doi: 10.1054/ bjoc.2001.1709, vilble online t http://www.idelibrry.com on http://www.bjcncer.com Brest cncer nd NSAID use: met-nlysis SA Khuder nd AB Mutgi Deprtment of Medicine, Medicl College of Ohio, 3120 Glendle Ave, Toledo, Ohio 43614-5809, USA Summry Recent epidemiologicl studies suggest tht non-steroidl nti-inflmmtory drugs (NSAIDs) reduce the risk of severl cncers including brest cncer. This met-nlysis exmined the studies on NSAID use nd brest cncer. The estimtors of reltive risk nd ssocited vrinces, which hve been djusted for the gretest number of confounders, were bstrcted nd included in the met-nlysis. Combined estimtors of reltive risk (RR) were clculted using either fixed or rndom effect models. Met-nlyses were performed on 6 cohort studies (number of cses rnged from 14 to 2414) nd 8 cse control studies (number of cses rnged from 252 to 5882). The combined estimte of reltive risk ws 0.82 (95% confidence intervl [CI] = 0.75 0.89). The combined estimte for cohort studies ws 0.78 (95% CI = 0.62 0.99) nd ws 0.87 (95% CI = 0.84 0.91) for cse control studies. The findings of this met-nlysis suggest tht NSAID use my be ssocited with smll decrese in the risk of brest cncer. However, the vilble dt re insufficient to estimte the dose response effect for durtion nd frequency of use of ny prticulr types of NSAID. http://www.bjcncer.com Keywords: NSAID; brest cncer; met-nlysis, femles Brest cncer is one of the most common cncers in women. The incidence of the disese hs incresed in recent yers in the United Sttes (Dinse et l, 1999). A substntil portion of the recent trend my be ttributble to screening (Wun et l, 1995), however, the long-term trend remins unexplined. The estblished risk fctors for brest cncer ccount for less thn hlf of ll brest cncer cses nd offer few opportunities for intervention (Kelsey et l, 1993; Mdign et l, 1995). Recently, non-steroidl nti-inflmmtory drugs (NSAIDs) hve received incresing ttention due to their potentil s chemopreventive gents ginst cncer. Animl studies showed inhibitory effect for NSAID on brest crcinogenesis (Ll et l, 1997; Robertson et l, 1998). However, severl epidemiologic studies hve exmined the reltion between NSAIDs nd brest cncer, with inconsistent results. In some studies (Lkso et l, 1988; Gridley et l, 1993) ptients with rheumtoid rthritis who use NSAIDs in high doses for symptom relief hd fewer thn expected cses of brest cncer. Numerous studies hve demonstrted tht the level of prostglndin is greter in brest cncer thn in norml tissue (Bennett et l, 1983). In prticulr, the inducible form of cyclooxygense (COX), the rte-limiting enzyme in prostglndin biosynthesis my be overexpressed in brest cncer (Hwng et l, 1998). NSAIDs re known to block COX ctivity (Robertson et l, 1998) nd thus become ttrctive gents for brest cncer prevention. A recent niml study (Hrris et l, 2000) confirmed the chemopreventive ctivity of NSAIDs ginst brest cncer through COX2 blocking. In this met-nlysis we exmined the epidemiologicl studies on NSAID use nd brest cncer. METHODS The studies were locted vi serch of the MEDLINE (from 1966 to 2000) nd Cncer Abstrct dtbses (from 1980 to 2000). Abstrcts of reserch presented t relted conferences (Society for Epidemiologic Reserch, Europen Cncer Reserch, British Cncer Reserch, nd Americn Assocition for Cncer Reserch) were lso serched. Studies were eliminted from the nlyses if