Useful Predictors of Bile Duct Stones Patients Undergoing ...

of 8 /8
ANNALS OF SURGERY Vol. 220, No. 1, 32-39 © 1994 J. B. Lippincott Company Useful Predictors of Bile Duct Stones in Patients Undergoing Laparoscopic Cholecystectomy Alan N. Barkun, M.D., Jeffrey S. Barkun, M.D., Gerald M. Fried, M.D., Gabriella Ghitulescu, M.D., Oren Steinmetz, M.D., Christine Pham, M.D., Jonathan L. Meakins, M.D., D.Sc., Carl A. Goresky, M.D., Ph.D., and the McGill Gallstone Treatment Group From the Divisions of Gastroenterology, General Surgery, and the Department of Radiology, McGill University, Montreal, Quebec, Canada Objective The authors determined the most useful predictors of common bile duct (CBD) stones as diagnosed by endoscopic retrograde cholangiopancreatography (ERCP) in patients who underwent laparoscopic cholecystectomy (LC). Methods Prospective and retrospective collection of historical, biochemical and ultrasonographic data was used. Receiver operating characteristics curve analysis was used to determine optimal biochemical cut-off values. Multivariate analysis using logistic regression with generation of the best model identifying independent predictors of CBD stones also was employed. Prospective validation of the model was performed on an independent group of patients. Results Endoscopic retrograde cholangiopancreatographies were performed before LC in 106 patients, and after LC in 33. Only four of ten clinical variables evaluated independently predicted the presence of CBD stones. The optimal model predicted a 94% probability of CBD stones in a patient older than 55 years of age who presented with an elevated bilirubin (over 30 ,umol/L) and positive ultrasound findings (a dilated CBD, and a CBD stone seen on ultrasound). This model was validated prospectively in a subsequent series of 49 patients in which the probability of CBD stone was only 8% when all four predictors were absent. Conclusions The identified independent clinical predictors of a CBD stone helps select a population of symptomatic gallstone bearers who benefit most from cholangiographic assessment. Laparoscopic cholecystectomy (LC) has become the now treated this way.2 However, 3% to 33% of all patients new therapeutic gold standard' in uncomplicated symp- with symptomatic gallstones may bear associated com- tomatic cholelithiasis, and at least 80% of all patients are mon bile duct (CBD) stones.3-7 The optimal approach to Presented in part at the Scientific Session of the Annual Meeting of the Address reprint requests to Alan Barkun, M.D., Room D7. 148, The Society of American Gastrointestinal Endoscopic Surgeons, Phoe- Montreal General Hospital, Montreal, Quebec, Canada, H3G 1 A4. nix, Arizona, April 1-3, 1993. Accepted for publication August 12, 1993. 32

Transcript of Useful Predictors of Bile Duct Stones Patients Undergoing ...

Page 1: Useful Predictors of Bile Duct Stones Patients Undergoing ...

ANNALS OF SURGERYVol. 220, No. 1, 32-39© 1994 J. B. Lippincott Company

Useful Predictors of Bile Duct Stonesin Patients Undergoing LaparoscopicCholecystectomyAlan N. Barkun, M.D., Jeffrey S. Barkun, M.D., Gerald M. Fried, M.D., Gabriella Ghitulescu, M.D.,Oren Steinmetz, M.D., Christine Pham, M.D., Jonathan L. Meakins, M.D., D.Sc.,Carl A. Goresky, M.D., Ph.D., and the McGill Gallstone Treatment Group

From the Divisions of Gastroenterology, General Surgery, and the Department of Radiology,McGill University, Montreal, Quebec, Canada

ObjectiveThe authors determined the most useful predictors of common bile duct (CBD) stones asdiagnosed by endoscopic retrograde cholangiopancreatography (ERCP) in patients whounderwent laparoscopic cholecystectomy (LC).

MethodsProspective and retrospective collection of historical, biochemical and ultrasonographic data wasused. Receiver operating characteristics curve analysis was used to determine optimalbiochemical cut-off values. Multivariate analysis using logistic regression with generation of thebest model identifying independent predictors of CBD stones also was employed. Prospectivevalidation of the model was performed on an independent group of patients.

ResultsEndoscopic retrograde cholangiopancreatographies were performed before LC in 106 patients,and after LC in 33. Only four of ten clinical variables evaluated independently predicted thepresence of CBD stones. The optimal model predicted a 94% probability of CBD stones in apatient older than 55 years of age who presented with an elevated bilirubin (over 30 ,umol/L) andpositive ultrasound findings (a dilated CBD, and a CBD stone seen on ultrasound). This modelwas validated prospectively in a subsequent series of 49 patients in which the probability of CBDstone was only 8% when all four predictors were absent.

ConclusionsThe identified independent clinical predictors of a CBD stone helps select a population ofsymptomatic gallstone bearers who benefit most from cholangiographic assessment.

