Pt)- bi-_ lb Thu Si: i@jil'
P r 7-,
r "@e!
Pv`tv Swu;CL n an. SC:Zr::.;_: ::SC:-,line be -a:c. Ccn-,::;n,.cc --;in :!@t nl`i-.::t .3i -.ne @iurr-.;-@
,V'Vilutlom Ar.-I con=J:0M. :!@: e-,:derr%ic c;:st en.-3ur.t-s se,-ert J:IT1c-_-:ts triv zo not -ritruct an cloenrn.:nts
stra,.ns 0l73r.:`n2:S U:!Ct! =rC"1jilv ccr@:rtilvj !atcrz:_-ryCor.C:!:Ons. In s@ite -c rc_@-
;r-. 7he Times. Novernzer scrnc sr:xins;y ;rcci" c!airns anout ::t:: num@e-s of zt2ins lunbutatic :0 Sr-.OK,.-.:.
z K,d. a 1!----iur-c doc-_@rncnt launc-ned by T @C "ta::n E-!L;raticn Ciurc.; !aintiv -,tr.
0..:zc c@aims art -3c: :n -,;-e It
Sr t@sh Mciiiczi Asscc:i,:ar.: -.-: wolur-c -e-- ssuc,_1 for t,.zn rcgionji hcaltn tma 'A
-kc,zording !o _6 sr-.zA;tr& :irnL;4..-, Kids 7 7-- :@Coplc -.5-5 IV mcr: and ta7 -cmcn; i it-.e -noie
in: Waics from licar-. -_;3C:St. 'Ur.; =2f`C:r arid Anc !:CS_2LSe Of 'n:!:r !OS
;nszzie are hosptti;izz-_' azcr: -ta. -1t,'i '.hwc @:;scziscs. T-,t!,t are uncenntcly tcrnfic :`;-_,rcs anz int-1 are ;.Ocjo;.. ;nlcnut@
!c -orrirv. Siv ccncer-@. @%c-e% tr. is less win -_'ne horror 2:--- more ---@n the sc:tnt;fic -_msis oi' :ne c:ojrns.
7-c biolovCaii.. rmaccr might for C0nV-.w'.;nj :!@_-L -:-.iccr!@iojc&y is not oni, A
Sz.V-.C_- .out an Xlcrzc;_-!@ one. in :r-n;ficz :.,or in rnortaitty or ;css th.., I part in 7- T.-.c
S,::-_r11lf`ic_-IIv .r2;r:c= --.:rson zzn asic: Ho- -:re the:.- irnpres-si, c =:Zero-'
T@e bu.%is of thc -!fit-.%
social se.-victs z:r-ts::cn:v-.; of The 7 -es -.c !s us. "The !%-_rzs ha,@c zetri zy ran& ,c-.::,c=z -,=Jtns '-C-
diseases ;ung 3-Z :rCricnills C-7pnysc-.3. arc :o smcit:ng -in: -,ro-_c-,.ions oy :r:
@_-rsi. how :S t@,.e -Z@zrCinij inc zcni6cat:cn )f the Cause or _-eztn7 '-k t 'live no -e@y rectnt
;r:--eStir,3Elor1 to 3ulce --S =@;- n .9.49 Heasmart and Lip-cr-n (1) surveyed :ec@oru .-T-n 75 ncs;tta;s of Natioral
Ser,ica in Engtand vid N4a.v:. They corrzaed the c.@n:c:ans' c:i3nosis of :1-.c ::,--se oi` oevn -mitt'. :ne paznoio&:s:s'
::::-opsy rc-'arz. C;)rs;r-.-- :ht :x2moic of :r:.-nary =nctf :f .he !un; -here. idmittzt!v. ;!,c:icrns ordi--Snosts are se%ert.
C'.:-.:c:ans d:aitn_-se= -"S czse! -h3ic pa(hc-cjists discovered 4 I 7 Zascs it ;:Os; Mcrcrn; :n only ::' irls-anc_-s. ho-e-,cr.
-is asrec-ncnt oinzin-ed. ir -.nz p3thcics:sz, rc-,orts -:rt corrcc- then : I '.. or of c:;.-.:c:ars' dizencscs --er:
cases (46:--i oCluri@; cznz_-r @ere rmsscd. A,;.houzn -e ruv arg,.;e '.hJLL Z221h Ctr'.01CZZ@C1,
in :986 is '.ikc;v -a oc more than it -as in J959 _e must acuct -hvncr pcr:'_-c:ion has be::-, achieved. %e ,lave n.-
--ti1ence -hcrezv -ccan csurnst: thc errors .hat retain ;.-. ccent arid can;:nworary 1-4Lz.
