Ora lncogy - unimi.it

8/5/2019 https://web8.elsevierproofcentral.com/en/index.html?token=SPS9da997bc819d5a81274b086f4680192c&type=AU

https://web8.elsevierproofcentral.com/en/index.html?token=SPS9da997bc819d5a81274b086f4680192c&type=AU 1/12

UNCORRECTED PROOF

Oral Oncology xxx (xxxx) xxx-xxx

Contents lists available at ScienceDirect

Oral Oncology

journal homepage: www.elsevier.com

Joint effects of intensity and duration of cigarette smoking on the risk of head and neckcancer: A bivariate spline model approach

Gioia Di Credico , Valeria Edefonti , Jerry Polesel , Francesco Pauli , Nicola Torelli , Diego Serraino ,Eva Negri , Daniele Luce , Isabelle Stucker , Keitaro Matsuo , Paul Brennan , Marta Vilensky ,Leticia Fernandez , Maria Paula Curado , Ana Menezes , Alexander W. Daudt , Rosalina Koifman ,Victor Wunsch-Filho , Ivana Holcatova , Wolfgang Ahrens , Pagona Lagiou , Lorenzo Simonato ,Lorenzo Richiardi , Claire Healy , Kristina Kjaerheim , David I. Conway , Tatiana Macfarlane ,Peter Thomson , Antonio Agudo , Ariana Znaor , Leonardo F. Boaventura Rios , Tatiana N. Toporcov ,Silvia Franceschi , Rolando Herrero , Joshua Muscat , Andrew F. Olshan , Jose P. Zevallos ,Carlo La Vecchia , Deborah M. Winn , Erich M. Sturgis , Guojun Li , Eleonora Fabianova ,Jolanda Lissowska , Dana Mates , Peter Rudnai , Oxana Shangina , Beata Swiatkowska ,Kirsten Moysich , Zuo-Feng Zhang , Hal Morgenstern , Fabio Levi , Elaine Smith , Philip Lazarus ,Cristina Bosetti , Werner Garavello , Karl Kelsey , Michael McClean , Heribert Ramroth , Chu Chen ,Stephen M. Schwartz , Thomas L. Vaughan , Tongzhang Zheng , Gwenn Menvielle , Stefania Boccia ,Gabriella Cadoni , Richard B. Hayes , Mark Purdue , Maura Gillison , Stimson Schantz , Guo-Pei Yu ,Hermann Brenner , Gypsyamber D'Souza , Neil Gross , Shu-Chun Chuang , Paolo Boffetta ,Mia Hashibe , Yuan-Chin Amy Lee , Luigino Dal Maso

Department of Statistics, Padua University, Padua, ItalyBranch of Medical Statistics, Biometry and Epidemiology G. A. Maccacaro , Department of Clinical Sciences and Community Health, Università degli Studi di Milano, Milano, ItalyCancer Epidemiology Unit, Centro di Riferimento Oncologico di Aviano (CRO), IRCCS, Aviano, ItalyDepartment of Economics, Business, Mathematics and Statistics, University of Trieste, Trieste, ItalyDepartment of Biomedical and Clinical Sciences, Università degli Studi di Milano, Milano, ItalyUniviverité Rennes, INSERM, EHESP, Irset (Institut de Recherche en Santé, Environnement et Travail), UMR_S 1085, Pointe-à-Pitre, FranceInserm, Center for Research in Epidemiology and Population Health (CESP), Cancer and Environment team, Université Paris-Sud, Université Paris-Saclay, Villejuif, FranceAichi Cancer Center Research Institute, Nagoya, JapanInternational Agency for Research on Cancer, Lyon, FranceInstitute of Oncology Angel H. Roffo, University of Buenos Aires, ArgentinaInstitute of Oncology and Radiobiology, Havana, CubaEpidemiology, CIPE/ ACCAMARGO, Sao Paulo, BrazilUniversidade Federal de Pelotas, Pelotas, BrazilHospital Moinhos de Vento, Porto Alegre, BrazilEscola Nacional de Saude Publica, Fundacao Oswaldo Cruz, Rio de Janeiro, BrazilUniversity of São Paulo School of Public Health, São Paulo, BrazilInstitute of Hygiene & Epidemiology, 1st Faculty of Medicine, Charles University, Prague, Czech RepublicLeibniz Institute for Prevention Research and Epidemiology, BIPS, Bremen, GermanyUniversity of Bremen, Faculty of Mathematics and Computer Science, Bremen, GermanyDepartment of Hygiene, Epidemiology and Medical Statistics, School of Medicine, National and Kapodistrian University of Athens, Athens, GreeceDepartment of Cardiovascular and Thoracic Sciences and Public Health, University of Padova, Padova, ItalyDepartment of Medical Sciences, University of Turin, Turin, ItalyTrinity College School of Dental Science, Dublin, IrelandCancer Registry of Norway, Oslo, NorwaySchool of Medicine, Dentistry, and Nursing, University of Glasgow, Glasgow, UKUniversity of Aberdeen Dental School, Aberdeen, UK

Abbreviations: CI, cred i ble in ter val; HNC, head and neck can cer; IN HANCE, In ter na tional Head and Neck Can cer Epi demi ol ogy; NUTS, No-U-Turn Sam pler; OCP, oral cav ity andphar ynx; OR, odds ra tio; WAIC, Wanatabe-Akaike In for ma tion Cri te rion.

Corresponding author at: Branch of Medical Statistics, Biometry and Epidemiology G. A. Maccacaro , Department of Clinical Sciences and Community Health, Università degli Studidi Milano, via Venezian 1, 20133 Milano, Italy.

Email address: valeria. edefonti@ unimi. it (V. Edefonti)

https://doi.org/10. 1016/ j. oraloncology. 2019. 05. 006Received 12 October 2018; Received in revised form 27 March 2019; Accepted 5 May 2019Available online xxx1368-8375/ © 2019.

a b, c d d c

e f g h i j

k l m n o

p q r, s t u

v w x y z

aa ab i ac p

c i ad ae af

b ag ah ah ai

aj ak al am an

ao ap aq ar as at

au av aw ax ay az

az az ba bb bc, bd

be, bf bg ag bh bi bj

bk, bl, bm bn bo bp bq

br bs c

a

b

c

d

e

f

g

h

i

j

k

l

m

n

o

p

q

r

s

t

u

v

w

x

y

z



UNCORRECTED PROOF

G. Di Credico et al. Oral Oncology xxx (xxxx) xxx-xxx

University of Hong Kong, Hong Kong, ChinaInstitut Catala Oncologia, Barcelona, SpainUniversity of São Paulo School of Medicine, São Paulo, BrazilPenn State College of Medicine, Hershey, PA, USAUniversity of North Carolina School of Public Health, Chapel Hill, NC, USADivision of Head and Neck Surgical Oncology in the Department of Otolaryngology/ Head and Neck Surgery at Washington University School of Medicine, USADivision of Cancer Epidemiology and Genetics, National Cancer Institute, Bethesda, MD, USAUT M.D. Anderson Cancer Center, Houston, TX, USARegional Authority of Public Health in Banska Bystrica, SlovakiaThe M. Skasodowska-Curie Memorial Cancer Center and Institute of Oncology, Dept. of Cancer Epidemiology and Prevention, Warsaw, PolandNational Institute of Public Health, Bucharest, RomaniaNational Institute of Environmental Health to National Public Health Institute, Budapest, HungaryCancer Research Centre, Moscow, RussiaInstitute of Occupational Medicine, Lodz, PolandRoswell Park Cancer Institute, Buffalo, NY, USAUCLA School of Public Health, Los Angeles, CA, USADepartments of Epidemiology and Environmental Health Sciences, School of Public Health and Department of Urology, Medical School, University of Michigan, Ann Arbor, MI, USAInstitut Universitaire de Medecine Sociale et Preventive (IUMSP), Centre Hospitalier Universitaire Vaudois and University of Lausanne, Lausanne, SwitzerlandCollege of Public Health, University of Iowa, Iowa City, IA, USAWashington State University College of Pharmacy and Pharmaceutical Sciences, Spokane, WA, USADepartment of Oncology, Istituto di Ricerche Farmacologiche Mario Negri IRCCS, Milan, ItalyDepartment of Otorhinolaryngology, School of Medicine and Surgery, University of Milano, Bicocca, Monza, ItalyBrown University, Providence, RI, USABoston University School of Public Health, Boston, MA, USAUniversity of Heidelberg, GermanyProgram in Epidemiology, Division of Public Health Sciences, Fred Hutchinson Cancer Research Center, Seattle, WA, USADepartment of Epidemiology, Brown University School of Public Health, Providence, RI, USASorbonne Université, INSERM, Institut Pierre Louis d Epidémiologie et de Santé Publique IPLESP, Department of Social Epidemiology, F75012 Paris, FranceDepartment of Woman and Child Health and Public Health, Public Health Area, Fondazione Policlinico Universitario A. Gemelli IRCCS, Roma, ItalySezione di Igiene, Istituto di Sanità Pubblica, Università Cattolica del Sacro Cuore, Roma, ItalyDipartimento di Scienze dell Invecchiamento, Neurologiche, Ortopediche e della Testa-Collo, Fondazione Policlinico Universitario A. Gemelli IRCCS, Roma, ItalyIstituto di Clinica Otorinolaringoiatrica, Università Cattolica del Sacro Cuore, Roma, ItalyDivision of Epidemiology, New York University School Of Medicine, New York, NY, USAThoracic/ Head and Neck Medical Oncology , The University of Texas MD Anderson Cancer Center, TX, USA

