Tobacco Documents Online

o Im.ruaunnal CliiJcm~olop¢al n....uznun 2000 riimnl 1n U1c n Ihnam e.ran.rn..uel ..... n..,r.y Fr~.nnirak..n' 2ooo.2V 9of Nasopharyngeal carcinoma in Malaysian Chinese: occupational exposures to particles, formaldehyde and heat R Warwick-Armstrong,a Peter B Itnrey,b Munn Sann Lyc,c M Jocelyn Armstrong,a Mimi C Yud and Sham Sanie Background During 1990-1992, 282 Chinese residents of Selangor and the Federal Territory, Malaysia with histologically confirmed nasopharyngeal carcinoma (NPC) were interviewed about occupational history, diet, alcohol consurrmption, and tobacco use, as were an equal number of Malaysian Chinese population controls, pair- matched to cases by age and sex- Methods Exposures to 20 kinds of workplace substances, solar and industrial heat, and cigarette smoke, were analysed by univariate and multivariate methods. . Results Nasopharyngeal carcinoma was associated with occupational exposures to con- struction, metal and wood dusts: motor fuel and oil; paints and varnishes: certain other chemicals; industrial heat; solar heat from outdoor occupauons: certain smokes: cigarette smoking; and childhood exposure to parental smoking. After adjustrnent for risk from diet and cigarette smoke, only wood dust (OR = 2.36; 95% CI : L33- 419), and industrial heat (OR = Z21; 95% CI : 1.12-4.33) remained clearly asso- dated- wood dust remained statistically significant after further adjustment for social class. No significant crude or adjusted association was found between NPC and formaldehyde (adjusted OR = 0.71: 95% CI : 0.34-1.43). Conclusions This study supports previous findings that some occupational inhalants are risk factors for NPC. The statistical effect of wood dust remained substantial after adjustment for diet, cigarette smoke, and social class. Intense industrial heat emerged as a previously unreported risk factor, statistically significant even after adjustment for diet and cigarette smoke. No association was found between NPC and formaldehyde. Keywords Nasopharyngeaf carcinoma, occupations, air particles, formaldehyde, heat, Malaysia Accepted 10 May 2000 Although comparatively rare worldwide, nasopharyngeal carcinoma (NPC) has substantial incidence and mortality in populations of southern Chinese ancestry in China and Southeast Asia. Nasopharyngeal carcinoma in southern Chinese is believed to result from a combination of genetic susceptibiliry. " Departmenr of Communiry Healih, Unlversiry of flllnms. Champaign, IL 61820, USA. ° Departmenr ot Srzrisria, and Depanmeni af Mediczl Infofmaiion Sciener. Universiry of Illinois. Champaign. f461820. USn. ` Institute fue Medical Research, Kuala Lumpur. 50588 Malaysia. 5 Dcpartmeni of Prevenrive Mcdi<ine. Univcrsiry of Southern CaGfornia School ol Mcdidnc. Los Angclcs. CA 90031. USA. ` Department of Geography. Univcrsiii Kebangsaan Malaysia. 8angi. 43600 Malaysia infection with Epstein-Barr virus, and regular consumption of salted fish beginning in childhood.f Nasopharyngeal carcinoma has been associated with salted foods.2-7 with alcohol coruump- tion.3'a cigarette smoking.8 and occupational exposures Z'4.7'9-IZ The epidemiology of NPC has been reviewed by Yu and Henderson.13 We report a case-control study of NPC in Chinese of the Federal Territory (Kuala Lumpur), and State of Selangor. Malaysia. The primary objective was to pursue indications from smaller studies of this population,2'14 and of others, that NPC risk dses with work-site inhalation of dust and smoke particles. formaldehyde, and certain aromatic hydrocarbons. We exam- ined associations of NPC with occupational exposure histories, active and passive cigarette smoking, and consumption of alcohol and various dietary components. Results on dietary exposures including alcohol have been reported.3 Here we focus on asso- dations with occupational exposures and cigarette smoke. 991

996 INTERNATIONAI. IOURNAI. r1F LPIUEMI(1LOGY Table 9 Logisric rcgressou odds rauu. :URi fur Icn(nW c_vposurc Increare nuh an.l snlli~ul adlu<linrcnl lur dficl and dgarcne slnokr indiccs Agent Unadjusted OR (95°/u Ctt Adjusted OR (95 % Cl) ------~- -- P fnr diet and cigarcar smoke adjusted OR" Dusts ConsVUCtion 1.14 (1 02 1 26) 195 (0.93-1 '0I 0.57 Fe[Iilizer 098 (0 64-I 19) 0A 2 t0 51-1 31~ 041 1 Mctal I.IS(102 1.3]) 1.14109a 1 ~. 