Second. hCw -c!! kno-ri sr: :!:e smCkin; -3bits o(the lopulation7 General Ho.;scniold 5@;rveys :_irsi inc!ucra Questions
on smoking ;n 197-- and re@'C.-tnct is made :a "...the d"'TcuRics a( coilcving rclia=ie data an Efte zopic ...... M-c Tobar:z
Rtsearch Caurcii (:? @:S ;u:@;sned cStimv_-s of the cons-impiion of manufactured z-%3re,.!es by men frcrn 1890 and
-cman from 1921 but it -&S :-.at until 1966 :nit survey data bec-3me. w-aiiaolc 'Or Ail 3CUIt age %ro-j:S orbc@:
s,-:-zs throu;ncut the Unitta K'ngdom. Altn2t_,Sh Elie tenzzicy ofpcopic lo underut;:-.rnace :nt:r c:girette consurrizition CZ7
'.-e 'and has been) by com-.aring survey :jta with 'Known cczl sails orcivarettes. .hcdurvion and his:c--.
oismok-ing on an inciviclu-3i '--asts is not avai.able. Even en sstr21j-t:_orkard caus.2i --.-oott-.e_-;s LrLe:_aLzil risXSZO Vi.-SMOKC.-S
sr:singrrom ttc:- t@c fe!isbiy cysi-.;3led
third. what referee shau:_1 be placed an the estimates of c:;arctte-=used dv%ths made by The Royal Colle;c of
Physicians (3)? This q-.;est:on @s. perhacs. .!t: -nos% diMc-.;;t ofall to inswerand %he sc=pefarc.-ror !s the !atgetst. extendin;
to several orders of magnituce. The Royal CaNege's cor-mittee S .nowed awareness of zctain dafric-.;Itics. The starting Pat.-.-
(cr 'heir calculations in 197, 1 Ls the survey or mortality amor.3 British male doctors -no voiUntecr-c-ti for
Cc:I-Hill study (-1) or the i-Tec-s of sm=King. The about or iivinz -note doctors. -ec nl!;L@;er
rczrc3critafive of tn: total pc=ujacion of I=or3 nor or the ;zncral male "c;.;:3won. The ;Royal coile5e made in
ac ustment for this!a,.1.crdm_-z=ancy ands;-,;cd-updca,..@s.orsrroKc.-sank: non-srr.GKc;rsa&z935-64vcarsin .11c: ger-cfal
.-czutation. over Itose arnon; docors. b, :he Consian. ratio. Some amazons arid limitaticns un;.::.!yinj; ;.@;b
CC.1.'cC1.:C1n wCfcnott= (3) anc t wasconsicv.-J thattr-c- ;opulai;zn
-.:--Id have t!t=n ........ lie rrvhod can !c--z only x an approximation".
-he next s:e= in :ne Rov3: Callegc's arz-.@.neqt is. perhaps. 0@c -must lr.1.-:7cs.:n;: "it :S not p-ossicic lo gt%c a prc-_:b:
rc ;:zn tic .,W
nate at inc -.romrtlor- c( :rise excess _!iIhs nrnunz srno@c@-s -h:ch are c3usc-_- :v snickini, 7"ne C
.31 !cast !ln'l, 31, :ne eS::r-_-:!_- 3 I ON Lx.;:@S rnz2e Brokers igt.@., _'J -o-I .11 ne Lnovj Were
to sr-1OKin;...- 7, h:s ;assalz: s:!---s a recc;.-T .:on tv the Fk:-al Coi::!;c thit net all at .ne ntn@v!n seeking --r.;
CZL;S,;. -s -nc@ccr an citim..it
.-@@7taficy is n Bruce-. nc :- ;! @f @ne -nak;njl-_,Uc Ol
ZZm;S-31 C0ntr:_-L::on rn:zn, ':c -lived jr.,1 :ne :Fort;= ",It -nu;:. !.CZ= :L. :@C jr-::@rzr,
J 'rare d:;zn::'tu- jC@_:!n!;On Choice in.; S 'L term L: S. Su-S:--. Ciner_ 1_- n @nis -nu a;
-i-.t.-n @:s "ci'r-s met ;r-.-:z:,,or..;:7:c @c CYN
'L C @7 L C 5 C 17
:II-- sr- 7! -,,;@C
Cr_%
BATCO LTD - MINNESOTATOBACCOLITIGATION

Forum
coronar) artery dise@.-se. These pron-ibiv Conservative assumptions to in c5airriatc of .1tiout 2J 000 dcalhs from these
three diseases zzusc%., t:y smoking among men ig---d 35 to ti-:-
The even greater JA7`iculE% uf %.stimjung smoking-caused Uciths a7n -, nV women was Ucs,:.,Ibcd (3) ind tht; Royal College
Concluded: "Hut it can reasonably be assumed that at least 40% c:' :Mc dc:ith% from IL:ng cancer. 60,vo at Those frineit
bronchitis. ine NS ol'tho@c frem :crnnurr heart disease in -omen 3Lt-, ;S - h4 tnjy well be due to cigarette smoking".
