ENVIRONMENTAL TOBACCO
SMOKE AND LUNG CANCER
by
Kenneth G. Brown
Douglas J. Crawford-Brown
with
Charles G. Humble
Chu-Chih Chen
Kenneth G. Brown, Ph.D.
310 Carl Drive
Chapel Hill, NC 27516
(919) 933-0789
• Research Statistics • Health Effects Assessment •
-------
ENVIRONMENTAL TOBACCO
SMOKE AND LUNG CANCER
by
Kenneth G. Brown
Douglas J. Crawford-Brown
with
Charles G. Humble
Chu-Chih Chen
Prepared for U.S. Environmental Protection Agency under Contract No. 68-08-0115 to
Battelle Memorial Institute. The views expressed in this report are those of the
authors. No endorsement by the Agency should be inferred.
-------
J -
TABLE OF CONTENTS
1. INTRODUCTION AND SUMMARY...... ""'''''' .......... ........................................... 1-1
2. EPIDEMIOLOGIC EVIDENCE OF LUNG CANCER FROM ETS........................ 2-1
3. DOSIMETRY OF ETS .. .......... """"""""'''''''' [[[ ~.. 3-1
4. ANNUAL EXCESS LUNG CANCER DEATHS ASSOCIATED WITH
EXPOSURE TO ETS ....... ...... """" .............. ....... .........,. ........................................4-1
5. INTERACTION OF ETS WITH RADON PROGENY.......................................... 5-1
APPENDIX: SUMMARY DESCRIPTIONS OF ELEVEN
-------
1. INTRODUCTION AND SUMMARY
In early 1981, the results of a case control study undertaken in Greece to explore
the relationship between lung cancer and passive smoking reported a significant increase in
lung cancer among nonsmoking women maxried to smokers (Trichopoulos, 1983). Almost
simultaneously, similax results were reported from a methodologically different cohort
study in Japan (Hirayama, 1981). 'Within the yeax, results from the cohort study of the
American Cancer Society appeared (Garfinkel, 1981). Evidence from the American study
was equivocal, but not incompatible with conclusions of the previous two studies. These
events in 1981 were probably the genesis of what a recent expert panel convened by the
International Agency for Reseaxch on Cancer (IARC) has described as the most actively
investigated segment of the whole smoking versus health domain of research during the
eighties - environmental tobacco smoke and lung cancer (Saxracci, 1989).
Three factors that contribute to the relevance of exposure to environmental tobacco
smoke (ETS) for public health include: (1) the ubiquity of ETS, as evidenced by
experimental measurements of airborne concentrations in public places and the presence of
tobacco-specific biological maxkers, such as cotinine, in nonsmokers; (2) the involuntary
aspect of exposure to ETS, in contrast to active smoking; and (3) the biological
plausibility of lung cancer risk from ETS, based on the identification of caxcinogens in
ETS and the lack of convincing evidence of a threshold tolerance level.
Additional epidemiologic studies in the eaxly 1980s continued to fuel concern about
potential health risks from ETS, with lung cancer risk probably receiving the most
attention, while discrepant conclusions and controvertible methods fueled controversy. In
the mid-1980s, the U.S. Surgeon General, the National Reseaxch Council, and IARC all
1 - 1
-------
convened scientists to consider the evidence of health risks associated with exposure to
ETS. As discussed by Samet (1988), all three groups concluded that ETS exposure is
associated with an increased risk of lung cancer, although they used somewhat different
approaches. The CS. Surgeon General (S.G., 1986) concluded that passive smoking is a
cause of lung cancer, based on (1) the evidence that active smoking is. a risk for lung
cancer, (2) the qualitative similarities between ETS and mainstream smoke, and (3) the
epidemiologic evidence. The National Research Council (NRC, 1986) considered the
biological plausibility of lung cancer from ETS and emphasized the epidemiologic
evidence. After an overall analysis of the data for ten case control and three cohort
studies, and carefully considering potential sources of bias (bias due to misclassification, in
particular), it was concluded that there is a positive association between ETS and lung
cancer. The International Agency for Research on Cancer (IARC, 1986) reviewed the
evidence available through the end of 1984 and emphasized issues related to the
physicochemical properties of ETS, the toxicological basis for lung cancer, and methods of
assessing and monitoring exposure to ETS. The report quotes the summary statement on
passive smoking of a previous IARC working group that found the epidemiological
evidence available at that time (1985) compatible with either the presence or absence of a
lung cancer risk. Based on other considerations related to biological plausibility, however,
it was concluded that passive smoking gives rise to some risk of cancer.
At the present time there are 21 case control studies suitable for analysis of a lung
cancer risk associated with ETS exposure. No additional cohort studies have appeared.
The more recent case control studies are described in the appendix of this report. In
Chapter 2 the weight of observational evidence of a lung cancer risk is statistically
assessed using the
large number of studies on ETS currently available. The data are
analyzed from several perspectives, with consist,ent results.
1 - 2
-------
Conclusion 1:
Based on analysis of epidemiological data, the occurrence of lung cancer
deaths in females classified as never smokers is positively associated with spousal smoking.
The weight of statistical evidence virtually precludes the possibility of occurrence by
chance.
Once a hazard has been identified, it is of interest to have some idea of the
magnitude of risk to the population. A tractable measure is the number of excess lung
cancer deaths (LCDs) associated with ETS exposure, often called the "population
attributable risk". The population in this case consists of U. S. women who are never
smokers of age 35 or above. Estimates in the literature can be classified under two
approaches: "cigarette equivalent", where ETS exposure is considered equivalent in risk to
some specified level of light active smoking, and "epi-data", \vhere the data from
epidemiologic data are utilized but no equivalence is made with active smoking. Several
authors have found the cigarette approach contentious. A biokinetic model has been
developed to shed some light on the biologically-based similarities and differences between
passive and active smoking that may affect lung cancer risk. That topic is reviewed next,
followed by estimation of excess LCDs.