011 Talc 1 14 10 96-1.511 (LA6(0G1-1'I. 0 19 Tin 1.19 (o.9n-15G) L16 (0.R3-I G 1, p 37 Textilc 1.14(0.96-1_36) 1 .I1) 10.94 142) 016 Wood 1 29 (1 13-I 46) I 29 il 07-I 41, 00018 Othere 1.09 I1 00-1.20) 1 U7 I(I_96-1 1 1l 0.23 Fumes Cooking I t)9 (0.91-L31I 0 117 9 10 ?M-I 21) 0 80 Engtneexhaust 1.o1 (0.91-L12) LUlI/q3N-I 1.1, 096 Wood I 20 (0 97-1.49) 1 0€ 10.84-I 3il 0.54° Other° 1 22 (I n2-1.45) 1 12 i0 .91 1.59i 0 27 Chemicals nloddes 0A8(0.80-1.20) 09310J41.17~ 0.40 Dyes L13 (0.801.58) 1 07 IOJ3-1.5F) 071 Formaldehyde 1.04 (0 8G-1 27/ 0.E8 10 70-I 121 0.2.9 Glues I.16 (0 97-1.38) 1.09 I!1.88-I 341 0 40 OOtrr, chemicalsd I 24 (0.99-1.55) 1.20 (0.92-1 561 0.15 Motnrfuelandoil I 1511D2-1_30) 1.05 (p.vl 12_'1 0.50 Paintsandvzrnishes 1.19(1-03-1.38) 1.08(091-I29i 0.35 Resins 0_8410.65-1_08) 074 i(1 54-I.0:) 0.047 Heat Solar 1_1411_01-1.29) 1 0210 89-11 8; 074 Industrial 1.33 (1_13 1.56/ I 21 I I 01-I 4S) 0025 ' n°a See Table I[nr dermilions. ` Statistically signlficanc quadralic rffen. .~. iexr confidence intervals for formaldehyde OR from these two studies overlap substantially, so Ihc apparent differences may not be matenal_ Considerably more data are needed to resolve the formaldehyde question. In I991-1992, we carried out baccline sampling of ambient air in 42 work sites in 10 industries in the study area. The industries were selected on the basis of occupatiunal histories of NPC cases recorded in this arld Iwo previous Malaysian studies.2,14 Air particle sampling confirmed the presence of high-risk pollution for the nasnpharynx of particles with dia- meter <10 µm in adhesives, metahvorking, ricemills, sawmill- ing. and shoemaking, with mean values exceeding [ 50 µg/rrti which is the US Environmental Prolection Agency's33 24-hour ambient air quality standard for PM10. Formaldehyde levels exceeded the American Conference of Governntental Industrial Hygienists (ACGIH)34 threshold linllt value (TLV-Cetling) of 0.37 mg1m3 in only the adhcsivcs industry but all others had mean 8-hour mncemrations betwec, 0.16 and 0.35 mghn3_ To our knowledge, intense indlulrial heat has not previously been examined in relation to NPC. Dry heat [rom suurres such as furnaces, welding machines, and rolling ntills, combined with the tropical Malaysian climate, yields high working temperatures that may intensity the vulnerabilily of thc nasopharyna to inspired dusts and lomes. The strrngth uf the Olt, and the persistent associauon after adjustment for other nsk factors. indicate that this exposure warrants further investigation. Analyses oI uur data by lateney time frames of 5, 10, 15 and 20 years prior Io diagnosis showed little difference in OR from those based on all exho.ure at least one year prior to diagnosis. Nor were dlete impnrl.ant differences by length of exposure (<10 vcrnus =111 ycars), by age of worker (quaniles, and <45 versus a45 years), or by Chinese subethnidty. This may indicate short latency aud thal acute high-level exposures are more dangerous than chronic low-level exposures. Alternatively, this could be due to sntall numbers of participating pairs in stratified cnmpansons, or to measurement problems. Cigarette sntoke has been associated with cancers of the oral cavity, respiralory tract, and distant organs_ Our data add to an accumulalion of evidence associating cigarettes with NPC, and confirm the finding of Yu u a1.5 that exposure to parental smok- ing ducing childhood p;ays a role. Social class has previously been associated with NPC_IS In this study. social class does not substannally ~nnfournd I he diet/smoking adjusted relationships lo NPC of any of the occupational exposures examined. Our study, as earlier case-control studies in high-risk NPC populations, is limiled by the fairly small numbers of cases and conlrols reporting various uncommon occupational cxposures. 2505587246

OCCUPATIONAI. IiXI'OSURES AND NPC IN MALAVSIA 995 Table 3 Estimatcd odds ratios tor anp vcrvtrs n0 histnry of 22 occupational exposures with and wilhuuo ad)uament for diet and cigarette smoke indices Substances Unadjusted Ingistic regression adjusted Dusts Constructlon I 71 (1 07-2.761 1 31 (0.77-2-24) Fertilizer I 00 (0.18-5. 371 t).52 (0-I0-2 71) Metal 1.90 (1 .07-3.45I 1.62 (0 90-3.08) Talc 1.431049JlA3i 0.49 (0-14-1.66) Tin 1.8010.54-6.84) 1 69 (0.45-6.30) Textile 1 _66 (077 374) 1 77 (0.76-1 .11) Wood 264(1 58-453) 2.36 (1.3" .19) Other6 1.53 (1.03-2.29J 1.39 (0.88-2.16) Fumes Cooking 1.50 (0.68-3.421 0.93 (0.38-2.27) Engine exhaust 1.08 (0.67-1 71 1 1.05 (0.61-1.79) Wood 2.20 (0.99-5-211 I 65 t069-3 92) Otherr 2.00 ( I 01 -i . I I I 1 46 (0.6(-3.23) Chemicals Biocides 1.00 /0.49-2.09) 091 (0.42-1.98) Dyes 1.2010_30--4.98) 1 03 (0.26-1 .02) Formaldehyde 1 24 (0.67-2 32) 0.71 (0.34-I_43) Glues 1 ]5 (0.91-347) 1.30 (0.61-2J6) Other chemlcalsd 2 37 (0.99~ 28) 2.36 (0.92-6.08) Motor fuel and oil 1.79 (L1~2.82) 1.33 (0.81-2.20) Paintsandvarntshes 2.00(t.lo-3J4/ 1.32 10.67-2.571 Resins 0.62 (0.22-1.611 0.41 10-14-1.211 Heat Solar 1.83 (1.09-3.16 1 .20 (0.67-2-14) Industrial 3.u5 1 1.64-6.08) 2.21 (1.12-1.33) u r A Sec Tablc I for dehmuons, in results In particular, wood dust results were essentially unaffected (OR = 1.22. P= 0.0057, 95% Ci : 1.05-1.42). while the estimated effect size for industrial heat was slightly attenu- ated, with