New York Eye and Ear Infirmary, New York, NY, USAMedical Informatics Center, Peking University, ChinaDivision of Clinical Epidemiology and Aging Research, German Cancer Research Center (DKFZ), Heidelberg, GermanyDivision of Preventive Oncology, German Cancer Research Center (DKFZ) and National Center for Tumor Diseases (NCT), Heidelberg, GermanyGerman Cancer Consortium (DKTK), German Cancer Research Center (DKFZ), Heidelberg, GermanyJohns Hopkins Bloomberg School of Public Health, Baltimore, MD, USADepartment of Head and Neck Surgery, Division of Surgery, The University of Texas MD Anderson Cancer Center, Houston, TX, USAInstitute of Population Health Sciences, National Health Research Institutes, TaiwanThe Tisch Cancer Institute, Icahn School of Medicine at Mount Sinai, New York, NY, USADivision of Public Health, Department of Family & Preventive Medicine and Huntsman Cancer Institute, University of Utah School of Medicine, Salt Lake City, UT, USADivision of Public Health, Department of Family & Preventive Medicine, University of Utah School of Medicine, Salt Lake City, UT, USA

A R T I C L E I N F O

Keywords:Bi vari ate spline mod elsCig a rette smok ing du ra tionCig a rette smok ing in ten sityHead and neck can cerIN HANCELa ryn geal can cerOral cav ity and pha ryn geal can cers

A B S T R A C T

Objectives: This study aimed at re-eval u at ing the strength and shape of the dose-re sponse re la tion ship be tweenthe com bined (or joint) ef fect of in ten sity and du ra tion of cig a rette smok ing and the risk of head and neck cancer (HNC). We ex plored this is sue con sid er ing bi vari ate spline mod els, where smok ing in ten sity and du ra tionwere treated as in ter act ing con tin u ous ex po sures.Materials and Methods: We used in di vid ual-level pooled data from 33 case-con trol stud ies (18,260 HNC casesand 29,844 con trols) par tic i pat ing in the In ter na tional Head and Neck Can cer Epi demi ol ogy (IN HANCE) consor tium. In bi vari ate re gres sion spline mod els, ex po sures to cig a rette smok ing in ten sity and du ra tion (com paredwith never smok ers) were mod eled as a lin ear piece wise func tion within a lo gis tic re gres sion also in clud ing poten tial con founders. We jointly es ti mated the op ti mal knot lo ca tions and re gres sion pa ra me ters within theBayesian frame work.Results: For oral-cav ity/ pha ryn geal (OCP) can cers, an odds ra tio (OR) >5 was reached af ter 30 years in cur rentsmok ers of 20 or more cig a rettes/ day. Pat terns of OCP can cer risk in cur rent smok ers dif fered across strata ofal co hol in ten sity. For la ryn geal can cer, ORs >20 were found for cur rent smok ers of 20 cig a rettes/ day for

30 years. In for mer smok ers who quit 10 years ago, the ORs were ap prox i mately halved for OCP can cers,and 1/ 3 for la ryn geal can cer, as com pared to the same lev els of in ten sity and du ra tion in cur rent smok ers.Conclusion: Re fer ring to bi vari ate spline mod els, this study bet ter quan ti fied the joint ef fect of in ten sity and dura tion of cig a rette smok ing on HNC risk, fur ther stress ing the need of smok ing ces sa tion poli cies.

Introduction

In re cent years, to bacco smok ing has con firmed its role as theworld wide lead ing cause of pre ventable dis eases and death. It is respon si ble for at least 12% of all deaths and of 22% of all can cer deathsworld wide [1]. To bacco has been strongly as so ci ated with head and

neck can cer (HNC, i.e., can cers of the oral cav ity, phar ynx, and lar ynx),which is the sixth most com mon can cer type, ac count ing for nearly900,000 new cases in 2018 [2].

A dose-re sponse re la tion ship with HNC risk has been re ported forsmok ing in ten sity (cig a rettes per day re ported dur ing the ex po sure period) and du ra tion (years of ex po sure), as well as for a cu mu la tivesmok ing ex po sure mea sured in pack-years [3,4]. Com pared with in ten

2

aa

ab

ac

ad

ae

af

ag

ah

ai

aj

ak

al

am

an

ao

ap

aq

ar

as

at

au

av

aw

ax

ay

az

ba

bb

bc

bd

be

bf

bg

bh

bi

bj

bk

bl

bm

bn

bo

bp

bq

br

bs



UNCORRECTED PROOF


sity, smok ing du ra tion was shown to pro vide a greater risk of de vel oping HNC [5,6], as well as, other to bacco-re lated can cers, such as lung[7,8] and blad der can cers [9,10].

So far, the as so ci a tion be tween to bacco smok ing and HNC risk waseval u ated by con sid er ing du ra tion and in ten sity as ei ther sep a rate orin ter act ing pre dic tors. In the lat ter case, the com bined ex po sure wasmod eled as ei ther cross-prod uct or pack-years [6,11,12]. Step-func tions[4, page 369] were widely used to pro vide risk es ti mates for the cross-prod uct of the cat e go rized ex po sures. This ap proach as sumed a constant rate within each com bined cat e gory of ex po sure; how ever, thisas sump tion may have been too rough and/ or lead ing to some ef ficiency loss [13 15]. As an al ter na tive, lin ear-ex po nen tial mod els es timated the ex cess odds ra tio (OR) of HNC within mod els in clud ingpack-years and smok ing in ten sity [6,9]. The pack-year ap proach wastra di tion ally based on a strong as sump tion about the ef fects of smok ingin ten sity and du ra tion; for in stance, the ef fect of smok ing ½ pack perday (10 cig a rettes) for 40 years was equiv a lent to smok ing 2 packs (40cig a rettes) per day for 10 years. In deed, both of them re flect a cu mu lative ex po sure of 20 pack-years [16]. How ever, the 2 ex po sures weremea sured on dif fer ent met rics.

Bi vari ate spline mod els have the po ten tial to pro vide a more re al istic rep re sen ta tion of the as so ci a tion be tween smok ing in ten sity and dura tion and HNC risk [15]. First, they al low to model in ten sity and dura tion in their orig i nal scale of con tin u ous ex po sures and, in this aspect, splines out per form step-func tion mod els. Sec ond, the two ex posures are al lowed to de ter mine dis ease risk in their sep a rate or in ter acting role. Third, a non-lin ear ity in the dose-re sponse re la tion ship maybe mod eled.

This study re-eval u ated the joint ef fect of in ten sity and du ra tion ofto bacco smok ing on HNC risk, by means of bi vari ate spline mod els,within the In ter na tional Head and Neck Can cer Epi demi ol ogy (INHANCE) con sor tium [11].

Materials and methods

Data source

The IN HANCE con sor tium was es tab lished in 2004 to elu ci date theeti ol ogy of HNC through pooled analy ses of in di vid ual-level data fromsev eral stud ies on a large scale [11]. Sev eral as pects of to bacco smoking and HNC risk have been pre vi ously in ves ti gated within the con sortium [5,6,11,17 19].

From the IN HANCE con sor tium pooled dataset (ver sion 1.5), we extracted all the avail able case-con trol stud ies (35 stud ies) that col lectedin for ma tion on cig a rette smok ing sta tus, in ten sity, and du ra tion at indi vid ual level (http: // www. inhance. utah. edu, last ac cessed March25th, 2019). Avail able data were har mo nized at the study co or di nat ingcen ter [11]. In all the stud ies, in for ma tion on smok ing his tory was self-re ported. While dif fer ent stud ies had used dif fer ent de f i n i tions of smoking sta tus, the cur rent study de fined as never smok ers those in di vid u alswho never smoked reg u larly, or smoked for a very short pe riod, i.e.,less than 12 months [11]; like wise, for mer smok ers were de fined asthose who had ab stained from any type of smok ing since at least12 months be fore can cer di ag no sis (cases) or in ter view (con trols).

De tails on in di vid ual stud ies, data har mo niza tion, and pool ingmeth ods are sum ma rized in Table 1. In formed con sent was ob tainedfrom all study sub jects. The in ves ti ga tions were ap proved by the rel evant Boards of Ethics, ac cord ing to the reg u la tion in force at data collec tion time.

Selection of subjects

The IN HANCE pro to col al lowed in clu sion of in va sive can cer casesof the oral cav ity, orophar ynx, hy pophar ynx, oral cav ity or phar ynx

Table 1Dis tri b u tion of cases of oral and pha ryn geal can cer, la ryn geal can cer, and con trols accord ing to se lected vari ables. IN HANCE con sor tium.