Passive and active smoking differ in a number of ways that may affect lung cancer
risk. ETS and mainstream smoke (MSS) differ in chemical constituency, with a wide
range in the relative concentrations of possible carcinogens. Exposure to ETS in passive
smoking and exposure to MSS in active smoking differ profoundly in terms of
concentration and temporal pattern of exposure. There is a wide range of variability in
exposure to ETS in nonsmokers alone, with the added complexity of an aging effect on
characteristics of ETS. For example as ETS ages, the proportion of tar in the vapor phase
increases to about 70% (Pritchard et aI., 1988) which may impact carcinogenic risk. This
contrasts with active smoking where all of the tar is in the particulate phase.
1 - 3
-------
The distribution of a chemical between the vapor and particulate phases affects
integral lung burden from both passive and active exposure (i.e., ETS exposure to
nonsmokers and MSS exposure to active smokers). In particular, the lung uptake and
deposition by lung region of a chemical in the particulate phase is largely influenced by
the distribution of the diameter and density of particulates to which the chemical attachs.
Chemical-specific characteristics are major determinants of dosimetry parameters for a
chemical in the vapor phase. ETS and MSS are complex mixtures of chemicals, with a
large number of identified (or suspected) carcinogens in common. The relative presence of
the chemicals in ETS and MSS (measured as a percentage of total mass) varies widely
across chemicals. Furthermore, the distribution of a chemical between the vapor and
particulate phases may differ for ETS and MSS.
The quantitative model derived in Chapter 3 relates exposure to any chemical
constituent of tobacco smoke (ETS or MSS) to integral lung burden and translocation to
systemic organs. It identifies parameters and their inter-relationships that distinguish
between active and passive smoking, the vapor and particulate phases, and major regions
of the lung (naseopharyngeal, tracheobronchial, and pulmonary). A number of possible
exposure measures to a given chemical in ETS are suggested that require increasing levels
of knowledge regarding biokinetic parameters. Calculations are demonstrated for nicotine
and cotinine.
While the precise meamng of "dose", and particularly "biologically significant
dose", has not been established in the literature, most theories rely on some measure of .
organ burden or integral organ burden as a measure of the rate or probability of transition
in a multistage process. This implies that the burden of a chemical in each cellular
subpopulation connected causally to lung cancer must be specified in order to make
accurate biologically-based predictions of risk. This is a severe requirement, compounded
by incomplete knowledge of which chemicals in ETS are the major contributors to lung
1 - 4
-------
cancer risk. This complexity makes exploration of potential dose surrogates attractive.
Two measures that have been suggested as dose surrogates for risk assessment are cotinine
and respirable suspended particulates (RSP). Lung burdens for RSP and nicotine/ cotinine
can be approximated. There is no support, however, for the assumption that burdens from
other chemicals in ETS (for which biokinetic parameters are not available) will scale
according to relative intakes.
Conclusion 2: Lung burdens cannot be calculated for most chemicals in ETS, even the
most significant ones. The major lack of data concerns the solubilization of chemicals from
respirable suspended particulates into lung tissue and the uptake of vapor phase
components. While limited data on clearance of RSP and nicotine from the lung are
available, retention functions for other chemicals are unavailable. As a result, it is not
possible to develop a surrogate measure of dose for chemicals existing simultaneously in
the vapor and particulate phases.
Conclusion 3: Even if lung burdens of all chemicals could be estimated, their individual
effects may be delivered to differing cellular subpopulations. Since the doses to cellular
subpopulations can differ dramatically within the lung, development of biologically
meaningful surrogates of dose must be tied to identification of the causal roles of the
various chemicals. This link is not available at present.
Conclusion 4: Any measure that is directly proportional to lung burden from exposure to
ETS will serve as an adequate surrogate, provided it is proportional to burden in all
important sub populations of cells, and the proportionality ratios are invariant of ETS
exposure concentrations, i.e., the proportionality ratios remain constant independent of
variations in ETS exposure.
1 - 5
-------
In Chapter 4. upper and lower confidence bounds (nominally 92%) are obtained for
the proportion of lung cancer deaths (LCDs) in 1988 associated with ETS exposure. The
population considered consists of U. S. female never smokers at least 35 years of age. The
population of women is used because the epidemiologic evidence is predominantly for
female never smokers and the data on males are relatively sparse. The confidence bounds~
based on the case control studies currently available, are applied to the 6,500 LCDs in
1988 among U. S. women classified as never smokers (estimate from the American Cancer
Society). The resultant confidence interval provides a range of values on the number of
LCDs associated with ETS exposure, i. e., the population attributable risk, that is
consistent with the epidemiologic evidence.
Estimates from the published literature are compared relative to the confidence
interval. Five estimates are based on the cigarette equivalent approach. Two of the five
equate risk from passive smoking to light active smoking based on relative cotinine levels
found in active and passive smokers; one determines an equivalent level of light active
smoking using RSP; the remaining two (by the same author) are hybrid procedures in
which the results are expressed as a range of values depending on unknown parameter
values. An additional five sources rely only on epidemiologic data (the epi-data approach).
One of the five estimates is based on a set of data independent of the case control studies
on ETS. The remaining four authors rely on case control studies for ETS, but differ in the
studies utilized and in various aspects of their analyses.
. ~ ='
~LL. -or=- ./( ~ '.,
Conclusion 5: The excess number of lung cancer deaths associated with ETS exposure in
U. S. female never smokers in 1988 is very likely between 850 and 2,400. This range is
based on an approximate 92% confidence interval derived from the results of sixteen case
control studies of ETS exposure, including two studies with a negative estimate of
attributable risk (below the parameter space [0,1]). The range has been increased by 20%
1 - 6
-------
'.
(based on the NRC report) to adjust for an estimated net effect from sources of bias.
