (defined by the presence of any of the

ollowing: history of congestive heart failure, pulmonarydema, or paroxysmal nocturnal dyspnea; physical ex-

mination showing bilateral rales or S3 gallop; or chest p

adiograph showing pulmonary vascular redistribution),istory of cerebrovascular disease (transient ischemicttack or stroke), insulin therapy for diabetes, serumreatinine level greater than 2 mg/dL, and high-riskurgery.

Although thoracic surgery was generically defined asigh-risk surgery in the original paper of Lee and col-

eagues [3], for the purpose of recalibrating the index weistinguished between lobectomy and pneumonectomy,onsidering only this latter operation as high-riskurgery.

Major cardiac complications were considered as thoseccurring during the admission or within 30 days and

ncluded acute myocardial infarction (diagnosed by elec-rocardiogram changes and increased serum troponinevel), pulmonary edema (confirmed by consistent find-ngs at chest roentgenography) ventricular fibrillation orrimary cardiac arrest, complete heart block, and anyardiac-related death.

Stepwise logistic regression and bootstrap resamplingnalyses were used to recalibrate the six factors in thetudy population. Bootstrap bagging with 1,000 samplesas used to assess the stability of the predictors. In theootstrap analysis 1,000 samples of the same size as theriginal population were drawn with replacement fromhe original dataset. Stepwise logistic regression wasepeated in each of these samples. The stability of thenal stepwise model can be assessed by identifying theariables that enter most frequently in the repeatedootstrap models and comparing those variables with theariables in the final stepwise model. If the final stepwiseodel variables occur in a majority (�50%) of the boot-

trap models, the original final stepwise regressionodel can be judged to be stable [6–8]. Only those

ariables with a probability value of less than 0.1 and aootstrap frequency of greater than 50% were retained in

he final score and proportionally weighted according toheir regression estimates, assigning 1 point to the lowestoefficient.

Patients were then grouped into different classes of riskccording to the original RCRI and the recalibrated scorehereby named for simplicity ThRCRI, where Th standsor thoracic), and the c indexes of the two scores werehen compared to verify their respective discriminativeapability. In the ThRCRI, patients were grouped intoisk classes according to their incremental risk of majorardiac events, maintaining the four-class structure of theriginal RCRI.The stability of the risk score across multiple popula-

ions was further tested in 1,000 bootstrapped samplesrawn with replacement from the original dataset. Alltatistical tests were performed on the statistical softwaretata 9.0 (StataCorp, LP, College Station, TX).

esults

he characteristics of the patients in this study areisplayed in Table 1. There were 69 major cardiac com-lications in 57 patients (cumulative incidence, 3.3%): 36

ulmonary edema, 11 acute myocardial infarctions, 6

cd

vat

fclp

agRmtt

Bcstbcc

C

RcccTTlmRwu2tcCgdpdl

T

V

ASDNMMFDICRDCP

a

uc

CfI

TA

V

IC

C

P

a

cpsu

I

TR

TS

12

T

TC

R

ABCDpc

a

t

201Ann Thorac Surg BRUNELLI ET AL2010;90:199–203 RECALIBRATED REVISED CARDIAC RISK INDEX

GEN

ERA

LT

HO

RA

CIC

ardiac arrests, and 16 cardiac-related deaths (of 48 totaleaths).Logistic regression showed that only four of the six

ariables present in the original RCRI were reliablyssociated with major cardiac complications in this set-ing (Table 2).

According to their regression estimates the reliableactors were proportionally weighted as follows: serumreatinine greater than 2 mg/dL (1 point), cerebrovascu-ar disease (1.5 points), cardiac ischemia (1.5 points), andneumonectomy (1.5 points).These weighted factors were then summed into an

ggregate score for each patient. Patients were thenrouped into four classes according to their recalibratedCRI (ThRCRI), predicting an incremental risk of cardiacorbidity (p � 0.0001; Table 3). Compared with the

raditional RCRI, the ThRCRI had a higher discrimina-ion (c indexes, 0.72 versus 0.62; p � 0.004; Table 4).

able 1. Characteristics of Patientsa

ariable Result

ge (y) 65 � 11.2ex (male, n, %) 1,398 (82%)iagnosis (n, %)onneoplastic 98 (6%)alignant primary 1,560 (92%)alignant secondary 38 (2%)

EV1 (%) 85% � 19%lco (%) 77.6% � 19.2%

HD (n, %) 185 (11%)erebrovascular disease (n, %) 69 (4%)enal comorbidity (n, %)b 57 (3.3%)iabetes (n, %) 188 (11%)HF (n, %) 15 (0.9%)neumonectomy (n, %) 270 (16%)

Results for 1,696 patients are expressed as mean � standard deviationsnless otherwise specified. b Renal comorbidity is defined as serumreatinine � 2 mg/dL.