Laparoscopic cholecystectomy (LC) has become the now treated this way.2 However, 3% to 33% ofall patientsnew therapeutic gold standard' in uncomplicated symp- with symptomatic gallstones may bear associated com-tomatic cholelithiasis, and at least 80% of all patients are mon bile duct (CBD) stones.3-7 The optimal approach to

Presented in part at the Scientific Session of the Annual Meeting ofthe Address reprint requests to Alan Barkun, M.D., Room D7. 148, TheSociety of American Gastrointestinal Endoscopic Surgeons, Phoe- Montreal General Hospital, Montreal, Quebec, Canada, H3G 1 A4.nix, Arizona, April 1-3, 1993. Accepted for publication August 12, 1993.

32

Page 2: Useful Predictors of Bile Duct Stones Patients Undergoing ...

CBD Stones and Laparoscopic Cholecystectomy 33

patients with suspected choledocholithiasis undergoingLC remains uncertain8 and may vary according to localexpertise.9 Endoscopic retrograde cholangiopancreatog-raphy (ERCP) before LC should not be performed sys-tematically on all patients because ofthe associated mor-bidity and low yield.9 Further scrutiny of the usefulnessof CBD stone predictors is, therefore, timely. In June1990, when LC was first introduced in the McGill Uni-versity teaching institutions, a common approach to pa-tients scheduled to undergo LC was adopted i.e., all pa-tients suspected of having a CBD stone were referred forpreoperative ERCP, with attempts at endoscopic sphinc-terotomy when choledocholithiasis was confirmed.'0This practice permitted us to study the usefulness ofdifferent clinical predictors of CBD stones diagnosed byERCP in a homogenous patient population. An optimalmodel ofCBD stone prediction was determined by logis-tic regression and validated prospectively in a subse-quent series of patients undergoing LC.

METHODSPatient PopulationThe study population consisted of patients who were

scheduled for or underwent LC, in whom ERCP was per-formed at one of four McGill University teaching hospi-tals between June 1990 and February 1992. PreoperativeERCP was performed if bile duct stones were suspected.Indications for ERCP included a history ofjaundice orpancreatitis; elevations in serum bilirubin, alkaline phos-phatase, gamma glutamyl transferase, aminotransfer-ases, and amylase; and ultrasound findings of a dilatedCBD or suspicion of a CBD stone. The criteria used forperforming ERCPs after LC were similar to thoseadopted in the preoperative setting, but also included theintraoperative findings of a dilated cystic duct or CBDstone. In addition to these, postoperative indications forERCP included the clinical suspicion of a bile leak orbile duct injury. Although not an indication for ERCP,ultrasonographic detection of multiple small gallbladderstones was noted and analyzed as a possible predictor ofCBD stones.

Data CollectionAll data were obtained from prospective surgical and

endoscopic databases. When necessary, data also werecollected prospectively and retrospectively from hospitalcharts, endoscopic reports, abdominal ultrasound, andERCP films. Age, gender, and a past history ofjaundiceor pancreatitis were recorded. Biochemistry results wereconsidered abnormal for a patient with any ofthe follow-ing: bilirubin > 17 Amol/L, alkaline phosphatase > 110

units/L, gamma glutamyl transferase (GGT) > 55 units/L, aminotransferases (aspartate transaminase [AST], al-anine transaminase [ALT]) > 40 units/L, and amylase >185 units/L. The highest serum levels within the 10 daysbefore the ERCP examination were recorded for each pa-tient and used for the analysis. Ultrasound studies wereperformed before ERCP. The common bile duct diame-ter (in millimeters) was measured in its mid-portion. Inaddition, a subjective assessment by a blinded examinerwas used to determine whether or not there was CBDdilatation considering only the patient's age and theoverall appearance of the biliary tree. The suspicion orpresence of bile duct stones seen on ultrasound was re-corded as was the presence of small, numerous gallblad-der stones. The ultrasound reports and films were subse-quently reviewed and standardized by an investigator(CP) blinded to the results of the ERCP examinations.Endoscopic retrograde cholangiopancreatography find-ings, with respect to the presence of bile duct stones andother biliopancreatic pathology, were reported by a cer-tified radiologist. Endoscopic retrograde cholangiopan-creatography films and reports were later reviewed inde-pendently by a blinded investigator (GG).

Statistical AnalysisThe corresponding proportion and total number of

patients in whom the information was available were ex-pressed for each clinical predictor. The unit of analysiswas a patient, not an ERCP examination. For patientswho had more than one ERCP, only the variables notedbefore the patient's first ERCP and the diagnosis at thedefinitive ERCP were used for the analysis. When dis-tributed normally, continuous variables were describedusing mean and standard deviation; if not, they were ex-pressed as median and range. Age was analyzed both as acontinuous and a categorical variable using the observedmedian of the patient population as a cut-off. Test char-acteristics were determined for all clinical parameters,including sensitivity, specificity, and positive and nega-tive predictive values. The ability of a biochemical testalone to differentiate between the absence and presenceof a bile duct stone was assessed using receiver operatingcharacteristics (ROC) curve analysis.' 1-13 An ROC curvedisplays the true-positive rate on the Y axis and the false-positive rate on the X axis for varying test thresholds,thus plotting the performance of a diagnostic test. 12 Foreach biochemical test, five possible cut-offs were chosenafter examination of its observed quartile values in thestudy population. The ideal cut-off for each biochemicaltest was chosen by determining the point closest to anideal test with 100% specificity and sensitivity (the upperleft most corner ofthe graph). 13 Thereafter, the biochem-ical tests were analyzed in multivariate analysis using the

Vot 220 - No. 1

Page 3: Useful Predictors of Bile Duct Stones Patients Undergoing ...