Conclusion 6: Published estimates of the number of excess LCDs associated with ETS~
standardized to a total of 6,500 LCDs~ are largely consistent with the range 850 - 2,400.
Four of the five sources using the epi-data approach are in the interval 1,300 - 1,900. The
fifth is slightly below the range (800). Of the five estimates following the cigarette
equivalent approach, the two based on cotinine concentrations are 1,800 and an interval
estimate of 1,600 - 2,150, well within the confidence bounds and consistent with four of the
five epi-data estimates. Results of the two hybrid methods are presented as plausible
intervals, 1,650 - 3,000, and 550 - 3,000. The fifth estimate using the cigarette equivalent
approach is based on exposure to RSP and is singularly small (5). It is the only aberrant
result relative to the confidence interval.
Conclusion 7: The number of excess LCDs associated with ETS exposure in U. S. female
never smokers in 1988 is conservatively about 1,500. The total for the whole U. S.
population, including male never smokers and ex-smokers of both sexes, is difficult to
assess. An additional increment of several hundred to the 1,500 is probably minimal. An
increase on that order is consistent with the limited epidemiologic evidence for males
alone. Unfortunately, data sources and biological knowledge regarding ETS exposure m
former smokers are both insufficient, even for speculation.
The NRC report suggests that ETS may modify the risk of lung cancer from radon
progeny in the home. Chapter 5 identifies eight possible risk-modifying effects and
addresses the potential influence of each. The overall impact of ETS on the risk from
radon progeny is quantified for an example home of specified characteristics (volume, air
exchange rate, average smoking rate in the home).
Conclusion 8: ETS changes both the equilibrium fraction and the unattached fraction of
1 - 7
-------
radon progeny, may change both the rate of mucous transport and mucous thickness, may
cause the breathing pattern to shift to more frequent shallow breaths, and may interact
synergistically with the radon progeny, increasing their effect.
Under the standard assumption that the unattached fraction of radon
Conclusion 9.
progeny contribute significantly to the lung dose, the net effect of ETS is a reduction in
dose-rate from radon progeny to about three-fourths of the initial (smoke-free) value. The
assumption, however, has been brought into question by very recent work suggesting that
the unattached progeny may be mostly removed by the naseo-pharyngeal region during
normal breathing (Swift et al., 1989). In that case, ETS would produce a small increase i~
dose-rate, instead of a net reduction.
Conclusion 10. Further research is needed to clarify the role of the unattached fraction in
the production of the lung dose from airborne radon progeny in the home, since this factor
determines whether ETS has a net effect of raising or lowering the risk from radon
progeny. Additional research is also needed to define the initial particle concentrations in
homes where ETS is likely to be produced, since the initial concentration is an important
determinant of the influence of ETS on lung dose.
1 - 8
-------
REFERENCES FOR CHAPTER 1
Garfinkel. L. (1981) Time trends in lung cancer mortality among nonsmokers and a note
on passive smoking. J Natl Cancer Inst 66:1061-1066.
Hirayama, T. (1981) Non-smoking wives of heavy smokers have a higher risk of lung
cancer: a study from Japan. Br Med J 282: 18~-185.
International Agency for Research on Cancer (IARC). (1987) Environmental carcinogens
methods of analysis and exposure measirement. Volume 9: passive smoking. (O'Neill, I.
K., Brunnemann, K. D., Dodet, B. and Hoffmann, D., eds.) Lyon: IARC Scientific
Publications No. 81.
National Research Council (NRC) (1986). Environmental tobacco smoke.
DC: National Academy Press.
'vVashington,
Pritchard, J.N.; Black, A.; McAughey, J.S. (1988) Physical behavior of sidestream smoke
under ambient conditions. Envir. Technol. Lett. 9(6): 545-552.
Samet, J.M. (1988) Environmental tobacco smoke and cancer. Report prepared for the
U.S. Environmental Protection Agency.
Saracci, R. (1989) Passive smoking and cancer risk. IRAC report of panel of experts.
Prepared at the request of the European School of Oncology through the Europe Against
Cancer program of the European Economic Community.
Swift, D.L. (1989) Unpublished data, presented at University of North Carolina.
Trichopoulos, D.; Kalandidi, A.; Sparros, L. (1983) Lung cancer and passive smoking: con-
elusion of Greek study. (Letter) Lancet 667-668. .
U.S. Surgeon General (SG) (1986) The health consequences of involuntary smoking. U.S.
Department of Health and Human Services, Public Health Service.
-------
2. EPIDEMIOLOGIC EVIDENCE OF LUNG CANCER
FROM ENVIRONMENTAL TOBACCO SMOKE
2.1. CASE CONTROL STUDIES: BACKGROUND
A list of case control studies for evaluating whether there is an elevated lung
cancer risk due to exposure to ETS, primarily in the home, is shown in Table 2-1. The
studies are denoted by the first few letters in the name of the first author for easy
reference. Additional descriptive characteristics of the studies are given in Tables 2-2,
2-3, and 2-4. The three cohort studies available (Hirayama, 1984; Garfinkel, 1981; and
Gillis, 1984) are discussed separately in Sections 2.6 and 2.7.
The report of the National Research Council (NRC, 1986) reviews and analyzes
ten of the studies shown in Table 1-1: AKIB, BUFF, CHAN, CORR, GARF, KABA,
KOO, LEE, PERS, and TRIC. The study designated as WU in the table is excluded by
the NRC because the raw data are not presented in the reference, a requirement of the
statistical method used by the NRC to combine results across all studies (meta-
analysis). The NRC excludes an earlier version of the KOO study, and the studies by
Knoth et al. (1983), Miller (1984), and Sandler el al. (1985) for assorted reasons (NRC,
1986, p. 227). Aside from WU, these same studies are also omitted from this report.
Also not included here is a small set of data on males in New Jersey that was pooled
with the data from BUFF and CORR for analysis by Dalager et al. (1986). The data
for the New Jersey males are not available separately, and the other two studies are
included in this report.
