Cancer risk among insulin users: comparing analogues with human insulin in the CARING five-country cohort study

ARTICLE Cance isk among insulin use s: compa ing analogues wi h human insulin in he CARING i e-coun y coho s udy Anna Bu 1 &Ma ie L. De B uin 2,3 &Ma loes T. Bazelie 2 &Vida Hjell ik 4 & Mo en Ande sen 5,6,7 &Anssi Au inen 8 &Jakob S a up-Linde 9 & Ma janka K. Schmid 10,11 &Ka i Fu u 4 &F ank de V ies 2,12,13,14 & Øys ein Ka ls ad 4 &Nils Eks öm 5 &Ja i Haukka 1 Recei ed: 13 Decembe 2016 /Accep ed: 24 Ap il 2017 /Published online: 1 June 2017 #The Au ho (s) 2017. This a icle is an open access publica ion Abs ac Aims/hypo hesis The aim o his wo k was o in es iga e he ela ionship be ween use o ce ain insulins and isk o cance , when add essing he limi a ions and biases in ol ed in p e i- ous s udies. Me hods Na ional Heal h Regis ies om Denma k (1996–2010), Finland (1996–2011), No way (2005–2010) and Sweden (2007–2012) and he UK Clinical P ac ice Resea ch Da alink da abase (1987–2013) we e used o conduc a coho s udy on new insulin use s (N= 327,112). By using a common da a model and semi-agg ega e app oach, we pooled indi idual-le el eco ds om i e coho s and applied Poisson eg ession models. Fo each o en cance si es s udied, we es ima ed he a e a ios (RRs) by du a ion (≤0.5, 0.5–1, 1–2, 2–3, 3–4, 4–5, 5–6 and >6 yea s) o cumula i e exposu e o insulin gla gine o insulin de emi ela i e o ha o human insulin. Resul s A o al o 21,390 cance cases occu ed du ing a mean ollow-up o 4.6 yea s. No end wi h cumula i e ea men ime o insulin gla gine ela i e o human insulin was obse ed in isk o any o he en s udied cance ypes. O he 136 associa ions es ed in he main analysis, only a ew inc eased and dec eased isks we e ound: among women, a highe isk was obse ed o colo ec al (RR 1.54, 95% CI 1.06, 2.25) and Elec onic supplemen a y ma e ial The online e sion o his a icle (doi:10.1007/s00125-017-4312-5) con ains pee - e iewed bu unedi ed supplemen a y ma e ial, which is a ailable o au ho ised use s. *Anna Bu anna.bu @helsinki. i *Ma ie L. De B uin M.L.deB [email protected] 1 Depa men o Public Heal h Clinicum, Uni e si y o Helsinki, Tukholmanka u 8B, P.O. Box 20, 00014 Helsinki, Finland 2 Di ision o Pha macoepidemiology and Clinical Pha macology, U ech Ins i u e o Pha maceu ical Sciences, U ech Uni e si y, Uni e si ei sweg 99, 3584 CG U ech , he Ne he lands 3 Copenhagen Cen e o Regula o y Science (CORS), Depa men o Pha macy, Uni e si y o Copenhagen, Copenhagen, Denma k 4 Depa men o Pha macoepidemiology, No wegian Ins i u e o Public Heal h, Oslo, No way 5 Cen e o Pha macoepidemiology, Ka olinska Ins i u e , S ockholm, Sweden 6 Resea ch Uni o Gene al P ac ice, Uni e si y o Sou he n Denma k, Odense, Denma k 7 Depa men o D ug Design and Pha macology, Uni e si y o Copenhagen, Copenhagen, Denma k 8 Depa men o Epidemiology, School o Heal h Sciences, Uni e si y o Tampe e, Tampe e, Finland 9 Depa men o Endoc inology and In e nal Medicine, Aa hus Uni e si y Hospi al THG, Aa hus, Denma k 10 Di ision o Molecula Pa hology, The Ne he lands Cance Ins i u e – An oni an Leeuwenhoek Hospi al, Ams e dam, he Ne he lands 11 Di ision o Psychosocial Resea ch and Epidemiology, The Ne he lands Cance Ins i u e –An oni an Leeuwenhoek Hospi al, Ams e dam, he Ne he lands 12 The Ne he lands Depa men o Clinical Pha macy and Toxicology, Maas ich Uni e si y Medical Cen e, Maas ich , he Ne he lands 13 The Ne he lands Resea ch Ins i u e CAPHRI, Maas ich Uni e si y, Maas ich , he Ne he lands 14 The Ne he lands MRC Li ecou se Epidemiology Uni , Uni e si y o Sou hamp on, Sou hamp on, UK Diabe ologia (2017) 60:1691–1703 DOI 10.1007/s00125-017-4312-5 endome ial cance (RR 1.78, 95% CI 1.07, 2.94) o ≤0.5 yea s o ea men and o malignan melanoma o 2–3 yea s (RR 1.92, 95% CI 1.02, 3.61) and 4–5 yea s (RR 3.55, 95% CI 1.68, 7.47]); among men, a lowe isk was obse ed o panc ea ic cance o 2–3 yea s (RR 0.34, 95% CI 0.17, 0.66) and o li e cance o 3–4 yea s (RR 0.36, 95% CI 0.14, 0.94) and >6 yea s (RR 0.22, 95% CI 0.05, 0.92). Compa isons o insulin de emi wi h human insulin also showed no consis en di e ences. Conclusions/in e p e a ion The p esen mul i-coun y s udy ound no e idence o consis en di e ences in isk o en cance s o insulin gla gine o insulin de emi use compa ed wi h human insulin, a