Expressions such as --Ouii.! not be unreasonable", .. Probably :_-nscrva6vc issumptions". "it can rcasonahh. he
assumed that". and "may -C!! he duc to". convey it least sorne )f the uncertainty, surrounding these c.%crcl.%cs in
judgement.
The BigKii'lAdopts the above percentages with the exception Of that for lung C3nCCt in -omen which -as raibcd from .10
to 30% on the authority or Profcssor Sir Richard Doll. That these ntcczly Founded pcrccnt,2Scs have culminated in Mc
estimiteclannual toll o(77 77.: dcaths (55 107 mcnand 22667 womeni. is not without a certain -himsecalcharm.
Wicare not informed about the procedure Used in The BrgKifitocst:.naic thcrsumberolsmoking-causcd deaths in those
aged 63 and above. The Royal College remarks "...this could net be done -i(h much conilidcrict. partly becaust:
certification of the cause of Jc2th in oldcr people. who may sulTcr a--Aricty or4mabilitics. is less accurate than in younger
people ..... no attempt has bc--,t mzdc to estimate the number of deaths _Iuc to cigarette smoking in older people." Mortality
ratios (smokc.-s versus non-smokc.-s) ror ischac-nic heart disc3sc rail - ith age and approach unitv at around 70-79 years:
they are likely to fill below Unity .1b0VC 80 )'cars ol'agc, thus providir.; a further complication for mcticulausactuaries. The
authors of The Stg Kill migrit have cn Ughtcrccl us about these 3wk-war: obstacles 10 COMPUMIOn and tl,.e way in which they
were circumvented.
Is a scientific analv%i% ltinssible'
We have to bear in mind ztat the reports of the Royal College c@ 71hysicians and of tne U.S, Surgeon General were
prepared by committees with a -,redcrninzintly medical background in: outlook. Their primary concern. therefore. is likely
to have been with the 3voidance of unnecessary sufTering and prema,@.:re death. No one can quarrel with these aims and the
good intentions Pc7rriezimr, the rczorts. The process or reaching sour.: conclusions about actusation is. however. more ori
scicnLiAC hall a medical task. Nledical skills arc required. orcourse. :c . each an accurate diagnosis of the cause of death and
a proper appreciation of limiLaucris in the evidence, but analysis oftt.- resulting $L3tiStiC3 Calls for ramiliarity and dexterity
with scientific :ogic. Tbc two skills are not incompatible but they ;re not always combined in the same person. We may
note that Feinstein. himself both a distinguished clinical cpidcmiolog:st and an expert medical statistician. protests (6) that
a "licensed" e;iderniologist ..... .an obtain and manipulate the C-,;,2 in diverse ways that are sanctioned not by the
delineated standards orsocricc. but by the traditional practiccor c par, ioiog ists ".
What arc these "traditional practices?" The Surgeon General re-.rc;;uccs in his 1982 Report (5) a pa5sjZc from his first
report about smoking and hczith published in 1964. It en=psula-.:s 3 meLhodology-curn-philosophy that enjoys wide
support among cptdcmiologists: causal significance or an issoc:arion isa matter or;udgmcnL which goes beyond any
statement of statistical probability. To judge or evaluate the causal vinificance of the association between an attribute or
agent and the disease. or the effect upon health. a numberorcriteria -.Just be utilized. no one of which is an all-sufficient
basis forjudgme-IL. T'nese criteria include: (a) the consistency of the association. (b) the strength or the association. (c) the
specificity or the association. (d) the temporal relationship of the association. and (e) the coherence or the association."
The inadequacy of these poorly defined criteria and the dubious manner of their application to the LSSOCiaLion between
smoking and lung cancer have beer discussed at length elsewhere (-,. 5). With reference to (c). Brownlee commented (7):
"the way it (the 196.4 Report] claims the facts are in conformity -!:.h the criterion is to Flatly ignore the facts". That
comment remains applicable not only to (c). but to all rive criter:a (8). Subjective judgment, on which the Surgeon
General places repeated emphiLsis. should play as limited role as possine in epidemiology as in other sciences. For how do
we distinguish between judgment and prejudice? Scientific analysis a;rns to replace subjective judgment by the objective
tc3tingorhypotheses.
Hypothesis testinr
Given a 'well est.2blished association in case-control and non-raricc-itzied prospective studies lb,cl-een a habit such as
smoking. and a disease such as lung Cancer. we are obliged by the 'Delineated standards or science" to consider all the
following Possibilities:
G) the smoking. or something closely connected with it such us the Means ofignition. causes the cancer:
(ii) a Prc-cancerous condition. such as carcinoma instant. causes the imoking.
(iii) some other factor (the "th3rd f2ctor") causes. or predisposes ic. both the habit of smaking. and the cancer,.
liv) because (i) to (iii) are not mutually exclusive any combination. afthern might be required to account for the -hole
oran observed association.
