902

Drug Toxicity and Dose

ACROSS-STUDY EVALUATION OFASSOCIATION BETWEEN STEROID DOSEAND BOLUS STEROIDS AND AVASCULAR

NECROSIS OF BONE

DAVID T. FELSON JENNIFER J. ANDERSON

Boston University Multipurpose Arthritis Center and theDepartment of Medicine, Boston City Hospital, Boston,

Massachusetts

Summary Studies investigating steroid dose andavascular necrosis of bone (AVN) have

found either a weak association or none at all. This

quantitative review of published studies has evaluated theeffects of steroid dose and bolus steroids on the risk of AVN.22 papers with sufficient information for analysis wereidentified. The mean steroid dose for the cohort was plottedagainst the percentage in whom AVN developed. Total dosewas divided into non-bolus (oral) and bolus dose, and doses1, 3, 6, and 12 months after beginning steroids were testedseparately for their association with AVN risk. There was astrong correlation between daily total dose and AVN rate(r = 0·61-0·80). Oral dose was strongly correlated with AVNrate (r = 0·70-0·86), but bolus dose was not associated withAVN risk. This strong association between AVN andsteroid dose contrasts with the weak relations found incase-control studies from individual centres in which casesand controls received similar steroid regimens and thereforedid not differ greatly in steroid dose. The method of derivinga single exposure level per study and comparing the amountof exposure across studies may be useful in assessingwhether a drug’s toxicity is dose dependent.

INTRODUCTION

AVASCULAR necrosis of bone (AVN), a common

complication of steroid therapy, occurs in between 0%1 and52%2 of patients receiving glucocorticoids. It causes painand disability, and treatment is unsatisfactory; some patientsneed total joint replacements.

Although a link between AVN and corticosteroid use hasbeen documented, the association between dose and AVNis unclear. Most studies of this association3-lO have found no

significant difference in steroid dose received by patientswith and without AVN. Even when the case-control dosedifference was significant," " it was small-10-20 % greaterin AVN patients.l1-13 Evidence suggesting a larger rolefor dose includes the high rates of AVN after renal

transplantation and in systemic lupus erythematosus (SLE)in which very high daily steroid doses are used, in contrast tothe low rates in asthma, inflammatory bowel disease, andrheumatoid arthritis, in which steroid doses used are lower.The association between high-dose pulse or bolus steroidtherapy and AVN has not been systematically studied,although two studies of renal-transplant patients showedthat instituting pulse therapy for transplant rejection did notchange14 or even reduced 15 the AVN rate.We decided to re-evaluate the association between AVN

and steroid dose, comparing AVN rates across studies andcapitalising on the large steroid dose differences betweenstudies. Our specific objectives were to determine whether

the risk of AVN is associated with steroid dose; whetherbolus steroids predispose to AVN; and whether certainunderlying disorders predispose to AVN in patientsreceiving steroids. In addressing these questions, wedeveloped a method that may be generally applicable toinvestigating whether a drug’s toxicity is dose dependent.

METHODS

We attempted to find all published reports on the long-termcourse of patients on glucocorticoids, regardless of diagnosis. Ourobjective was to determine from each the percentage of patients inwhom AVN developed and the mean steroid dose for patientswithin the group.

To be included in the analysis a study had to meet the followingcriteria. All patients had received steroids. The group of patientshad been followed up long enough for most cases of AVN to occur.16Either mean, median, or midrange of the follow-up time for thegroup had been at least 2 years, with a minimum follow-up of 3months for all patients. The report had to include the percentage ofpatients in whom symptomatic AVN developed. We accepted thestudy’s designation of patients as having symptomatic AVN, butthis diagnosis had to include both symptoms and X-ray changes.The report had also to include the mean dose received by the wholegroup of patients. Specifically, daily or cumulative dose 1 month, 3months, 6 months, or 12 months after beginning steroids had to bereported. We thought that information such as the usual doseregimen or the steroid dose received by those in whom AVNdeveloped was not a sufficiently accurate estimate of the wholegroup’s dose. To ensure further that inadequate follow-up of somecohorts did not affect our findings, we selected a subset of studies inwhich the average follow-up was at least 4 years and minimumfollow-up 1 year (long-follow-up group).