ollow-up exceeding 5 yea s. Keywo ds Cance isk .Coho s udy .Common da amodel . Cumula i e ea men ime .De emi .Gla gine .Human insulin .Insulinanalogues .Newinsulinuse .Poissonmodel . Ra e a io .Semi-agg ega e .Si e-speci ic Abb e ia ions ATC Ana omical The apeu ic Chemical CARING Cance isk and insulin analogues CPRD Clinical P ac ice Resea ch Da alink DDD De ined daily dose HRT Ho mone eplacemen ea men NIAD Non-insulin an idiabe ic d ug NSAID Nons e oidal an i-in lamma o y d ug RR Ra e a io In oduc ion Diabe es melli us and cance a e common diseases wi h ising incidence and p e alence globally [1,2]. Diabe es is associa - ed wi h an inc eased isk o ce ain cance s [3]and hepa e n and magni ude o he excess isk a e gene ally simila o ype 1 and 2 diabe es [4,5]. I has been sugges ed ha ce ain diabe es isk ac o s, as well as glucose-lowe ing medica ions, may con ibu e o his associa ion [6]. In 2009, he publica ion o ou obse a ional s udies [7–10] spa ked conce ns abou insulin gla gine as a po en ial modi ie o cance isk [11]. The oncogenic po en ial o a ious insulin analogues has been sugges ed by p eclinical sa e y e alua ions showing ha IGF and insulin ecep o signalling pa hways, which a e es- sen ial o mi ogenic po ency, a e a ec ed by ligand-speci ic ecep o dynamics, depending on he cell ype [12]. Ini ial obse a ional s udies [7–10] we e c i icised o limi a ions and biases [11,13,14] such as sho ollow-up, inclusion o p e alen insulin use s and ime-lag bias [15]. Fu he a emp s o ule ou o con i m he associa ion yielded inconsis en indings, emphasising he impo ance o p ope ly designed and conduc ed obse a ional s udies [16]. The e idence om he obse a ional s udies emains incon- sis en [17,18], pa icula ly due o he in ol emen o me hod- ological d awbacks, such as ime- ela ed biases and selec ion bias [15]. Mo eo e , he indings om s udies ha igno e dose– e ec aspec s, ime- a ying cha ac e o exposu e, si e-speci ic isks o long la ency o cance a e limi ed in hei biological plausibili y [16,19]. I was hough ha RCTs, he gold s an- da d o ea men compa isons, would disen angle he associa- ion. Howe e , he ORIGIN ial, which showed a null e ec [20] and cons i u es pe haps he s onges RCT e idence conce ning he use o insulin analogues and cance ou comes, has been c i icised o e alua ion o cance isk wi h espec o he ini ial insulin gla gine alloca ion bu no he cumula i e insulin gla gine use [21]. O e all, he e is li le obus e idence om obse a ional s udies o RCTs o d aw i m conclusions. The p esen obse a ional s udy, which is a pa o he Cance Risk and Insulin Analogues (CARING) p ojec , sough o in es iga e he e ec o exposu e o insulin gla gine o insulin de emi on cance isk as compa ed wi h ha o human insulin, mi iga ing h ough s udy design and analy ical app oaches he limi a ions and biases in ol ed in he p e ious s udies. Me hods O e iew This coho s udy on new insulin use s was conduc - ed using na ionwide da a om he No wegian, Swedish, Danish and Finnish Na ional Heal h Regis ies, as well as da a om UK gene al p ac i ione s in he Clinical P ac ice Resea ch Da alink (CPRD). P ojec pa ne s ob ained e hics app o al om hei espec i e au ho i ies. The s udy p o ocol, whe e da a sou ces and s udy coho s a e desc ibed in mo e de ail, was egis e ed in he Eu opean Ne wo k o Cen es o Pha macoepidemiology and Pha maco igilance (ENCePP) elec onic egis e o s udies [22]. To synch onise de ini ions o demog aphics, exposu es, ou comes and con ounde s, a common da a model and concep dic iona y we e de eloped. Table 1shows he s udy design and me hodological app oaches o he da a analysis ha we implemen ed o mi iga e di e en ypes o selec ion, in o ma ion and ime- ela ed biases o en inhe en in he obse a ional esea ch [15,23–25]. Da a sou ces, selec ion and ollow-up o pa icipan s Na ional heal h egis ies in he No dic coun ies comp ise compu e ised eco ds o he en i e popula ion o 26 million people, each o whom a e assigned a unique pe sonal iden i- ica ion numbe . The cance egis ies ha e a long adi ion o p o iding compa able and high-quali y da a wi h almos 100% co e age o inciden cance cases [26]. P esc ip ion egis ies, es ablished in 1995 in Denma k and Finland, in 2004 in No way and in 2005 in Sweden, ha e p o ided ample da a o pha maco-epidemiological esea ch [27]. The CPRD, 1692 Diabe ologia (2017) 60:1691–1703 Table 1 Me hodological sho comings and biases mi iga ed, add essed by checking he obus ness o he esul s o disen angled by he design and analy ical ea u es used in he s udy Design/analy ical ea u e Selec ion bias In o ma ion bias Time- ela ed bias Con ounding bias Heal hca e access bias a P e alen use bias b Misclassi ica ion o exposu e c Misclassi ica ion o ou come c , de ec ion bias d P o opa hic bias ( e e se causa ion) e Immo al ime bias Time-lag bias g , ime-window bias h Con ounding by indica ion i Residual con ounding j Adjus men o ime since s a o insulin use ––Mi iga ed –– –Mi iga ed g –Mi iga ed Ac i e-compa a o app oach ––– Mi iga ed d –Mi iga ed Mi iga ed g Mi iga ed Mi iga ed Cumula i e exposu e de ini ion ––Mi iga ed –Disen angled –– Mi iga ed – Na ionwide No dic d ug egis e s Mi iga ed –– – – –– – – Na ionwide No dic cance egis e s ––– Mi iga ed c ––––– New-use design –Mi iga ed Mi iga ed –– Mi iga ed Mi iga ed h Mi iga ed – Sensi i i y analysis Checked k –– Checked c,k –– –Checked l Time- a ying exposu e de ini ion ––Mi iga ed –Disen angled Mi iga ed Mi iga ed h,m Mi iga ed a Heal hca e access bias: di e en ial deg ee o access o he heal hca e among pa ien s b P e alen use bias: isk o ou come is conside ably highe o lowe du ing he ea ly pe iod o d ug he apy; inclusion o p e alen use s may dis o he associa ion be ween he use o d ug and ou come (including p e alen use s may also in oduce con ounding) c Misclassi ica ion (measu emen ) bias: inaccu a e measu emen o classi ica ion o key s udy a iables, such as exposu e, ou come o con ounde s: misclassi ica ion o exposu e may a ise om use o oo simple (bina y) exposu e de ini ion, especially o he complex pa e n o use; misclassi ica ion o ou come may occu due o he use o incomple e eco ds o iden i y e en s d De ec ion bias: di e en p obabili y o ou come de ec ion du ing he ollow-up in he compa ed g oups e P o opa hic bias: symp oms ea ed by a d ug a e he mani es a ion o he ye -undiagnosed disease o in e es Immo al ime bias a ises because o exclusion o misclassi ica ion o he ollow-up ime be ween he coho en y and i s exposu e o a d ug, be o e which he ou come o in e es canno occu g Time-lag bias: compa ed ea men s a e commonly used a di e en s ages o he disease ( i s -line he apy s second- o hi d-line he apy) h Time-window bias: unequal oppo uni y o become exposed be ween he compa ed g oups owing o he ime-window di e en ial i Con ounding by indica ion: di e ences be ween compa ed ea men s wi h espec o hei indica ions (o con aindica ions) j Residual con ounding: he dis o ion ha emains a e con olling o con ounding due o he unmeasu ed/uncon olled con ounde s k Res ic ion o he No dic coho s l Res ic ion o heindi iduals wi h ype 2 diabe es m Res ic ion o he calenda pe iod om 2000 onwa ds Diabe ologia (2017) 60:1691–1703 1693 a la ge compu e ised da abase es ablished in he UK in 1987, con ains anonymised medical eco ds, including demo- g aphics, p esc ip ions and cance diagnoses, ha a e consid- e ed o be o good quali y [28]. Cu en ly, 4.4 million indi id- uals, 6.9% o he UK popula ion, mee he quali y c i e ia and a e b oadly ep esen a i e o he en i e popula ion wi h ega d o demog aphic cha ac e is ics [29]. Wi hin he s udy pe iod (Table 2) de ined as he pe iod o alid da a collec ion [22], we iden i ied all indi iduals ha ing a leas one insulin pu chase (No dic coun ies) o p esc ip ion (CPRD). No dic coho s we e linked wi h he egis e ed da a on cance , dea h and emig a ion; da a o he B i ish coho we e compiled om in o ma ion on cance and dea h a ail- able om he CPRD. New insulin use s, who we e de ined based on a 1 yea lead-in pe iod, we e included i hey had no his o y o cance (excep non-melanoma skin cance ) and we e aged ≥18 yea s on he i s p esc ip ion o any insulin (index da e). Follow-up s a ed a index da e and ended a he da e o emig a ion (Sweden, Denma k and No way) o ans- e ou o he CPRD, diagnosis o any cance (excluding non- melanoma skin cance ), dea h o end o ollow-up, whiche e occu ed i s . Cance ou comes, insulin ea men s and po en ial con- ounde s We elied on coding dic iona ies, compiled acco d- ing o di e en coding sys ems (ICD-7, ICD-9 [www. icd9da a.com/2007/Volume1], ICD-10 [www.who.in / classi ica ions/icd/en/] and