In practice the situation is further complicated because "smoking" s not a homogeneous habit and "lung cancer- is not
a homogeneous duicasc. But cv%:,, ir _c -C. c able to confine our aLtiintion to a highly specil IC form or smoking and to a
prccisely-dcrinerl type or lung cancer. the implications of liv) arc ;.s6;n0rtg. ir the evaluation of an association is to be
,Comprehensive - and the U.S. Surgcon General (5) makes that c:-,:.,7. - then the relative contributions of (i). Iii) and
(iii), toilethc.- with their rcsvective confidence limits. have to be --ss-zssed (8) So fur as I am aware no one his Yet
succceucd in iiis ij@o@ and. so long as sv,.ii:-naiic errors rscrust in it: -3- dita. the problem is likcl.v to remain unsolved.
BAT Co LTD MINNESOTA TOBACCO LITIGATION

Foru III
Fbitimmes uniccomainit;d Z)y con.-Iwc-ice :imits snould bt tremet-, with extreme cauttun. %c%eflhclcss there irc keys, and
means -hercoy we can ;rv to jr7proach 3 solut:ort to the Causal proolcm i1though. following Bettie. perfcction will almost
certainly elude us.
Very often. our main c-Tort rc,.Iuccs :C c5timvinj; alit: rclolivc contributions a( It) ind (m). The American Stutistical
Assoctation invited Brownlee to re,.,Ie- I..-.C first report of -.he Sur6con Gcneral on Smoking ind he3lth. lie concluded (7)
that the Mst rot causality bc,.wet:n smoi,.nj jnU lung cinccir hud not been cbtabbstictl. But neither did he believe that he
had falsified it: he lick! that it -as not :men posbibic to reach definitive conclusions because. amongst other things. the
genetic hypothesis". a version o(hypoinesis iiii). had not bccn diblirovcCl.
Many pcople rnight feel inclined to dismiss hypothesis (ii) Out ol hind -here lung cancer is concerned but in 1959.
Fisher. who was not without a certain cc-npe-ence in these and related mutters. was not prepared to do so (9). It was no(
until 1962 that evidence bearing directly on this issue was publizilicd. Hypothesis (ii) predicts that the age or onset of lung
cancer will relate directly to the age ofcommencing smoking. P2sscy (10) found that the mean age oronset oflungcancer
is independent both or the daily rate of smoking and of the age 31 which Smoking bc&3n. Hcrrold U 1) confirmed Flasscy'3
findings. we may conclude. therefore, ,h2t hypothesis 60. the converse causal hypothesis. can accounli for little. or no
part. of the association between smoking and lung cancer. It would. nevertheless. be j herculean task to place conriUcncc
limits on this conclusion and specific assumptions would nt:cd lo be adopted jbtiut the age-Paticrin or the pre-canccrous
condition that @hypothctically) rrccipitites the smoking. Conridcrice limits would be modcl-dcpcndent. I should add that
the observations off"asscy and lierrold jiso pl3cc severe restrictions ( 12) on possibility it).
Temporal.renas
Al first sight. once or %:ie -nost atiric:ive - 3%-s o(discrimin3ting bctwc::n G) and Gii) might be to anjiybe ternporni trends in
smoking anct mortality. The inierprv;iuon of trends in a large, clTectivcly clos%;d. population 3voids the complications
connected -file seir-se!evion. (It is pcinLICSS to Compare trends in voluntary ex-smokers with those in voluntary
continuing smoke-s because many stuC-.:s have shu-n that ex-smokc.,s dirrcred on the average from continuing smokers
in v3rious resvects before they gave L;-- the habit. To draw valid inicrences about cause in epidemiology. as in other
disnpimes. we leave to ensure thot t!%t ;rouv exposed to the causal agent shall be as ne2r as possible identical to the
unexposed group. in all other pertinent -MmcLs.) For an entire. for example national. population consisting orsmokers.
cx-smokem and nonsmokers. hypothesis (i) predicts. other thing being equal, that an increase in smoking will be rollo-ed
by a rise in mor.-3fity and 3 decretise wtil Le followed. sooner or later, by a fall in mortality.
In favourabIc circumstances lots kint; of analysis can indeed provide a powerful test. Mormlity ratios (smokers versus
non-smokers) for ischac-nic heart disease dc:vnd on sex and fall markedly with age in both Sexes: these features can be
exploited in studies of the temporal trcnds in Sex- and age-specillic mortality from the disease. Unless some (unknown)
fac:or has consistently opcratect so as lo counter the changes in the putwive elfecLs of smoking - other things are not
always cQual - we can inrcr that little o- no causal action of Smoking is implic2tcd in the pathogenesis of is-chaemic heart
disease ( 13). Systematic errors in death =rtific:ition are perhaps the main barrier to the estimation orcionficicrice HmiLs.