Methylprednisolone, prednisolone, hydrocortisone, anddexamethasone were converted to prednisone equivalents.17 Themean prednisone dose in mg/day was used as the measure of steroiddose. The conversion factor for mg/kg/day was 70 kg (includingpaediatric studies) and that for mg/m2 was 1 ’72 m2. Bolus steroidswere defined as :;:’500 mg/day prednisone equivalent for 1 or

more days.

We carried out MEDLINES searches and used bibliographicreviews for all English and French language studies published from1970 to 1985 on the topics of AVN, glucocorticoids, renal

transplantation, SLE, asthma, inflammatory bowel disease,neurosurgery, and cancer treatment. We found 157 articles which

reported either long-term follow-up of steroid-treated patients orevaluations of the association between steroids and AVN. Morethan half were case-reports or literature reviews. Although many ofthe others addressed the association between steroids and AVN,they contained insufficient data on steroid dose. We found 23papers which met the inclusion criteria. We excluded 118 because a

histogram showing steroid dose seemed to be inaccurately labelled,giving doses more than three times the prescribed regimen. Theremaining 22 papers included 20 long-term observational studies ofsteroid-treated patients’ 5,91-12,19 33 and 2 randomised clinical trials inwhich patients were assigned to two different steroid-treatmentregimens.34,35 For these trials, each treatment group was analysed asa separate study group. One of two groups in Harrington andcolleagues’ study32 was not followed long enough to meet inclusioncriteria. Thus, we found 24 patient cohorts (20 observational studiesand 2 clinical trials, each providing 2 study groups). 19 were groupsof renal-transplant patients, 3 SLE patients, 1 a series of patientsfrom a neurosurgical service, and 1 a group of patients withHodgkin’s disease.

Most of the studies were attempts to examine the associationbetween steroid dose and complications such as AVN. Most werecase-control studies, comparing the steroid doses in groups withand without AVN. Mean doses for case and control groups were

generally provided, but we needed the mean dose received by the

903

TABLE 1-22 STUDIES USED IN META-ANALYSIS (24 COHORTS)

RT = renal transplant; HD = Hodgkin’s disease.*Bolus given to patients but bolus dose not reported.tPaediatric studies.

t206/234 had steroid dose reported.§AVN percentage estimated; 5% had hip AVN.

whole cohort. To estimate this, we assumed that the control groupdose was that for all cohort patients without AVN. If AVN

developed in a proportion A of subjects, the mean cumulative dosefor the whole cohort was A x the mean cumulative dose of cases

plus (1-A) x the mean cumulative dose of controls. The steroiddose for each cohort was defined as the mean dose received bypatients in the cohort. Total steroid dose per day was defmed as themean bolus and non-bolus (oral) steroids per day. For example, ifthe mean cumulative amount of oral prednisone in the first monthwas 2000 mg and, on average, each member of the cohort received1000 mg methylprednisolone bolus over this time (converted to1250 mg prednisone), the mean total steroid dose per day was 2000mg/30 days + 1250 mg/30 days =108-33 mg/day. The meannon-bolus (oral) steroid dose per day was also calculated. In theabove example it is 2000 mg/30 days = 66-67 mg/day. The meanbolus steroid dose was also calculated for each treatment period.

To determine whether this dose estimate for the cohort was ajustifiable approximation, we carried out an analysis of variance toevaluate the relative contributions of between-study and within-study dose variation to total variation. The F-statistics for thewithin-study variation between case and control means ranged from41 to ll’l for the four treatment periods analysed (p < 0-0 in allcases), whereas the between-cohort F-statistics ranged from 98 4 to557-7 (p< 0-0001 in all cases). We concluded that summarisingcohort doses with a single number was justified.

One study30 reported only the proportion of patients in whom hipAVN developed (5%). We estimated from all other studies thenumber of all AVN patients given only those with hip AVN (7%).Later analyses using the 5% figure and an analysis without thisstudy did not affect the results.