they included subjects used in other more-inclusive studies. The estimtors of reltive risk nd ssocited vrinces, which hs been djusted for the gretest number of confounders, were bstrcted nd included in the met-nlysis. A series of met-nlyses ws conducted nd the results were evluted in the context of the published literture. The homogeneity of the estimtors of reltive risk ws tested using Cochrn s Q sttistics (Cochrn, 1954). This is chi-squre test with degrees of freedom equl to the number of studies minus one, nd tests the null hypothesis tht the within-study estimtes of reltive risk re homogeneous cross studies. The fixed-effect model ws used to obtin the combined estimtor of reltive risk nd its stndrd error (SE). The rndom-effects model (DerSimonin nd Lird, 1987) ws used in sitution when we detected significnt heterogeneity within the groups of studies. The potentil for publiction bis in published reports ws investigted by constructing funnel plots of log odds rtio ginst the size of the study. A Kendll tu rnk correltion test (Begg nd Mzumdr, 1994) ws used to test for the sttisticl significnce of publiction bis. Received 19 October 2000 Revised 11 Jnury 2001 Accepted 16 Jnury 2001 Correspondence to: Sdik A. Khuder RESULTS We identified 15 studies (Friedmn nd Ury, 1980; Pgnini-Hill et l, 1989; Rosenberg et l, 1991; Thun et l, 1993; Schreinemchers nd Emerson, 1994; Hrris et l, 1995; Rosenberg, 1995; Egn et l, 1996; Hrris et l, 1996; Neugut et l, 1998; Coogn et l, 1999; Hrris et l, 1999; Cotterchio et l, 2000; 1188
NSAID use nd brest cncer 1189 Tble 1 Cohort studies used for met-nlyses of NSAIDs use nd brest cncer Reference NSAID Ttype Number of cses OR 95% CI Hrris et l, 1999 Aspirin 76 0.60 0.47 0.77 Acetminophen 36 0.84 0.60 1.18 Ibuprofen 37 0.51 0.36 0.72 Egn et l, 1996 Aspirin 2414 1.01 0.80 1.27 Schreinemchers nd Everson, 1994 Aspirin 79 0.70 0.50 0.96 Thun et l, 1993 Aspirin 0.94 0.80 1.10 Pgnini-Hill et l, 1989 Aspirin 68 0.96 0.75 1.21 Friedmn nd Ury, 1980 Aspirin, 2 0.20 0.05 0.80 Indomethcin 12 0.50 0.28 0.88 Tble 2 Cse control studies used for met-nlyses of NSAIDs use nd brest cncer Reference Number of cses NSAID type Type of control OR 95% CI Shrpe et l, 2000 5882 Any Generl popultion 0.90 0.84 0.97 Cotterchio et l, 2000 2681 Any Generl popultion 0.74 0.65 0.85 Lngmn et l, 2000 3105 Any Cncer 1.00 0.92 1.09 Coogn et l, 1999 6558 Any Cncer 0.80 0.70 1.00 Non-cncer 0.70 0.60 0.90 Aspirin Cncer 0.70 0.60 0.90 Non-cncer 0.70 0.50 0.80 Neugut et l, 1998 252 Aspirin Non-cncer 0.80 0.35 1.80 Hrris et l, 1996 106 Any Generl popultion 0.66 0.52 0.83 Aspirin, Generl popultion 0.69 0.46 0.99 Ibuprofen Generl popultion 0.57 0.36 0.91 Roenberg, 1995 4485 Any Cncer 0.90 0.60 1.20 Non-cncer 0.80 0.60 1.00 Hrris et l, 1995 744 Any Cncer 0.96 0.67 1.39 Non-cncer 0.81 0.63 1.03 Nested cse control study. Tble 3 Combined nlysis of studies used for met-nlysis of NSAID use nd brest cncer Reference NSAID type Number of studies Test of heterogeneity OR 95% CI χ 2 P vlue All studies Any 16 38.1 0.001 0.80 0.73 0.87 Cohort studies Any 6 22.8 <0.001 0.78 0.62 0.99 Aspirin 6 17.8 <0.001 0.79 0.59 1.06 Cse control All Any 10 15.2 0.09 0.83 0.79 0.88 Cncer 3 1.0 0.621 0.84 0.72 0.97 Non-cncer 7 14.2 0.03 0.79 0.72 0.86 All Aspirin 4 0.10 0.99 0.70 0.61 0.81 Lngmn, 2000; Shrpe et l, 2000) evluting the ssocition between NSAIDs nd brest cncer tht were published between 1980 nd 2000. One study (Rosenberg et l, 1991) ws excluded from the nlysis becuse of lck of dt on the estimtor of reltive risk. The remining studies were 6 cohort studies (Tble 1) nd 8 cse control studies (Tble 2). The number of cses rnged from 14 to 2414 for cohort studies. For cse control studies the number of cses rnged from 252 to 5882 nd the number of controls rnged from 42 to 89528. 