StudiC41ormortality from all =uses. -hich are especially rcievant to estimates in The BigKill. have one great advantage.
they are independent of the diagnosis a( the actual =use of death. With this considerable advantage in mind I compared
the ternpor-al .rends in sex- and age-spec;rjc mormlity (all causes) in England and Wales. with corresponding trends in
cigarette contraption (14). The period covered -a3 1950 to 1976 during which smoking. in terms of "constant tar
cigarette consumption". rose to a peak then re.11. Conditions seemed propitious ror the detection or the presumed raLal
impact of cigarette smoking bUL the v@-:ts expected. an the basis of mortality ratios rrom non-randomized prospective
studies. could not be discerned. The trends in mortality - generally downwards. particularly -hen cigarette consumption
was rising - were obviously dominated b,y cn3nges in factors, other than smoking. Over the age range 20 to 44 yczrs the
cau.se or the rail in over-all mamility %&s easy to idcntiry: it was almost entirely ;accounted for by the near-eradication of
mortality from tu bercuicisis (I 4).
Age-patterns
The curve or agc-specific incidence or -tonality. versus age. is the oge-pattem for the disease. Although the actiological
importance of temporal trends is wci! recognized by epidemiologists very little attention has been given to a related
corollary of hypothesis M: differenc-cs in the consumption per head of a lethal agent with age will affect the age-patiernor
the associated disease. This corollary may be tested if we can assess the age-pattern of the disease in a population that does
not consume the lethal agent. By these .means I have demonstrated that the sex. and Zille-specific death rates for
ciescriliageal cancer in England ind Wales and U.S. Whites (as well as the temporal trends in England and Wales) ate
consistent with the hypothesis that the consurnotion or alcohol helps to cause the discust: ( 15). The demonstration exploits
an unricinular unified theoryortheage-Cependenct: of discuse and will not. thcrcibre, command universal acceptance.
The age-pattern or mortality frnm iscnaemic heart disease in males and females. as well as the temporal trends. rails to
suprort the h%.-,nthcsis that the v3rious "risk f:clors". other than the familial (genetic) luctor. arc causal in origin
silldirs
We might hore to contrc5l for the int"Lcrtce ur inticritancc hy baudying monozygotic twins in @hich one twin has markedly C-N
ditlcrcni srnr.,N!nf; habits from the co---n Such sludiu% -.u ll,:r from t-o limitations. one pricticul. the o(hcr theoretical
BAT Co LTD i@INI@A@TA TOBACCO LITIGATION

Forum
(14). The Practical limituitom is mat ol'srnjJI numbers: manozygotic twins. slightl)- undcf P@ Ofthe Population. tend to be
highly concordant for thcirsmoking habits. The thcorettcal limitation resides in the assumption that -nunozygotic twins are
prictically-it1c.iticai at birth. Indeed. cases or monor)gotic twins discordant for sex. 4:hrumosnml ibnormulitics and
developmental (genetic) zonorrn:htics have been rc-,oricd ( 16- 191. Hence. POSI-LYfiOtIC g--nc %:har.%-. might render one
twin - but not the co-t-in - icrictically predisPOSCd to. sj% . smoking and lung cancer.
The Swedish study or twins by ccacridr er ai. (:) ollrvrs some cviUcnc,. about 4ssuciaijons bet-ccri smoking and
mortality. The largest nurriters ocitainabic from their study are given in their Table 7.10 for the Twat Cohort orSmoking
Discordant Groups -here snioKLng discordjr?cc within pairs is defined by contracts bct-ccri -'Lo-" (including Larci) and
"High" lcve!S of smoking. Combining the data ror males and rernalcS. some 46 deaths *crc recorded among
-Low--smoking monozygotic twins and 53 among their "High"-smoking co-twins (p - 0.26. onc-iail. XI test). In the
series or dizygotic twin pairs similarly discordant for smoking, 98 deaths -ere traced among -Low- smokers and 142
among "High" smokers (p - 0.002). Ignoring the theoretical difficulties described above. and disregarding hypothie3is
(ii), these data favour either the constitutional hypothesis (iii) in isolation. or hypothesis (iii) combined with a small
component of the causal hy.-otl,.esis W.