TABLE II-STEROID DOSES USED IN STUDY COHORTS

We calculated Pearson correlation coefficients to evaluate theassociation between steroid dose and the rate of AVN. Total andoral doses 1,3,6, and 12 months after beginning steroids were testedseparately for their association with AVN rate.To assess whether bolus steroids were associated with AVN, we

carried out a multiple regression analysis with AVN rate as thedependent variable and oral and bolus doses as the two independentvariables. Bolus steroid use was confined to renal-transplantationcohorts. Because data on bolus doses were scanty, and we wished tomaximise our chance of finding a relation between it and AVN, weestimated both the expected cumulative bolus dose and expectedoral dose at 6 months for four studies in which such data wereprovided only at 3 or 12 months, thus increasing the sample size to11. Thus, if total bolus dose over the first 1 and 3 months was 1 g, weassumed it was 1 g at 6 months. This assumption appeared accurate,because almost all renal-transplant patients received bolus steroidsto prevent rejection within the first few months of transplantation.To evaluate the contribution of underlying disease to AVN risk,

we carried out a multiple regression analysis with disease (renaltransplant against all others) and oral steroid dose as independentvariables.

The regression analyses were repeated after a variance-stabilisingtransformation of the dependent variable (ie, the arc sine of thesquare root of the percentage of patients with AVN). Essentiallyidentical results were obtained. Regressions including both oraldose and the square of oral dose showed no evidence of non-linearityin the dose-response relation.

Fig 1-AVN risk and steroid dose.

904

TABLE HI&mdash;CORRELATION OF STEROID DOSE WITH AVN RATE*

*Correlation coefficients weighted by cohort size.

Of the 24 patient cohorts, 2 studies1,32 had been published toreport a reduction in AVN rate with lower steroid doses. We wereconcerned that publication might be less likely for reports of noreduction in AVN rates with lower steroid doses.36 These twocohorts were therefore excluded from the analysis, although theresults were essentially the same when they were included.

All correlations and multivariate analyses were weighted bycohort size. A two-tailed p value of 0 05 was taken as significant.

RESULTS

Table I lists the studies analysed, and table II shows themean steroid dose at various times after starting steroids andthe minimum and maximum mean doses among thecohorts. Since most studies did not provide information ondose at all four times (eg, deGraafet al28 reported only oralprednisone dose during the first 3 months of therapy), thenumber of cohorts for each time sample is less than 24.

Plots of cumulative AVN incidence versus total and oraldose for the first month and first 6 months of therapy areshown in fig 1. Plots for other times after starting steroidswere similar. In almost all cases, renal-transplant cohortshad the largest mean doses, particularly of total steroids. Forall times after starting steroids, steroid dose (both total andoral) was highly correlated with the risk of AVN (table ill).If the "publication bias" studies were included in the

analysis,1,32 the correlation between oral steroid dose andAVN prevalence remained greater than 0-6 (p < 0-01) at alltimes. These strong associations persisted when any onecohort was excluded from the analysis. In the long-follow-up cohorts, AVN risk was also highly correlated with total

Fig 2-Between-cohort and within-cohort variability of oral steroiddose during first 12 months of therapy.

TABLE IV-EFFECTS OF ORAL DOSE, BOLUS DOSE, AND

UNDERLYING DISEASE ON AVN RISK: WEIGHTED REGRESSION

RESULTS FOR FIRST 6 MONTHS

*Defined as 1 if renal-transplant patients, 0 if not renal transplant (includesSLE and others). Negative value means non-renal-transplant patients athigher risk.

steroid dose (r= 0-72-0-87) and with oral dose (r=074-0-91).

Bolus steroid dose, after adjustment for oral dose, was notsignificantly correlated with AVN risk for any time afterstarting steroids. In these multiple regression analyses, oraldose remained a powerful predictor (p < 0-01) at all times.We found no significant association between underlyingdisease and AVN.With imputed 6-month dose data, which increased the

sample size to 11, there was no association between bolusdose and AVN (table iv). However, with these data, wefound that renal-transplant cohorts had a significantly(p = 0-05) lower risk of AVN than other cohorts given thesame oral dose (table iv).

Fig 2 compares within-cohort variation in steroid doseand between-cohort variation during the first 12 months oftreatment in 11 case-control studies.

DISCUSSION

This quantitative review strongly suggests that steroiddose is the major predictor of the risk of AVN. The strengthof the association (R2 = 0’75) suggests that for patients oncorticosteroids, the daily dose of steroids may be the onlymajor risk factor. The oral dose effect amounts to a 4.6%increase in the risk of AVN for every 10 mg/day rise in oralsteroids during the first 6 months of therapy (table iv).