The NRC, after carefully assessing the epidemiologic evidence from the three
cohort studies and the ten case control studies listed above and then adjusting for
2 - 1
-------
potential bias from misclassification, concludes that "The weight of evidence derived
from epidemiologic studies shows an association between ETS exposure of nonsmokers
and lung cancer that, taken as a whole, is unlikely to be due to chance or systematic
bias." The Surgeon General's report on passive smoking (SG, 1986), based on the same
epidemiologic studies except with WU included and BUFF omitted, concludes that
"The epidemiologic evidence that involuntary smoking can significantly increase the risk
of lung cancer in nonsmokers is compelling when considered as an examination of low-
dose exposure to a known carcinogen (i.e., tobacco smoke)."
The report of the
International Agency for Research on Cancer (IARC, 1987) includes the three cohort
studies above and nine case control studies in its assessment of passive smoking. The
case control studies differ from the list for NRC by inclusion of Knoth et al. (1983) and
WU, and omission of BUFF, PERS, LEE.
The IARC report quotes the summary
conclusion on passive smoking of an IARC working group, which met in February, 1985:
"The observations on nonsmokers that have been made so far are compatible with
either an increased risk from 'passive' smoking or an absence of risk". The reason that
fewer case control studies are included in the work of IARC appears to be attributable
to study availability. The IARC committee began meeting in 1984, even though the
report did not appear in the IARC scientific publications until 1987.
Of the three major reports (NRC, SG, IARC), the NRC places the greatest
emphasis on technical evaluation of the epidemiologic studies. In particular, it a.nalyzes
the raw data from each study and then combines the evidence across studies to estimate
an overall relative risk with a confidence interval. A recent article by Wells (1988b)
contains a similar analysis that includes several studies subsequent to the NRC report
(BROW, HUMB, and LAMT). Wells' explicit criteria for studies to be a.nalyzed
excludes CHAN because it uses only current status of spouse smoking, but includes WU
and Sandler et al. (1985), the latter of which has only a small number of lung cancer
2 - 2
-------
cases. Wells' estimate of an average relative risk for lung cancer in nonsmoking females
exposed to ETS is a little higher than the value obtained by the NRC, but both give
95% confidence intervals strictly above the value one. The location of the confidence
interval in both references implies that the observed association between lung cancer
incidence and exposure to ETS is statistically significant (P< 0.05 for a two-tailed test;
P<0.025 for a test against the alternative of only a positive association).
The NRC
approach of estimating an overall relative risk from raw data in the studies, which
appears in Blot and Fraumeni (1986) and Wald (1986) in addition to Wells, is extended
in the next section to case control studies currently available (Table 2-1).
Data on
nonsmoking males married to a smoker are relatively sparse, so this report focuses on
data for females where an association of passive smoking with lung cancer should be
more detectable.
The three cohort studies are treated separately here instead of
including them in the overall analysis due to characteristic differences between case
control and cohort studies.
Primary references for the 21 case control studies analyzed in this report are in
Table 2-1. The Surgeon General's report (SG, 1986) contains particularly good summary
reviews of the studies available at that time. The studies are selectively described or
compared in several additional sources, e.g., NRC, 1986; IARC, 1987; Balter et al.,
1986; Blot and Fraumeni, 1986; Correa, 1986; Eriksen et aI., 1988; Kuller et aI., 1986;
Repace and Lowrey, 1985; Riboli, 1987; Samet, 1988a, b; Saracci and Riboli, 1989;
Weiss, 1986; Wells, 1988b; 'Oberla, 1987; Varela, 1987. Descriptions of the studies
subsequent to the NRC report, two of which are unpublished dissertations (LAMW and
V ARE), are given in the Appendix. In addition to LAMW and V ARE, the studies sum-
marized there include BROW, GAO, GENG, HUMB, INOU, LAMT, SHIM, SVEN and
WU.
2 - 3
-------
2.2. THE NRC APPROACH TO AN OVERALL ESTIMATE OF RISK AND
CONFIDENCE INTERVAL EXTENDED TO THE 21 CASE CONTROL
STUDIES OF TABLE 2-1
The NRC (1986) and Wald (1986) use a common approach to estimate an overall
relative risk across all studies using the method described by Yusuf et al. (1985). Blot
and Fraumeni (1986) and Wells (1988b) achieve the same objective using the Mantel-
Haenszel method (Mantel and Haenszel, 1959; Mantel, 1963), a standard approach for
the combination of information from 2-by-2 contingency tables. As noted in Yusuf et al.
(1985), however, the method described there is ba.sically equivalent to the procedure of
Mantel-Haenszel (M-H). The M-H method will be applied to the raw data for females
available in the current ca.se control studies (Table 2-5). The available data for males is
also shown in 'rable 2-5.
Studies have primarily been concerned with exposure to
women in the home since nonsmoking males with wives who smoke are relatively
uncommon. Additionally, a much smaller percentage of men than women stay at home
during the day in some of the countries where the studies were conducted.
As in the NRC report, the raw data for the 2-by-2 classification (exposed or
unexposed)-by-( case or control) will be used for each study. Several differences in the
studies, however, need to be noted first. The meaning of "unexposed" differs between
studiesua zero-exposure group is not always used for comparison with exposed groups.
The unit of measure of exposure, e.g., cig./day or total years of exposure, varies across
studies.
Also, exposure environments may differ, e.g., smoking spouse at home, all
smokers at home, exposure outside the home (such as at work). A few studies count
former .smokers as nonsmokers if they have abstained from tobacco usage for some
specified time period, citing the commonly held view that the risk of lung cancer for an
ex-smoker is approximately the same as for a never smoker of equivalent age within a
few years after smoking is terminated. That belief has been brought into question
2 - 4
-------
recently, however, in a reanalysis of the British doctors' data on smoking and lung
cancer (Moolgavkar et al., 1989). It is readily apparent from Tables 2-5 that the case
control studies are heterogeneous with respect to numerous characteristics aside from a
variety of nationalities and cultures. (Statistical heterogeneity as in the sense of Bres-
low and Day (1980) is not intended here.)