Randoma-rd trwls
Finaliv. we must consider the scope for the andomi-cd trial. described by Feinstein (21) as the "Sold standard" of
epidernicilogy. It is not feasible to randomize children or adolescents to one group forbidden to smoke and to another
condemned to smoke. Ho-ever. it has proved Possible to randomize middle-aged male smokers to one of two groups: the
interrention group. that is subjec:cd to intensive advice by the ;nvcstigator to desist rrom smoking: and the frormal care
group, that rezz-ves no such special udvicc. Dv these -nein s. the decision to abandon the habit in the mtervcnticin group is
ortcn strongly influenced by the investigator: v large and sustained difference in the average levels of smoking between the
two groups is the objective in randomized controlled intervention trials. The attjinmcni ofa substantial difference shows
that the bias of self-sclevion h2s; been overcome. Such trials represent the closest approach to rigorous experimental
conditions that has been achieved so far in epidemiology. Ncycrthcicss. Sczbc's verdict remains intact. The -perfect
study" requires 311 pertinent (actors other than smoking to be identical. an the average. in the two groups. Any Stress
resulting fror-n giving up smoking or any consequent c.@anges ordict and body weight might. in principle at least. introduce
uncontrolled pertinent difTerences between the two groups. In any case. discrimination between hypotheses G) and (iii)
will depend on the SMISIX21 POWCr Or the trial ever when uncontrolled differences are deemed to be irrelevant. But
because we can never be quite certain in epiderniology that we have controlled. or concerned. for all pertinent factors the
gold ofour "gold standard" is likc!y to have fewer thin 24 carats.
Only one large r2ndomi:Ld trial has beef rccortcd in which intervention was confined to smoking. the Whitehall Study
or London male civil se-variLs (2:2). the Oslo trial addcd dietary intervention (23) and the US Multiple Risk Fuctor
Intervention Trial (24) added treatment or hypertension to smoking and dietary counsel in the intervention group.
The Whitchall Study succeeded in achieving an average rate or smoking, over the 10-year period for the trial. or 7.6
cig&rc-tcs a day lower in the intervention than in the normal =re group. Some 123 deaths (17.2% of entrants) were
recorded in the intervention group and 128 deaths (I 7.51-') or entrariLs) in the normal care group. The trivial difference
amounted to 2% of the mortality and confidence limits were given as -22% to + 23% (22). This result happens to be in
almost perfect agreement with the constitutional hypothesis (hypothesis iii) but numbers are much too small ror a
definitive test- (Although more than twice as many deaths -ere recorded in the Multiple Risk Factor IntervcnLicn Trial -
265 in the intervention group (4.12%) and 260 in the usual care group (4.0411i:i) - the statistical power or this auLcomc.
vin=lly identical water, that of the Whitehall Study. is also inadequate and the complexity of multialc intervention
complicates interpretation).
It must be emphasized that the Whitehall Study (22) provides the only direci epidemiological evidence from which
estimates of smoking-causcd deaths in England and Wales might legitimately be drawn. Nevertheless. the study (14)
mentioned above or secular trends in mortality and smoking in England and Wales also avoided the bias or self-sefection
and railed to detect any causal connexion: although more indirect in nature it is fully consistent with the Whitehall Study.
Justification for the claims in 7-he 3ig Kill. based on mere association (3. 5), is not. forthcoming.
Conclusions
The realization that epidemiology faces peculiar difficulties is far from new. In the First edition of his Principles al'Medical
Slarislics. Published in 1937. Bradford Hill (75) issued a salutory warning about the interpretation of a positive association
-Merely to presume that the relationship is one orcausc and effccl is rittally easy. to secure satisfactory procircir disproor. ir
it be possible 3i all. is often a task of very great complexity." Under the influence or Popper we no longer aspire to "proof-':
in emptric3l science but the more modest objective or corroborating the claim that smoking kills 77, 774 people annually in
England and Wales (ram heart disease. lung cancer and bronchitis/emphysema remains "i task of very great ComPlexitY".
I even feel tempted to discard scientific caution and to suggcsi that this lesser task lies beyond the collective efforts or the
Health Educatinn Council and the British Medical Association. This is evidently not their %;icw Ind -we must presume that
they wcrc not fully irpraiscd of the ditrwultics. the cilculition or the annual toll to the nearest death from percentages
rounded to nuircst 6% c bc:rz@ s a certain innoccrIL-2. I But -as it surmi%cd that 77 774 would retire a more dramatic and
persuasive :rn-,ict thin itout 000" -hich might horit it some doubt and irnhiguity') Althousth he outcome might be
cc
BAT Co LTD MINNESOTA TOBACCO LITIGATION

Forum
$CCII asi Ir:3;urUI[v 1,;r@e Of L:S '.nC blame !IC$ not -me I n :he carnmicrSuf Itc I!,WUMCS. A @u;&Ancss-itrlin
the sc:w1::,,.: ::In! -Iu!ltlv I.,rrjntf,.n-
Fisner. -iro-mcu- anu Ycrusr;.-..fn-. . of' the Nignest have Vrawn &I:Cn:;0n to in ire
CVIJ4:nc-. aml MI: mchods. incl conspicuous hrivoic-s" 19). but warnings -c:-.t Jar;--!y.
uni-ccUvd jnU the PrON92nUa fn3C!!:ftu 6z!n%:J m0mc-'Itum 10 pursue Ri, victorious criuric. Oliver oti%ervcs (26i-
-C.jrdimo,;-i,ts Lrd phy,:oicians t?trouiat)ut :ne world 4re lic-r.1; persuai.lcd by health educationalists. and some
emdcrniuiowtits who hivic also assurnct! :his roic. that the only rez-fly cjTcc%svc way to prevent --oronary heart disease i2. to
endorse 4ru prornufl;@:tc chanScs :n he !ifcs;i.-;;c orthc public as large. Such is the force ol'tha 1-uggemaut set forth by these
pror-agandisis,zhat IiLtlfz circaencc :s ever to ;%.r:uinc scientaricsccnicsm".