HF � congestive heart failure; Dlco � diffusing capacity of the lungor carbon monoxide; FEV1 � forced expiratory volume in 1 second;HD � ischemic heart disease.

able 2. Results of the Stepwise Logistic Regressionnalysisa

ariables Coefficients SEM p Value Bootstrap %b

HD 1.4 0.3 �0.0001 98%reatinine � 2mg/dL

0.97 0.5 0.06 54%

erebrovasculardisease

1.32 0.4 0.003 82%

neumonectomy 1.46 0.3 �0.0001 99%

Dependent variable is defined as the occurrence of major cardiacomplications. A parsimonious model was used (only the significantredictors are shown). b Bootstrap % is the frequency of bootstrapamples in which the variable has a p � 0.1 (1,000 bootstrap samplessed).

HD � ischemic heart disease; SEM � standard error of the mean.Rr

ootstrap analysis showed that the risk of major cardiacomplications in class A was less than 3% in 100% ofamples, whereas in class B the cardiac risk was lowerhan 10% in 100%, but greater than 5% in 78% ofootstrapped samples. Class C had a risk of cardiacomplications greater than 10% in 84% of samples, andlass D had a risk greater than 20% in 73% of samples.

omment

ecently published guidelines from international medi-al and surgical societies have recommended the use ofardiac risk scores as screening tools for stratifying theardiac risk of patients undergoing noncardiac surgery.he European Respiratory Society/European Society ofhoracic Surgeons task force for evaluating fitness of

ung resection candidates with lung cancer recom-ended the application of the RCRI [4] in this setting. AnCRI lower than 2 has been reported to be associatedith a low cardiac risk, and the patient would not need tondergo additional tests. However, an RCRI greater thanhas been associated with an increased cardiac risk, and

he task force recommendation is to obtain a cardiaconsultation with noninvasive testing as per Americanollege of Cardiology/American Heart Associationuidelines [4]. Therefore, the RCRI has a pivotal role ineciding further cardiac tests before proceeding with aulmonary evaluation. However, this score was originallyeveloped in a generic surgical population including only a

imited number of thoracic surgery patients. Of the 2,893

able 3. Distribution of Patients in Each Class of theecalibrated Revised Cardiac Risk Index

hRCRIcore

RiskClass

Numberof Cases

Major CardiacComplications

0 A 1,173 18 (1.5%)–1.5 B 468 27 (5.8%)–2.5 C 16 3 (19%)

�2.5 D 39 9 (23%)

hRCRI � thoracic revised cardiac risk index.

able 4. Comparison of Traditional and Thoracic Revisedardiac Risk Index

isk ClassesTraditional

RCRI ThRCRI

. . . 1.5%2.3% 5.8%6% 19%7.5% 23%

value 0.001 �0.0001index 0.62 (0.55–0.68) 0.72 (0.65–0.78)a

p � 0.004 (comparison between the two receiver operating characteris-ics curves).

CRI � revised cardiac risk index; ThRCRI � recalibrated thoracicevised cardiac risk index.

pi

tpf

tfwwrsgcrt0

stHepc

iocc

igpisc

pbtpaecm

rbiifi

opirrwe

nmi

pcocbba

mdpphtetarw

ttTli

pppr

pcswm

atimisia

TC

R

202 BRUNELLI ET AL Ann Thorac SurgRECALIBRATED REVISED CARDIAC RISK INDEX 2010;90:199–203

GEN

ERA

LT

HO

RA

CIC

atients comprising the derivation set, only 346 (12%) werentrathoracic, not-otherwise-defined, procedures [3].

The objective of this investigation was to recalibratehe RCRI in a large sample of patients undergoing majorulmonary resection to obtain a more specific instrument

or preoperative cardiac workup in this set of patients.By using logistic regression and bootstrap resampling

echniques, we found that only four of the original sixactors comprising the RCRI remained reliably associatedith major cardiac complications. These four factorsere then proportionally weighted according to their

egression estimates, and their respective scores wereummed to derive the total score. Patients were thenrouped into four classes showing an incremental risk ofardiac complications. When the original RCRI and theecalibrated ThRCRI were applied to the same popula-ion, the latter showed a superior discrimination (c index,.72 versus 0.62; p � 0.004).