ICD-O-3 in he No dic coun ies; Read code sys em in he CPRD), o iden i y inciden cance cases de ined as he i s occu ence o any cance ype [22]. Mul iple cance s diagnosed on he same da e we e conside ed as dis inc si e-speci ic endpoin s. Ou p ima y in e es was in Table 2 Baseline and ollow-up cha ac e is ics o he s udy coho s o new use s o insulin Cha ac e is ic Denma k (N= 66,698) Finland (N= 105,945) No way (N= 21,541) Sweden (N=85,319) UK (CPRD) (N= 47,609) S udy pe iod a 1996–2010 1996–2011 2005–2010 2007–2012 1987–2013 Male sex, n(%) 38,292 (57) 57,691 (55) 12,053 (56) 48,931 (57) 25,589 (54) Age yea s, mean (SD) b 60.1 (16.0) 61.6 (15.9) 57.3 (17.8) 64.0 (16.4) 59.3 (16.3) Age yea s, n(%) b 18–30 3193 (4.8) 4481 (4.2) 1634 (7.6) 3121 (3.6) 2555 (5.4) 30–40 5075 (7.6) 7317 (6.9) 2653 (12.3) 4924 (5.8) 4540 (9.5) 40–50 8334 (12.5) 11,095 (10.5) 3088 (14.3) 8383 (9.8) 5825 (12.2) 50–60 14,432 (21.6) 22,928 (21.6) 4248 (19.7) 14,870 (17.4) 9351 (19.6) 60–70 16,306 (24.5) 25,254 (23.8) 4220 (19.6) 21,378 (25.1) 11,661 (24.5) 70–80 12,527 (18.8) 22,110 (20.9) 3194 (14.8) 17,176 (20.1) 9505 (20.0) 80+ 6831 (10.2) 12,760 (12.0) 2504 (11.6) 15,466 (18.1) 4172 (8.8) Follow-up ime, yea s Mean (SD) 5.3 (3.9) 5.6 (3.9) 2.7 (1.8) 2.7 (1.8) 5.7 (4.3) Median (in e qua ile ange) 4.5 (1.9, 7.8) 4.7 (2.3, 8.3) 2.5 (1.1, 4.1) 2.6 (1.1, 4.1) 4.7 (1.9, 8.4) No. o pe son-yea s/1000, all (male sex) 331.2 (184.4) 589.1 (316.9) 57.8 (32.4) 226.6 (131.3) 265.3 (141.6) E e -use, n(%) c Human insulin 54,216 (81) 68,894 (65) 17,579 (82) 48,976 (57) 23,183 (49) Insulin gla gine 7151 (11) 43,741 (41) 1447 (7) 15,138 (18) 15,374 (32) Insulin de emi 9520 (14) 24,593 (23) 868 (4) 4367 (5) 7373 (15) O he insulin 33,388 (50) 48,280 (46) 14,376 (67) 53,810 (63) 27,491 (58) Baseline use, n(%) HRT d 5187 (18) 6546 (14) 1641 (17) 6621 (18) 1530 (7) NSAID e 17,800 (27) 29,609 (28) 5437 (25) 16,485 (19) 8935 (18) Any o al glucose-lowe ing he apy 49,569 (74) 83,935 (79) 15,051 (70) 62,522 (73) 37,239 (78) S a in 22,948 (34) 38,493 (36) 9309 (43) 39,635 (46) 24,188 (51) a S a o s udy pe iod de ined acco ding o he s a o p esc ip ion egis y (No dic coun ies) o s a o alid da a collec ion (CPRD) b Age a baseline c E e -use o speci ic insulin du ing he ollow-up d Female sex only e Based on p esc ip ions only 1694 Diabe ologia (2017) 60:1691–1703 en si e-speci ic cance s. Based on NORDCAN da a (cance s a is ics om No dic coun ies) [26], we selec ed he eigh cance ypes wi h he highes incidence a es (ICD-10 codes): achea and lung (C33, C34), melanoma o skin (C43), blad- de (C67), colo ec al (C18-21), non-Hodgkin lymphoma (C82-86, C88.4), b eas (C50), endome ial (C54) and p os a e (C61). Li e (C22) and panc ea ic (C25) cance s we e also included because o hei s ong associa ion wi h diabe es. As a seconda y ou come o in e es , we in es iga ed he i s occu ence o any cance . Based on Ana omical The apeu ic Chemical (ATC) classi- ica ion codes [30] (B i ish Na ional Fo mula ion codes o he CPRD), we iden i ied use s o human insulin (A10AC01, A10AB01, A10AD01, A10AE01, A10AF01) and he insulin analogues insulin gla gine (A10AE04) and insulin de emi (A10AE05). Any o he insulins and analogues we e consid- e ed as a single g oup. P esc ip ion da a o m he No dic eg- is ies included he da e and amoun pu chased, in de ined daily doses (DDDs) [30], bu no in o ma ion on indi idual dosage. Fo he CPRD coho , we de i ed DDDs om he dosage in o ma ion (subs ance s eng h and amoun ) con ained wi hin p esc ip ion da a. We assumed a daily con- sump ion o 1 DDD pe day and ans o med each d ug eco d in o a pe iod co e ed by he numbe o DDDs. Fo each insulin ype o in e es , we de ined insulin expo- su e ime-dependen ly as a cumula i e ea men ime. A e spli ing he indi idual ollow-up pe iod in o in e als o 120 days, he exposu e a he beginning o each in e al was upda ed. The exposu e began on he da e o i s p esc ip ion/ pu chase, a e which poin an indi idual was conside ed ex- posed. Cumula i e ea men ime acc ued un il exposu e s opped and emained unchanged, unless ea men was e- sumed (see elec onic supplemen a y ma e ial [ESM] Me hods). We hen di ided cumula i e ea men ime in o hal -yea ca ego ies o he i s yea and 1 yea ca ego ies o longe exposu e; he las ca ego ies we e >6 yea s o he b oadly ca ego