By transiorming a rather, ds1M;;;11 sc:cn;i0c problem into a simple ind essentially moral i3m.;c suc.'s as: -Smaxing is the
largest singic cause a( prevvitabic. death in this country". the crusaders have disarmed many a poie.-SLIal CrItiC.
What bright but ambitious young would d2rc to challenge the esiablishmen: I 11 well be. and has becin.
contentled Mat :n mat'C.-3 o[:i(c -nd Cloth scmt relaxation or scicm:fic rigour is justiriablc@ or,-osition to orthodoxy. with
the w3rii ot'sorinistit.:ation. cz-n be rt:rctcnt= is obscur2ritism. Hcz-v:h educationisLs shouid, rc:!-.-.t that the c:caii0n or in
unncc"sjry wnsc of guilt and lc2r in smokc:-S is itself rrossly Privsponsibic.
My oncrung question re-nains: "C--n c.-:CC-iioioj;y bccomt: a --;orous scicnt;c?- IL is :far. I hope. thzt human
variability and the impossibility of exerz:xing legal control over the environment will always ;rcvcni cindempotogy from
@ttaining, the high degree or rigour assoctamt.1 with the experimental sciences. On the other hand. the grower absurdities
that accarnrizny so many provagand:si claims can be and should be avoidcd. Carcrully conducted cwc-control and
non-randommcd studies lead to the iiacn-irt":.-on or "risk factors- And our troubles generally ttcvin at that suilfc. A strong
dOSC Of SC-C-ItIrIC 5Ce:!L3CI3M is letter'. :a cornzat the persu3sivc infI;.;;-nCc of the positive allocation. I met .2 director of a
mediczl research unit (not an e:!eern:clogis;. I hasten to add) who scemcd incapable or uneerstanding that association
does not necessarily impiv cz-usauorz. l:t'e!ltVU3l limitations. coucled with a sincere COMMitrnCrE to the ideals of medicine.
Cart be 3 po-e.-fuf combination. Bu-. Can,:,_-.'zed with a reproduc:blc positive assoc:.2tion we have. as Scientists. no
alternative Blue a suspend to ;rocczd to the objective esting ol'hypotheses (I) to I ivl above.
In spite or limitations the mricornizetj controlled interventiots tnil orfars the most direc- test 2vail2biC. IL probably
jUSLiries Fvnste-n's "gold standard" iccol2da provided we bear in mind the sub-2-2 carat composition or the average
safripic. I-rc outstanding prac,.ical ;rcc;c-n is J-c cost: the Multiple Risk Factor Intervention Triji cost about St 15 000 000
and was riot r-c--riv lar;c cnaugn. Mc.-covcr. nat trial gave the "wrong- result and strenuous clTorts have been made to
rescue scirnc:hing rrnm the wrccka;e. Alt."u;:n Stallones risked thc creation of' many personzi enernies when he -rote
(27).- ccinc!usion is that :hc .1cst cx;ian@:::on for the failure to detect a beneficial vlec: :n %.IRFrr is that no benefit
accrued. No arnount or sqt;;r.-nin; on :.-::: h(30ii, afterS the fact that rcr every IOUO test subiccts 41.2 died and for every 1000
control su!:jects 'O.4 died.- Enthusiasm ror "Micicnify largc. and ,ftcrcrorc very expense .vc. randomized trials is not going
to be unbounded.
Studies or-wins have suffered frar- sr.,.2fl numbers and although they can offer ilo-erful carmbomi,ion, orconstitutional
hypotheses they cannot. for thccre-icc; but %wefl founded reasons. rcec' such hypotheses.
Given good information about tize Age-de:endence or the mortali:v ratio for "a risk factor-. or the Sex- and agc-specific
consumption Ora putative cousalagent such as alcohol. analysis or tte age-paitcrn orthe ss;soc;2ted disease can offerstronf.
corroboration or -he constitutional hvpothasis. (iii). as in reference (13). and of the cousal hypothesis. W. as in reference
(IS). Suc!h demonstrations depend to some t.,ttent on theory and they are unlikely to convince sceptics. p2rticilarly those
UnenCUMDUad by mathematics.