Heterogeneity across studies is reflected in the proportion of the control group
that is exposed to ETS. A plot of the proportions is shown in Figure 2- 1, which covers
a range of about 43% to 84%, a two-fold difference, with the study BROW a low excep-
tion at,15%. The differences in proportions are more than can be accounted for by statis-
tical variation, which would be the case if the referent populations (sampled control
populations) were identical with respect to the meaning of "unexposed". The referent
populations are not solely determined by the source of controls and their characteristics,
such as being hospitalized or not, and whether subjects are alive or dead. Other charac-
teristics playa role as well: study design, protocol, analysis, and interpretation of data;
the definition of "exposure"j inclusion/exclusion of former smokersj source of infor-
mationj potential confounding factors included in the matching and/or data analysis;
and degree of confirmation of primary lung cancer in cases. Study differences in the
observational meaning of "exposed" do not invalidate testing the hypothesis that expo-
sure to ETS does not increase the risk of lung cancer, however, either for testing in a
single study or for combining the test results across studies. This is because a test is
based on the relative risk for each study, which is a within-study comparison between
cases and controls. The overall (combined) test is valid because under the hypothesis
tested the (true) relative risk equals one in each study, even if the referent populations
differ. The statistical power to detect a small but meaningful increase in lung cancer
risk from a single case control study is often fairly small, while the total power from
assimilating test results may be considerably enhanced.
2 - 5
-------
For estimating an overall relative risk, however, the study differences create
some difficulty of interpretation. Judging from Figure 2-1, the true values of relative
risk are not constant across studies.
Qualitative differences are suggested by the
variable sources of controls and the discrepant meanings of "unexposed" evident i~
Tables 2-1 to 2-4. What, then, is the interpretation of the parameter being estimated
by the relative risk? Does it apply to a population of interest? It can be said with
assurance that the parameter equals one if there is no risk of lung cancer associated with
ETS exposure, is greater than one if there is a positive association of lung cancer risk
with ETS, and is less than one if there is a negative association. The parameter will be
referred to as the overall relative risk parameter (ORRP), and its estimate will continue
to be called the overall relative risk estimate. Technically, it may be noted that the
value of ORRP depends on the set of studies considered, sa its (true) value may not be
quite the same for this report and others. An estimate of the ORRP and an associated
confidence interval are useful for hazard identification, i.e., in deciding if there is a lung
cancer risk associated with ETS and in evaluating the strength of the evidence. It is
not clear, however, to what extent an estimate of the ORRP may be useful for quanti-
fying the degree of hazard (risk) for any real population" of interest. This latter point is
discussed further in Section 4.
The M-H estimate of relative risk (odds ratio) and its associated confidence inter-
val are given in Table 2-5 for females of each case control study where the raw data are
available. Our calculated values are shown in boldface type on the first line for each
entry. The study author's estimate of the relative risk, most often the odds ratio, and
the reported confidence interval are on the following line for comparison, as available.
The M-H estimate of the overall relative risk parameter is 1.40 with 95% confidence
interval (1.22,1.61), based on our calculated values for the 18 studies in Table 2-5 for
which the raw data are available. The corresponding values in the NRC report for
2 - 6
-------
females of ten case control studies and three cohort studies are 1.32 and (1.16; 1.51).
(For reference, it is noted here that the NRC gives values of 1.62 and (0.99, 2.64) for
males.)
Wells (1988b) obtained an overall relative risk estimate of 1.50 with 95%
confidence interval (1.3,1.8) for females from 14 case control studies, with some
differences in the choice of studies compared to NRC and this report, as previously
noted. The higher estimate may be attributable, at least in part, to Wells' exclusion of
the CHAN study due to its use of only current smoking habits as a measure of exposure.
The CHAN study was the first and probably most simplistic of the four Hong Kong
studies. It produced the lowest estimated relative risk among the studies in Table 2-5,
namely 0.75.
Aside from the difficulty of interpreting the overall relative risk estimate for
inference on populations of interest, the method of calculating the confidence interval is
contentious when there is significant between-study variability. The confidence interval
produced by the M-H (or Yusuf) method of combining evidence across studies takes into
account within-study variability, but not between-study variability. The "within"
variability determines the confidence interval for a study by itself, while the "between"
variability refers to variation of the relative risk across studies. Paul Meier (19$7) has
noted the potential for error if between-study variability is non-negligible but not taken
into account. Meier's comments were made in the context of applying Yusuf's method
to meta-analysis of clinical trials, but they are also relevant to the problem considered
here.
A statistic referred to as "S" is included in Tables 2-5 and 2-6. It is the square-
root of the M-H chi-squared statistic with the sign (+ or -) of (R -1), the odds ratio
minus 1, attached. Equivalently, S is the estimated log-odds (In(R)) divided by its
estimated standard error. The estimator S is approximately normally distributed under
the null hypothesis that the true relative risk for a study equals one (Woolf, 1955). The
2 - 7
-------
values of Ps in Table 2-5 and Table 2-6 are the (one-tailed) significance levels of S for
testing the null hypothesis against the alternative of elevated risk from ETS. Since S is
standardized to account for sample size under the null hypothesis, its values are
comparable between studies as evidence against the null hypothesis. The wide range of
S values in Figure 2-2 suggests that between-study variability is probably too large to
ignore, which casts some -doubt on the actual size of the nominal 95% confidence
intervals for the ORRP described above. This would need to be taken into consideration
in any inference based on an overall confidence interval.