concomitant increase in P-value to a conventionally suggestive level (OR = 1.18, P=0.07[, 95% CI : 0.98-1.41)- Discussion Since the nasopharynx serves to trap primarily medium size particles (5-lO pm) in inspired air25 and absorbs soluble chemicals, inhaled carcinogens are biologically plausible risk factors for NPC. Dust particles from wood, metal, construction sites, tin mining, talc, and textiles, and smoke particles from incomplete combustion of wood and other materials, all occur frequently in Malaysian occupational environments and are of sizes and weights deposited mostly in the nasopharynx. Several studies13 have associated wood dust with adenocarcinoma of the nasal cavities and paranasal sinuses, and in 1994 the Inter- national Agency for Research on Cancer (IARC)26 classified wood dust as a human carcinogen. Demers er a(.27 performed a pooled analysis of data from one UK and four US cohort studies on wood dust, and found excess NPC among furniture and plywood workers. But evidence relating NPC to other occupational risk factors has remained less clcar. Irt California, Henderson et a/9 found NPC significantly associated wnh fumcs. smoke. and chemicals, but nut wilh dusts. In A1alaysia, Armstrungetrt/.2 found dose-response relationships of NPC to dust and sntukc, but not to chemicals. In China. Ning er a1.2 associated NPC with smoke, dust, and chemical tumes, as did Yu eral-4 in Hong Kong residenls ¢35 years of aga Chen er aLlo tound an associalion in Taiwan with sntoke bttl not dust, as did Yu et rrLll in China- West er al.12 found an association in the Philippines with dust and engine exhaust. Dnmestic exposure5 to snwke from burning wood, incense, or anti-mosquito coils are suggested risk factors for which case-control studics provide little supporr13 An exception is the finding by vVest er al.12 in the Philippines of an assocLattnn with anlt-ritoCqntttl cndG- Our results confirm earlier observations that NPC in Malaysia is crudely associated with occupational exposures to chemical fumes, smoke and dusts, particularly fronl wood and metals- The persistence of the association with wood dust after adjusl- ment for cigarette smoke, diet, and social class, strengthens the epidemiologic plausibility of a causal pathway involving that ex- posure. Howcver, cigareuc smoke and diet partially confounded the relationships of NPC to other inhalants. We did not find dose-response effects of other inhalants with substantial explan- atory power beyond that of presence or absence. This leaves open the question of whether these inhalanls are biologically active contributors or simply comnron companions of other active agents. Formaldehyde has been suggested as a possible cause of NPC since 1980, when animal studies showed that high doses cause nasal and paranasal cancers in rats. Epidemiological studies have since sought associations between occupational exposure to fonnaldehyde artd various canccrs in low-risk NPC populations in Europe and Noah America. Most of these found an elevated risk for NPC in association with lotmaldehyde,le.2g-3n but not all.S1 The evidence for an association between NPC and formaldehyde has recently been rcvicwed by the US Agency for Toxic Substances and Disease Registry_32 In 1987, the US Environmental Protectinn Agency,32 and in 1995 the IARC.26 classified formaldehyde as probably carcinogenic for humans. In moderate-risk populations, to date the only case-control study to report an association between NPC and formaldehyde is that by west et a(.,12 in a non-Chinese Filipino population. Using occupational histories of 257 subjeets, of whom 60 (23-3%) were exposed to formaldehyde, they found OR of 3.5 for cases first exposed 25 or more years prior to NPC diagnosis, and 3.2 for cases tirst exposed before the age of 25, relativc to those never exposed. These analyses incorporatcd a 10-year latency period. In contrast, our study in Malaysia is the largest case-control investigation of the formaldehyde-NPC relation- ship in a high-risk population, and conveys no suggestion that occupational exposure confers NPC risk. Other differences between our data and those of West er al., c.gc regarding con- sumption of salted and fresh fish, suggest that these populations may indeed differ with respect to NPC risk factors. Ifowever, we identified formaldehyde exposure in only 51 of 564 subjects (9.0 %) of our Malaysian Chmesc sample, of whom only eight had accumulated ~ 10 years of exposure outside a 1 O-year latenp? period. The Malaysian occupations included those where ex- posure to formaldehyde would be expcc7ed, namely: adhesives, foundoes, latex processing, metalworking and welding, plywood manufacture, rubber (ire manufacture, sawmilling, shoe making (glues), and textiles (permanent Press fabrics/. Funhcrmore.

