,K667
    REVIEW 400 CONFIDEITIAL
    (of Tl"4
    Subject ref.8b
    . Smoking, passive smoking and histological
    types in lung cancer in Hong Kong Chinese women"
    T.H. Lam at al
    British Journal of Cancer (1987). 56, 673-678
    0
    A casa-control study is described in which a comparison is made
    of 445 Chinese females with confirmed lung cancer and 445
    individually age and residence neighbourhood controls. Questions
    were asked of each subject on their own smoking habits, chose of
    their spouse and also on demographic and other variables. The main
    findings were:
    (i) That the risk of lung cancer of smokers relative to non-smokers
    was 3.81 overall, the relative risk being highest for squamous
    call carcinoma (8.10) and small call carcinoma (12.00) and
    lowest for adencearcinoma (1.87), though it was statistically
    significant for each cell tv
    ype. The risk increased with amount
    smoked.
    (ii) That the risk of lung cancer of non-smokers married to smokers
    relative to other non-smoker was 1.65 overall. The relative
    risk was only significantly raised for adenocarcinoma, 2.12,
    (Jri
    r__1
    CD
    Ui
    Uli
    Co
    13ATCo document for Legal Services : Health Canada 19 October 1999

    




    the most common form of lung cancer among non-smokers, the next
    2 most common groupings. squamous cell carcinoma and "others
    and unclassified" showing risks about 1. A significant trend
    in relation to amount smoked was noted for all cell types and
    for adenocarcinoma.
    (LIL) That, since 25% of the patients with adenocarcinoma. were
    neither smokers themselves nor passive smokers, the well-known
    high incidence of adenocarcinoma among Hong Kong females is not
    fully accounted for by these factors.
    The results in relation to active smoking are broadly consistent
    with those from a number of studies which have found much stronger
    relationships for squamous cell carcinoma than for adenocarcinoma.
    The results in relation to passive smoking are based on a
    relatively very large set of lung cancer cases among non-smokers. The
    overall relative risk for all cell types of 1.65, with confidence
    limits of 1.16-2.35, are not inconsistent with meta-analyses which
    have shown relative risks of the order of 1.30-1.35. The higher
    increase for adenocarcinoma is similar to the original Hirayama
    results but was not found in a number of other studies. for example,
    che recent study by Pershagen found an effect only for squamous cell
    carcinoma and the Trichopoulos study also found a large increase in a
    study where adenocarcinornas were excluded.
    In considering the -validity of the results from this study,
    particularly in respect of passive smoking, a number of points have
    to be considered:
    N:
    \Z
    BATCo document for Legal Services: Health Canada 19 October 1999

    




    -3-
    Plausibilit-v The results for squamous cell carcinoma seem
    plausible, but the results for adenocarcinoma do noc- Relative to
    non-smokers not married to a smoker, risks of adenocarcinoma are
    increased by a factor of 2.69 --or smokers and 2.12 for passive
    smokers. In other words, the apparent excess risk in relation to
    passive smoking is some 67% of that in relation to active smoking.
    It is dificult to conceive that a non-smoker would be exposed to a
    dose of smoke constituents anything like chat of accive smokers.
    even if attention is restricted to components of smoke in the
    sidestream. To invoke a cause and effect relationship, one would
    have to posit that mainstream smoke had some protective element so
    that non-smokers were more sensLzive to the effects of passive
    smoking than vere smokers.
    Diagnosis of lung cancer This appears to have been carefuly done
    with only confirmed cases included.
    Dose-relationshi,~s The chi-squared tests for trend used include
    results from the non-exposed groups, both in Table III and
    Table V. It is in fact evident from inspection of these tables
    that although, within exposed groups. risk of squamous cell
    carcinoma clearly rises with amount of active smoking, there is no
    real dose relationship at all for adenocarcinoma in relation to
    either active or passive smoking.
    U7
    cc
    BATCo document for Legal Services : Health Canada 19 October 1999

    




    -4-
    Age grandardtsarion The authors have fallen into a standard
    statistical trap in all their analyses of relative r-sk. They
    have assumed falsely that, because the cases and controls are
    individually age-matched, there is no need to rake age into
    account in the analysis. There are two underlying sources of
    error in this approach.
    The first, which is more obvious, relates to subset analyses such
    as Table IV. While, overall, cases and controls have been forced
    to have the same age distribution, there is clearly no reason to
    believe that never smoking cases and never smoking controls
    necessarily have the same age distribution.
    The second relates to the fact that even -where relative risk
    estimates within each age group are identical, an estimate based
    on data combined over age groups . may differ from this identical
    value.
    To illustrate these two poLnts, suppose chat the true distribution
    in a hypothetical survey is as follows:
    0
    BATCo document for Legal Services : Health Canada 19 October 1999