34 Barkun and Others

ideal cut-offs determined at ROC analysis. Common bileduct diameter at ultrasonography was analyzed categor-ically using the subjective assessment ofenlargement anda cut-off value of >6 mm,'4,'5 with no difference in re-sults. Chi square testing, or Fisher's exact test (where ap-propriate),'6 were used to determine those clinical vari-ables that were associated with the presence of a CBDstone, when examined individually. Multiple stepwiselogistic regression then was used to construct the modelthat would best predict the presence of a CBD stone atERCP, considering all clinical predictors together.Goodness of fit ofthe model was assessed using the Pear-son's chi square test.'7 Then a probability tree was devel-oped to show the probability ofCBD stones according tothe presence or absence of the independent predictorsidentified. Prospective validation of the logistic regres-sion model was performed on a subsequent series of pa-tients scheduled for LC-who first underwent ERCP-by comparing their observed CBD stone prevalence withthat estimated by the model using chi square testing.'7 Inaddition, the parameter coefficients of the initial modelwere compared with those obtained after adding the dataof the subsequent series of patients.

RESULTSPatient Characteristics

The prospective McGill Laparoscopic Cholecystec-tomy Registry identified 1300 patients undergoing LC atfour McGill University teaching hospitals during the 18-month study period.'0 During this time, 127 ERCP ex-aminations were performed before LC in 106 patients(mean = 1.2 ERCP per patient). After LC, 49 ERCPswere performed in 33 patients (mean = 1.5 ERCP perpatient). All patients who underwent ERCP examina-tions were included in the analysis. The cohort mean agewas 54.9 ± 17.7 years (range = 17-91 years), and 63%of the study population were women (87 patients). Twopatients required ERCP examinations before and aftersurgery (preoperative ERCP examinations were unsuc-cessful in one patient, and falsely negative in the other).During the study period, the intraoperative cholangiog-raphy rate was 4%. ' Further details on the outcome ofpatients who underwent intraoperative cholangiographyare discussed elsewhere."'

Description of Clinical PredictorsThe mean ages were 57.5 ± 16.8 years in the preoper-

ative and 48.1 ± 15.2 years in the postoperative patientpopulations. The frequency of the different predictorsboth preoperatively and postoperatively are displayed in

Table 1. PATIENTS WITH ABNORMALCLINICAL PREDITORS

Patients(Proportion of Patients)

Clinical Predictor Preoperative Postoperative

Age (>55 yrs)Female GenderHistory of jaundiceHistory of pancreatitisBilirubin > 17,mol/LAlkaline phosphatase > 110 units/LGGT > 55 units/LAST > 40 units/LALT > 40 units/LAmylase > 185 units/LDilated CBD on U/SCBD stone on U/SMultiple, small gallbladder stones

66 (63%)66 (62%)42 (41%)18 (12%)33 (33%)69 (75%)47 (85%)75 (77%)74 (77%)24 (28%)39 (39%)22 (22%)77 (80%)

12 (36%)21 (64%)8 (26%)4 (14%)15 (56%)16 (62%)16 (84%)21 (81%)20 (80%)17 (68%)10 (38%)7 (25%)

GGT = gamma glutamyl transferase; AST = Aspartate aminotransferase; ALT =

Alanine aminotransferase; CBD = common bile duct; U/S = abdominal ultrasonog-raphy.Dilated CBD on ultrasound was reported as such or over 6 mm in diameter.

Table 1. Postoperative indications for ERCP included asuspected retained stone in 25 patients (76%), and suspi-cion of a bile leak or CBD injury in 8 (24%). Mean eleva-tions of biochemical predictors were 96 ± 130 ,umol/Lfor bilirubin, 243 + 143 units/L for alkaline phospha-tase, 307 ± 185 units/L for gamma glutamyl transferase,258 ± 261 units/L for AST, 308 ± 264 units/L for ALT,and 1484 ± 1841 units/L for amylase. The intraoperativefinding of a dilated cystic duct could not be analyzed as apost-LC indication for ERCP because of missing or in-complete data. The time interval between the laboratoryor imaging results and the initial ERCP examinationranged from 1 to 10 days for all patients.

ERCP FindingsCommon bile duct stones were found in 50 of 106

(47%) patients preoperatively and 17 of 33 (52%) postop-eratively (including 15 ofthe 25 [60%] patients in whomCBD stones were suspected post-LC). Endoscopic retro-grade cholangiopancreatography did not alter patientmanagement. Among the preoperative ERCPs, addi-tional pancreaticobiliary information was obtained infive patients (15%). Pancreas divisum was diagnosed inthree patients, and unsuspected chronic pancreatitischanges were found in three. In one of the patients withpancreas divisum, changes of chronic pancreatitis werefound in the dorsal duct. Postoperative ERCP diagnosisof leaks and CBD injury are detailed elsewhere.'8

Ann. Surg. *-July 1994

Page 4: Useful Predictors of Bile Duct Stones Patients Undergoing ...