998 INTLRNATIONAL JOt112NAL OF CPIDHMIOLOGY 2binternational Agency for Rrzesr.h un Cancer- Wood Dusr mrd Farmaldelrl'de. ManograPlu arr r)rr Sraivurrmr ef rlre Curcrnogaur Rsk in Hnurnns. Yal 62. Lyon' IARC, 199s. 2l Demers PA, Boffetta P. Kogevinas M rr al Pooled rranalysis of cancer mortality anrong five cohons of wnrkrrs in woodrelated indusrrivs. Scand J Work Environ HealJe 1995,21: 179 -90. 28 Hardcli L. Johansson B. Axeisou C) Fpdemlologlcal study of nasal and nasoplrarynReal cancer and iheir relanon to phenoxy acid chlorophrnul expnsure. Am J lnd Mud 1982;3:247-57. 39ROUSh GC, Walrurh J, Stayner 1.]'. ICaplan SA, Flannery JT. Blair A Nasopharyngeal cancer. sinonasal rancer, and occupations related to formaldehyde' a case-conrrnl srudy. J Vnrl Can.erlrsst 1987:99:1221-24. snBlair A, Sacaca R. Stewan PA. Hayrs RB. Shy C. Epidemiologic evidence on the rclauonshrp hawerrn formaldrhyde cxposure and cancer Scand J v/ork Environ Healu. 1990,16:381-93. Appendix In the context of this paper, a couditional logistic regression model adjusting for diet and cigarette smoke, while incorporating exposure time quantitatively, employs a linear predictor (3DzD+ fiGZ,+(fsx, where x is an exposure measure f,e.g., IogID(l+I), as used for Tablc 3)- Such a model has likelihood function L(A)=11it+exp(f)n(r,lu-zioD) r (k:lrrlC-ziocl+IJx(xil-xra)~- for matched pairs i= I to n(hcr n- 282). whcrc i indcxcs the matched pairs, and I or 0 in the snbscript distinguishes cases from controls.15 Statistical interence for this model is based on the permutation distribution ol f,(#) obtained from joint exchanges, within cach of the 3" possible subsets of pairs, of observed values of the predictor values zdN, zd, and xd for a case ir Purchasc IPIi. Paddlr CM. Docs Inrrnaldchydc cauar nasopharynGCal cancer in man] Cmrrer LeH 19F9:4609-R5. )zUS Dcpanmenr of lieabh and Human Scrvicrs. Public Hcald) Servicc, Agency for Toxic Substances and Disrase Registry. Toxirnlogrra! Profile for FannalJeLydn Atlantz, 19977 rsUS Envuonm<ntal P:n:c.unn Agency. National Ambient Air Qual- rty Standard>. IOn-lu cl (hup-//wwsv-epa.gnv/airs/criterla.htmll 1999 oAnrerican Conferencr uf Goremmcntal Industrml Hygremsts. L)ocumrnrauon of rlee Tlvesdald f inrle va[ua and Rlologiml Exposurr Indicu. 6rlr Edn. Washmgton, DC. 1994. 350rrzlnw Nr, Dey VIl Sra[urrral Mcrhodr m Can¢r Researdr. Ynl l: T/re Anc/:ars o(Case r:m:rel Smdiar [ynm Imernational Ageny for Research on Cancer 1930. p.253 with d)c predictor values zio, z/oo and xio for the corresponding malched contrnl- The adjusted sign test is obtained by simply replacing f)r(xl -xr0) In L(fi) by (1rsi, where si=s(gn(xg _.rio), and vgn(xrl-x[o1 = I if x, >xio,=0 ifxtl=xro, and =-I if sry < xi) - Note then that if L(P) is the correct specificatlorr <d the conditional likelihood, then fix = 0 implies that (3r=0 in ihe modiFied likelihood /. +cx3r r_ ~-t '(91- R I I do'd~ 'ron) i Pcadc-~iacl+Rrsl . which then is identical to 1.(fl). Thus. a test of (3r=0 in L'(,?) is also a test of /3r=0 in L(p). The usual null large-sample chi- square distribution (if the likelihood ratio statistic applies because its derivaor)n is conditional on Ihc values of the predictors, and docs not dcpend on the manner of their azsigmrn-ul-

994 INTERNATIONAI. IOUI2SA:. OF EPIDEMIOLOGV Table I Associarions of nasophar) ngcal camcF with historics of 22 oc<-upatiunal cxpo un,. Exposure (%) - P-value (diet and Substance <:ases Controls F,spused pairs Median di0erence in hours exposed (toos) P-value (sign test) cigarette smoke adjusted sign tesqa Dusts Construction 22.0 14 2 93 -1 5 0_012 0.30 Fenrlizer 1.4 1.4 8 --~ 1 00 044 Metal 16.0 9 6 65 1 1 0 0025 0.06 Talc 3.6 2.5 17 eJ 0.65 0_26 Tin 3_2 1.8 14 /6 6 041 0.41 Tezllle 7 .8 5_0 34 4/, ' 0.12 0_12 Wood 24.8 12.1 92 lo I 0000014 U.0028 Otherb 39.7 31.2 157 , 1 5 0.07 0. 31 Fumes Cooking 67 4_6 31 G10 7 0.47 0_82 Engine exhaust 211.9 199 98 61 0.92 072 Wood 9.6 5_3 37 3 5 00.099 0.54 Othere 10.3 5.3 43 1.5 5 0.031 0.27 Chemlcals Riocides 6 7 67 37 04 1.00 0.63 Uyes 2 1 1.A I I '_ 3 1.00 0 97 Formaldehyde 99 8.2 49 0 o 0.5] 025 Glues I I 0 6.7 47 i x 0.09 0.43 Other chemicakd 51 32 28 Io~ 0.036 0.05 Motor fuel and oil 29.4 199 117 4 0 0_0053 0}6 Parnts and vamishes 16.0 9.6 63 17 .1 0.023 0.46 Resins 11 8 4.6 21 -]C.2 38 0.1(1 Heat Solar 18 1 11.0 75 39.N 0.037 0 75 Industrial 170 6.7 62 174 0.00050 0.027 a Adjusted for diet usrng xn=0 n6 x Ifi r t. -IS) t 0_23 x (f4-ts) _ 086 x fa . 0.29 x 17, where fl - -1 Ci I a~ording to whether subjecl .....sumed sahed fish less than monthly, monlhly bur Irss ihen weekly. or at least weekly 5 years pnoi to dragnoyis ol Jir arc_ f_u defined similarly to fl, bui fnr pork and/ur beef livee IS = 0, 1 or 2 acmrdfing to whrYner Chinese Oowering cabbage was cnnvumed Icss Ihan .veckip bn!h 5 ycars prior to diagnosis and au age 10, ar least weekly at one of thosc umes. anI ai Irasr weekly a1 botb of rhose timer, ly = fa i + 1yi. whue f~ i n ecflucd as II for salted eggs 5 years prior to diagnosis nd f6z is de[ined similarly ar age 10, fs is dehned as fa, but for Iresh shrimp; fa = I or 0 according io wheibr o: oranges and/or tangerlnes were ronsumed at least monthly at either nme: and f- - 1. 