    




    Passive Distrib of ReLative D i s t - ib Dis z: rib
    A 7_e Smoking Smoking Livine Pou Risk of cases o~ controts-
    2 2 il
    + 1 i 1 5.5
    + 3 10 30 16.5
    Total To tal 6 33 33
    + 2 10 20 50
    + 3 10 30 75
    + 1 100 100 25
    + Total Total 6 150 150
    Total 4 22 6
    + 4 31 8
    + 4 130 4
    Total Total Total 12 183 18
    Here we assume that, compared with non-exposed individuals,
    smokers are younger and passive smokers older. Also. that age and
    active smoking. multiply risk by 10 but passive smoking has no
    effect. Multiplying the relative distribution of the I-ving
    population by the relative risk gives the relative distr--Ibut~on of
    the cases. We now choose the controls from the living population
    to have the age distribution of the cases by multiplying ahe
    living population distribution by 33/6 and 150/6 respectIvely. The
    study data are now the right hand two columns. one can see that
    the relative risk in relation to both active smoking and pass~ve
    smoking are correctly estimated as 10 and 1 respeczive~y
    age group, by taking the appropriate cross-producc ratio. -he
    estimate based on the data combined over age group are. howe-:er.
    both biassed. For passive smoking. the cross-product rac-o of
    (3lx6l)/(22x8O.5) - 1.07 and not 1.00 as Ls correct. whi-,e for
    U-I
    0:1
    CD
    Uri
    (__j
    UI;
    N)
    BATCo document for Legal Services : Health Canada 19 October 1999

    




    -6-
    act:;-.e smokinZ the cross -prod-.;ac ratio of (130 x '--'-.5)/k53 x
    41.5) - 8.36 and not 10.00 as is correct.
    Treatment of never married never s-mokers It: is noted zhazi analyses
    include never married never smokers in the non-exposad never
    smokers groups. This seems inadvisable, partly as the age
    distribution of never married and ever married people is very
    different and partly as there will then be an inevitable
    confounding between effects of factors related to marital status
    and factors related to passive smoking in the analysis. However,
    since the authors note that excluding never married never smokers
    from the analysis did not affect the answers much, this may not be
    so important.
    Confounding variables Hatching for age and residence made the
    cases and controls similar in respect of demographic variables
    such as place of birth, durat-on of stay in Hong Kong, level of
    education, marital status and husband's occupat.on, but,
    theoretically, they may still have caused bias particularly in
    comparison of never smoking cases and never smoking controls.
    There are of course other possible confounders that were not
    studied at all.
    Hisclassification of active smoking status The authors considered
    this possibility. They claimed to have to some extent eliminated
    it by asking an aditional question on lifetime smoking after the
    original questions based on the MRC questionnaire. Because of
    CC
    BATCo document for Legal Services : Health Canada 19 October 1999

    




    'very few posizive replies to this additional question" they -.:e:e
    sac sf,-ed thaz under- reporting of the smoking habit ~as -.o:: a
    maj o : prob I er. " This seems dubious as subjecrs could easil,.r
    denied s-moking consistently. Biochemical validation would ha-.e
    been an advantage.
    Hisclassilfication of husband's smoking No attempt was made zo
    cross-check this with statements made by the husband or to use
    biochemical validation. Recall bias is an obvious important
    possibility here, partly because cancer cases may be more keen
    than random individuals to try to find a reason for their disease,
    and partly because, according to Linda Koo (personal
    communication), many of Lam's controls were gathered in "on zhe
    streec" interviews. As she says,
    "If I were a 60 year old woman, and walking down the
    street, and approached by an inter-viewer who wanced to
    ask ma about various exposures in the past and I had
    sm-oked for a few years 30 years ago, even if i did
    recall it. I would just say that I was a never-smoker
    and be finished with the interview, Wouldn't you? The
    same is true if I received a telephone call after a
    hard da-, at work, or receiving a mailed 1-15 page
    questionnaire.."
    This 'lack of comoacibilicy of circumstances of interview of cases
    and controls =av be a particularly important source of bias, and
    is my =ajor worry about: this study.
    0
    F.N.Lee
    23.3.88
    BATCo document for Legal Services : Health Canada 19 October 1999