Controls (%)

Oral andpharyngealcancer cases (%)

Laryngealcancercases (%)

29,844 13,317 4943SexFemale 9599 (32) 3786 (28) 730 (15)Male 20,245 (68) 9531 (72) 4213 (85)Age (years)<40 2063 (7) 708 (5) 94 (2)40 to 44 2045 (7) 823 (6) 207 (4)45 to 49 3018 (10) 1692 (13) 471 (10)50 to 54 4231 (14) 2321 (17) 787 (16)55 to 59 5044 (17) 2627 (20) 1025 (21)60 to 64 4726 (16) 2185 (16) 1011 (20)65 to 69 4181 (14) 1489 (11) 754 (15)70 to 74 3044 (10) 922 (7) 413 (8)

75 1492 (5) 550 (4) 181 (4)RaceWhite 21,462 (72) 9076 (68) 3627 (73)Black 976 (3) 683 (5) 169 (3)Hispanic 421 (1) 149 (1) 41 (1)Asian and

PacificIslanders

3849 (13) 1293 (10) 77 (2)

Others andBrazilians

3136 (11) 2116 (16) 1029 (21)

StudyAviano 802 (3) 288 (2) 128 (3)Baltimore 163 (1) 123 (1) 29 (1)Beijing 377 (1) 322 (2) 0 (0)Boston 473 (2) 339 (3) 75 (2)Buffalo 863 (3) 282 (2) 113 (2)Central Europe 730 (2) 238 (2) 295 (6)France

Multicen.(1989 1991)

255 (1) 163 (1) 247 (5)

FranceMulticen.(2001 2007)

2964 (10) 1343 (10) 366 (7)

Germany-Heidelberg

644 (2) 0 (0) 172 (3)

Germany-Saarland

83 (0) 53 (0) 22 (0)

HOTSPOT 59 (0) 63 (0) 0 (0)Houston 738 (2) 561 (4) 119 (2)International

Multicenter1450 (5) 1053 (8) 0 (0)

Iowa 600 (2) 344 (3) 58 (1)Italy

Multicenter2506 (8) 685 (5) 421 (9)

Japan (20012005)

2817 (9) 392 (3) 71 (1)

Latin America 1446 (5) 981 (7) 612 (12)Los Angeles 905 (3) 273 (2) 70 (1)Milan (1984

1989)1413 (5) 161 (1) 215 (4)

Milan (20062009)

669 (2) 118 (1) 162 (3)

MSKCC 115 (0) 61 (0) 20 (0)New York

Multicenter1246 (4) 818 (6) 202 (4)

North Carolina(1994 1997)

154 (1) 91 (1) 31 (1)

North Carolina(2002 2006)

982 (3) 594 (4) 283 (6)

Puerto Rico 410 (1) 182 (1) 0 (0)Rome 350 (1) 98 (1) 173 (3)Sao Paulo 1519 (5) 1042 (8) 406 (8)Seattle (1985

1995)465 (2) 317 (2) 0 (0)

Seattle-Leo 371 (1) 264 (2) 123 (2)Switzerland 824 (3) 311 (2) 104 (2)Tampa 789 (3) 115 (1) 48 (1)US Multicenter 936 (3) 710 (5) 0 (0)

3

a

a

a



UNCORRECTED PROOF


Table 1 (Continued)

Controls (%)



Western Europe 1726 (6) 932 (7) 378 (8)EducationNo education 1078 (4) 726 (5) 118 (2)

Junior highschool

10,456 (35) 4674 (35) 2371 (48)

Some highschool

5330 (18) 2751 (21) 922 (19)

High schoolgraduate

3883 (13) 1883 (14) 717 (14)

Technicalschool, somecollege

4825 (16) 1989 (15) 496 (10)

Collegegraduate

4272 (14) 1294 (10) 319 (6)

Drinking statusNever user 8068 (27) 2279 (17) 578 (12)Former user 3072 (10) 2521 (19) 833 (17)Current user 14,210 (48) 6943 (52) 2442 (49)Missing 4494 (15) 1574 (12) 1091 (22)Alcohol drinking intensity (number of drinks/day)0 <1 17,207 58) 4899 37) 1433 (29)1 <5 8913 30) 4099 31) 1757 (35)

5 2925 10) 3814 29) 1626 (33)Missing 799 3) 505 4) 127 (3)Alcohol duration (years)Never drinkers 8068 (27) 2279 (17) 578 (12)0 <20 2984 (10) 1182 (9) 297 (6)20 <40 9427 (32) 5422 (41) 1848 (37)40 <60 6013 (20) 2920 (22) 1435 (29)

60 412 (1) 176 (1) 123 (2)Missing 2940 (10) 1338 (10) 662 (13)

ABBREVIATIONS: MSKCC: Memorial Sloan Kettering Cancer Center.We excluded subjects with missing information on age, sex, and race (see Fig. 1).

Missing values for education were imputed according to the INHANCE protocol.

not oth er wise spec i fied, lar ynx, or un spec i fied HNC. Cases with can cersof the sali vary glands or of the nasal cav ity/ ear/ paranasal si nuses wereex cluded [11]. The orig i nal study sam ple in cluded 25,865 HNC casesand 37,248 con trols, giv ing a to tal of 63,113 sub jects.

We con ducted sub jects se lec tion ac cord ing to the fol low ing mainsteps (eAppendix, text and Fig. 1), ex clud ing: 1. cases with un spec i fied(95 sub jects) or over lap ping HNC (331 sub jects); 2. sub jects re port ingsmok ing to bacco prod ucts other than cig a rettes (i.e., cigar, pipe, andcig a r illo), to avoid risk dis tor tion due their use [5] (6255 sub jects); 3.sub jects with miss ing in for ma tion on du ra tion and/ or in ten sity of cig arette smok ing (1897 sub jects); 4. all sub jects from stud ies that in cludedonly never (i.e., Japan 1988 2000) (822 sub jects) or cur rent (i.e.,France 1987 1992) smok ers (457 sub jects), af ter the pre vi ous se lec tionsteps.

In or der to pre vent po ten tial es ti ma tion dis tor tion at the high estlev els of the ex po sure dis tri b u tions (due to small num bers of sub jectsor in for ma tion bias in heavy to bacco con sumers), fur ther ex cludedwere those sub jects re port ing the high est 5% of cig a rette smok ing inten sity (>40 cig a rettes per day) or du ra tion (>51 years) (14% HNCcases and 6% con trols ex cluded).

Af ter all the de scribed se lec tion steps (eAppendix, text and Fig. 1),the analy sis in cluded 33 stud ies [20 53] with 48,104 sub jects (18,260HNC cases and 29,844 con trols) (Table 1). When a study re ported tohave con ducted a case-con trol match ing, sep a rate sets of con trols werematched for oral cav ity and phar ynx (OCP) can cers com bined and forla ryn geal can cer cases. In de tail, the analy sis in cluded: 5423 can cers ofthe oral cav ity; 6261 pha ryn geal can cer cases (4648 oropha ryn geal and1613 hy popha ryn geal can cers cases); 1633 un spec i fied oral cav ity/phar ynx can cers (giv ing a to tal of 13,317 OCP can cer cases com bined),and 4943 la ryn geal can cers.

Statistical analysis

The bi vari ate re gres sion spline mod els [54] used to in ves ti gate thedose re sponse re la tion ship be tween HNC and the joint ex po sure tosmok ing in ten sity and du ra tion was de scribed ex ten sively in eAppendix Sta tis ti cal Analy sis and in Di Credico s PhD The sis [55]. We as sumed

a gen er al ized semi-para met ric lo gis tic re gres sion model where the twoex po sures were en tered as a joined piece wise poly no mial of a lin earde gree with con straints for con ti nu ity at each join point (called knot),to gether with po ten tial con founders (i.e., age, sex, race, study, ed u cation, drink ing sta tus, drink ing in ten sity, and drink ing du ra tion) [15].Knots rep re sented change points in the slope of the risk sur face. The setof spline re gres sion pa ra me ters de scribed the shape of the risk sur face.We fur ther as sumed that the knot lo ca tions for any of the two ex posures were un known pa ra me ters to be es ti mated, up to a max i mum of2 knots al lowed for each ex po sure. At the max i mum level of com plexity, the risk sur face was al lowed to have 2 change-points in the slopefor any ex po sure and 9 dif fer ent ar eas with a pos si bly chang ing slope.