ised exposu e and 9–10 yea s o he inely ca ego ised exposu e. In addi ion, each exposu e a iable in- co po a ed a non-exposed ca ego y assigned o indi iduals emaining unexposed o he speci ic insulin. We conside ed only con ounde s a ailable in all i e da ase s [22]. In addi ion o age, sex and calenda ime, his included use o non-insulin glucose-lowe ing d ugs (ATC code A10B), s a ins (C10A), nons e oidal an i-in lamma o y d ugs (NSAIDs; M01A) and ho mone eplacemen ea men (HRT; G03), de ined as a leas one p esc ip ion wi hin 1 yea be o e he index da e. We also de i ed se e al o he po en ial con ounde s: ype 1 diabe es melli us was assigned o hose aged ≤30 yea s wi h no non-insulin glucose-lowe ing d ug on he index da e; ype 2 diabe es melli us was assigned o hose aged ≥40 yea s wi h o wi hou non-insulin glucose-lowe ing d ugs; unspeci ied diabe es ype was assigned o he es o he coho . We speci ied he du a ion o insulin- ea ed diabe es as ime since he index da e (in 1 yea in e als) and de ined menopausal s a us ime-dependen ly based on cu -o o 50 yea s o age. Fu he mo e, he coun y o he da a o igin se ed as a co a ia e. S a is ical me hods The indi idual-le el da a om he i e coho s we e s anda dised by each esea ch pa ne locally using he common da a model. We hen conduc ed cen alised analyses by uploading he uni ied da a o a se e a S a is ics Denma k, whe e o each coho we cons uc ed he indi idual- le el da ase o assess insulin exposu e and o he a iables in exac ly he same way. We employed a semi-agg ega e le el app oach [31] o combine he da ase s, which we e abula ed by cance si e as he numbe o cance cases and pe son-yea s agg ega ed by ca ego ical a iables. To es ima e he incidence a es, we i ed mul i a iable Poisson eg ession models o he e en numbe s wi h he na u al log o pe son-yea s as an o se . Each model included all h ee ime-dependen insulin exposu e a iables and was adjus ed o ime-dependen age and du a ion o insulin- ea ed diabe es, sex (no in he sex-s a i ied analysis), baseline calenda ime, use o non-insulin an idiabe ic d ugs (NIADs), o he co-medica ion (when ele an [22]) and coun y. We conduc ed an ac i e-compa a o analysis [25], whe e he d ug o in e es is compa ed wi h ano he d ug commonly used o he same indica ion a he han wi h no ea men . Inclusion o all h ee insulin exposu es in he same model allowed us o calcula e he a e a ios (RRs) and 95% CIs o a pa icula exposu e ca ego y by con as ing he incidence a es, which we e es ima ed o each insulin ype and du a ion. In he p ima y analyses, we examined sex- and si e-speci ic cance endpoin s wi hou sepa a ing be ween diabe es ypes and using insulin exposu es wi h a b oade ca ego y (>6 yea s) o he longe cumula i e ea men ime. Fo he seconda y analyses, we pe o med simila e alua ions wi hou s a i ying on sex and using insulin exposu es wi h ine ca ego ies. Sensi i i y analysis We also pe o med se e al sensi i i y analyses. We es ic ed he analyses o hose who me ype 2 diabe es c i e ia o check whe he he esul s change by diabe- es ype (da a o indi iduals wi h ype 1 diabe es we e limi - ed). In Eu ope, ma ke ing au ho isa ion o he long-ac ing insulin analogues insulin gla gine and insulin de emi was g an ed in June 2000 and June 2004, espec i ely [32,33]. Coinciden ally, usage o wo al eady-app o ed apid-ac ing insulin analogues, insulin lisp o and insulin aspa , gained popula i y in he ea ly 2000s. To accoun o he changes in he p o iles o new insulin use s, we excluded hose en e ed be o e 2000. By excluding he CPRD, we add essed he po- en ial o unde es ima ing cance incidence due o case asce - ainmen h ough he CPRD only wi hou linkage o he na- ional cance egis a ion da a. Fo b eas and endome ial can- ce , we u he adjus ed o menopausal s a us. Diabe ologia (2017) 60:1691–1703 1695 We used e sion 3.2.2 o R (www.R-p ojec .o g)[34] o pe o m all s a is ical analyses, he Epi package, e sion 1.1. 