5 studies (Pgnini-Hill et l, 1989; Thun et l, 1993; Schreinemchers nd Emerson, 1994; Egn et l, 1996; Neugut et l, 1998) were restricted to spirin use nd three studies (Hrris et l, 1999; Pgnini-Hill et l, 1989; Friedmn nd Ury, 1980) provided dt strtified by NSAID type. Twelve studies reported reduction in the risk of brest cncer with NSAID use. The estimtor of reltive risk for cohort studies rnged from 0.20 to 1.01 nd 6 of these estimtors were significnt. Only one cohort study (Egn et l, 1996) reported nonsignificnt increses in the risk of brest cncer with spirin use. The estimtor of reltive risk for cse control studies rnged from 0.57 to 1.00 nd 7 of these estimtors were significnt. None of the cse control studies reported n increse in the risk of brest cncer with ny NSAID use. The results of met-nlyses re presented in Tble 3. Significnt heterogeneity ws detected mong the studies (χ 2 = 53.0, P = 0.001). The combined estimte of reltive risk using the rndom-effect model ws 0.82 (95% CI = 0.75 0.89). Heterogeneity mong studies ws significntly reduced when studies were combined within specific design nd type of control. The combined estimte of reltive risk for cohort studies ws 0.78 (95% CI 0.62 0.99) with ny NSAIDs nd ws 0.79 (95%
1190 SA Khuder nd AB Mutgi Tble 4 Durtion of use reported in studies used in the met-nlysis of NSAID use nd brest cncer Study Mesure Durtion OR 95% CI Coogn et l, 1999 NSAID regulr use (yers) begun 1 yer before dmission Never 1 < 1 0.90 0.50 1.70 1 < 2 1.10 0.70 1.70 2 < 5 0.70 0.50 1.00 5 < 10 0.70 0.40 1.00 10 < 20 0.70 0.40 1.10 20+ 0.60 0.30 1.00 Unknown 0.40 0.20 0.70 Hrris et l, 1995 NSAID use, yers 0 1 1 4 1.09 0.80 1.50 5 0.63 0.50 0.90 Egn et l, 1996 Aspirin use, yers <5 0.89 0.76 1.05 5 9 0.98 0.81 1.19 10 19 1.11 0.85 1.46 20 1.00 0.71 1.41 Hrris et l, 1996 NSAID use, yers <5 0.65 0.47 0.91 5 0.60 0.40 0.91 Lngmn et l, 2000 Prescription of NSAID before dignosis, months 13 24 1.03 0.93 1.13 25 36 1.00 0.91 1.11 Shrpe et l, 2000 Highest level of NSAID exposure before dignosis, yers 1/2 1.05 0.91 1.23 1/2 1 1.20 1.02 1.40 2 5 0.76 0.63 0.92 6 10 1.13 0.92 1.39 11 15 0.83 0.63 1.11 Significnt dose response reltionship. Tble 5 Frequency of NSAID use in studies used in the met-nlysis of NSAID use nd brest cncer Study Mesure Frequency OR 95% CI Shrpe et l, 2000 Sum of NSAID mg dy 1 dispensed mximum mg dy 1 0 1.0 recommended (Σp) in 2 5 yers before dignosis 0 < Σp 1 0.93 0.85 1.01 0.1 < Σp 0.3 0.91 0.79 1.06 Σp >0.3 0.76 0.63 0.92 Egn et l, 1996 Number of spirins per week 0 1.00 1 3 0.99 0.89 1.11 4 6 0.94 0.80 1.10 7 10 1.00 0.84 1.20 11 14 1.11 0.91 1.37 >14 1.05 0.89 1.23 Pgnini-Hill et l, 1989 Aspirin use None 1 <dily 0.95 0.68 1.34 dily 0.96 0.69 1.34 Hrris et l, 1996 NSAID dose/week 3 6 0.73 0.46 1.13 7 0.63 0.49 0.81 Thun et l, 1993 Aspirin frequency/month occsionlly 0.93 0.73 1.19 1 15 0.98 0.76 1.26 16+ 0.88 0.62 1.24 Hrris et l, 1999 NSAID pills per week 0 < 1 1.00 1 3 0.64 0.50 0.82 4 0.57 0.44 0.74 Lngmn