Confidence intervals have an equivalence in hypothesis testing. For example, an
hypothesis that the ORRP equals a certain value, e.g., one, would be rejected with
(two-sided) P
-------
The two tests just applied are both statistically significant, but only the raw data
have been examined.
Several studies provide an adjusted statistical analysis, where
"adjusted" refers to inclusion of potential confounding variables among the possible
explanatory variables (along with exposure to ETS). An adjusted analysis is generally
preferable to an analysis of the odds ratio from the raw data, even when the data are
matched (Schlesselman, 1982, p. 190).
A second reason for examining the adjusted analyses is the large number of
studies that appear not to be matched for passive smoking. Although most studies were
originally matched on several variables, as shown in Table 2-1, many of the studies
included issues in addition to ETS exposure and thus may be unmatched over the
sample data for passive smoking. For example, several studies included active smokers
which were then removed for analysis of passive smoking. Also, a study may not be
matched due to the addition of supplementary data at a later time to increase the
sample size or to collect different information, or just may be poorly matched due to
study limitations or subject availability. A study originally designed for a purpose other
than passive smoking, or designed for objectives in addition to passive smoking, could
still be matched for analysis of lung cancer risk from exposure to ETS if
smoking/nonsmoking is one of the variables used for matching, but no mention of that
was found for any of the 18 studies considered here. Unfortunately, most studies do not
state explicitly whether the data used to assess exposure to ETS alone are matched. If it
appeared that a study may be unmatched for ETS, an entry of "No" is in Table 2-1.
Using that rule, most of the studies are listed as unmatched for ETS, although the
number could be overstated somewhat due to the general lack of specific information in
several cases.
A combined test for an effect across studies with an adjusted statistical analysis
is the topic of the next section.
2 - 9
\
'<
-------
2.3. META-ANALYSIS OF CASE CONTROL STUDIES THAT INCLUDE
AN ADJUSTED STATISTICAL ANALYSIS
Table 2-1 identifies the studies in which an adjusted method of statistical
analysis was used, generally logistic regression when reported. In most of the studies the
relative risk and confidence interval are given for two or more levels of exposure, e.g., 1-
20, 21-40, or 41+ cig./day smoked by the spouse, corresponding to 1 or less, 2 or less,
and more than 2 packs/day. Outcomes of adjusted analyses are displayed in Table 2-6.
The values of the relative risk, R, and the confidence intervals, are from the individual
study reports except that reported 90% confidence intervals have been converted to 95%
intervals. The table includes a comparison of unexposed with exposed when it is
available in the study report. Some reports give results by exposure level, in which case
thehighestexposure level is used here since the power to detect an association of ETS
and lung cancer should be highest there (other factors such as sample size aside). In
some cases a report gives an estimated risk differential for some level of exposure, such
as 20 cig./day smoked by the spouse, which is used in Table 2-6. The results recorded
from adjusted statistical analyses are not directly comparable in most cases because of
differences in what authors have reported. It is still valid, however, to test the
hypothesis of interest, i. e., that lung cancer and ETS exposure are not associated, using
all studies. The value of the statistic S in Table 2-6 was calculated for a study either
from information on the estimated logistic regression function or from the confidence
interval, in the latter case back-calculating to find the value of S that would produce
the same confidence interval provided. Thus a value of S is being used that is
consistent with an author's results in some sense, irregardless of how they were
obtained. This approach provides some common ground for comparison.
In some studies where an extreme exposure level was used, there was more than
one choice of what constitutes an extreme exposure case. The choices were generally due
2-10
-------
to more than one exposure unit, e.g., cig.jday, total years, etc. (see Table 2-4), or
because more than one source of exposure to ET5 is evaluated in a report, e.g., solely
from spouse smoking in the home, from smoking by the spouse and others in the home,
from one or more environments outside the home, etc., as indicated in Table 2-3. As a
rule-of-thumb, preference was given exposure measured as cig.Jday in the home,
generally due to spouse smoking alone. In none of the cases was the reported statistical
outcome, for example the value of R, considered in the construction of Table 2-6.
The values of 5 for studies with an adjusted analysis are plotted in Figure 2-3.
Of the significance values of 5, Ps in Table 2-6 where Ps refers to the one-tailed
significance level, five are 0.05 or less. If the hypothesis being tested is true, that the
true relative risk value is one in each study, then five of the eleven tests are making a
Type I error, i.e. are incotrectly finding a significant effect (at the 0.05 level). Assuming
that the studies are independently conducted, which seems reasonable, the chance of
observing five or more Type I errors in eleven independent studies is less than 0.001. If
one chooses to exclude one of the studies for which Ps is 0.05 or less (for any reason),
the corresponding probability for the remaining studies is 0.001 (figures are rounded). If
two of the significant studies are excluded, leaving three 'of nine with Ps equal to 0.05 or
less, the probability is still only 0.008. It is highly unlikely that so many significant test
results would be observed if there is, in fact, no association between ET5 exposure and
lung cancer incidence. (Technical note: No multiple comparison adjustments are
necessary for studies in which results for a single extreme exposure level are used
because the choice of exposure level does not depend on statistical significance reported
in the study. Test results reported at other exposure levels other than the one used are
not relevant, and hence there is no implicit multiple comparisons to adjust for.)
The 5 statistics of Table 2-6 may be used in the Wilcoxon signed rank test, as
conducted previously with the raw data, to provide another statistical test of the
2 - 11
-------
hypothesis of interest. The test is significant with P=0.014. In this test the magnitude
of the evidence from each study is a factor, in distinction to the dichotomization of
studies as significant or not at the 0.05 level in the preceding calculation. Both
statistical tests indicate that the cumulative evidence that lung cancer increases with
ETS exposure would be very unlikely to occur by chance alone.