Interestingly, the traditional RCRI classes in this serieshowed an incidence of cardiac complications similar tohe one reported in the paper by Lee and associates [3].

owever, the ThRCRI showed more-pronounced differ-nces between these classes. In the recalibrated system,atients in class D had a threefold higher rate of cardiacomplications than patients in the traditional class D.

The recalibrated score looks also more balanced as itncludes all four classes of risk, contrary to the traditionalne, which, by definition (as all thoracic procedures wereonsidered as high-risk surgery), included only threelasses (B, C, and D).

In fact, contrary to the original RCRI, in which allntrathoracic procedures were classified as high-risk sur-ery, we chose to differentiate between lobectomy andneumonectomy to have a more precise and specific

nstrument. Indeed, pneumonectomy turned out to be aignificant and reliable predictor of major cardiacomplications.

Bootstrap analysis was used to assess reliability of bothredictors and final score. This technique has alreadyeen shown to be superior to the traditional training and

est splitting of the population to assess reliability ofredictors and risk models [8]. One of the greatestdvantages of bootstrap in circumstances in which thevents are rare is to have the possibility to use the entireohort of patients to recalibrate the model and obtainore robust estimates.To our knowledge there is no similar specific cardiac

isk score in thoracic surgery. Several other scores haveeen developed and tested in this setting, but they

ncorporated cardiac as well as pulmonary morbidity,ncluding also minor cardiac complications such as atrialbrillation [9–12].Other more sophisticated noninvasive or invasive tests

r investigations can be used in clinical practice toredict the occurrence of major cardiac events. However,

n this cost-conscious era we are asked to allocate and useesources appropriately. This recalibrated score may rep-esent a cost-effective instrument to screen those patientsho really need more sophisticated, high-tech cardiac

xamination (ie, patients in classes C and D).

Although this analysis was performed in a homoge-eous sizable group of patients undergoing major pul-onary resections applying rigorous statistical methods,

t may have potential limitations.Although in line with that reported in the original

aper, limited to the thoracic population [1, 8], theumulative incidence of major cardiac complications inur series was relatively low (3.3%). This low incidenceould be influenced by uncontrollable patient selectionias. The present findings need therefore to be confirmedy other independent studies with larger populationsnd consequently a higher number of events.The retrospective and multicenter nature of the studyay entail inherent problems of variable and outcomes

efinitions and recording. This is important to noticearticularly because the original RCRI was developedrospectively [3] and methodological differences mayave played a role in determining differences between

he models. However, given the severity of the cardiacnd points chosen as outcomes, it would appear unlikelyhat these types of events were overlooked in this studynd were not prospectively recorded in the patients’ecords. Certainly, an independent prospective studyould be desirable to confirm these results.The recalibration of the score was performed on pa-

ients undergoing major pulmonary resections (lobec-omy or pneumonectomy). The generalization of thehRCRI to other thoracic procedures (including minor

ung resections) needs to be verified by othernvestigations.

The score has been developed and recalibrated toredict the risk of major cardiac events. Its use forredicting the occurrence of other frequent cardiac com-lications such as atrial fibrillation is not warranted andequires specific evaluation.

The cardiac morbidity rates were not different amongatients in classes C and D. If future independent studiesonfirm this finding, the scoring system could be furtherimplified by collapsing these two classes into one. Thisas not done in the present study as we wanted toaintain the original four-class structure of the RCRI.In conclusion, we were able to recalibrate the RCRI inlarge series of major pulmonary resections. Although

he present findings need to be confirmed by otherndependent prospective investigations, we think they

ay warrant the use of the ThRCRI as a first-line screen-ng instrument during cardiologic risk stratification forelecting those patients needing further cardiologic test-ng from those who can proceed with pulmonary evalu-tion without any further cardiac tests.

he authors wish to thank Stephen D. Cassivi, MD, from Mayolinic, MN, for his help in reviewing this work.

eferences

1. Fleisher LA, Beckman JA, Brown KA, et al. ACC/AHA 2007guidelines on perioperative cardiovascular evaluation and

care for noncardiac surgery: a report of the American Col-lege of Cardiology/American Heart Association Task Force

1

1

1

D

Dw

D

Ddma1

i

fgi

Dp

l

203Ann Thorac Surg BRUNELLI ET AL2010;90:199–203 RECALIBRATED REVISED CARDIAC RISK INDEX

ENER

AL

TH

OR

AC

IC

on Practice Guidelines (Writing Committee to Revise the2002 Guidelines on Perioperative Cardiovascular Evaluationfor Noncardiac Surgery). Circulation 2007;116:e418–99.