et l, 2000 Number of prescriptions of NSAID received in 13 24 months before dignosis 0 1 1 0.99 0.87 1.13 2 6 0.96 0.83 1.11 7 1.10 0.92 1.30 Significnt dose response reltionship. CI = 0.59 1.06) for spirin use. The cse control studies were reltively homogeneous within cncer nd non-cncer controls. The combined estimte of reltive risk for these studies ws 0.87 (95% CI 0.84 0.91) with ny NSAID nd ws 0.70 (95% CI = 0.61 0.81) for spirin use. The combined estimte of reltive risk for studies with non-cncer controls ws 0.79 (95% CI 0.72 0.86) nd ws 0.96 (95% CI = 0.89 1.03) for studies with cncer controls.
NSAID use nd brest cncer 1191 Six studies provided results on durtion of use of spirin nd tht of other NSAIDs (Tble 4). 9 studies provided results strtified by mesures of NSAID use (Tble 4). The risk reduction for highest durtion of use (20+ yers) rnged from 40% (Coogn et l, 1999) to zero (Egn et l, 1996). Trend tests for dose response reltions in only one of these studies ws significnt. The vilble dt in Tble 4 re insufficient to estimte the combined dose response effect for durtion of use of ny prticulr types of NSAID. Seven studies provided results on frequency of use of spirin nd tht of other NSAID (Tble 5). In one study (Hrris et l, 1999) four or more pills of NSAID per week ws ssocited with 43% reduction in the risk of brest cncer (RR = 0.57, 95% CI 0.44 0.74). In nother study (Pgnini-Hill et l, 1989) dily use of spirin ws ssocited with only 4% reduction in the estimted risk of brest cncer (RR = 0.96, 95% CI 0.69 1.34). The vilble dt in Tble 5 re insufficient to estimte the combined dose response effect for frequency of use of ny prticulr types of NSAID. There ws no evidence of publiction bis in studies included in this met-nlysis. The Kendll tu correltion coefficient for the stndrd error nd the stndrdized log odds rtio ws 0.96 (P = 0.34). DISCUSSION This met-nlysis showed tht NSAID use my decrese the risk of brest cncer. This is evident by the consistently reduced reltive risk in the mjority of studies included in the nlysis. The effect observed ws similr in most studies regrdless of design or type of cses (incident or ftl cses). The only negtive study (Egn et l, 1996) my hve been confounded by reproductive fctors. In this met-nlysis, regulr use of NSAIDs ws ssocited with n 18% reduction in the risk of brest cncer. The reduction in risk ws higher in cohort studies (21%) thn cse control studies (13%). Within cse control studies, the reduction in risk ws smller in studies with cncer controls thn in those with non-cncer controls. Although this finding is consistent with studies on NSAID use nd colon cncer (Hrris et l, 1995), it my rgue ginst true effect ginst brest cncer since this should be consistent cross control groups. It is possible tht some cncer subtypes (for exmple, gstrointestinl) were relted to NSAID use nd these ptients discontinued the drug. If so this my overestimte the odds rtio nd my bis the estimte of reltive risk wy from the null vlue. It is possible tht the results of these studies re bised by higher prevlence of preexisting medicl conditions commonly ssocited with NSAID use mong non-cncer controls. On the other hnd popultion-bsed cse control studies (Neugut et l, 1998) used s control subjects who underwent screening mmmogrphy, nd their use of NSAID could hve overestimted the prevlence of use in the study bse. With