71 (h ps://c an. -p ojec .o g/web/packages/Epi/index.h ml) [35] o ca y ou exposu e calcula ions and he o es plo package, e sion 1.7 (h ps://c an. -p ojec .o g/web/ packages/ o es plo /index.h ml)[36] o he g aphical ou pu . Resul s In he i e coho s o alling 327,112 new insulin use s, men p edomina ed and he mean age a baseline a ied be ween 57 and 64 yea s (Table 2). Fo he combined da a, he mean ollow-up ime was 4.6 yea s (median 3.7, in e qua ile ange 1.7–6.3). A he end o ollow-up, he e was 212,848, 82,851 and 46,721 e e -use s o human insulin, insulin gla gine and insulin de emi , espec i ely. In all coho s, human insulin p edomina ed in e e -use pa e ns. E e -use o insulin gla gine and insulin de emi was mos common in he Finnish coho , as was he baseline use o non-insulin glu- cose-lowe ing he apy. Baseline use o o he medica ion also di e ed be ween he coho s. A o al o 1.47 million pe son-yea s accumula ed and 21,390 new cance cases occu ed du ing he ollow-up. Table 3shows coun y- and sex-speci ic c ude incidence a es o he en si e-speci ic cance s and any cance . P os a e cance in men and b eas cance in women showed he highes inci- dence a es in all coho s excep he No wegian, whe e pan- c ea ic cance was he mos common cance . Abou 32% o all cance cases and he majo i y o panc ea ic cance cases (63%) we e diagnosed du ing he i s yea o insulin ea men . The esul s o he sex- and si e-speci ic analyses showed no sys ema ic di e ences ac oss si es and exposu e ca ego ies (Figs 1,2and ESM Table 1). In women, a highe isk o colo ec al (RR 1.54, 95% CI 1.06, 2.25) and endome ial (RR 1.78, 95% CI 1.07, 2.94) cance was obse ed o he i s hal -yea o cumula i e ea men ime on insulin gla gine ela i e o ha on human insulin and a highe isk o mela- noma o skin o 2–3 yea s (RR 1.92, 95% CI 1.02, 3.61) and 4–5 yea s (RR 3.55, 95% CI 1.68, 7.47). In men, simila compa isons yielded a lowe isk o panc ea ic cance o 2–3 yea s o exposu e (RR 0.34, 95% CI 0.17, 0.66), and o li e cance o 3–4 yea s (RR 0.36, 95% CI 0.14, 0.94) and >6 yea s (RR 0.22, 95% CI 0.05, 0.92). In addi ion, he esul s sugges ed an inc ease in he isk o bladde (RR 1.41, 95% CI 0.92, 2.17) and colo ec al cance (RR 1.28, 95% CI 0.94, 1.75) in men o <0.5 yea s and o b eas cance o <0.5 yea s (RR 1.32, 95% CI 0.98, 1.79) and 0.5–1 yea s (RR1.32, 95% CI 0.95, 1.85) in women. In simila analyses pe o med using he 1 yea ca ego ies o longe du a ion o exposu e (≥6 yea s) and bo h sexes combined (no sex-speci ic cance s), esul s emained simila (ESM Fig. 1): RR (95% CI) 1.41 (1.11, 1.79) o colo ec al cance o <0.5 yea , 0.67 (0.43, 1.03) o panc ea ic cance o 2–3 yea s, 0.44 (0.21, 0.91) o li e cance o 3–4yea s and 1.60 (1.05, 2.43) o melanoma o he skin o 4–5yea s o insulin gla gine s human insulin use. Compa isons o insulin de emi s human insulin and insulin gla gine s insulin de emi also showed no consis en di e ences in sex- and si e-speci ic incidence a es (ESM Table 1)aswellasin he analyses combining bo h sexes (ESM Figs 2and3). Fo any cance in women, we ound an ele a ed isk o 0.5 yea o insulin gla gine use ela i e o human insulin (RR 1.17, 95% CI 1.03, 1.32); in men, he e was a lowe isk o 0.5–1 yea (RR 0.87, 95% CI 0.77, 1.00), 1–2yea s(RR0.84, 95% CI 0.73, 0.95) and >6 yea s (RR 0.61, 95% CI 0.48, 0.78) o exposu e (Figs 1,2and ESM Table 1). Analysis pe o med wi hou s a i ying on sex (Fig. 3) yielded an ele a ed isk o any cance o insulin gla gine use ela i e o human insulin o 0.5 yea (RR 1.12, 95% CI 1.03, 1.20) and a lowe isk o 1–2 yea s (RR 0.90, 95% CI 0.83, 0.98), 6–7y ea s(RR0.72, 95% CI 0.56, 0.91) and 7–8 yea s (RR 0.62, 95% CI 0.44, 0.86). O he analyses yielded a lowe isk o any cance in men o 0.5–1, 2–3 and >6 yea s o insulin de emi use ela i e o ha o human insulin (ESM Table 1), and an inc eased isk in men and women combined (RR 1.18, 95% CI 1.05, 1.33) o <0.5 yea s o insulin gla gine use ela i e o ha o insulin de emi (Fig. 3). Resul s we e obus ac oss a ange o sensi i i y analyses (ESM Tables 2, 3). Inclusion o he popula ion wi h ype 2 diabe es (1.31 million pe son-yea s [90%], 21,151cance cases [99%]) yielded simila esul s o hose o he p ima y analysis. The esul s did no change a e ei he es ic ion o he s udy pe iod o he ime a e insulin gla gine’s launch (1.12 million pe son-yea s [77%], 16,838 cance cases [79%]) o es ic ion o he s udy popula ion o he No dic coho s. Fo b eas and endome ial cance s, he esul s emained unchanged a e u he adjus men o menopausal s a us. Discussion In he coho s udy o 327,112 new insulin use s om i e Eu opean coun ies, we add essed he ela ionship be ween insulin use and cance incidence o en si e-speci ic cance s and any cance , when mi iga ing me