Other studies have suggested that steroid dose is related toAVN. For example, the use of low-dosage regimens forrenal-transplant patients has resulted in very low rates ofAVN (0-2%).37 Many of these studies did not meet ourinclusion criteria, however. Other evidence includes thefalling AVN rates associated with the trend toward lowerdaily prednisone dose in renal-transplant treatment and thefact that the disorders most commonly linked to AVN arethose in which long-term high-dose steroids have beenused-renal transplantation and SLE.Our analysis suggests that bolus steroids pose little risk of

AVN. The findings are corroborated by two other studies.Susan et al14 found that adding bolus therapy for rejection inrenal-transplant patients caused no change in AVN rate; andMussche et al15 reported a lower AVN rate when transplantrejection was treated with pulse methylprednisolone insteadof longer-term high-dose oral prednisone. In the studies weanalysed, bolus steroids were given less often than once amonth, thus we cannot say whether bolus steroids givenmore often are safe with respect to AVN. Furthermore,bolus steroid data were available only in renal-transplantstudies; generalising that bolus steroids are safe in all

patients may be unwarranted.Our analysis of the association between underlying

disease and the risk of AVN was limited by the small

905

number of non-renal-transplant cohorts in the analysis andthe tendency for renal-transplant cohorts to receive higherdoses than cohorts with other disorders. Nonetheless, thestrong association between AVN risk and dose suggests thatthe high rates of AVN experienced by renal-transplantpatients were due, partly, to the huge doses of steroids theyreceived.Although our study suggests a straight-line relation

between oral steroid dose and AVN, we cannot say forcertain whether the line is straight or curved. That theassociation between oral dose and AVN was strong even

among cohorts on high-dose steroids suggests that there isno threshold dose of increased risk.

Cohorts in the analysis had all been followed for anaverage of 2 years after starting steroids. The averagefollow-up of renal-transplant cohorts was 48-7 months(range of averages 24-108 months). The mean time ofoccurrence of AVN in renal-transplant cohorts in which allpatients were followed for long periods ranged from 12months26 to 18-6 months’&deg; after starting steroids. Thus, therenal-transplant studies included the majority of AVN casesthat would have occurred in these cohorts. In contrast,patients with SLE may have AVN of later onset; the meantime from starting steroids to occurrence of symptomaticAVN was 3-4 years38to 5-5 years.2 1 of the 3 SLE studies24with mean follow-up of 31 months may not have capturedmost AVN cases, but its exclusion from the analysis did notchange the results. We found the strong relation betweendose and AVN even in the long-follow-up cohorts.

It is possible that different studies used different diag-nostic criteria for symptomatic AVN. In most of thesestudies, X-rays were taken only when symptoms occurred.However, in a few,9,ll,28,29,33 X-rays were taken regularly andthe patients were asked about symptoms. The latter cohortshad not necessarily received higher (or lower) doses.Most of the studies analysed investigated the association

between steroid dose and AVN by case-control methods. Ofthe 15 case-control evaluations included,5,9-12,16,21-26,28-30 11found no significant difference in steroid dose between casesand controls, whereas 4/11,12,23,28 found that AVN patientshad received significantly higher doses than controls.

Although we excluded a few other case-control studies,2,4-7,39they did not differ in their findings from the ones included.There are two possible explanations for our finding a very

strong relation between steroid dose and the risk of AVN,when only some of the case-control studies found a

significant association. First, the studies may have hadinsufficient power to demonstrate a significant differencebetween cases and controls. All the case-control

comparisons had case/control dose ratios that were close toone (for the 15 case-control studies analysed, range 0-78-1 88, median 1-13; 50% had ratios between 1-0 and 1-23).Also, 4 of the studies reported an inconsistent relationbetween dose of AVN cases and controls (greater for sometreatment periods, less for others).Given these case/control dose ratios, if we compared

pooled cases and pooled controls, no strong associationbetween AVN and dose would be found, even though thedifference between cases and controls might reachsignificance. Therefore, increased statistical power inherentin pooling does not explain our results. We suggest anotherexplanation, that within-cohort variation was too small toallow a large case-control dose difference.Each of the case-control studies was done at a single

centre, on patients with one disease. In general, all patients

in a cohort received the same steroid regimen. Since allpatients in a given centre were treated similarly, both.casesand controls were receiving essentially the same steroiddose. We found that across-cohort dose variation was much

greater than within-cohort variation (fig 2).In this investigation, we derived point estimates from

single studies and compared these across studies. Thisapproach assumed homogeneous dosage within cohorts, anassumption suggested by the single-disease, single-centrenature of each study. Homogeneity of dosage within a cohortis not necessary, although a heterogeneous cohort dosage islikely to dilute any association found in across-studyevaluations. This feature may be useful in evaluating otherdrugs to see whether dose affects toxicity.