The op ti mal knot lo ca tions and re gres sion pa ra me ters were jointlyes ti mated within the Bayesian ap proach. Prior dis tri b u tions ex pressed apri ori knowl edge (here vague) on plau si ble val ues of knot lo ca tions andre gres sion pa ra me ters [56,57]. Their pos te rior dis tri b u tion com binedprior in for ma tion and avail able data through the Bayes the o rem. Ajoint pos te rior dis tri b u tion was sim u lated for each can cer site, smoking-sta tus stra tum (cur rent/ for mer smok ers), and com bi na tion of number of knots us ing the Markov Chain Monte Carlo-type NUTS (No-U-Turn Sam pler) [58] al go rithm. For each com bi na tion of site and stratum, the best model was se lected among the con ver gent mod els withsen si ble knot lo ca tions ( 95 per centile of ei ther ex po sures) as the onethat min i mized the Wanatabe-Akaike In for ma tion Cri te rion (WAIC)[59,60]. The op ti mal knot lo ca tions, the ORs, and their 95% cred i blein ter vals (CIs) were de rived from the pos te rior dis tri b u tion cor re sponding to the best model. The ORs were pre sented through three-di mensional mesh plots that dis played the sur face of risk for any com bi na tionof cig a rette smok ing in ten sity and du ra tion, and through two-di mensional con tour plots that showed iso-risk curves iden ti fy ing the com bina tions of the two ex po sures with the same OR. Also ex plored weretrade-offs be tween in ten sity and du ra tion in two-di men sional con tourplots by com par i son of ORs for a fixed cu mu la tive ex po sure (i.e., pack-years).

To eval u ate po ten tial mod i fy ing ef fects of some co vari ates, analy seswere car ried out in strata of al co hol drink ing in ten sity (i.e., <1, 1 <5,

5 drinks per day) for cur rent smok ers, and years since quit ting (i.e.,<10, 10 years) for for mer smok ers. Sep a rate re sults for the risk of laryn geal can cer across al co hol drink ing in ten sity strata were pre sented,due to ab sence of het ero gene ity of the risk sur faces across strata forcur rent smok ers. As a com par i son, risk es ti mates were fur ther es timated ac cord ing to the Bayesian step-func tion re gres sion model.

When pos si ble, all the mod els were fit ted with the full set of po tential con founders shown in Table 1; more over, Never smok ers wereas sumed as the ref er ence cat e gory. Cal cu la tions were car ried out us ingthe open-source Stan pro gram [61] within the open-source R pro gram[62,63].

Results

Table 1 shows se lected char ac ter is tics of cases and con trols, ac cording to the vari ables in cluded in the mod els as po ten tial con founders.Ap prox i mately 70% of the sub jects were white. Stud ies from Eu ropecon tributed with ap prox i mately 44% of sub jects; 31% of sub jects werefrom the United States, whereas the re main ing ones were from Latin

4

a

a

11

th



UNCORRECTED PROOF


Fig. 1. Flow chart of sub jects se lec tion.

5



UNCORRECTED PROOF


Amer ica (14%) and Asia (11%). Six stud ies pro vided only cases of OCPcan cer.

Table 2 shows the dis tri b u tion of cig a rette smok ing habits for OCPcan cer, la ryn geal can cer, and con trols. Never smok ers were 21% ofOCP and 7% of la ryn geal can cers, ver sus 45% of con trols. Cur rentsmok ers were 66% of la ryn geal can cer cases, 57% of OCP can cer cases,and 26% of con trols. The preva lence of for mer smok ers was sim i lar incases and con trols, but the per cent age of sub jects who quit cig a rettesmok ing 10 years ago was 52% among HNC cases and 72% amongcon trols.

Fig. 2 shows the mesh and con tour plots for can cers of the OCP andlar ynx among cur rent cig a rette smok ers. Knot lo ca tions were in di catedby thicker black lines; the in ter sec tions of the lines de fined the ar easwhere the risk sur face had dif fer ent slopes. For both can cer sites, thebest model was char ac ter ized by one knot for du ra tion (at 33 years forOCP can cer and 30 years for la ryn geal can cer) and one knot for in tensity (at 16 and 25 cig a rettes/ day, re spec tively). Smok ing du ra tion modi fied OCP can cer risk at any lev els of in ten sity (Fig. 2A). In con trast, forany du ra tion up to 10 years, the ORs were al ways <2, re gard less of thein ten sity of cig a rette smok ing. In ad di tion, at a fixed value of 20 pack-years, an in ten sity of 40 cig a rettes/ day and a du ra tion of 10 years ledto an OR of 2, whereas the OR was equal to 4 with a du ra tion of40 years in smok ers of 10 cig a rettes/ day.

For la ryn geal can cer, ORs >20 were found for in ten si ties of >20cig a rettes/ day and du ra tions of >28 years (Fig. 2B). More over, ORs>10 of la ryn geal can cer were reached by cur rent smok ers of >20 ciga rettes/ day only when du ra tion was >20 years; how ever, ORs >10were not reached for any du ra tion <15 years and any level of in tensity. Fi nally, the OR was 6.2 for smok ers of 40 cig a rettes/ day for10 years, but it was higher (be tween 9 and 10) for 20 cig a rettes/ daysmoked for 20 years, or for 10 cig a rettes/ day smoked for 40 years.

The shape of the risk sur face and the val ues of the ORs were alsovery sim i lar across ma jor OCP can cer sub sites (i.e., oral cav ity andoropha ryn geal can cers) (eFig ure 2).

Fig. 3 shows the joint ef fect of cur rent smok ing in ten sity and du ration in strata of al co hol con sump tion. Among never drinkers (<1drink/ day), the shape of the risk sur face was sim i lar to the one pre

Table 2Dis tri b u tion of cases of oral and pha ryn geal can cer, la ryn geal can cer, and con trols accord ing to se lected as pects of cig a rette smok ing habits. IN HANCE con sor tium.

Controls (%)



Cigarette smoking statusNever

user13,347 (45) 2791 (21) 330 (7)

Formeruser

8792 (29) 2909 (22) 1353 (27)

Currentuser

7705 (26) 7617 (57) 3260 (66)

Cigarette smoking intensity (number of cigarettes/day)Never

user13,347 (45) 2791 (21) 330 (7)

1 15 7199 (24) 2814 (21) 1054 (21)>15 25 6166 (21) 4483 (34) 2083 (42)>26 40 3132 (10) 3229 (24) 1476 (30)Cigarette smoking duration (years)Never

user13,347 (45) 2791 (21) 330 (7)

1 25 7214 (24) 2019 (15) 604 (12)>25 35 4360 (15) 3370 (25) 1330 (27)>35 51 4923 (16) 5137 (39) 2679 (54)Time since quitting cigarette smoking (years) (for formers only)

1 <10 2300 (26) 1310 (45) 626 (46)10 3655 (72) 1522 (52) 711 (53)

Missing 137 (2) 77 (3) 16 (1)

Proportions were calculated among former smokers only.

sented for all al co hol in ten si ties to gether (Fig. 2A), but the ORs weregen er ally lower; all the ORs were <2 for du ra tions 15 years and anyin ten sity, whereas an OR >5 was ob served only af ter 25 years of dura tion or more (Fig. 3A). How ever, the shape of the sur face and/ or theORs of the joint ef fect of du ra tion and in ten sity were dif fer ent whenlight (Fig. 3B) and heavy (Fig. 3C) drinkers were con sid ered.

As a com par i son, the ORs and their cor re spond ing 95% CIs were esti mated for each can cer site in cur rent smok ers within the Bayesian logis tic re gres sion model that as sumed the pres ence of step func tions(Table 3, main ORs). These es ti mates were com pared with the range ofOR es ti mates de rived from the spline mod els for each joint cat e gory ofdu ra tion and in ten sity (Table 3, brack eted ORs). For both can cer sites,all Min-Max ranges in cluded the OR es ti mates ob tained from theBayesian lo gis tic re gres sion, and this repli ca tion re as sured that thespline model was valid. In ad di tion, within the ex am ined cat e gories,the step-func tion in ter vals widely over lapped with the Min-Max rangesof the ORs from the spline mod els, but failed to cap ture the ORs variabil ity. For in stance, for OCP can cer in cur rent smok ers of 26 40 cig arettes/ day for 36 51 years, the cat e gor i cal OR was 8.4 (95% CI: 8.08.9), whereas the OR var ied from 7.1 to 10.6 un der the spline modelap proach for the same com bined cat e gory (Table 3). The same pat ternemerged at any com bi na tion of du ra tion and in ten sity for both can cersites for for mer smok ers who quit 10 years ago (eTable 2).

Among for mer smok ers who quit 10 years ago, no ORs >4 wereob served for OCP can cer (Fig. 4A), and the ORs were ap prox i matelyhalved, as com pared to the same lev els of du ra tion and in ten sity in current smok ers (Fig. 2A). A mean risk re duc tion of 2/ 3 was also found forla ryn geal can cer in long-term for mer smok ers, with ORs de clines thatvar ied from 1/ 3 to 4/ 5 de pend ing on the dif fer ent com bi na tions of inten sity and du ra tion (Fig. 4B). These es ti mates were de rived from models that in cluded one knot for in ten sity (27 cig a rettes/ day) for la ryngeal can cer and no knots for OCP can cer. Re sults on the com bined setof for mer smok ers (re gard less of time since quit ting) were sim i larlypre sented in eFig ure 3.