5 depending on frequency and amoum of beer coruumpris ears prior ro diagnosis of rhe c s dascobed m Armstrnng rral.l Adjusted [or ciga~vuo s n0ke usmg z,- = I if subjca ever smnked for ai Iean 6 munih• oi s.as exposed lu parunal smaking rn ~he homc during childhood, and 0 other.vise I' Includes cemenL nce. Rour. fenilrznr .and gyhum dusts. ` Includes smokes from oiL tarsm grass a:rd otLcr non-metalGc soumes, excludes smokes [rom wun,f bunmir ~nd cooking. engine exlsaust, and cigarene srnokee d tntludes a variety of acids, bases. solvcnis, detergcnes and soaps_ Table 2 Unadjusted and diet-adjustcd simdtaneously estimated odds rarlos (Ott) for smuking au. i p,vemal smoking among snrokcrs and non-smokers, with P-values and 95% O Exposed pairs .. Cigarette smoking history >6 months Parental smoking for smokersa Parental smoking for non-smokersa __ Unadjusted for dietary Index Adjusted for dietary index OR P 95% CI OR P 95% CI 2.86 0.002 1.92-5J6 1.82 0 16 0.78-1 .23 0.88 0.70 0A7-1.65 0.74 043 0.34-1.59 2.27 0 002 1.32--3.88 2 28 0.008 L21-1 28 a Ovcrall P-values [or rhe hill mdels mnlaming smoking, parental smoking and Iheir mmranlou n~e un I and a 040In Ihe unadjusrcd ar:d ad)uaed mudnlk respeaively, with corresponding P valur. for mierzaion leftea rnndificaUOn of anukmg and poonial si~ o. ngl of 0.018 and 0025. absolute dose. It does nol rcmain significant altcr [9onferrnni by wood drrsl, dieta-s' . and cigarette smoke exposures. Only correction for multiple comparisons industrial heat relamed a statisticaliy significant adjusted Dose-response effects of indusfrial heat and other chemicals, dose-response effecl /OR - 1.23, P = 0.021). Simultaneous metal, tin and textile dusts, and wnod smoke were further adjustment of all dusr n~sponse effects lor dietary and cigarette adjusted to account for posslble snnultaneous confoundiog smoke risks, and .urrenl sncial class pruduced little change 250LJ.872'T4

A dietary and cigaret[c risk-adjusted OR and asymptutic 95% CI were obtained from [he conditional logistic rcgreSsiun model with linear predictor (iozD +(fca, +OSS•, where s• = l or 0 depending on whether the subject was ever or never exposed. To examine possible doseresponse rclanonships, (Iccupational exposure was represented by x= logt,t(f +/(. where t= estimaled hours exposed, since f is a surrogate for cumulative dose of an hypothetically toxic agent. and the logistic tolerance distribu- tion is typically more successful in descrihing quantal responses to log rather than absolute doses of toxic agcnts.24 Logistic regression models were then fit using x, with and without the dietary index zp and cigarette smokee index _r as covariates. Non-linearity was examined by sequential addition of quadratic and, as necessary, cubic terms in x to the modcl. Likelihood ratio tests were used for these and other logistic model curnparisons. Upon choice of a doseresponse curve of order k= I, 2 or 3 as described. effect modification by diet and cigarette smoke were examined by separate sequential addition of imeraction terms z,cI and zc-s~, i= 1....k. until the first non-significant interaction term of each type was reached. In these models, a one unit increase in the exposure measure x= fogfp(t+ 1) represents, to a very close approximation. a tenfold increase in the estimated h exposed. and estimated OR from logistic regressions are presented corresponding to this exposure ratio. Because an association uf wood dust with NPC has previously been reported and is supported by our data. othcr uccupational exposures with OR > 1.5 after adjustment for diet and cigarette smoke were also simultaneously adjusted for diet, ngarette smoke, and wood dust. This was done by adding zty, the log-transformed hours of exposure to wood dust to the models derived above. We also checked for confounding of each occupational exposure by social class by adding three dummy variables for social class to the diet and cigarette adjustment models. To account for a possible long NPC latency, cumulative ex- posures were restricted to each of five time frames: > 1, 5, 10, 15 and 20 years prior to NPC diagnosis- We also classified par- ticipan[s by exposures above each of three thresholds: 10+ years of exposure, 5+ years of high level exposure, and.20+ years of low level exposure, at any time in working lifc The impacts of latency and exposure thresholds were examined by calculating unadjusted OR for NPC using each of the eight risk dichotomies implied by the five latency periods and three thresholds defined above, and screening for trends with time frame or exposure threshold. Results All 282 NPC cases were squamous cell carcinomas. Cases included 195 males aged 19-72 years (mean 45 6 versus 45.1 for controls), and 87 females aged 24-74 (mean 44.6 versus 44_2 for controls). Case and control groups had similar distlibutions of binhplace and marital status, but differed in social class, Chinese subethnicity, and education. Twenty per cent of cases and 8% of controls belonged to the 'poor' social class /P < 0.00005(, reflecting among other factors fewer years of education among cases ( P= (A 1 1). The NPC OR for poor versus al l others was 4.1 (P < 0,01). Childhood social class was not significantly related to NPC after adjustment for current social class. Hokkien Chinese were underrepresented among cases (23%) as compared to con- trols (32%). with Khek and smaller subethnicilies tnoderately PCCUPATIONAI- 12XPOSUR[5 ANL) NPC IN MALAYSIA 993 overrepresen[ed (P = 0 U5), bttt the assnciation of sube[Irnicity with NPC vanished after adjustment for diet.t Table I compares cases with controls for each occupational exposttre. including resolts of crude and diet-adjusled sign tests for statistically significant association. Although 282 case-control pairs were studicd, Ihe numbcr of informative pairs relativc to each individual expusurc (pairs with at least one exposed subject), varicd widely from a low of I I (dyes) to a high of 157 (other fumes), with ntedian of 45 pairs. The power to