CBD Stones and Laparoscopic Cholecystectomy 35

.8

> 30cf 0 6zwca 50CD> 0.4 70- ..........................................

0

tL o2 0

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

FALSE POSITIVES (1- SPECIFICITY)Figure 1. Determination of the optimal bilirubin serum cut-off value usingROC curve analysis. The sensitivity and specificity of different bilirubin cut-off values (30, 50, 70, 90, 100 umol/L) were plotted. The optimal cut-offwas 30 Mmol/L (arrow) because it is the point of the curve lying closest toan ideal test with 100% specificity and sensitivity (the upper left most cor-

ner of the graph).

Determination of Optimal Serum Cut-OffValues for Biochemical Predictors

Using ROC curve analysis, optimal predictive serumcut-off levels were determined to be 30 ,umol/L for bili-rubin (Fig. 1), 500 units/L for amylase, 120 units/L forAST (similar results were found for ALT, which was

available in fewer patients and, therefore, not studiedfurther), and 300 units/L for alkaline phosphatase (Fig.2) (GGT was not studied further because it was missingin many patients because the biochemical laboratories ofparticipating institutions measure this serum value onlyif the alkaline phosphatase is elevated).

Accuracy of the Clinical PredictorsTable 2 details the performance of each possible pre-

dictor of CBD stone analyzed, including individual sen-sitivity, specificity, and positive and negative predictivevalues. When considered individually, the significantpredictors ofcholedocholithiasis included hyperbilirubi-nemia, a dilated CBD on ultrasonography, and the ultra-sonographic suspicion of a CBD stone (p < 0.01). A his-tory of pancreatitis and hyperamylasemia were associ-ated significantly with the absence ofa CBD stone.

Determination of the Multivariate ModelUsing logistic regression, the best model for predicting

bile duct stones at ERCP included the following inde-pendent predictors: age (continuous or dichotomized,

1 -m-Amylase*Alkaline phosphatase*AST

.8

C,)

wco

COwg04 120 MMM

co00.2 * ...........................

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

FALSE POSITIVES (1- SPECIFICITY)Figure 2. Determination of the optimal biochemical serum cut-off valuesfor alkaline phosphatase, AST, and amylase using ROC curve analysis.The sensitivity and specificity of different cut-off values for alkaline phos-phatase (150, 300, 450, 600, 750 units/L), AST (80, 120, 160, 200, 240units/L), and amylase (300, 500, 700, 900, 1100 units/L) were plotted,leading to three ROC curves. The optimal cut-off for each test was deter-mined by the point of the respective ROC curve lying closest to an idealtest with 100% specificity and sensitivity (the upper left most corner of thegraph). Only the ideal cut-offs for each test are shown and include 300units/L for alkaline phosphatase, 120 units/L for AST, and 500 units/L foramylase.

using a cut-offof 55 years), an elevated bilirubin (over 30gmol/L), the presence ofa dilated CBD on ultrasonogra-phy (> 6 mm), and suspected or detected bile duct stone(at ultrasonography). The respective coefficients for eachof these were 0.02 (0.88 when dichotomized), 0.83, 0.86,and 1.80. The choice of predictors did not change,whether the post-LC patients were excluded or includedin the analysis. The two ultrasonographic predictors ex-hibited the largest coefficients and thus, were the strong-est predictors of choledocholithiasis. Depending on thepresence or absence of these four parameters, the modelyielded probabilities offinding a bile duct stone at ERCPthat ranged from 18% (no predictor present), to 94% (allfour predictors present). When a stone was seen on ultra-sound, the overall probability offinding a stone at ERCPranged from 58% to 94%, depending on the presence orabsence of the other criteria. Figure 3 shows the set ofprobabilities expected according to the presence or ab-sence ofclinical predictors for patients in whom no CBDstone was noted on ultrasound.

Prospective Validation of the Model ThatPredicts Choledocholithiasis at ERCPThe multivariate model was validated prospectively in

a subsequent consecutive series of 49 patients suspectedof having CBD stones and scheduled for LC. Common

Vol. 220 - No. 1

JOL

Page 5: Useful Predictors of Bile Duct Stones Patients Undergoing ...

36 Barkun and Others

Table 2. PERFORMANCE OF CLINICAL PREDICTORS IN UNIVARIATE ANALYSIS

Predictive values

Predictor Sensitivity Specificity Positive Negative

Age (>55)Female genderHistory of jaundiceHistory of pancreatitistBilirubin > 30 pmol/L*Alkaline phosphatase > 300 units/LAST > 120 units/LAmylase > 500 units/LtDilated CBD*CBD stone U/S*Multiple, small gallbladder stones

57%57%43%7%74%79%81%28%53%36%83%

49%32%66%75%48%35%25%51%73%90%29%

51%43%54%20%57%53%49%33%64%78%52%

56%46%55%47%66%65%60%45%64%58%64%

GGT = gamma glutamyl transferase; AST = Aspartate aminotransferase; ALT = Alanine aminotransferase; CBD = common bile duct; U/S = abdominal ultrasonography.The optimal biochemical values used as cut-offs were determined by ROC curve analysis (see Figures 1 and 2).Although not an indication for pre-operative ERCP, the presence of multiple small gallbladder stones was examined as a possible predictor of bile duct stone at ERCP. DilatedCBD on ultrasound was reported as such or over 6 mm in diameter.* Significant predictors of a CBD stone (p < 0.01).t Significant predictors of the absence of a CBD stone on univariate analysis (p < 0.01).