Discussion

Our large pooled analy sis showed that cig a rette smok ing du ra tionand in ten sity did not in crease HNC risk to the same ex tent, but the effect was greater for longer du ra tions. At the high est com bined lev els ofcig a rette in ten sity and du ra tion, sub jects may reach ORs >10 of OCPcan cer and ORs >40 of la ryn geal can cer, as com pared to never smokers. Bi vari ate re gres sion spline mod els have been proven suc cess ful inex plor ing the sep a rate and joint ef fects of in ten sity and du ra tion of ciga rette smok ing. There fore, when pos si ble, mod els that al low this dif feren tial im pact of in ten sity and du ra tion on the risk should be con sid ered[8,12,16].

Re sults from the pre sent study were con sis tent with pre vi ous findings show ing no thresh old ef fect on HNC risk of shorter du ra tions orlower in ten si ties of to bacco smok ing [13,64]. In deed, the lit er a turesup ported the pres ence of a dose-re sponse re la tion ship be tween cig arette smok ing and HNC risk over the en tire range of con sump tion[5,13]. Our con tri bu tion ad di tion ally sug gested that the dose-re sponsere la tion ship with OCP and la ryn geal can cer risk was still far from be inglin ear, with a steeper in crease for in ter me di ate con sump tions and apos si ble plateaux in di cat ing a sat u ra tion ef fect in smok ers with>20 years of du ra tion and >30 cig a rettes/ day.

Our re sults also strengthen the ev i dence [18,65] on the im pact ofsmok ing ces sa tion on HNC can cer risk, with the OR in for mer smok erswho quit 10 years ago be ing more than halved for OCP can cer and

1/ 3 lower for la ryn geal can cer, in com par i son with cur rent smok ers.These re sults con vey to the gen eral pop u la tion and to pub lic healthpro fes sion als the valu able mes sage that it is never too late to give upcig a rette smok ing, as quit ting pro vides far smaller risks of de vel op ing

6

a

a



UNCORRECTED PROOF


Fig. 2. Odds ra tios of oral and pha ryn geal can cer and la ryn geal can cer in cur rent smok ers, for the joint ef fect of in ten sity (cig a rettes/ day) and du ra tion (years) of cig arette smok ing es ti mated through bi vari ate spline mod els. IN HANCE con sor tium. Fit ted mod els in cluded ad just ment for age, sex, race, study, ed u ca tion, drink ing sta tus, drink ingin ten sity, and drink ing du ra tion. The ref er ence cat e gory was de fined as Never smok ers . On the grid, black thicker lines rep re sent knot lo ca tions: 16 cig a rettes/ day and 33 years ofdu ra tion for oral and pha ryn geal can cer and 25 cig a rettes/ day and 30 years of du ra tion for la ryn geal can cer, re spec tively. Dark grey lines in con tour plots in di cate iso-risk curves at defined lev els of risk.

HNC can cer for OCP and, most of all, for la ryn geal can cers. This is inac cor dance with an other re cently pub lished meta-analy sis of ob ser vational stud ies on la ryn geal can cer based on 14,292 cases from 3 co hortand 15 case-con trol stud ies [65].

When the com bined ef fect of smok ing in ten sity and du ra tion wasmod elled as a cross-prod uct term, an in ter ac tion term be tween al co holcon sump tion and cig a rette smok ing was tra di tion ally added to themod els [3,4]. In our study this ef fect was ex plored by strat i fied re gression spline mod els es ti mated within cat e gories of al co hol con sump tion;we com pared the shape of the risk sur face and the ORs val ues from theiso-risk curves across the three strat i fied analy ses. How ever, our re sultsalso showed that al co hol acts as a sub stan tial mod i fier of the as so ci ation be tween OCP can cer risk and the joint ef fect of cig a rette in ten sityand du ra tion [4,17].

Ma jor strengths of our study in cluded a large sam ple size, in ter national rep re sen ta tive ness, and valid in for ma tion on po ten tial con founding fac tors. In ad di tion, we ap plied a novel Bayesian ap proach to

jointly es ti mate the op ti mal knot lo ca tions and the ORs of HNC for thejoint ef fect of in ten sity and du ra tion in a bi vari ate con text. Af ter ex amin ing var i ous fre quen tist so lu tions [66,67], we opted for the Bayesianap proach that al lowed to put the knots where the data sug gested, oncewe had taken the en tire set of con found ing vari ables into con sid er ation. Within the Bayesian frame work, we were also able to choose theop ti mal num ber of knots (up to 2) and knot lo ca tions by model compar i son based on in for ma tion cri te ria, in stead of just com par ing mod elswith fixed num ber and lo ca tion of the knots. Fi nally, we could in cor porate in the model se lec tion process ev i dence from the lit er a ture on themax i mum num ber of changes in the risk pat tern, which is un likely tobe higher than 2. In deed, cig a rette smok ing is sup posed to have a protec tive or null ef fect on the risk at lower lev els, a sat u ra tion ef fect atthe high est in ten sity lev els [68], as well as the ex pected in crease pre viously de scribed at the in ter me di ate lev els of con sump tion. The proposed Bayesian ap proach is ap plic a ble to other epi demi o logic sce nar ios

7

a,b

a

b



UNCORRECTED PROOF


Fig. 3. Odds ra tios of oral and pha ryn geal can cer in cur rent smok ers by al co hol drink ing in ten sity, for the joint ef fect of in ten sity (cig a rettes/ day) and du ra tion (years) ofcig a rette smok ing, es ti mated through bi vari ate spline mod els. IN HANCE con sor tium. Fit ted mod els in cluded ad just ment for age, sex, race, study, ed u ca tion, and al co hol drink ingsta tus. The ref er ence cat e gory was de fined as Never smok ers , in each strata of al co hol drink ing in ten sity. On the grid, black thicker lines rep re sent knot lo ca tions: 32 years of du ration for never drinkers, and 12 cig a rettes/ day and 25 years of du ra tion for heavy drinkers, re spec tively. Dark grey lines in con tour plots in di cate iso-risk curves at de fined lev els of risk.

8

a,b

a

b



UNCORRECTED PROOF


Table 3Odds ra tios (ORs) and 95% cred i ble in ter vals (CIs) of oral and pha ryn geal can cer and la ryn geal can cer in cur rent smok ers, for the joint ef fect of in ten sity (cig a rettes/ day) and du ra tion(years) of cig a rette smok ing es ti mated through step-func tion as com pared with re sults from bi vari ate spline mod els. IN HANCE con sor tium.

Intensity (cigarettes/day)

1 15 16 25 26 40

Cancer type Ca:CoOR (95%CI)[Min-Max] Ca:Co [Min-Max] Ca:Co

OR (95%CI)[Min-Max]

Oral cavity and pharynxDuration (years)1 25 322:1046 1.5 (1.4 1.7) 371:663 2.7 (2.4 2.9) 191:270 3.3 (2.8 3.9)

[0.9 3.1] [0.9 4.2] [0.8 5.9]26 35 583:911 2.9 (2.7 3.1) 1138:954 5.1 (4.9 5.4) 849:466 7.2 (6.7 7.8)

[1.4 4.8] [3.5 6.9] [4.6 11.0]36 51 912:1175 3.7 (3.5 3.9) 1838:1409 6.0 (5.8 6.2) 1413:694 8.4 (8.0 8.9)

[1.6 5.8] [5.2 7.1] [7.1 10.6]LarynxDuration (years)1 25 75:908 3.8 (3.3 4.4) 112:576 9.4 (8.3 10.5) 70:247 14.8 (12.2 17.8)

[1.1 8.8] [2.3 20.1] [2.8 23.1]26 35 183:812 7.3 (6.9 7.9) 423:845 17.2 (16.8 17.6) 319:415 28.5 (27 30.2)

[2.9 12.6] [10.1 29.6] [22.4 37.7]36 51 394:1064 8.9 (8.8 9.1) 966:1274 19.5 (19.3 19.8) 718:616 33.9 (33.5 34.3)

[3.9 15.3] [13.9 31.0] [30.9 43.2]

Fitted models included adjustment for age, sex, race, study, education, drinking status, drinking intensity, and drinking duration. The reference category was defined as Neversmokers . The reference category included 2791 cases and 13,139 controls for the analysis on oral and pharyngeal cancer and 330 cases and 11,433 controls for that on laryngealcancer.

Min and Max represent the lowest and the highest OR values estimated for any combinations of intensity and duration by bivariate spline models.

where con tin u ous ex po sures in their po ten tial in ter ac tion af fect dis easerisk.