2. Poldermans D, Bax JJ, Boersma E, et al. Guidelines forpre-operative cardiac risk assessment and perioperativecardiac management in non-cardiac surgery: the Task Forcefor Preoperative Cardiac Risk Assessment and PerioperativeCardiac Management in Non-cardiac Surgery of the Euro-pean Society of Cardiology (ESC) and endorsed by theEuropean Society of Anaesthesiology (ESA). Eur Heart J2009;30:2769–812.

3. Lee TH, Marcantonio ER, Mangione CM, et al. Derivationand prospective validation of a simple index for prediction ofcardiac risk of major noncardiac surgery. Circulation 1999;100:1043–9.

4. Brunelli A, Charloux A, Bolliger CT, et al. ERS/ESTS clinicalguidelines on fitness for radical therapy in lung cancerpatients (surgery and chemo-radiotherapy). Eur Respir J2009;34:17–41.

5. Eagle KA, Berger PB, Calkins H, et al. ACC/AHA guidelineupdate for perioperative cardiovascular evaluation for non-cardiac surgery—executive summary: a report of the Amer-ican College of Cardiology/American Heart Association

Task Force on Practice Guidelines (Committee to Update the

imited to the thoracic population in the old one, may be quite

ddad

Daf

ewt

Dtciti

1996 Guidelines on Perioperative Cardiovascular Evaluationfor Noncardiac Surgery). J Am Coll Cardiol 2002;39:542–53.

6. Blackstone EH. Breaking down barriers: helpful break-through statistical methods you need to understand better.J Thorac Cardiovasc Surg 2001;122:430–9.

7. Grunkemeier GL, Wu YX. Bootstrap resampling method:something for nothing? Ann Thorac Surg 2004;77:1142–4.

8. Brunelli A, Rocco G. Internal validation of risk models inlung resection surgery: bootstrap versus training and testsampling. J Thorac Cardiovasc Surg 2006;131:1243–7.

9. Epstein SK, Faling LJ, Daly BD, et al. Predicting complica-tions after pulmonary resection. Preoperative exercise test-ing vs a multifactorial cardiopulmonary risk index. Chest1993;104:694–700.

0. Birim O, Maat AP, Kappetein AP, et al. Validation of theCharlson Comorbidity Index in patients with operated pri-mary non-small cell lung cancer. Eur J Cardiothorac Surg2003;23:30–4.

1. Brunelli A, Fianchini A, Xiume F, et al. Evaluation of thePOSSUM scoring system in lung surgery. Physiological andoperative severity score for the enumeration of mortality andmorbidity. Thorac Cardiovasc Surg 1998;46:141–6.

2. Ferguson MK, Durkin AE. A comparison of three scoringsystems for predicting complications after major lung resec-

tion. Eur J Cardiothorac Surg 2003;23:35–42.

G

ISCUSSION

R PAUL HENRY SCHIPPER (Portland, OR): I think diabetesas one of those things that fell out of the model?

R BRUNELLI: Diabetes and congestive heart failure, yes.

R SCHIPPER: I’m wondering, that first cardiac model that waseveloped before and published in 1999, and your model isuch more recent. At our institution we have instituted fairly

ggressive insulin protocols and this has all happened since999.Number one, do you have such an insulin protocol at your

nstitution?And, number two, do you think that’s maybe why diabetes has

allen out of this risk model? That we are managing the bloodlucose much better and therefore diabetes has been neutral-

zed so to speak as a cardiac risk?

R BRUNELLI: We don’t have any standardized diabetesrotocol.I agree with you. I think the case mix between the two studies,

ifferent. I think we see now much more patients with insulin-ependent diabetes. Perhaps we are a bit more careful inssessing them in the postoperative period even though weon’t have a standardized pathway in this regard.

R SETH D. FORCE (Atlanta, GA): Usually I just get a dobut-mine stress echo on pretty much every patient with cardiac riskactors, going to surgery. So I found this very interesting.

Did you go back and look at the type of diagnostic cardiacvaluations these patients had presurgery to find out which onesere useful in predicting these events that you saw, especially in

he patients in the higher risk group?

R BRUNELLI: We just wanted to recalibrate the six factors ofhe original RCRI score. We didn’t assess additional factors. Ofourse, it is possible to integrate this score with other variables,ncluding, for example, ergometric parameters. And certainlyhis may be a subject of future possibly prospective

nvestigations.