Tcrnpo-31 trends == afre: valuable ducs and. -hen diagnosis is -liable. they can define conort and per-od efTects. Sex.
and 2ge-speciric data for the trends in mcr;alicy and pumfivc causal agents can render plausible both causal and
constitutional hypotheses although it is usually open to objtctors ;a argue that an unknown factor changed in such away as
to cancel the e.,Tcc:s orthe putati%;e =usai agent. When such objections c2n be overcome ternooral trends provide powerful
tests or hypotheses W to fiv). Such studiesare inherently quantitative in character. they after exploit generally available
datz and -hen they do they are very c@-Z:tp. But. other things are net always equal. and the pvt'ectionist will always suffe'
pangs of doubt anikconscience about the unknown factor.
what or my subsidiary question: "How Big is The Big Ul!- The notorious T 7 '774 deaths per year have to be consigned
to the realm of fantasy but I have to confess to being unable to produce a neat alternative. Findings from randomized
intervention trials. the studies of twins. and the temporal trends of mortality in England and Wailes are 211 Consistent with
the null hypothesis orno fatal action ofctgare--c smoking but conlidjence limits are %wide for the first two types of study wW
cannot be rchably estimated for the third. N-:e sound scientific estimate is yet attainable. 'Whereof one cannot speak.
thereon one must be silent."
Finally. do Critical csuy% like the Presents one serve any useful purpose? Past experianct: indicates that they are largely
inefTeLlual znd changt: rcw opinions: but if '.he condition offrecdorn is eternal vigilance thcn the delineated standards of
-itul4uum
sciencusurc!v call for occ3sional reci.
1. licair.-an. N1 A ind Lip--treh. L. 4 1 f Ca"z.- W/ Drath. Ccricnol Rcpsier Olrwc. Studws an Medit;-41 -and
. .... .).bms. %.. :U. L-01
C Y aper 1. Ire Cddlofr
T.in-cn R,% Jlch Counkil II" N . ttl-). ka@currn V
CD
BAT Co LTD - MINNESOTA TOBACCO LITIGATION

Furum
J.Royal College at Physmans q:97 I)5mr*&mwmWJieanft-Vo-. MrnanM6dicaiJLScsena&e. London
4.Doll. R. and Ifill, A.D. (1964, J.-Med. I.. 1. 13".1410 ant! 14.80-146,
5. Surgeon General (198:) US DemrimentorMeshh. Eduemtom and Welfare
6.Feirmcm. A.A. C 19?9) J Ormn. DiL. 31- 15 -41
7.Brownlee. 1,LA. (1%5) J.Am- SJW.,4&sw.. 60. T2!- 7-39
1.Burch. PAJ. (19131 1. Cwem. DU- 36. 8.11-836
9. Fisher. R.A. (1959) Smokim, Me Cancer Comroverry. Some Aftemm 10 ASWIS tire Ewdemw. Oliver lloyc. Eomomegn, A Lonam
10.Pamy. R.D. (I %2) Lsacrt. ' . 107-1 1 2
11.Harrold. K.MCD. (1972) A:..-04 Annual. 7.45-79
12-Bunt. F.Rj. (9978) Y it. s,.,. si,,. 141. 437-A77
13.Burch. P.RJ. (19110) commuese-Re2_24.307.131
14.Burch. P.KJ. C 1911) J. Ckmar. 40#.s.. 34.17-103
Ij. am"ji. P.Rj. C 1914).I. C'hrojv. Dis.. 37. 79J-804
16.K&m. L. Bryant. J.L. Tsaa=. C.. Gible-t. F. and Fialkow. PJ. (1975) 1_41rd. Cram. 11. M71
17.SCUMidt. R.. SOW. LA.. M90-skr. N.M.. Der. H. and Allen. F.H. (I 976) J. Med. G;cm@. 33. 6A42
IL Cgmkusnausky. G. &M llermw. C. ( 197 6) J. MM. Grow.. 13. Al)&-tft
19.Pedersen. I.K.. Pkilia. J.. Sale. V. anu Swum J. (I 980) Arjo Obster. Gyprcaj. sedw.. 51. 459.462
20.Caderldf. R.. Friberg. L and Lundbcfg. T. (19-o7) Area Afed. 3,on.L. S.PP4. 61.
21.Feutstan. A.R. ( 1985) 1. CAmm. Om. 38.177-133
22.Aose- G.. Hamsiton. P-J.S.. Cco-ell. L and Shaplay.M.J. C 1922) 1. Zparnmol. Commum Meath. 36. 102. 1 08
23.Hiemamt. I- Velve9wre. IL. Holme. 1. and Laren. P. (I 94 1) Lancer. -I IJ0J - I 3 I 0
24. M AFMIMukirsle Risk Fm=ar !ntervocrigion Trial (I 982) J. Am. .14K. 1465-1477
25.HiLL A.A. U 937) P@vxmkt W-AfriftemI.Twaxom Is& edition. The Lancet W. London
26-Oliver. M.F. (1986) Cocuianaa. 73. 1-9
27.StaWnet. I-A. (1923) AM. 1. Zpairmpul.. 117. 647-650
BAT Co LTD MINNE@OTA TOBACCO LITIGATION