Among study lim i ta tions, the ret ro spec tive study de sign and theself-re ported smok ing his tory were the most rel e vant ones. Even if alarge amount of lit er a ture has sug gested ac cept able cor re la tions between self-re ported smok ing in ten sity and co ti nine lev els in blood orurine [69], in ac cu rate self-re port ing may oc cur. Dis crep an cies be tweenself-re ported and ob jec tive in for ma tion were more likely among long-term heavy smok ers [70]; higher val ues of in ten sity and du ra tion werethere fore more prone to in ac cu rate re port ing. Fur ther more, smok ingin ten sity may have var ied over time and by age of ex po sure. How ever,its es ti mates were of ten based on the self-re ported av er age num ber ofcig a rettes per day; these two as pects may have led to ap pre cia ble errors in mea sur ing the true mean in ten sity of ex po sure over one s lifetime. These is sues may well be se ri ous here, given the con tin u ous nature of the ex po sures con sid ered in spline mod els. To re duce in for mation bias and resid ual con found ing at the ex treme val ues of the ex posure dis tri b u tions, we ex cluded sub jects re port ing higher (>95th percentiles) cig a rette in ten sity and/ or du ra tion from the pre sent analy sis[13,15,71]. In ad di tion, to avoid bias due to the use of other to baccoprod ucts, we ex cluded sub jects re port ing use of to bacco prod ucts otherthan cig a rettes. We were also obliged to com bine dif fer ent sub sites ofOCP can cer, al though dif fer ences in ae ti ol ogy (i.e., the causal role ofhu man pa pil lo mavirus in oropha ryn geal can cer) have been demonstrated [34]. How ever, re sults were sim i lar for ma jor sub sites, al thoughbased on less sta ble mod els. Fi nally, our Bayesian ap proach was compu ta tion ally time con sum ing, ask ing for sev eral hours of server comput ing for each model fit ted.

Re sults from the pre sent study con firmed pre vi ous ev i dence on themain con tri bu tion of cig a rette smok ing du ra tion in in flu enc ing HNCrisk and on the im por tant re duc tion in HNC risk ob served in for mersmok ers who quit 10 or 15 years ago. Both re sults strongly sup port aneven wider dif fu sion of pub lic in ter ven tions to en cour age smok ers toquit and help them suc ceed.

Acknowledgements

The au thors would like to thank Xavier Castell sagué who col lecteddata in the IARC In ter na tional Mul ti cen ter study and passed away in2016. We thank Mrs Luig ina Mei for ed i to r ial as sis tance.

Funding

This work was sup ported by grants from the: Na tional In sti tutes ofHealth (NIH) [no grant num ber pro vided for the IN HANCE Pooled DataPro ject, grant num bers P01 CA068384, K07 CA104231 for the New YorkMul ti cen ter study, grant num bers R01 CA048996, R01DE012609 forthe Seat tle (1985-1995) study, grant num ber TW001500 for the Fog arty In ter na tional Re search Col lab o ra tion Award (FIRCA) sup port ing theIowa study, grant num ber R01 CA061188 for the North Car olina (1994-1997) study, grant num bers P01 CA068384, K07 CA104231,R01DE013158 for the Tampa study, grant num bers P50 CA090388,R01 DA011386, R03 CA077954, T32 CA009142, U01 CA096134,R21ES011667 for the Los An ge les study, grant num bers R01ES011740,R01 CA100264 for the Hous ton study, grant num bers R01 CA078609,R01 CA100679 for the Boston study, grant num ber R01 CA051845 forthe MSKCC study, grant num ber R01 CA030022 for the Seat tle-Leostudy, grant num ber DE016631 for the Bal ti more study]; Na tional Cancer In sti tute (NCI) at the Na tional In sti tutes of Health (NIH) [grantnum ber R03 CA113157 for the IN HANCE Pooled Data Pro ject, no grantnum ber pro vided for the In tra mural Pro grams sup port ing the PuertoRico and the US Mul ti cen ter stud ies, grant num ber R01 CA90731-01 forthe North Car olina (2002-2006) study]; Na tional In sti tute of Den taland Cran io fa cial Re search (NIDCR) at the Na tional In sti tutes of Health(NIH) [grant num ber R03DE016611 for the IN HANCE Pooled Data Project, grant num bers R01DE011979, R01DE013110 for the Iowa study,no grant num ber pro vided for the In tra mural Pro gram sup port ing thePuerto Rico study]; Ital ian As so ci a tion for Re search on Can cer (AIRC)[no grant num ber pro vided for the Mi lan (1984-1989) study, for theAviano study, for the Italy Mul ti cen ter study and for the Rome study,grant num ber 10068 for the Mi lan study (2006-2009)]; Ital ian Leagueagainst Can cer [no grant num ber pro vided for the Aviano and Italy

9

a

b b b

a

b



UNCORRECTED PROOF


Fig. 4. Odds ra tio of oral and pha ryn geal can cer and la ryn geal can cer in for mer smok ers who quit 10 years ago, for the joint ef fect of in ten sity (cig a rettes/ day) and dura tion (years) of cig a rette smok ing es ti mated through bi vari ate spline mod els. IN HANCE con sor tium. Fit ted mod els in cluded ad just ment for age, sex, race, study, ed u ca tion,drink ing sta tus, drink ing in ten sity, and drink ing du ra tion. The ref er ence cat e gory was de fined as Never smok ers . On the grid, the black thicker line rep re sents the knot lo ca tion: 27cig a rettes/ day for la ryn geal can cer. Dark grey lines in con tour plots in di cate iso-risk curves at de fined lev els of risk.

Mul ti cen ter stud ies]; Ital ian Min istry of Re search [no grant num berpro vided for the Aviano and Italy Mul ti cen ter stud ies]; the Swiss Research against can cer/ On co su isse [grant num bers KFS-700, OCS-1633for the Swiss study]; World Can cer Re search Fund [no grant num berpro vided for the Cen tral Eu rope study]; Eu ro pean Com mis sion [grantnum ber IC18-CT97-0222 (INCO-DC Pro gram) for the Latin Amer icastudy, grant num ber IC15-CT98-0332 (INCO-COPER NI CUS Pro gram)for the Cen tral Eu rope study]; Vet er ans Af fairs Merit Re view Funds [nogrant num ber pro vided for the Iowa study]; Na tional In sti tute of En viron men tal Health Sci ences (NIEHS) [grant num ber P30ES010126 forthe North Car olina (1994-1997) study, grant num ber P30ES010126 forthe North Car olina (2002-2006) study]; Alper Re search Pro gram forEn vi ron men tal Ge nomics of the UCLA Jon s son Com pre hen sive Can cerCen ter [no grant num ber pro vided for the Los An ge les study]; Fondopara la In ves ti ga cion Ci en tifica y Tec no log ica Ar gentina (FON CYT) [nogrant num ber pro vided for the Latin Amer ica study]; In sti tut Hos pi taldel Mar d In ves ti ga cions Mediquès (IMIM) [no grant num ber pro videdfor the Latin Amer ica study]; Fundação de Am paro à Pesquisa no Estado de São Paulo (FAPESP) [grant num ber 01/ 01768-2 for the LatinAmer ica study, grant num bers GEN CAPO 04/ 12054-9, 10/ 51168-0 for

the Sao Paulo study]; Fondo de In ves ti ga ciones San i tarias (FIS) of theSpan ish Gov ern ment [grant num ber FIS 97/ 0024, FIS 97/ 0662, BAE01/ 5013 for the In ter na tional Mul ti cen ter study]; In ter na tional UnionAgainst Can cer (UICC) [no grant num ber pro vided for the In ter na tionalMul ti cen ter study]; Ya m agiwa-Yoshida Memo r ial In ter na tional Can cerStudy Grant [no grant num ber pro vided for the In ter na tional Mul ti center study]; Eu ro pean Com mu nity (5th Frame work Pro gramme) [grantnum ber QLK1-CT-2001-00182 for the West ern Eu rope study]; Min istryof Sci ence, Re search and Arts Baden-Wurt tem berg [no grant num berpro vided for the Ger many-Saar land study]; Ger man Min istry of Ed u cation and Re search [grant num ber 01 GB9702/ 3 for the Ger many-Heidel berg study]; Sci en tific Re search grant from the Min istry of Ed u cation, Sci ence, Sports, Cul ture and Tech nol ogy of Japan [grant num ber17015052 for the Japan (2001-2005) study]; Third-Term Com pre hensive 10-Year Strat egy for Can cer Con trol from the Min istry of Health,La bor and Wel fare of Japan [grant num ber H20-002 for the Japan(2001-2005) study]; Johns Hop kins Richard Gelb Can cer Pre ven tionAward [no grant num ber pro vided for the HOTSPOT study]; Ital ianFoun da tion for Can cer Re search (FIRC) [no grant num ber pro vided forthe Mi lan study (2006-2009)]; Ital ian Min istry of Ed u ca tion - PRIN

10

a,b

a

b



UNCORRECTED PROOF


2009 Pro gram [grant num ber X8Y CBN for the Mi lan study (2006-2009)]; VE was sup ported by Uni ver sità degli Studi di Mi lano YoungIn ves ti ga tor Grant Pro gram 2017 .

Appendix A. Supplementary material

Sup ple men tary data to this ar ti cle can be found on line at https: //doi. org/ 10. 1016/ j. oraloncology. 2019. 05. 006.

References

[1] World Health Or ga ni za tion. WHO global re port: mor tal ity at trib ut able to to bacco.Geneva: WHO;2012.

[2] Fer lay J, Ervik M, Lam F, Colom bet M, Mery L, Piñeros M, Znaor A, So er jomataram I, Bray F. Global Can cer Ob ser va tory: Can cer To day. Lyon, France: In terna tional Agency for Re search on Can cer; 2018. Avail able from: https: // gco. iarc. fr/today, ac cessed [21 March 2019].