hodological sho com- ings and biases in ol ed in p e ious s udies. Compa isons o cance incidence by cumula i e ea men ime using ac i e compa a o s showed no consis en di e ences in he cance isk o insulin gla gine o insulin de emi use ela i e o ha o human insulin use. Al hough we obse ed inc eased and dec eased cance isks o some si es and ea men du a ions, no ends in he isk wi h du a ion o use we e seen. The indings o p e ious obse a ional s udies on he ela- ionship be ween cance isk and use o insulin gla gine a e 1696 Diabe ologia (2017) 60:1691–1703 Table 3 Sex- and si e-speci ic numbe s o cance cases, c ude incidence a es wi h 95% con idence in e als Cance ype Denma k Finland No way Sweden UK (CPRD) No. IR (95% CI) No. IR (95% CI) No. IR (95% CI) No. IR (95% CI) No. IR (95% CI) Li e cance Men 144 0.78 (0.66, 0.92) 308 0.97 (0.87, 1.09) 15 0.46 (0.26, 0.76) 89 0.68 (0.54, 0.83) 74 0.52 (0.41, 0.65) Women 28 0.19 (0.13, 0.28) 102 0.37 (0.31, 0.45) <6 NS 23 0.24 (0.15, 0.36) 21 0.17 (0.11, 0.26) Panc ea ic cance Men 315 1.71 (1.52, 1.91) 531 1.68 (1.54, 1.82) 76 2.35 (1.85, 2.94) 238 1.81 (1.59, 2.06) 129 0.91 (0.76, 1.08) Women 233 1.59 (1.39, 1.80) 417 1.53 (1.39, 1.69) 64 2.52 (1.94, 3.21) 199 2.09 (1.81, 2.40) 109 0.88 (0.72, 1.06) Lung cance Men 466 2.53 (2.30, 2.77) 623 1.97 (1.81, 2.13) 55 1.70 (1.28, 2.21) 192 1.46 (1.26, 1.68) 218 1.54 (1.34, 1.76) Women 244 1.66 (1.46, 1.88) 174 0.64 (0.55, 0.74) 33 1.30 (0.89, 1.82) 112 1.17 (0.97, 1.41) 116 0.94 (0.77, 1.12) Melanoma o skin Men 59 0.32 (0.24, 0.41) 130 0.41 (0.34, 0.49) 27 0.83 (0.55, 1.21) 80 0.61 (0.48, 0.76) 69 0.49 (0.38, 0.62) Women 53 0.36 (0.27, 0.47) 77 0.28 (0.22, 0.35) 9 0.35 (0.16, 0.67) 45 0.47 (0.34, 0.63) 35 0.28 (0.20, 0.39) Bladde cance Men 106 0.57 (0.47, 0.70) 281 0.89 (0.79, 1.00) 43 1.33 (0.96, 1.79) 183 1.39 (1.20, 1.61) 117 0.83 (0.68, 0.99) Women 34 0.23 0.16, 0.32) 77 0.28 (0.22, 0.35) <6 NS 40 0.42 (0.30, 0.57) 28 0.23 (0.15, 0.33) Colo ec al cance Men 423 2.29 (2.08, 2.52) 484 1.53 (1.39, 1.67) 68 2.10 (1.63, 2.66) 312 2.38 (2.12, 2.66) 219 1.55 (1.35, 1.77) Women 258 1.76 (1.55, 1.99) 366 1.34 (1.21, 1.49) 46 1.81 (1.32, 2.41) 145 1.56 (1.31, 1.83) 102 0.82 (0.67, 1.00) Non-Hodgkin lymphoma Men 52 0.28 (0.21, 0.37) 180 0.57 (0.49, 0.66) 14 0.43 (0.24, 0.73) 68 0.52 (0.40, 0.66) 31 0.22 (0.15, 0.31) Women 47 0.32 (0.24, 0.43) 132 0.48 (0.41, 0.58) 7 0.28 (0.11, 0.57) 46 0.48 (0.35, 0.64) 35 0.28 (0.20, 0.39) B eas cance a 462 3.15 (2.87, 3.45) 729 2.68 (2.49, 2.88) 51 2.00 (1.49, 2.64) 250 2.62 (2.31, 2.97) 301 2.43 (2.17, 2.72) Endome ial cance 152 1.04 (0.88, 1.21) 289 1.06 (0.94, 1.19) 29 1.14 (0.76, 1.64) 107 1.12 (0.92, 1.36) 57 0.46 (0.35, 0.60) P os a e cance 501 2.72 (2.48, 2.97) 1339 4.23 (4.00, 4.46) 100 3.09 (2.51, 3.76) 582 4.43 (4.08, 4.81) 290 2.05 (1.82, 2.30) Any cance Men 2733 14.82 (14.27, 15.39) 5315 16.77 (16.33, 17.23) 542 16.74 (15.36, 18.21) 2272 17.31 (16.60, 18.04) 1918 13.54 (12.94, 14.16) Women 1950 13.29 (12.70, 13.89) 3618 13.29 (12.86, 13.73) 333 13.09 (11.73, 14.58) 1361 14.27 (13.52, 15.05) 1348 10.89 (10.32, 11.49) a Women only IR, incidence a e; NS, no shown (in cells whe e numbe o e en s <6 he IR and CI a e no shown acco ding o con iden iali y p inciples) Diabe ologia (2017) 60:1691–1703 1697 con lic ing and may in ol e me hodological limi a ions and biases [15,17]. Mos o he cu en e idence is based on sho ollow-up and use o an elemen a y ep esen a ion o expo- su e, igno ing he dose o du a ion o insulin exposu e. The la e may esul in he inadequa e isk es ima es, especially when ea men du a ions o widely di e ing leng hs a e con- side ed equi alen [15,19]. O he ecen s udies wi h a sho ollow-up as a main limi a ion [15], only ou we e buil on a new-use coho [37–40], and wo assessed cance isk by ea men du a ions using an ac i e-compa a o app oach [37,38]. In he obse a ional s udy on new use s o insulin gla gine (n= 43,306) and human insulin (n= 9147) en olled in a US heal h plan, no associa ion wi h he isk o p os a e, b eas , colon and any cance was ound o ea men du a ions o 0– 6, 6–12, 12–24 o ≥24 mon hs [38]. Howe e , he isk es i- ma es epo ed in his s udy we e imp ecise due o he small e e ence g oup. An obse a ional s udy on a coho o 70,027 new insulin use s in F ance ound no di e ences in isk a median ollow-up <3 