bile duct stones were found at ERCP in only 1 of 12 (8%)patients in whom none of the four predictors were pres-ent, in contrast to 10 of 15 (66%) in whom two or morepredictors were present. There were no significantdifferences between the observed and predicted probabil-ities in the subgroups of this patient population when

broken down according to the different possible predic-tor combinations. In addition, the parameter coefficientsof the model did not vary significantly when the datafrom the initial patients and subsequent 49 patients werecombined.

AGE OVER ELEVATED CBD DILATATION55 BILIRUBIN ON ULTRASOUND

(>30,pmoi/L)

=(+)

(-)+)

(-)

(-)

(-)

(+)

(-)

(+)

PROBABILITY OFCBD STONEAT ERCP

72%

50%

61%

38%

49%

28%

38%

19%Figure 3. Probabilities of CBD stone at ERCP depending on the pres-ence or absence of three independent predictors identified using logisticregression analysis for patients in whom ultrasonography does not dem-onstrate a CBD stone. The probabilities vary between 58% and 94% whena stone is seen at ultrasonography.

DISCUSSIONThe advent of LC has, rightly or wrongly, rekindled

the debate of the optimal management of patients withsuspected CBD stones who undergo cholecystec-tomy.1920 One reason has been the desire to maintainthe benefits of minimally invasive surgery. In addition,laparoscopic techniques to clear the CBD are evolving,21but require a commitment of time, resources, and spe-cialized equipment that is not widely available at pres-ent. The morbidity and success of many of these tech-niques have not been defined clearly or compared inlarge series to endoscopic sphincterotomy (before, dur-ing, or after LC) or open common bile duct explora-tion.2229 Predictors of CBD stones previously had beenstudied extensively in the era of open cholecystectomy.The reports, however, remained disparate in their con-clusions, with different sets ofpredictors performing var-iably.30 This inconsistency can be attributed to many fac-tors, including differences in patient selection, differ-ences in predictors studied and their timing with respectto the diagnosis of CBD stones, differences in gold stan-dards used to detect the presence of choledocholithiasis,and the use of poorly adapted statistical techniques towhat are often highly intercorrelated clinical predictors.

Ann. Surg. *-July 1994

Page 6: Useful Predictors of Bile Duct Stones Patients Undergoing ...

CBD Stones and Laparoscopic Cholecystectomy 37

Some authors examined only preoperative predictors,and others examined a combination of preoperative andintraoperative variables. Furthermore, many ofthe stud-ies were carried out before the availability of modem ul-trasonographic equipment.

All studies were performed before the era ofLC-per-haps an important consideration when taking into ac-count the recently reported prevalence of CBD stonesthat appears to have dropped since the advent of LC.3'32This decline may be real, possibly reflecting a differentpatient population that presents earlier in the course ofsymptomatic gallstone disease, or may be artificial andattributable to decreased detection.3'Common bile duct stones are found in 3% to 9% of

gallstone patients who have never been jaundiced.20'33-35No patient with normal liver function tests was foundwith choledocholithiasis in two series totalling over 400patients using intraoperative cholangiography (IOC) per-formed on patients at low risk for carrying CBDstones.3637 Saltztein et al. demonstrated that only thecombination of an elevated bilirubin and alkaline phos-phatase within the 14 days before open cholecystectomywould significantly predict a CBD stone.38 Many studieshave found ultrasonographic criteria to be necessary forthe generation ofa good predictive model39-4' because oftheir stronger predictive abilities when compared withother preoperative clinical criteria.4' Although sugges-tive ofcholedocholithiasis,42 a dilated CBD alone may bean unreliable indicator,43 perhaps because of the abilityof the CBD diameter to change over a short period oftime.44'45 Combining the CBD diameter with other ab-normalities, such as an elevated bilirubin, increases itspositive predictive value, perhaps by reflecting ongoingbiliary obstruction.'5 Conversely, CBD diameter lessthan 3 mm almost rules out a CBD stone.46 Lacaine etal. noted that a CBD diameter over 12 mm, with a raisedalkaline phosphatase, minimally raised bilirubin, or his-tory ofjaundice gave a 90% to 100% predictive value ofCBD stones.39 In their series, patients with none ofthesesymptoms exhibited a less than 5% risk of choledocholi-thiasis. They also reported, as others have,'5'4' a signifi-cantly older mean age in the patients with choledocholi-thiasis, but did not otherwise examine this finding. Lim-itations of this study included the exclusion of allpatients with bilirubin levels greater than 3 mg/dL, theabsence of multivariate modelling, and the unknown er-ror for each subgroup point estimate. Wilson et al., usingIOC, suggested a similar predictive index that includedelevated levels ofalkaline phosphatase or bilirubin in thepast 6 months, an abnormal CBD, and a dilated cysticduct at surgery.43 Taylor et al., using multivariate analy-sis, correctly classified 90% of patients when consideringonly a dilated CBD at intraoperative cholangiographyand the presence of numerous small gallbladder stones