regrd to type of NSAID, spirin ws the mjor type used in the studies included in this met-nlysis. In generl, the reduction in risk of brest cncer with spirin use ws similr to other NSAID type. Two studies reported higher reduction in risk with ibuprofen in comprison to spirin. However, vilble dt were not dequte enough to test this in met-nlysis. Nine studies evluted dose response reltion of NSAID use nd brest cncer but only two studies reported significnt dose response reltion for durtion (Coogn et l, 1999) nd frequency (Shrpe et l, 2000) of NSAID use. In one study (Coogn et l, 1999) the highest reduction in brest cncer risk ws reported for the ctegory unknown yers of use. Although publiction bis is possible becuse of the possibility of filure of investigtors to submit negtive results or filure of journls to publish negtive studies our nlysis did not suggest this. The effect of NSAIDs on brest cncer risk reduction is biologiclly plusible. A number of niml studies hve suggested protective effect of NSAIDs ginst mmmry cncer. A potentil mechnism for nti-tumour effect of NSAIDs involves inhibition of the synthesis of prostglndins. NSAIDs block the enzyme cyclooxygense nd in turn inhibit prostglndin biosynthesis. Prostglndins my serve s cofctors in crcinogenesis with potentil effects rnging from direct mutgenesis to tumour promotion nd immune suppression (Lupulescu, 1978; Mellemkjer et l, 1996). One study (McCormick nd Wilson, 1986) suggests tht the cncer inhibitory effects of NSAIDs my be independent of their effects on prostglndin synthesis. There is evidence from niml studies tht indomethcin inhibits the effects of oestrogen in the pituitry glnd (Neugut et l, 1998). Invitro studies of humn brest cncer cells indicte tht cetylslicylic cid my inhibit direct binding of oestrdiol to oestrogen receptor (Thompson et l, 1995). The mjority of studies included in this met-nlysis djusted for known risk fctor for brest cncer. Our inclusion of estimtors of reltive risk, which were djusted for the gretest number of confounders, would hve reduced the possibility of confounding effect. The combined estimte of this study supports protective effect for NSAIDs ginst brest cncer. Other support for the protective effect of NSAIDs ginst brest cncer comes from studies on ptients with rheumtoid rthritis who use NSAIDs in high doses for symptom relief. 2 studies (Gridley et l, 1993; Bron, 1995) reported tht these ptients hd less thn expected occurrence of brest cncer. The risk pttern for NSAID users found in this met-nlysis is consistent with the pttern found in studies on ptients with rheumtoid rthritis. The limittions of this study stem from the studies included in the met-nlysis. All the studies included re observtionl studies nd therefore subject to bises. Some re cse control studies nd we cnnot rule out the possibility of selection nd informtion bises. It is possible tht the reltion between NSAIDs nd brest cncer my reflect recll bis by the cses or controls. Misclssifiction of exposure is potentil problem in observtionl epidemiologicl studies. None of the studies included in this met-nlysis utilized n objective method of exposure ssessment. In ll studies the NSAID use ws self-reported nd therefore subject to recll bis. These drugs re often tken spordiclly in pttern of intke tht my be difficult to remember or summrize for some subjects. For exmple, some widely