2.4. EVIDENCE OF TRENDS IN CASE CONTROL STUDIES WITH MORE
THAN ONE EXPOSURE LEVEL
Data for studies that report relative risk by levels of exposure are given in Table
2-7, along with the results of statistical tests for trends when available. The relative
risks are plotted by exposure and shown in Figure 2-4 (WU was inadverten~ly excluded
from the figure, but the values for the plot are in Table 2-7). Some of the relative risks
are estimated by adjusted statistical methods and some are not, as indicated in the
table. Some observations are apparent from the plots. For example, the estimated rela-
tive risks never decrease from one exposure level to the next higher one in seven studies:
AKIB, CaRR, GAO, GENG, INOU, PERS, and TRIC; are monotonically decreasing
(slightly) in one case: LEE; and are variable in the remaining five plots: GARF, HUMB,
KOO, LAMT, and V ARE. Trend tests depend on more than just the relative risk
values shown in the plots on Figure 2-4, so one cannot make conclusions from the plots
alone.
If relative risk (R) is not associated with ETS exposure, the observed values at
all exposure values in a study should be equally likely to be greater or less than one
since differences can be ascribed to chance variation instead of a causal mechanism,
such as exposure to ETS. or its correlates. Under the hypothesis of no association, the
minimum R value at (positive) exposure levels in a study equals one or less with
probability at least one-half. Only two studies out of the thirteen have an R value of
2 - 12
-~
-------
one or less in an exposed group (LEE and V ARE), an event that would occur by chance
with probability less than 0.012. If the number of exposure levels could be taken into
account this figure would be smaller since there is more than one exposure level in each
study, but the calculations are prohibitive because the R values within a study are
correlated but unknown.
It can be concluded, however, that the plots for trend are
consistent with the previous statistical tests conducted on the case control studies.
2.5. BASIC ISSUES IN POTENTIAL BIAS FROM MISCLASSIFICA TION
IN CASE CONTROL STUDIES
Bias, and the potential source of it that has received the most attention,
misclassification of subjects, is not limited to case control studies. Most of the attention
in the literature, however, has been with respect to evaluation and interpretation of the
case control studies, so the topic is reviewed here in that context.
2.5.1. Background
Estimation bias is due to study design, protocol, or method of analysis that
apriori makes the expected outcome too large (positive bias) over too small (negative
bias). Sample size and dispersion in the population sampled contribute to outcome
variability, usually measured for an estimate by its standard error, but do not affect
bias per se. If bias is zero, then a sufficient number of independent and identical
repetitions of a study will assure that an estimate is arbitrarily close to the true value
(on average). Alternatively, that same end can be achieved by making the sample size
for the single study sufficiently large. Neither repeated sampling nor increasing the
sample size affect bias; the estimate (on average) will simply becomes arbitrarily close
to the unknown value of interest, i.e., the true relative risk plus the bias. In other
words, sufficient data will overcome the statistical uncertainty due to sampling, but will
2 - 13
-------
not affect bias because it is characterized by the methodology, not the observed data.
In practice, each case control study has its own bias (possibly zero) and true relative.
risk, neither of which is observable. If bias is largely random over a set of studies, some
averaging effect toward zero would be expected. If there is a consistent source of bias in
studies, however, sometimes referred to as "systematic bias", then it cannot be expected
to disappear as the number of studies available increases.
2.5.2. Sources of Bias
The NRC devoted considerable attention to the subject of misclassification
before concluding that "while the epidemiologic studies show a consistent and, in total,
a highly significant association between lung cancer and ETS exposure of nonsmokers,
the excess might, in principle, possibly be explained by bias.' However, detailed
consideration of the nature and extent of the bias shows that given some reasonable
assumptions, the bias would be insufficient to explain the whole effect. In fact, there are
some types of bias that lead to underestimates of the effect. It must be concluded,
therefore, that some, if not all, of the effect reported in spouse studies is causal" (NRC,
1986, p.242).
As reported recently by Saracci and Riboli (1989), two sources of bias may act to
decrease the observed relative risk among nonsmoking women exposed to ETS by
smoking spouses. First, this group of women is compared with other nonsmokers who
are not "pure" subjects unexposed to ETS, as some of them may be exposed to other
unrecorded sources of ETS, e.g., at work or in public places. Second, random misclassifi-
cation of exposure tends to dilute any existing effect and its relative risk. Two other
sources of bias would operate in the opposite direction of spuriously increasing the
estimate of relative risk. First, it has been shown that smoking (nonsmoking) wives tend
to be associated with smoking (nonsmoking) husbands. This would not bias the results
2 - 14
-------
if there was not at the same time misreporting of the nonsmoker status. However, if
some smoking women (or ex-smokers) are incorrectly reported as lifelong nonsmokers
then a bias would be introduced. The size of this bias also depends on the relative risk
among women misreported as nonsmokers. In addition, the magnitude of the bias
depends on the proportion of smokers among males and females in the population under
consideration. A further complication is introduced if the rate of misreporting of
nonsmoking status is not the same for women exposed to spouses' ETS and women who
are not exposed. The effect of differential misreporting has received very little atten-
tion, however, relative to the concern over inclusion of smokers or ex-smokers among
the never smokers, and the misclassification of never smokers as exposed.
That subjects classified as unexposed are rarely purely unexposed is supported by
data on cotinine measurements of body fluids indicating the ubiquity of ETS, which the
NRC took into account in adjusting an overall relative risk for a "net" bias. Recent
survey results of Cummings et al. (1989b) provide additional evidence of background
exposure in never smokers. Detectable levels of cotinine were found in 132 of 162 (81%)
of nonsmokers who reported no exposure in the four days preceding the interview. A
mean urinary cotinine level of 8.8 ng/ml was found among nonsmokers. Although the
study is based on self-selected volunteers, the authors note that the results are consis-
tent with reports from other studies. Cummings and colleagues conclude that exposure
to ETS is extremely prevalent, even among those not living with a smoker.