delccl existing associations with NPC thus varied substantially among the exposun-s. Exposures to construcrlon, metal and wood dusa, solar heal from outdoor occupations, industrial heat (from furnaces, rolling nrills, welding machines, etc.), motor fucl and oil, paints and varnishes, other chemicals (primarily acids. bases, solvents, de[ergents, and soaps), and other smoke (from oil, tars, grass and other non-metallic sources), show statistically significant (a = 517,) excesses among cases. The median hourly case-control differences in exposure were betwcen 1000 and 8000, with the exccpGon of mo[or fuel and oil where the differ- ence was 460 hours. Ifowever, only wood dust, other chemicals, and industrial Iteat rernain statistically significant (ce = 5%) after adjustment for risk from diet and cigarette smoke. History of active cigarette smoking (>6 monrhs) was asso- ciated with NPC (OR = 1.66. P = 0.012). After adjustment for such history. exposure to passive parental smoking during child- hood was associated with NPC as well (OR = 1.54, P= 0-0401- fn this multiple logistic regression model, non-smokers exposed to parental smoking in childhood exhibited similar excess risk of NPC as did active smokers, whether the latter were passively exposed in childhood or not- Table 2 shows these results with and without adjustment for die[. Passive exposure to smoke from spouse or other household member shosved no association with NPC after adjustment for active smoking. Models including estimated pack-ycars of active smoking and years of exposure to parental smoking (not reported) added no significant predictive power to the models in Table 2. Table 3 shows crude and diet/smoking-adjusted OR for any versus no history of each occupational exposure. Wood dust (OR = 2.36), other chemicals (OR = 2.36), and industrial heat. (OR = 2.21) have the highest apparent diet-adjusted OR, followed by mctal, textile, and tin dusts. and wood smoke, with essentially equal OR between 1-65 and 1.77. Only wood dust and industrial heat achieve statistical significance after adjustment_ Table 4 represents dose-response relationships in terms of OR associated with a tenfold ratio of hours exposed. Wood dust, industrial heat, and other chemicals again have the highest crude and diet/smoking-adjusted OR. Adjusted OR for metal, tin, and textile dusts, and for other smoke. form a lower tier. Based on adjusted dose-response relationship, only wood dust (observed median case-control difference of 4610 h) and in- dustrial heat (observed median difference of 3740 h) retain stat- istically significant positive associations with NPC. Exposure to resins was moderately negatively associated with NPC (OR = 074(, with a marginally statistically significant P-value- Neither effect modification by diet or cigarette smoke nor non-linear dose-response relationships were found for these variables- A quadratic effect nf exposure to wood smoke was found in ex- ploratory analysis. This is likely anifaaual, as it occurs without a statistically significant linear trend and suggests a biologically implausible decline in impact ot progressively larger increases in

992 INTERNATIONAL JOLiRNAL OF EPIDGMLOLO(.1' Methods Cases were ascenained through records of NPC diagnosis and/or treatuunt at four centres with radiotherapy in the study area of Selangor and the Federal Territory General Hospital, 'Nng Shin Hospital, Pantai Medical Ccnue( and Compuler Radionicti Centre, Kuala Luntpur. 6etween l July 1990 and 30 June 1992 wc identified 530 Chinese cases with histologically confirnrcd NPC who had resided in the study area for at Ieast 5 years, and been diagnosed between I January 1987 and 30 June 1992-Of these. 121 (23 %) had died, 63 (12%) could not be located. 4(t%I were too ill for interview, and 60 (11 %) declined partiripation, leaving 282 cases (53%) for study. Of the 530 eligible cases, 282 were prevalent (diagnosed before 1990) and 245 were incident cases (1990 to mid-1992). The 282 NPC cases for study comprised 119 (42%) prevalent and 163 (58 %) incident cases. The modest pantdpation rate was largely due to the attempt to obtain a population-based sample, mdudtng non-hnspitalized cases and prevalent cases with onset as much as 3.5 years prior to initiation of data collection. Specifically, of the 125 cases who could not be interviewed because they were too ill or had died. 104 (83 %) were prevalent cases when ascertained. So were 41 (65 %) of those who could not be located, some of whom may also have died. Each case was matched by iex and age (within 3 years) to one control partieipant in good health with no history of cancer of the head, neck, or respiratory system, selected front the general Chinese population of the study area using a standard procedurc of multistage area sampling- For each case, an interviewer began at a randomly chosen house in a randomly selected postal code district of a Chinese neighbourhood, and proceeded house-to- house by a standard algorithm15 until a qualified control, also resident in the study area for at least 5 years, was found. The overall refusal rate among eligible controls was 10%, but in affluent neighbourhoods it was 20%. Data were collected from each participant during two 40-50-ntinute in-home, structured interviews by specially trained full-time Chinese interviewers fluent in all local dialects. The interviews requested complete residential and occupational history, information on use of alcohol and tobacco, and frequencies of consumption of 55 food items at age 10 and at 5 years prior to diagnosis of [YPC for the case (for matched controls, same calendar year as the index case). For each job in the occupational history, the interview covered job description, work performed, calendar time, machines, tools and substanccs use4 size and type of work- place, exposure to dusts, smoke, gases and chemicals. Exposures to 20 inhalants and heat from two sources were recorded by trade or profession with calendar years. frequency (days per week), and duration (hours per day) of exposure. Questions foi- Iowed the format of Germ er a! 16 and Gerin and Siemiatycki. 