We thank Eliana Vilbrun for secretarial assistance and Dr R. F. Meenan for

helpful suggestions. This study was supported by Multipurpose ArthritisCenter grant AM-20613 from the National Institutes of Health. D. T. F. is a

recipient of an Arthritis Foundation arthritis investigator award.

Correspondence should be addressed to D. T. F., ACC 3E-09, Boston CityHospital, 818 Harrison Avenue, Boston, MA 02118, USA.

REFERENCES

1. Elmstedt E. Incidence of skeletal complications in renal graft recipients. Acta OrthopScand 1982; 53: 853-56.

2. Zizic TM, Marroux C, Hungerford D, Dansereau JV, Stevens MB. Corticosteroidtherapy associated with ischemic necrosis of bone in systemic lupus erythematosus.

Am J Med 1985; 79: 596-604.3. Bergstein JM, Wiens C, Fish AJ, Vernier RL, Michael A. A vascular necrosis of bone

in systemic lupus erythematosus. J Pediatr 1974; 85: 31-35.4. Cruess RL, Blennerhassett J, MacDonald R, MacLean LD, Dossetor J. Aseptic

necrosis following renal transplantation. J Bone Joint Surg 1968; 50-A: 1577-90.5. Arfi S, Moreau F, Heuclin C, Kreis H, Paolaggi JB, Auquier L. L’Osteonecrose

aseptique da la transplantation renale. Rev Rhum 1975; 42: 165-76.6. Gottlieb MN, Stephens MK, Lowrie EG, et al. A longitudinal study of bone disease

after successful renal transplantation. Nephron 1978; 22: 239-48.7. Troch R, Rombouts JJ, Ypersele Gv, Vincent A, Alexandre JP. Epiphyseal

osteonecrosis in transplanted patients: effect of surgical treatment. Proc EDTA1972; 9: 376-87.

8. Ruderman RJ, Poehling GG, Gray R, Nardone M, Goodman W, Seigler HF.Orthopedic complications of renal transplantation in children. Transplant Proc1979; 11: 104-06.

9. Levine E, Erken EH, Price HI, Meyers AM, Solomon L. Osteonecrosis followingrenal transplantation. Am J Roentgenol 1977; 128: 985-91.

10. Ibels LS, Alfrey AC, Huffer WE, Weil R. Aseptic necrosis of bone following renaltransplantation: experience in 194 transplant recipients and review of the literature.Medicine (Balt) 1978; 57: 25-45.

11. Pierides AM, Simpson W, Stainsby D, Alvarez-Ude F, Uldall PR. Avascular necrosisof bone following renal transplantation. Quart J Med 1975; 44: 459-80.

12. Harris RR, Niemann KMW, Diethelm AG. Skeletal complications after renaltransplantation. South Med J 1974; 67: 1016-19.

13. Briggs WA, Hampers C, Merrill JP, et al. Aseptic necrosis in the femur after renaltransplantation. Ann Surg 1972; 175: 282-89.

14. Susan LP, Braun WE, Banowsky LH, Straffon RA, Bergfeld JA. Avascular necrosisfollowing renal transplantation. Urology 1978; XI: 225-29.

15. Mussche MM, Ringoir SMG, Lameire NN. High intravenous doses of

methylprednisolone for acute renal allograft rejection. Nephron 1976; 16: 287-91.16. Metselaar HJ, van Steenberge EJP, Bijnen AB, et al. Incidence of osteonecrosis after

renal transplantation. Acta Orthop Scand 1985; 56: 413-15.17. Axelrod L. Glucocorticoid therapy. Medicine (Balt) 1976; 55: 39-65.18. Feletti C, DiFelice A, Scolari MP, Bonomini V. Glucocorticoids and avascular bone

necrosis in renal transplantation. Adv Exp Med Biol 1984; 171: 361-68.19. Gluckman JC, Aime B, Aime F, Legrain M, Kuss R. Utilisation du dossier

informatique pour la surveillance des malades traites par transplantation renale.Ann Med Intern 1971; 122: 317-26.