yea s o bladde , b eas , colo ec al, head and neck, li e , lung and kidney cance be ween new use s o insulin gla gine and o he basal insulins, when excluding he i s yea o use and de ining exposu e as e e -use o cumula i e dose [40]. A s udy o a coho o 19,337 inciden insulin use s om he Ne he lands ound a dec eased isk o o e all and colon cance bu no di e ence in isk o bladde , espi a o y ac and p os a e cance , when compa ing ime-dependen ly de ined cumula i e ime using insulin gla gine o ha using human insulin, hough wi hou u he dis inc ion be ween di e en ea men du a ions [39]. In he p esen s udy, we ound ha 22–63% o cance cases we e diagnosed wi hin he i s yea a e s a ing insulin ea - men . Exclusion o 0.5–1 yea o insulin use, o analysis o sho - e m use only, may p eclude obse ing he ac ual dynam- ics o cance incidence among insulin use s and hus may hin- de a be e unde s anding o he na u e o he link be ween diabe es and cance . Fo he i s hal -yea o cumula i e ea - men ime on insulin gla gine ela i e o ha on human insulin, we ound an inc eased isk o colo ec al and any cance in women and bo h sexes combined, and o endome ial cance in women. These indings sugges possible in ol emen o de ec ion o p o opa hic bias [41].Thela e ismo elikely o Fig. 1 RR (adjus ed o age, calenda ime, NIADs, du a ion o insulin- ea ed diabe es, coun y; o li e and colo ec al cance s, addi ional adjus men o ele an co-medica ions) wi h 95% CI o si e-speci ic cance s and any cance in men by cumula i e ea men ime (yea s) on insulin gla gine s human insulin 1698 Diabe ologia (2017) 60:1691–1703 a ec he p esen s udy, whe ein use o insulin gla gine was less common han ha o human insulin and ini ia ion o insulin gla gine was o en p eceded by use o o he insulins, p edom- inan ly human insulin. Acco ding o he cu en guidelines, swi ching om human insulin o insulin gla gine should be conside ed i an indi idual has hypoglycaemia o ails o each he a ge glucose le el [42]. Poo glycaemic con ol may be a sign o unde lying cance and hus swi ching insulins because o highly a iable blood glucose could be associa ed wi h mo e equen ly de ec ed cance . Al hough he esul s o he p esen s udy sugges a shi owa ds inc eased isk o b eas cance o he ini ial yea o insulin gla gine s human insulin use, no di e ences we e ound o longe du a ions o ea men , when using app op i- a e compa a o s and adjus ing o he o e all ime on insulin. In con as o ou s udy, h ee ecen s udies ound an associ- a ion be ween use o insulin gla gine and inc eased isk o b eas cance [37,39,43]. Howe e , hose indings may e- lec an imbalance in compa a o and exposu e cha ac e is ics a he han di e ences in cance isk due o insulin use i sel . A s udy on he UK’s Gene al P ac ice Resea ch Da abase coho e ealed an ele a ed isk o >5 yea s since he s a o insulin gla gine, when compa ing insulin gla gine use s wi h p e ious use o insulin s p e alen use s o o he insu- lins, ma ched on p io insulin du a ion a baseline [37]. Compa ison by ime since s a o insulin gla gine use is likely o p o ide an unbiased es ima e o sho - e m use bu no o longe use whe e he ac ual ime on insulin and ime since he ini ia ion may di e no iceably be ween compa a o s. A s udy om he Ne he lands epo ed an ele a ed isk o b eas can- ce when using cumula i e du a ions wi hou di e en ia ion be ween hem, hus igno ing an imbalance be ween compa a- o s in he ollow-up imes (median o 2.2 yea s o insulin gla gine use s and 3.8 yea s o human insulin use s) [39]. A ecen s udy on a coho o 12,468 new insulin use s om he UK’s CPRD also epo ed an inc eased isk o b eas cance among new use s o insulin gla gine wi h ex ensi e pas ex- posu e o o he insulins, when compa ing insulin gla gine use o >3 yea s wi h e e -use o o he insulins [43]. Compa ison o speci ic du a ion wi h e e -use may yield a biased esul , especially when he ime pe iod co e ed by speci ic du a ion di e s conside ably om he pe iod co e ed by e e -use. Fig. 2 RR (adjus ed o age, calenda ime, NIADs, du a ion o insulin- ea ed diabe es, coun y; o li e , colo ec al, b eas and endome ial cance s, addi ional adjus men o ele an co-medica ions) wi h 95% CI o si e-speci ic cance s and any cance in women by cumula i e ea men ime (yea s) on insulin gla gine s human insulin Diabe ologia (2017) 60:1691–1703 1699