from a list of 36 possible preoperative and intraoperativevariables.4' Hauer-Jensen et al. observed 393 patients for6 to 8 years after surgery.40 Logistic regression recognized5 of 11 possible clinical criteria as independent, signifi-cant predictors. The three preoperative variables wereage, hyperbilirubinemia, and the presence of a CBDstone on preoperative imaging. A dilated cystic duct anda palpable CBD stone were the two intraoperative pre-dictors also identified. However, no prospective valida-tion of the model was carried out. Alkaline phosphatasedid not add independent preoperative prediction beyondthat of bilirubin and ultrasonography, yet the authorschose to keep this variable in their model because ofother groups' findings.39'4' Positive and negative predic-tive values of the model were 28% to 36% and 97% to100%, respectively.When interpreting the performance of the models de-

veloped, it must be remembered that intraoperative cri-teria carry the strongest predictive abilities.4' 42'47'48 Un-fortunately, their availability may have decreased in theera of LC (inability to palpate a CBD stone at laparos-copy). The model generated in the present study, whichexamines no intraoperative variables, includes all threepreoperative predictors determined in the study byHauer-Jensen40 and ultrasonographic CBD diameter,found by many groups to be useful.39-4' The superiorpredictive abilities of ultrasonographic criteria also areconfirmed.4' Hyperamylasemia and a history of pancre-atitis were not significant predictors of a CBD stone, asreported previously.'5'40'49 Some have suggested furtherthat hyperamylasemia may lower the predictive abilityof other tests.38 Clinical series suggest that unsustainedhyperamylasemia reflects recent stone migration acrossthe sphincter of Oddi.50'5' Moreover, patients with pan-creatitis are at high risk for further episodes during thenext few months.52'53 Thus, hyperamylasemia may bestbe used as a predictor of recurrent pancreatitis ratherthan as a predictor of CBD stone discovery.52'53 In pa-tients presenting with symptoms other than pancreatitis,the timing of laboratory findings in relationship to cho-langiography and surgery remains poorly studied. None-theless, it is plausible that the shorter the time intervalbetween biochemical abnormalities and cholangiogra-phy, the higher the positive prediction while the likeli-hood of subsequent stone migration across the sphincterof Oddi (a cause for false-positive prediction) decreases.The time intervals in our study between the onset ofacute symptoms, laboratory data sampling, and ERCP(1-10 days) reflect real life conditions and make the ob-served predictors' performance useful, as reported. Thepotential for further gallbladder stone migration is pres-ent, however, and consideration must be given to the op-timal timing of LC after a negative ERCP. Certaingroups have recommended that this interval not exceed

Vol. 220 - No. 1

Page 7: Useful Predictors of Bile Duct Stones Patients Undergoing ...

38 Barkun and Others

24 to 48 hours.23'25 These suggestions are based on asmall number ofanecdotal observations, and our experi-ence in 106 patients does not support such recommen-dations.'0

Strengths of the present study lie in the use of multi-variate modelling and the assessment of a homogenouspatient population undergoing LC. This may differ fromother endoscopic series, which might also include the re-sults of ERCP examinations performed in patients longafter open cholecystectomy. Such a population subsetmay not be biologically similar with respect to CBD dis-tensibility and enzyme elevation in response to transientor persistent obstruction by retained or recurrent stones.The present study results are limited by partial relianceon retrospective data and the short follow-up time. Thismay have resulted in an underestimation ofthe true neg-ative predictive value because ERCP was not, and shouldnot be performed routinely on all patients preopera-tively.9'28 An approximation of the true-negative predic-tive value in the absence of long-term follow-up can bederived only from studies using routine intraoperativecholangiography. Indeed, only then does the study pop-ulation include patients exhibiting no predictors whatso-ever. The strength ofthe present model, therefore, lies inits ability to predict the presence of a CBD stone, not itsabsence.Cotton et al. and others have proposed different diag-

nostic or therapeutic approaches for patients stratifiedaccording to varying likelihoods ofcarrying a CBD stonebefore LC.954 In the present series, based on widely usedpreoperative criteria, the overall positive predictive valueof a preoperative ERCP approach was 47%, a perfor-mance comparable to those of other endoscopic seriesusing similar selection criteria (32-61%22-26). The modelwe have generated predicts a CBD stone with a probabil-ity ranging from 19% to 38%, in which none or only onecriterion is present, to 49% to 94%, in which three or fourcriteria are fulfilled. A selection based on these predictorsshould help tailor an appropriate cholangiographic ap-proach, given an individual's risk of CBD stone. Thepresent study design does not permit the identification ofan optimal method of cholangiography (ERCP pre-LCor post-LC vs. IOC) because it fails to examine other im-portant considerations, such as the morbidity ofthe pro-cedure, the implications of a false-positive cholangio-graphic examination, or a determination of the impactof undetected CBD stones. More precise and unbiasedquantifications of these associated factors, which mayvary from center to center according to local expertise,9are required urgently. Until then, based on this reportand our previously published experience,10 we wouldrecommend that centers with similar available expertiseperform preoperative ERCP in patients exhibiting oneor more of the aforementioned strong predictors (other

Ann. Surg. -July 1994

than age > 55 years alone). Patients with "weaker" pre-dictors might best be investigated with other methods,such as IOC.55 The lack of pertinent, comparable dataunderscores the importance of, and need for long-termprospective trials to better define the optimal roles ofdifferent approaches to patients suspected of choledo-cholithiasis who undergo LC.