20. Kuss R, Legrain M, Gluckman JC, Guedon J, Poisson J. Allotransplanlation renalechez l’homme. Resultats et prospective a partir de 72 observations. La Presse Med1971; 79: 1889-94

21. Bewick M, Stewart PH, Rudge C, Farrand C, McColl I. Avascular necrosis of bone inpatients undergoing renal allotransplantation. Clin Nephrol 1976; 5: 66-71.

22. Smith FE, Sweet DE, Brunner CM, Davis JS. Avascular necrosis in SLE. Ann RheumDis 1976; 35: 227-32.

23. Abeles M, Urman JD, Rothfield NF. Aseptic necrosis of bone in systemic lupuserythematosus. Arch Intern Med 1978; 138: 750-54.

24. Dimant J, Ginzler EM, Diamond HS, et al. Computer analysis of factors influencingthe appearance of aseptic necrosis in patients with SLE. J Rheumatol 1978; 5:136-41.

25. Potter DE, Genant HK, Salvatierra O. Avascular necrosis of bone after renaltransplantation. Am J Dis Child 1978; 132: 1125-29.

906

26. Elmstedt E. Avascular bone necrosis in the renal transplant patient: a discriminantanalysis of 144 cases. Clin Orthop 1981; 158: 149-57.

27. Prosnitz LR, Lawson JP, Friedlaender GE, Farber LR, Pezzimenti JF. Avascularnecrosis of bone in Hodgkin’s disease patients treated with combined modalitytherapy. Cancer 1981; 47: 2793-97.

28. deGraff P, Van Hooff JP, Boekhout M, Achterberg J, Pauwells EKJ, Kalff MW.Hyperparathyroidism and avascular necrosis of bone after kidney transplantation.Neth J Med 1982; 25: 230-36.

29. Hely D, Fennell RS, Petty W, Hudson T, Richard GA. Osteonecrosis of the femoralhead and condyle in the post transplantation courses of children and adolescents.Int J Pediatr Nephrol 1982; 3: 297-303.

30. Haajanen J, Saarinen O, Laasonen L, Kuhlback B, Edgren J, Slatis P. Steroidtreatment and aseptic necrosis of the femoral head in renal transplant recipients.Transplant Proc 1984; 16: 1316-19.

31. Sambrook PN, Hassall JE, York JR. Osteonecrosis after high dosage, short termcorticosteroid therapy. J Rheumatol 1984; 11: 514-16.

32. Harrington KD, Murray WR, Kountz SL, Belzer FO. Avascular necrosis of boneafter renal transplantation. J Bone Joint Surg 1971; 53A: 203-15.

33. Stern PJ, Watts HG. Osteonecrosis after renal transplantation in children. J Bone JointSurg 1979; 61: 851-56.

34. Morris PJ, Chan L, French ME, Ting A. Low dose oral prednisolone in renaltransplantation. Lancet 1982; ii: 525-27.

35. DeVecchi A, Rivolta E, Tarantino A, et al. Controlled trial of two different

methylprednisolone doses in cadaveric renal transplantation. Nephron 1985, 41:262-66.

36. Light RJ, Pillemer DB. Summing up: The science of reviewing research. Cambridge,Massachusetts: Harvard University Press, 1984: 65-69

37. McGeown MG, Douglas JF, Brown WA, et al. Low dose steroid from the dayfollowing renal transplantation. Proc EDTA 1979; 16: 395-99.

38. Klipper AR, Stevens MB, Zizic TM, et al. Ischemic necrosis of bone in systemic lupuserythematosus. Medicine (Balt) 1976; 55: 251-57.

39. Nielsen HE, Melsen F, Christensen MS. Aseptic necrosis of bone following renaltransplantation. Acta Med Scand 1977; 202: 27-32.