17 Inhalant selection was limited to those dusts, smoke, and gascs associated with deposition or absorption in the nasopharynx, with special attention to fotmaldehyde- Exposure to inhalants was suhsequently coded by one of us (RWA) who is familiar with Malaysian industries and hygiene. Coding was conducted blind to case-control status. Codes were adapted from Hoar er al.,t 8 Gerin cr af.16 and Gedn er al 19 Jobs were classified using official Malaysian occupational codes-zo Lcvcl of exposure to inhalants (cvcr/never: low, mediurn. high) was a.scssed .+ath referutce to kind of jub, work perfuretede mode of coniact (respiratory and/ur cutaneous), respondcnt'~ reporting of exposure to paniadar inhalants, years of exposure. frequency, aud duralion. Parlicipants were asked about history of active smoking and expusurc w cigareuc smoke from spause and other family ment- bcrs, and from parents while growing up. They were also asked about cducational ievcl, occupation, job status (employed, retircd, etc.), spouse's and parental occupation, and house type as a basis fur establishing social class. Data analysis Consistenl with nwv other sludies, we focus on duration nf exposure. Estimated hours of each work-site exposure was calculaied by subject Cry suntming 52 x (calendar years in job) r. (workdays pet week) x (hours per day) over jobs where tht exposure .vas prrsent. Stausncal sigmficance of each exposure's crude association with NPC was examined by sign test of estimated hours ex- posed, cxduding tied pairs. The sign test was used because of ils robustness and good performance with heavy-tailed distribu- tions.21,'2 Cigarette smoking was studied by multiple logistic regression- Cigarette smoking history was defined as present if the subject reported ever having smoked for a period of 6 months or more, and abscnl otherwise. For subjects with smoking historirs, .umulative pack-vears were estimated as reported years of srnoking multipFied by scores of 025, 0.75, 1.25, and 2.00 for sutokers, respectively, averaging <10, 11-20, 21-30 and >30 dgarettes dail}. Sve examined models using dichotomous variables for cigarette smoking history >6 months, exposure to parental cigarette smoking in chlidhood, and exposure to cigar- ette smoking by spouse or other household ntember. We con- sidered estimated accumulated pack-years of smoking, years of living with smoking parents, and years of living with smoking spouse or other household member as crude indices of dose, and these were also evaluated as quaruitative predicYors. Social class categurnes (poorr lower middle, upper middle, high) based on t,iberatos et aL23 were constructed for each partieipant I'or time of tnterview. and childhood (age 10). Social class coding was blind In case-control status. To account for possible confounding, a dietary and cigarette risk-adjustcd sign test (Appendix) was obtained from a con- ditional logistic regtession model with linear predictor (fi~Zp+ J3,zc+ (3ss), whcre z/, is a dietary risk index derived previously from these data3 which summarizes past consumption of salted fish, salted egg, purk or beef liver, shrimp, Chinese flowering cabbage, oranges or tangerines, and beer; zC is a cigarette smoke risk index, where :, = I if the case or control had smoked cigarettes for >6 rnunths andlor had been exposed to parental smoking, a:id zr = 0 if ncither: and s= L or 0 is an indicator variable tdeutifyumg the member of each matehed pair with highest eslimatcd hours of past exposure. The likelihood ratio test of No: J3s = 0 Lrom thc model-based conditional likelihood gives the adj:utcd tesl- To mdirate the svength of observed statistical relationships, the medtan case-<'ontrol difference in hourly exposures was calculated after excloding unexposed pairs. The maximum (condiiionall likelihood point estimate of the odds ratio (Oli) for any expusurr versus none, with exact 95% CI, was obtainrd fronr tho subsci of I,airc wlth cxanly one exposed subject 2505587242

Combined with random exposure nteasurement and occasional misclassification errors, this otay have produced low power to detect statistically significant dose-response relationships- Also, in view of limited power, we did not adjust our significance testing for multiple comparisons; thus, occasional Type I errors would not be surprising. (However, the resuh for wood dus( in Table 4 persists even after Bonferroni adjustntent.) On the other hand, although standardized interviews were carried out by trained professional interviewers using stmaured questionnaires, recall and exposure-suspicion biases may have contributed to the predominance of posirive over negative associations in our results. In addition, it should be noted that h is relatively easy to assess wood dust exposure in populalion-based casecontrol studies whtle it is much more diffintlt to asscss formaldehyde exposure, and this may be a factor in our negative findings for the latter. Finally, 24% of the cases we ascertained had died or were too ill for interview at time of contact, most of these being prevalent cases, contributing to a 47% non-participation