[3] IARC, IARC mono graphs on the eval u a tion of car cino genic risks to hu mans, Volume 38. To bacco smok ing, IARC Press, Lyon, 1986.

[4] IARC, IARC Mono graphs on the eval u a tion of car cino genic risks to Hu mans, Volume 83: To bacco smoke and in vol un tary smok ing, IARC Press, Lyon, 2004.

[5] J.H. Lu bin, M. Pur due, K. Kelsey, et al., To tal ex po sure and ex po sure rate ef fectsfor al co hol and smok ing and risk of head and risk of neck can cer: a pooled analy sisof case-con trol stud ies, Am J Epi demiol 170 (2009) 937 947.

[6] J.H. Lu bin, M.M. Gaudet, A.F. Ol shan, et al., Body mass in dex, cig a rette smok ing,and al co hol con sump tion and can cers of the oral cav ity, phar ynx, and lar ynx: model ing odds ra tios in pooled case-con trol data, Am J Epi demiol 171 (2010) 12501261.

[7] R. Doll, An epi demi o log i cal per spec tive of the bi ol ogy of can cer, Can cer Res38 (1978) 3573 3583.

[8] J. Vlaan deren, L. Porten gen, J. Schüz, et al., Ef fect mod i fi ca tion of the as so ci a tionof cu mu la tive ex po sure and can cer risk by in ten sity of ex po sure and time since expo sure ces sa tion: a flex i ble method ap plied to cig a rette smok ing and lung can cer inthe SYN ERGY Study, Am J Epi demiol 179 (2014) 290 298.

[9] J.H. Lu bin, M.C. Ala vanja, N. Ca po raso, et al., Cig a rette smok ing and can cer risk:mod el ing to tal ex po sure and in ten sity, Am J Epi demiol 166 (2007) 479 489.

[10] D. Baris, M.R. Kara gas, C. Ver rill, et al., A case-con trol study of smok ing and bladder can cer risk: emer gent pat terns over time, J Natl Can cer Inst 101 (2009) 15531561.

[11] M. Hashibe, P. Bren nan, S. Ben hamou, et al., Al co hol drink ing in never users of tobacco, cig a rette smok ing in never drinkers, and the risk of head and neck can cer:pooled analy sis in the In ter na tional Head and Neck Can cer Epi demi ol ogy Con sortium, J Natl Can cer Inst 99 (2007) 777 789.

[12] J.H. Lu bin, N.E. Ca po raso, Mis un der stand ings in the mis con cep tion on the use ofpack-years in analy sis of smok ing, Br J Can cer 108 (2013) 1218 1220.

[13] J. Pole sel, R. Ta lamini, C. La Vec chia, et al., To bacco smok ing and the risk of upper aero-di ges tive tract can cers: a re analy sis of case-con trol stud ies us ing splinemod els, Int J Can cer 122 (2008) 2398 2402.

[14] L. Desquil bet, F. Mar i otti, Dose re sponse analy ses us ing re stricted cu bic splinefunc tions in pub lic health re search, Stat Med 29 (2010) 1037 1057.

[15] L. Dal Maso, N. Torelli, E. Bian cotto, et al., Com bined ef fect of to bacco smok ingand al co hol drink ing in the risk of head and neck can cers: a re-analy sis of case-con trol stud ies us ing bi-di men sional spline mod els, Eur J Epi demiol 31 (2016)385 393.

[16] J. Peto, That ef fects of smok ing should be mea sured in pack-years: mis con cep tions4, Br J Can cer 107 (2012) 406 407.

[17] M. Hashibe, P. Bren nan, S.-C. Chuang, et al., In ter ac tion be tween to bacco and al cohol use and the risk of head and neck can cer: pooled analy sis in the In ter na tionalHead and Neck Can cer Epi demi ol ogy Con sor tium, Can cer Epi demiol Bio mark ersPrev 18 (2009) 541 550.

[18] M. Mar ron, P. Bof fetta, Z.F. Zhang, et al., Ces sa tion of al co hol drink ing, to baccosmok ing and neck can cer risk, Int J Epi demiol 39 (2010) 182 196.

[19] A. Wyss, M. Hashibe, S.C. Chuang, et al., Cig a rette, cigar, and pipe smok ing andthe risk of head and neck can cers: pooled, Am J Epi demiol 178 (2013) 679 690.

[20] S. Franceschi, R. Ta lamini, S. Barra, et al., Smok ing and drink ing in re la tion to cancers of the oral cav ity, phar ynx, lar ynx, and esoph a gus in north ern Italy, Can cerRes 50 (1990) 6502 6507.

[21] E. Ne gri, C. La Vec chia, S. Franceschi, A. Ta vani, At trib ut able risk for oral can cerin north ern Italy, Can cer Epi demiol Bio mark ers Prev 2 (1993) 189 193.

[22] C. Bosetti, S. Gal lus, A. Tri chopoulou, et al., In flu ence of the Mediter ranean diet onthe risk of can cers of the up per aerodi ges tive tract, Can cer Epi demiol Bio mark ersPrev 12 (2003) 1091 1094.

[23] F. Levi, C. Pasche, C. La Vec chia, F. Luc chini, S. Franceschi, P. Mon nier, Foodgroups and risk of oral and pha ryn geal can cer, Int J Can cer 77 (1998) 705 709.

[24] M. Hashibe, P. Bof fetta, D. Zaridze, Ev i dence for an im por tant role of al co hol- andalde hyde-me tab o liz ing genes in can cers of the up per aerodi ges tive tract, Can cerEpi demiol Bio mark ers Prev 15 (2006) 696 703.

[25] J.E. Mus cat, J.P. Richie Jr., S. Thomp son, E.L. Wyn der, Gen der dif fer ences insmok ing and risk for oral can cer, Can cer Res 56 (1996) 5192 5197.

[26] K.A. Rosen blatt, J.R. Dal ing, C. Chen, K.J. Sher man, S.M. Schwartz, Mar i juana useand risk of oral squa mous cell car ci noma, Can cer Res 64 (2004) 4049 4054.

[27] E.M. Smith, H.T. Hoff man, K.S. Sum mers gill, H.L. Kirch ner, L.P. Turek, T.H. Haugen, Hu man pa pil lo mavirus and risk of oral can cer, Laryn go scope 108 (1998)1098 1103.

[28] A.F. Ol shan, M.C. Weissler, M.A. Wat son, D.A. Bell, GSTM1, GSTT1, GSTP1,CYP1A1, and NAT1 poly mor phisms, to bacco use, and the risk of head and neckcan cer, Can cer Epi demiol Bio mark ers Prev 9 (2000) 185 191.

[29] A. Elahi, Z. Zheng, J. Park, K. Eyring, T. Mc Caf frey, P. Lazarus, The hu man OGG1DNA re pair en zyme and its as so ci a tion with oro la ryn geal can cer risk, Car cino gen esis 23 (2002) 1229 1234.

[30] Y. Cui, H. Mor gen stern, S. Green land, et al., Poly mor phism of xe ro derma pig mento sum group G and the risk of lung can cer and squa mous cell car ci no mas of theorophar ynx, lar ynx and esoph a gus, Int J Can cer 118 (2006) 714 720.

[31] Z. Zhang, Q. Shi, Z. Liu, E.M. Stur gis, M.R. Spitz, Q. Wei, Poly mor phisms of methio n ine syn thase and me thio n ine syn thase re duc tase and risk of squa mous cellcar ci noma of the head and neck: a case-con trol analy sis, Can cer Epi demiol Biomark ers Prev 14 (2005) 1188 1193.

[32] R.B. Hayes, E. Bravo-Otero, D.V. Klein man, et al., To bacco and al co hol use andoral can cer in Puerto Rico, Can cer Causes Con trol 10 (1999) 27 33.

[33] K. Szymańska, R.J. Hung, V. Wünsch-Filho, et al., Al co hol and to bacco, and therisk of can cers of the up per aerodi ges tive tract in Latin Amer ica: a case-con trolstudy, Can cer Causes Con trol 22 (2011) 1037 1046.

[34] R. Her rero, X. Castell sagué, M. Pawlita, et al., IARC Mul ti cen ter Oral Can cer StudyGroup. Hu man pa pil lo mavirus and the risk of Hu man pa pil lo mavirus and oral cancer: the In ter na tional Agency for Re search on Can cer mul ti cen ter study, J NatlCan cer Inst. 95 (2003) 1772 1783.

[35] E.S. Pe ters, M.D. Mc Clean, M. Liu, E.A. Eisen, N. Mueller, K.T. Kelsey, The AD H1Cpoly mor phism mod i fies the risk of squa mous cell car ci noma of the head and neckas so ci ated with al co hol and to bacco use, Can cer Epi demiol Bio mark ers Prev14 (2005) 476 482.