AcknowledgmentsThe authors thank Christine Wickham and Diane Thibeault for their

assistance in data collection, and Mary Muccino for her assistance inthe preparation of the manuscript.

References1. Barkun JS, Barkun AN, Sampalis JS, et al. Laparoscopic versus

mini-cholecystectomy: a randomized controlled trial. Lancet1992; 2:1116-1119.

2. Gallstones and laparoscopic cholecystectomy: NIH consensus de-velopment panel on gallstones and laparoscopic cholecystectomy.JAMA 1993; 269:1018-1024.

3. Glenn F, McSherry CK. Calculous biliary tract disease. In RavitchMM, ed. Current Problems in Surgery. Chicago: Yearbook Pub-lishers, 1975, pp 1-38.

4. Crump C. The incidence ofgallstones and gallbladder disease. SurgGynecol Obstet 1931; 53:447-457.

5. Lieber MM. Incidence of gallstones and their correlation withother diseases. Ann Surg 1952; 135:394-405.

6. Isch JH. Operative cholangiography: the case for broadened use inbiliary tract surgery. Arch Surg 1973; 107:106- 110.

7. Arnold DJ. 28,621 cholecystectomies in Ohio: results of a surveyin Ohio hospitals by the Gallbladder Survey Committee, OhioChapter, American College of Surgeons. Am J Surg 1970; 119:714-717.

8. McEntee G, Grace PA, Boucheir-Hayes D. Laparoscopic cholecys-tectomy and the common bile duct. Br J Surg 1991; 78:385-386.

9. Cotton PB, Baillie H, Pappas TN, Meyers WS. Laparoscopic cho-lecystectomy and the biliary endoscopist. Gastrointest Endosc1991; 37:94-97.

10. Barkun JS, Fried GM, Barkun AN, et al. Cholecystectomy withoutoperative cholangiography: implications for common bile duct in-jury and retained common bile duct stones. Ann Surg 1993; 218:371-379.

11. McNeil BJ, Keeler E, Adelstein SJ. Primer on certain elements ofmedical decision making. N Engl J Med 1975; 293:211-215.

12. Swets JA, Pickett RM. Evaluation ofDiagnostic Systems: Methodsfrom Signal Detection Theory. Academic Press: New York, 1982,pp 64-65.

13. Hanley JA, McNeil BJ. A method of comparing the areas underreceiver operating characteristics curves derived from the samecases. Radiology 1983; 148:839-843.

14. Coopergerg PL. High-resolution real-time ultrasound in the evalu-ation of the normal and obstructed biliary tract. Clin Radiol 1978;129:477-480.

15. Reiss R, Deutsch AA, Nudelman I, Kott I. Statistical value of var-ious clinical parameters in predicting the presence of choledochalstones. Surg Gyn Obstet 1984; 159:273-276.

16. Armitage P, Berry G. Statistical methods in medical research. 2nded. Oxford: Blackwell, 1987.

17. Hosmer DW, Lemeshow S. Assessing the fit of the model. In

Page 8: Useful Predictors of Bile Duct Stones Patients Undergoing ...

Vol. 220 . No. 1

Hosmer DW. Lemeshow S, eds. Applied Logistic Regression. NewYork: Wiley Interscience Pub, 1989, p 137.

18. Rezieg M, Barkun JS, Fried GM, et al. The management of bileleaks in the era of laparoscopic cholecystectomy. (abstract) Gas-trointest Endosc 1993; 39:A330.

19. Corder AP, Scott SD, Johnson CD. Place of routine operative cho-langiography at cholecystectomy. Br J Surg 1992; 79:945-947.

20. Cranley B, Logan H. Exploration of the common bile duct - therelevance of the clinical picture and the importance of operativecholangiography. Br J Surg 1980; 67:869-872.

21. Hunter JG. Laparoscopic transcystic common bile duct explora-tion. Am J Surg 1992; 163:53-56.

22. Lillemoe KD, Yeo CJ, Talamini MA, et al. Selective cholangiogra-phy. Ann Surg 1992; 215:669-676.

23. Soper NJ. Stockmann PT, Dunnegan DL, Ashley SW. Laparo-scopic cholecystectomy - the new 'gold standard'? Arch Surg1992; 127:917-923.

24. Aliperti G. Edmundowicz SA, Soper NJ, Ashley SW. Combinedendoscopic sphincterotomy and laparoscopic cholecystectomy inpatients with choledocholithiasis and cholecystolithiasis. Ann IntMed 1991; 115:783-784.

25. Boulay J. Schellenberg R, Brady PG. Role ofERCP and therapeu-tic biliary endoscopy in association with laparoscopic cholecystec-tomy. Am J Gastroenterol 1991; 87:837-842.