rate among diagnosed cases. Therefore, we cannot exclude the possibility of prevalence-incidence (Neyman) or other selection biases, though we are unaware of any specific presumptive rationale or evidence for their presence. In summary, our data support previous findings that some occupational inhalants are risk factors for NPC. Tlte statistical effect of wood dust remained substantial after adjustment for diet, cigarette smoke, and social class. Intense industrial heat emerged as a previously unreported risk facaor. statistically significant even after adjustment for diet and cigarette smoke. No association was found between NPC and formaldehyde. Base- line sampling of ambient air conditions in Malaysian worksites confirms the presence of potentially high-risk particle pollution. Acknowledgements We thank, in Kuala Lumpur, Mr KIi Chiew, Mrs PF Chow, Miss CM Lim, Mr HC Loke, Mr KC Ong, and Mr KS Wu for interviewing; Miss KE Chua and Mrs PL Ng for secretarial support; the Director, Institute for Medical Research for admin- istrative support; Drs A Lim, Chief, Institute for Radiotherapy and Nuclear Medicine. General Hospital, MR Tan. Cancer Treatment Centre, Tung Shin Hospital. S Narasimha, Computer Radionics Centre, and S Ganesan, Pantai Medical Centre for case referral; and in Champaign, Miss L Davis, Mr D Grammenos, and Miss R Moorman for assistance In data preparation. This work was supported by the US Nationa] Cancer Institute; Grant number: R01-CA46567. References I Ho lHC- Nasopharyngeal carcinoma (NPCI. Adr Cancer Res 1972; 15:54-]-82. 2 Armstrong RW. Armstrong MJ, Yu MC, lienderson BE. Salted fish and inhalants as risk factors for nasopharyngeal rarnnuma in Malaysian Chinese. Cancer Res 1983;43:2962-T0. tArmstrong RW, Imrey PB, Lye MS, Armstrong MJ. Yu MC, Sani S. Nasopharyngeal carcinoma in Malaysian Chinesre salted fish and other dietary exposures. fnt J Cancer 1998;Y7:22A-15. 4 Yu MC, Ho IHC, Li S-H, Henderson BE Cantonese-srvle salted fish as a cause of nasopharyngeal carcinoma- report of a case-control study in Hong Kong. Cancer Res 1986.46:956-61 OCCUI'ATIONAL E%POSURL'S AND NPC IN MALA1'SIA 997 5 Yu MC. Mo C-C. Chong W-X. Ych F-S. Hendersun BE Preserved foods and nasupharyngeal carcimm. a: a case-convol study in Guangxi, China. Canro Res 1988;48:1954-59. 6 Yu MC. Huang TB, Henderson BE. Diet and nasopharyngeal caninoma: a case-control study in Guangzhou. China. Inr J Canar 19A9;43: I0]2-82 7Ning lP, Yu MC, WangQS. Henderson BE. Consumption of sahed fish and other risk factors for nasnpharyngeal carcinoma in Tianjin, a lowrisk reglon fur NPC in the People's Republic of China l NarlCancofnst 1990;82:291-96. BNam JM, Mclzughlin JK, Blot WJ. Cigarette smoking, alcohol. arrd nasopharyngeal carcinoma: a case-comrol study among US Whites J Nml C'aucer lusE 1992;84:619-22 9Henderson BE. Louie E, ling lS, Buell P Gardner MB Risk factors associated with nasopharyngeal carcinoma. N Deg! J Med 1976, 295:I101-06. I0Chen C-J. Wang YP Shieh T. Chen l-1', Lio M-Y. Multifacturial etiology of nasopharyngeal carcinoma. EpsteinBarr virus, familial lendency and enviromnental cofaanrs. InWulf GT. CareY TE /edsl Head and Neck Onmlogy Research. Amsterdam: Kugler 1988, pp. 469-'/6 11 Yu MC, Garabranr DH, Huaug TB, Henderson BE. Occupational and other non-dietary risk factors for nasopharyngeal carcinoma in Guangzhou, China. Inr J Cancer 1990:45:1033-39. 12 West S. Hildesheim A, Dosemeci M. Nonviral risk facmrs for naso- pharyngeal carcinoma in the Philippines: results from a case-control smdv. fnr J Cmrcrr 1993:55:]22-27. 11Yn MC, Henderson BE. Nasopharyngeal cancer In: Schottenfeld D, Fraurneol JF (eds) . Cancer Epidemiology and Prevention. 2nd Edn. New York: Oxford Universiry Press, 1996, pp.603-18. taArmstrong RW, Kannan Kutty M, Armstrong MJ. Self-specific environments associated with nasopharyngeal carcinoma in Selangor, Malaysia. .Soc Sri Med 19J8;12D:149-56. Is Kelsey JL, Douglas Thompson W, Evans AS. Methods in Observarional Epidemiology New Yurk Oxford University Press. 1986. pp.264-,s5. t6Gerin M. Siemiatycki J. Kemper H, Begin D. Obtaining occupauonal exposure histur(rs in epidemiologic case-control srudies. l OcruP Med 1985;27:420-26. t7 Germ M, Siemiatycki 1 The occupational questionnaire in rerro- spective epidemiologi< studies: recent approaches in community- based studies. APplOrcup Environ Hyg 1991;6:495--501. taHoar SK. Morrison AS, Cole P Silverman UT. An occupation and exposure linkage system for the study of occupational caranugenesis. J Ouup Med 1980:22:7 22-26. 19Gerin M. Siemiatycki l, Nadon L Dewar R, Krewski D. Cancer risk due to occupational exposure to formaldehyde: results from a multi- site case-conrrol srudy in Monveall lnt l Cancer 1989.44:53-58. 20Malaysia Kementerian Buruh dan Tenaga Rakyat. TerjemahnnTajuk- Tajuk Pnrke jaan Dagi Kamus Pengke/usun Peke jaan (Occupational Titles for the Dictionary of Occupational Classification). Kuala Lumpur, 1988. ~t Pratt JW, Gibbons JD. Concepts of Nonparamerric Theory. New York: Springer-Verlag, 1981. P.384. 12Randles RH. Wolle DA. Inrroduaicn to tlte Theory oJ Nonparametrir Smtisua. New Ynrk: Wilry, 1979, p.116. 23 Ltberatos P. Link BO, Kclsey JL. The measurement of social class in epidemiology. Epidemio! Rev 1988;10:8T-1 21 24Gaylor DW. Dose-respome models in risk analysis In: Armitage P Calton T(eds). Eoryclopedia of Riostatisucs. l o7. 2. Chichester'. Wilcy. 1998,pp.1219-23. 25 Menzel OB. Amdur MO. Toxic responses of the respiratory sysrem. In. Klaassen CD. Amdur MO, Duull J(edsl. Casarerr and Do.dl s Taximlogy. 9rd Edn. New York. Macmillan, 1986, Pp33t}58