used brnds or combintion product my not be recognized s contining spirin (Hrris et l, 1995). It is possible tht NSAID use my reflect helth consciousness mong the control group. However, Hrris et l (1995) reported no ssocition between NSAID use nd level of eduction, which cn be tken s proxy for helth consciousness. Currently known risk fctors ccount for less thn hlf of ll brest cncer cses nd offer limited opportunities for intervention. Therefore, ny preventive mesure identified will be importnt. This met-nlysis suggests tht NSAIDs hve wek chemopreventive vlue ginst brest cncer. However, we cnnot rule out the possibility of n lternte explntion for this finding due to the limittions of the studies included in the nlysis. There is need for more studies tht prospectively evlute the reduction in risk of brest cncer utilizing better mesure of NSAID dose. There is need to estblish whether NSAIDs re
1192 SA Khuder nd AB Mutgi efficcious in preventing brest cncer nd type nd optiml dose. This cn be ccomplished using rndomized clinicl tril on different types of NSAID. REFERENCES Bron JA (1995) Aspirin nd cncer. Prev Med 24: 121 124 Bennett A, Berstock DA, Crroll MA, Stmford IF nd Wilson AJ (1983) Brest cncer, its recurrence, nd ptient survivl in reltion to tumor prostglndins. In: Smuelsson B, Poletti R, Rmwell P (eds) Advnces in prostglndin, thromboxne, nd leukotriene reserch, Vol 12. Rven Press: New York Begg CB nd Mzumdr M (1994) Operting chrcteristics of rnk correltion test for publiction bis. Biometrics 50: 1088 1101 Crter CA, Milhollnd RJ, She W nd Ip MM (1983) Effect the prostglndin synthetse inhibitor indomethcin on 7, 12-dimethylbenz()nthrcene-induced mmmry tumorigenesis in rts fed different levels of ft. Cncer Res 43: 3559 3562 Cochrn WG (1954) The combintion of estimtes from different experiments. Biometrics 8: 101 129 Coogn FP, Ro RS, Rosenberg L, Plmer JR, Strom BL, Zuber AG, Stolley PD nd Shpiro S (1999) The reltionship of nonsteroidl nti-inflmmtory drug use to the risk of brest cncer. Prev Med 29: 72 76 Cotterchio M, Kreiger N, Steingrt A nd Buchn G (2000) Non-steroidl ntiinflmmtory drugs (NSAID) use nd brest cncer. SER Abstrct. Am J Epidemiol 151: S72 DerSimonin R nd Lird N (1987) Met-nlysis in clinicl trils. Controlled Clin Trils 7: 177 188 Dinse GE, Umbch DM, Ssco AJ, Hoel DG nd Dvis DL (1999) Unexplined increses in cncer incidence in the United Sttes from 1975 to 1994: possible sentinel helth indictors? Ann Rev Public Helth 20: 173 209 Egn KM, Stmpfer MJ, Giovnnucci E, Rosner BA nd Colditz GA (1996) Prospective study of regulr spirin use nd the risk of brest cncer. J Ntl Cncer Inst 88: 988 993 Friedmn GD nd Ury HK (1980) Initil screening for crcinogenicity of commonly used drugs. J Ntl Cncer Inst 65: 723 733 Gridley G, McLughlin JK, Ekbom A, Klreskog L, Admi H, Hcker DG, Hoover R nd Frumeni Jr JF (1993) Incidence of cncer mong ptients with rheumtoid rthritis. J Ntl Cncer Inst 85: 307 311 Hrris RE, Nmboodiri K, Stellmn SD nd Wynder EL (1995) Brest cncer nd NSAID use: heterogeneity of effect in cse-control study. Prev Med 24: 119 120 Hrris RE, Nmboodiri KK nd Frrr WB (1996) Nonsteroidl ntiinflmmtory drugs nd brest cncer. Epidemiology 7: 203 205 Hrris RE, Ksbri S nd Frrr WB (1999) Prospective study of nonsteroidl ntiinflmmtory drugs nd brest cncer. Oncol Rep 6: 71 73 Hrris RE, Alshfie GA nd Seibert K (2000) Chemoprevention of brest