26. Flowers JL, Zucker KA, Graham SM, et al. Laparoscopic cholan-giography - results and indications. Ann Surg 1992; 209-216.

27. Deslandres E, Gagner M, Pomp A, et al. Intraoperative sphinctero-tomy for common bile duct stones during laparoscopic cholecys-tectomy. Gastrointest Endosc 1993; 39:54-57.

28. Kozarek RA. Laparoscopic cholecystectomy: what to do with thecommon bile duct. Gastrointest Endosc 1993; 39:99-101.

29. Manoukian AV, Schmalz MJ, Geenen JE, et al. Endoscopic treat-ment ofproblems encountered after cholecystectomy. GastrointestEndosc 1993: 39:9-14.

30. Johnson AG, Hosking SW. Appraisal of the management of bileduct stones. Br J Surg 1987; 74:555-560.

31. Meyers WC, Branum GD, Farouk M, et al. A prospective analysisof 1518 laparoscopic cholecystectomies. N Engl J Med. 1991; 324:1073-1078.

32. Larson GM. Vitale GC, Casey J, et al: Multipractice analysis oflaparoscopic cholecystectomy in 1983 patients. Am J Surg 1992;163:221-226.

33. Yip A, Lam KH. An evaluation of routine operative cholangiogra-phy. Aust N Z J Surg 1988; 58:391-395.

34. Schulenberg CAR. Operative cholangiography: 1000 cases. Sur-gery 1969: 65:723-739.

35. Faris 1, Thomson JPS, Grundy DJ, et al. Operative cholangiogra-phy: a reappraisal based on a review of 400 cholangiograms. Br JSurg 1975; 62:966-972.

36. Pernthaler P, Sandbichler P, Schmid Th, Margreiter R. Operativecholangiography in elective cholecystectomy. Br J Surg 1990; 77:399-400.

CBD Stones and Laparoscopic Cholecystectomy 39

37. Del Santo P, Kazarian KK, Rogers JF, et al. Prediction ofoperativecholangiography in patients undergoing elective cholecystectomywith routine liver function chemistries. Surgery 1985; 98:7-11.

38. Saltztein EC, Peacock JB, Thomas MD. Preoperative bilirubin, al-kaline phosphatase and amylase levels as predictors of commonduct stones. Surg Gynec Obstet 1982; 154:381-384.

39. Lacaine F, Corlette MB, Bismuth H. Preoperative evaluation ofthe risk of common bile duct stones. Arch Surg 1980; 115:1114-1116.

40. Hauer-Jensen M, Karesen R, Nygaard K, et al. Prospective ran-domized study of routine intraoperative cholangiography duringopen cholecystectomy: long term follow-up and multivariate anal-ysis of predictors of choledocholithiasis. Surgery 1993; 113:318-323.

41. Taylor TV, Torrance B, Rimmer S, et al. Operative cholangiogra-phy: is there a statistical alternative? Am J Surg 1983; 145:640-643.

42. Taylor TV, Armstrong CP, Rimmer S, et al. Prediction of choled-ocholithiasis using a pocket microcomputer. Br J Surg 1988; 75:138-140.

43. Wilson TG, Hall JC, Watts J McK. Is operative cholangiographyalways necessary? Br J Surg 1986; 73:637-640.

44. Lygidakis NJ. The incidence and significance ofcommon bile ductdilatation in biliary calculous disease. World J Surg 1984; 8:327-334.

45. Mueller PR, Ferucci JT, Simeone JF, et al. Observations on thedistensibility of the common bile duct. Radiology 1982; 142:467-472.

46. Hunt DR, Reiter L, Scott AJ. Pre-operative ultrasound measure-ment of bile duct diameter: basis for selective cholangiography.Aust N Z J Surg 1990; 60:189-192.

47. Hauer-Jensen M, Karesen R, Nygaard K, et al. Predictive abilityof choledocholithiasis indicators - a prospective evaluation. AnnSurg 1985; 202:64-68.

48. Hampson LG, Fried GM, Stets J, et al. Common bile duct explo-ration: indications and results. Can J Surg 1981; 24, 455-457.

49. Hashmonai M, Arisson R, Schramek A. Indications for explora-tion of the bile ducts. Int Surg 1980; 65:239-245.

50. Acosta JM, Ledesma CL. Gallstone migration as a cause of acutepancreatitis. N EngI J Med 1974; 290:484-487.

51. Kelly TR. Gallstone pancreatitis: pathophysiology. Surgery 1976;80:488-492.

52. Frei GJ, Frei VT, Thirlby RC, et al. Timing ofsurgery for gallstonepancreatitis. Am J Surg 1986; 151:170-175.

53. Steer ML. Timing of surgery for gallstone pancreatitis. Gastroen-terology 1986; 91:780-781.

54. Voyles CR, Petro AB, Meena AL, et al. A practical approach tolaparoscopic cholecystectomy. Am J Surg 1991; 161:365-370.

55. Leitman IM, Fisher ML, McKinley MJ, et al. The evaluation andmanagement of known or suspected stones of the common bileduct in the era of minimal access surgery. Surg Gynecol Obstet1993; 176:527-533.