Occasional Survey

CHOLECYSTECTOMY AND LARGE BOWELCANCER

GARY D. FRIEDMAN MARILYN K. GOLDHABERCHARLES P. QUESENBERRY, JR

Division of Research, Kaiser Permanente Medical Care Program,3451 Piedmont Avenue, Oakland, California 94611, USA

Summary The records of 5898 patients with colo-rectal cancer and 27 687 controls were

examined for previous cholecystectomy. The estimatedrelative risks (and 95% confidence intervals) of develop-ment of any cancer of the large bowel and cancer of the rightcolon after cholecystectomy were 1&middot;0 (0&middot;8-1&middot;2) and 1&middot;1

(0&middot;8-1&middot;5) in women and 1&middot;1 (0&middot;9-1&middot;5) and 1&middot;2 (0&middot;8-1&middot;9) inmen, respectively. Although these data do not rule out asmall increase in risk, it is proposed that the associationfound in some other studies is, at least in part, an artifact.Intense diagnostic effort and treatment aimed at mildabdominal symptoms, encouraged by some patients andsome medical care settings, could increase the detection andremoval of gallstones and the early detection of colorectalcancer.

INTRODUCTION

SEVERAL investigations have suggested that cholecystec-tomy raises the risk of cancer of the large bowel,1-5 but othershave not.6-9 The proposed biological mechanism is thecontinuous secretion of bile into the gut that occurs in theabsence of a gallbladder. This continuous secretion leads toenhanced formation of secondary bile acids, such as

lithocholic and deoxycholic acid, because of increasedenterohepatic circulation and degradation of primary bileacids by intestinal bacteria.10,11 Secondary bile acids can actas cocarcinogens in chemically induced cancer of the colonin rats;12 an increased incidence of such cancers of the colonhas also been reported in cholecystectomised mice.13 Theseacids are found in greater amounts in patients with cancer ofthe colon than in those without cancer.14 Furthermore,stools of people from Western industrialised societies with ahigh incidence of colon cancer tend to contain more

secondary bile acids than those from African and FarEastern populations where the incidence is low.15The association between cholecystectomy and large

bowel cancer has been observed most often in women and in

patients with cancers of the right side of the colon.1-5

Whether or not this association is confined to these groups a

heightened risk of a common cancer due to a frequentlyperformed operation would have important public healthimplications. We therefore investigated2he relation betweencholecystectomy and large bowel cancer in a large popu-lation of subscribers to a health maintenance organisation.

SUBJECT AND METHODS

SettingThe Kaiser Permanente Medical Care Program (KPMCP) in

Northern California is a prepaid comprehensive medical servicethat began in 1945. The group of subscribers is racially andsocioeconomically diverse16 and- has grown to over two million.Subscribers receive almost all of their hospital and outpatient care inKPMCP facilities, and their records can be linked together by themedical record number recorded in all medical contacts with theservice.

Case Ascertainment

A total of 5898 patients with large bowel cancer, initiallydiagnosed in KPMCP facilities in 1971 or later, were identifiedfrom two sources-a computer file of all Northern CalifomiaKPMCP hospital admissions to the end of 1984 and the files ofCalifornia Resource for Cancer Epidemiology (RCE), which wascomplete for the period up to 1982 and included some of the casesdiagnosed in 1983. The RCE records all cases of cancer in the fivemost populous counties adjacent to San Francisco Bay for thenational Surveillance, Epidemiology, and End Results (SEER)programme. If there was disagreement about subsite or date weused information from the following sources, listed in order ofpriority: review of records done for a previous study; RCE; andhospital records. If the subsite was unknown or unspecified, or ifboth colon and rectal cancers seemed to be present, then the medicalcharts were reviewed to establish the site of the cancer. If separateprimary cancers were confirmed in both the colon and rectum, thepatient was included in the unknown or unsited subgroup. Rightsided colon cancers were those sited at, or to the right of, the splenicflexure.We validated the identification of right sided colon cancers in

women as follows. We accepted as accurate all diagnoses in whichthere was exact agreement between hospital and RCE records withrespect to the four digits of the ICD code indicating the subsite ofthe large bowel (eg, ascending colon, sigmoid colon) or diagnosesthat had been confirmed by chart review for this or previousinvestigations. 435 (45 %) of 976 right-sided colon cancers inwomen were confirmed by this means. We reviewed the medicalrecords of a sample of 183 unverified cases. 157 (85-8%) wereconfirmed to have right-sided colon cancer and the 26 unconfirmedcases were excluded. Applying the 85-8% verification rate to theremaining 358 unconfirmed and unvalidated cases we estimate that