cncer in rts by celecoxib, cyclooxygense 2 inhibitor. Cncer Res 60: 2101 2103 Hwng D, Scollrd D, Byrne J nd Levine E (1998) Expression of cyclooxygense-1 nd cyclooxygense-2 in humn brest cncer. J Ntl Cncer Inst 90: 455 460 Kelsey JL, Gmmon MD nd John EM (1993) Reproductive fctors nd brest cncer. Epidemiol Rev 15: 36 47 Lkso M, Mutru O, Isomki H nd Koot K (1988) Cncer mortlity in ptients with rheumtoid rthritis. J Rheumtol 15: 1319 1322 Ll PK, Al-mutter N nd Orucevic A (1997) Effects of chronic indomethcin therpy on the development nd progressions of spontneous mmmry tumors in C3H/HEJ mice. Int J Cncer 73: 371 380 Lngmn MJS, Cheng KK, Gilmn EA, Lncshire RJ (2000) Effect of ntiinflmmtory drugs on the overll risk of common cncer: cse-control study in generl prctice reserch dtbse. Br Med J 320: 1642 1646 Lupulescu A (1978) Enhncement of crcinogenesis by prostglndins. Nture 272: 634 636 Lynch NR, Cstes M, Astoin M nd Slomon JC (1978) Mechnism of inhibition of tumour growth by spirin nd indomethcin. Br J Cncer 38: 503 512 Mdign MP, Ziegler RG, Benichou J, Byrne C nd Hoover RN (1995) Proportion of brest cncer cses in the United Sttes explined by well-estblished risk fctors. J Ntl Cncer Inst 87: 1681 1685 McCormick DI nd Wilson AM (1986) Combintion chemoprevention of rt mmmry crcinogenesis by indomethcin. Cncer Res 46: 3907 3911 Mellemkjer L, Linet MS, Gridley G, Frisch M, Moller H nd Olsen JH (1996) Rheumtoid rthritis nd cncer risk. Eur J Cncer 32A: 1753 1757 Neugut AI, Rosenberg DJ, Ahsn H, Jcobson JS, Whid N, Hgn M, Rhmn MI, Khn ZR, Chen L, Pblos-Mendez A nd She S (1998) Assocition between coronry hert disese nd cncers of the brest, prostte, nd colon. Cncer Epidemiol Biomrker Prev 7: 869 873 Pgnini-Hill A, Cho A, Ross RK nd Henderson BE (1989) Aspirin use nd chronic diseses: cohort study of the elderly. Br Med J 299: 1247 1250 Robertson FM, Prett ML, Jorder FS, Ross M, Abo-Iss, Alshfie G nd Hrris RE (1998) Ibuprofen-induced inhibition of cyclooxygense isoform gene expression nd regression of rt mmmry crcinom. Cncer Lett 122: 165 175 Rosenberg L (1995) Nonsteroidl nti-inflmmtory drugs nd cncer. Prev Med 24: 107 109 Rosenberg L, Plmer JR, Zuber AG, Wrshuer ME, Stolley PD nd Shpiro S (1991) A hypothesis: nonsteroidl nti-inflmmtory drugs reduce the incidence of lrge-bowel cncer. J Ntl Cncer Inst 83: 355 358 Schreinemchers DM nd Everson RB (1994) Aspirin use nd lung, colon, nd brest cncer incidence in prospective study. Epidemiology 5: 138 146 Shrpe CR, Collet JP, McNutt M, Belzille E, Boivin JF nd Hnley JA (2000) Nested cse-control study of the effects of non-steroidl nti-inflmmtory drugs on brest cncer risk nd stge. Br J Cncer 83: 112 120 Thompson HJ, Briggs S, Prnk NS, Pizz GA, Brendel K, Gross PH, Sperl GJ, Pmukcu R nd Ahnen DJ (1995) Inhibition of mmmry crcinogenesis in rts by sulfone metbolite of sulindc J Ntl Cncer Inst 87: 1259 1260 Thun MJ, Nmboodri MM, Clle EE, Flnders WD nd Heth CW (1993). Aspirin use nd risk of fetl cncer. Cncer Res 53: 1322 1327 Vn Aswegen C, vn Schlkwyk ICD, Roux LJ, Becker PJ nd Du Plessis DJ (1992) A novel study on the effect of cetylslicylic cid on the binding cpcity of estrogen receptors from MCF-7 cells. Clin Physiol Biochem 9: 145 149 Wun LM, Feuer EJ nd Miller BA (1995) Are increses in mmmogrphic screening still vlid explntion for trend in brest cncer incidence in the United Sttes? Cncer Cuse Control 6: 135 144