[36] P. Gallì, G. Cadoni, M. Volante, et al., A case-con trol study on the com bined ef fectsof p53 and p73 poly mor phisms on head and neck can cer risk in an Ital ian pop u lation, BMC Can cer 9 (2009) 137.

[37] W.J. Blot, J.K. McLaugh lin, D.M. Winn, et al., Smok ing and drink ing in re la tion tooral and pha ryn geal can cer, Can cer Res 48 (1988) 3282.

[38] A.F. Bo ing, J.L. Fer reira An tunes, M.B. de Car valho, et al., How much do smok ingand al co hol con sump tion ex plain so cioe co nomic in equal i ties in head and neck cancer risk?, J Epi demiol Com mu nity Health 65 (2010) 709 714.

[39] S.P. Schantz, Z.F. Zhang, M.S. Spitz, M. Sun, T.C. Hsu, Ge netic sus cep ti bil ity tohead and neck can cer: in ter ac tion be tween nu tri tion and mu ta gen sen si tiv ity,Laryn go scope 107 (1997) 765 781.

[40] M.A. Rogers, D.B. Thomas, S. Davis, T.L. Vaughan, A.E. Nevissi, A case-con trolstudy of el e ment lev els and can cer of the up per aerodi ges tive tract, Can cer Epidemiol Bio mark ers Prev 2 (1993) 305 312.

[41] P. La giou, C. Georgila, P. Mi naki, et al., Al co hol-re lated can cers and ge netic suscep ti bil ity in Eu rope: the ARCAGE pro ject: study sam ples and data col lec tion, EurJ Can cer Prev 18 (2009) 76 84.

[42] K.S. An der son, J.E. Ger ber, G. D Souza, et al., Bi o logic pre dic tors of sero logic responses to HPV in oropha ryn geal can cer: the HOTSPOT study, Oral On col51 (2015) 751 758.

[43] D. Twardella, M. Loew, D. Rothen bacher, C. Stegmaier, H. Ziegler, H. Bren ner, Thedi ag no sis of a smok ing-re lated dis ease is a promi nent trig ger for smok ing ces sa tionin a ret ro spec tive co hort study, J Clin Epi demiol 59 (2006) 82 89.

[44] A. Di etz, H. Ram roth, T. Ur ban, W. Ahrens, H. Becher, Ex po sure to ce ment dust,re lated oc cu pa tional groups and la ryn geal can cer risk: re sults of a pop u la tionbased case-con trol study, Int J Can cer 108 (2004) 907 911.

[45] T. Suzuki, K. Wakai, K. Mat suo, et al., Ef fect of di etary an tiox i dants and risk oforal, pha ryn geal and la ryn geal squa mous cell car ci noma ac cord ing to smok ing anddrink ing habits, Can cer Sci 97 (2006) 760 767.

[46] K. Di varis, A.F. Ol shan, J. Smith, et al., Oral health and risk for head and necksqua mous cell car ci noma: the Car olina Head and Neck Can cer Study, Can cerCauses Con trol 21 (2010) 567 575.

[47] G. Men vielle, D. Luce, P. Gold berg, A. Leclerc, Smok ing, al co hol drink ing, oc cu pational ex po sures and so cial in equal i ties in hy popha ryn geal and la ryn geal can cer,Int J Epi demiol 33 (2004) 799 806.

[48] V. Jayaprakash, N.R. Rigual, K.B. Moysich, et al., Chemo pre ven tion of head andneck can cer with as pirin, Arch Oto laryn gol Head Neck Surg 132 (2006) 12311236.

[49] D. Luce, I. Stücker, ICARE Study Group. In ves ti ga tion of oc cu pa tional and en vi ronmen tal causes of res pi ra tory can cers (ICARE): a mul ti cen ter, pop u la tion-basedcase-con trol study in France, BMC Pub lic Health 11 (2011) 928.

[50] G. D' souza, A.R. Kreimer, R. Vis cidi, et al., Case-con trol study of hu man pa pil lomavirus and oropha ryn geal can cer, N Engl J Med 10 (356) (2007) 1944 1956.

[51] T.Z. Zheng, P. Boyle, H.F. Hu, et al., Den ti tion, oral hy giene, and risk of oral cancer: a case-con trol study in Bei jing, Peo ple's Re pub lic of China, Can cer Causes Control 1 (1990) 235 241.

[52] F. Bravi, C. Bosetti, M. Filomeno, et al., Foods, nu tri ents and the risk of oral andpha ryn geal can cer, Br J Can cer 109 (2013) 2904 2910.

[53] J.H. Lu bin, M. Ko gev inas, D. Sil ver man, et al., Ev i dence for an in ten sity-de pen dentin ter ac tion of NAT2 acety la tion geno type and cig a rette smok ing in the Span ishBlad der Can cer Study, Int J Epi demiol 36 (2007) 236 241.

[54] D. Rup pert, M. Wand, R. Car roll, Semi para met ric re gres sion (Cam bridge Se ries inSta tis ti cal and Prob a bilis tic Math e mat ics), Cam bridge Uni ver sity Press, Cam bridge,2003.

[55] G. Di Credico, Some de vel op ments in semi para met ric and cross-clas si fied mul tilevel mod els, Ph.D. The sis Padua Uni ver sity, Italy, 2018.

[56] A. Gel man, H.S. Stern, J.B. Car lin, D.B. Dun son, A. Ve htari, D.B. Ru bin, Bayesiandata analy sis, Chap man and Hall/ CRC, 2013.

[57] A. Gel man, A. Jakulin, M.G. Pit tau, Y.S. Su, A weakly in for ma tive de fault prior distri b u tion for lo gis tic and other re gres sion mod els, Ann Appl Stat 2 (2008) 13601383.

[58] M.D. Hoff man, A. Gel man, The No-U-turn sam pler: adap tively set ting path lengthsin Hamil ton ian Monte Carlo, J Mach Learn Res 15 (2014) 1593 1623.

11



UNCORRECTED PROOF


[59] S. Watan abe, As ymp totic equiv a lence of Bayes cross val i da tion and widely ap plic able in for ma tion cri te rion in sin gu lar learn ing the ory, J Mach Learn Res 11 (2010)3571 3594.

[60] A. Gel man, J. Hwang, A. Ve htari, Un der stand ing pre dic tive in for ma tion cri te ria forBayesian mod els, Stat Com put 24 (2014) 997 1016.

[61] Stan De vel op ment Team. Stan mod el ing lan guage users guide and ref er ence manual. Ver sion 2.17.0;2017.

[62] R. Ihaka, R. Gen tle man, R: a lan guage for data analy sis and graph ics, J Com putGraph Stat 5 (1996) 299 314.

[63] R. De vel op ment Core, R. Team, A lan guage and en vi ron ment for sta tis ti cal comput ing, R Foun da tion for Sta tis ti cal Com put ing, Vi enna (Aus tria), 2018.

[64] J. Berthiller, K. Straif, A. Agudo, et al., Low fre quency of cig a rette smok ing and therisk of head and neck can cer in the IN HANCE con sor tium pooled analy sis, Int JEpi demiol 45 (2016) 835 845.

[65] J.J. Zuo, Z.Z. Tao, C. Chen, et al., Char ac ter is tics of cig a rette smok ing with out alco hol con sump tion and la ryn geal can cer: over all and time-risk re la tion. A meta-analy sis of ob ser va tional stud ies, Eur Arch Otorhi no laryn gol 274 (2017) 16171631.

[66] N. Moli nari, J.F. Du rand, R. Sabatier, Bounded op ti mal knots for re gres sion splines,Com put Stat Data Anal 45 (2004) 159 178.

[67] Wenxin Mao, Linda H. Zhao, Free-knot poly no mial splines with con fi dence in tervals, J Royal Sta tis ti cal Soc B 65 (4) (2003) 901 919, https: // doi. org/10. 1046/ j.1369- 7412. 2003. 00422. x.

[68] H. Schöllnberger, M. Manuguerra, H. Bi jwaard, et al., Analy sis of epi demi o log i calco hort data on smok ing ef fects and lung can cer with a multi-stage can cer model,Car cino gen e sis 27 (2006) 1432 1444.

[69] R.S. Cara ballo, G.A. Giovino, T.F. Pechacek, P.D. Mow ery, Fac tors as so ci ated withdis crep an cies be tween self-re ports on cig a rette smok ing and mea sured serum co tinine lev els among per sons aged 17 years or older: Third Na tional Health and Nutri tion Ex am i na tion Sur vey, 1988 1994, Am J Epi demiol 153 (2001) 807 814.

[70] N.A. Morales, M.A. Ro mano, K. Michael Cum mings, et al., Ac cu racy of self-reported to bacco use in newly di ag nosed can cer pa tients, Can cer Causes Con trol24 (2013) 1223 1230, https: // doi. org/10. 1007/ s10552- 013- 0202- 4.

[71] K.J. Roth man, S. Green land, T. Lash, Mod ern epi demi ol ogy, Third Edi tion, Lip pincott Williams & Wilkins, Philadel phia, 2008303 327.

12

Ora lncogy - unimi.it

Documents

Transcript of Ora lncogy - unimi.it