The NRC report's overall estimate of relative risk, i.e., estimate of the overall
risk parameter (ORRP), places the increased risk of lung cancer from a smoking spouse
at about 34%. An adjustment downward for possible misclassification of the never-
smoker status brings the value to 25%. A further adjustment to make the risk relative
to a purely unexposed subject, i.e., to take into account a background level of exposure,
raises the increased risk to 42%. Consequently, the net adjustment for bias is upwards.
2 - 15
-------
The reports of the Surgeon General (S.G., 1986) and IARC (1987) do not calculate an
overall risk estimate and then adjust it for possible bias.
Wald et al. (1986) and the NRC committee followed a similar course in
estimating overall risk and then adjusting for bias. For the same ten case control studies
and three cohort studies (see Section 2.1), Wald and colleagues estimate the overall
increased risk at 35%, with adjustments. to 30% and then 53%, corresponding to the
25% and 42% figures above from the NRC report. In the more recent analysis of Wells
(1988b) that includes 14 case control studies, the overall increased risk for females is
44%, which is then adjusted for bias in steps to 43% and then to 48%. Wells analyzes
cohort studies separately for corroboration of the estimate from the case control studies.
The methods of analysis and adjustment for possible bias by NRC and Wells are similar
in approach but not identical in detail. Blot and Fraumeni (1986) calculated an
increased risk for the same studies as the NRC, except with WU included but not
BUFF or the cohort study by Gillis et al. (1984). The estimate of the overall risk
parameter for females obtained by Blot and Fraumeni is 1.3, giving an increased risk of
30% compared to NRC's value of 34%. No adjustments were made for possible misclassi-
fication.
The diagnosis of lung cancer in cases may be a source of bias, e.g., a cancer that
originated at another primary site and then spread to the lung may be incorrectly
diagnosed as a primary cancer of the lung (Samet, 1988b). This misclassification tends
to be random and therefore biases relative risk estimates toward unity (Copeland et al.,
1977, as cited in Eriksen, 1988). As an example, Garfinkel et al. (1985) report that
about 12% of lung cancer patients identified through hospital records were reclassified
after histological review. Some studies addressed this. issue by including only
pathologically confirmed lung cancers or by considering histological cell type in their
analyses (CORR, GARF, PERS, and others) (Eriksen, 1988).
2 - 16
-------
Bias due to respondent by a proxy in place of the subject has. also been raised as
an issue by Mantel (1987b) and by Kilpatrick (1987), although they cite evidence from
only two studies. Our review of V ARE left us wondering about Figure No.1 (Varela,
1985), wherein it appears that one aberrant point associated with surrogate respondents
might be having an undue 'influence (a not uncommon occurrence in regression
methods) .
As reported in Eriksen et al. (1988), respondent bias can be a source of
negative or positive bias (Sackett, 1979), but in general the information provided by
surrogates has been fairly comparable to that provided by the individuals themselves
(Blot et al., 1985). The recent study by Cummings et al. (1989a) of the passive smoking
histories of 380 never smokers further supports that conclusion. They report good
agreement between subjects and -surrogates on most exposure measures.
Vandenbroucke (1988) and Mantel (1987a) have questioned whether there may
be a publication bias, studies with non-significant results being less likely to be
published. Vandenbroucke constructed a quantitative approach but found that publica-
tion bias was only found compatible for the studies on men. Wells (1988a) reviewed the
subject and concluded that it is unlikely that publication bias, i.e., the suppression of
work with high standard errors by authors or reviewers, has any substantial effect on
the relative risks that have been calculated from published reports for passive smoking
for either men or women.
That misclassification might fully account for an observed increase risk of lung
cancer from exposure to ETS remains a contention in some camps. Judging from the
literature, the most ardent researcher is P. N. Lee (1984, 1986, 1987a, 1987h, 1988). His
procedures and related quantitative assumptions, some of which lie outside the range
considered likely in the NRC report, have evoked responses in the literature. For
example, Wells (1988c) challenges some critical assumptions of Lee with evidence and
data to the contrary. Doll and Peto (1986), in a letter responding to Lee, accuse him of
2 - 17
-------
consistently selecting data and making choices otherwise that minimize the predicted
risk of exposure to ETS. It is not our purpose here to fully review the arguments and
cross-arguments on an issue that cannot be strictly proved or disproved, but requires
reasoned judgment of the evidence. The recent report of an expert panel on passive
smoking and lung cancer (Saracci, 1989) contains the following consensus opinion:
"While for a few authors (Lee, 1988) bias can essentially account for the whole of the
observed effect, it appears that this would demand the occurrence of a rather extreme
combination of values of the relevant parameters. For instance, to wholly account for an
observed relative risk of 1.35 requires as much as 10% of women reporting themselves as
nonsmokers to be actually active smokers, with a relative risk of eight for lung cancer,
and with an aggregation factor (odds ratio) of spouse smoking habits of 3.5. It seems
thus unlikely that this type of bias could be the sole [underscored in source] explanation
of the observed elevation in relative risk."
To give the reader some feeling for the sensitivity of the overall relative risk esti-
mate to the misclassification of subjects, it was re-estimated from the raw data for
females in the 18 studies in Table 2-5. The M-H approach was used, just as in Section
2.2, except with varied percentages of misclassification of subjects assumed. The
percentage of misclassification among cases and among controls were addressed sepa-
rately. For cases, a percentage X of the exposed group was reclassified as unexposed,
after rounding to the nearest integer, and then the overall relative risk estimate was
calculated. Values of X=1,2,3,...were tried until the significance level (p-value) of the
M-H chi-squared statistic reached a value of 0.05. This was for a two-sided test. so the
corresponding p-value of a test against a one-sided alternative would be one-half that
value (0.025). For cases, the two-sided p-value remained below 0.05 for values of X up
to 7, i. e., up to 7 percent of the exposed cases in each study can be shifted (reclassified)
as unexposed before the significance level exceeds 0.05. Among controls, 10 percent of
2 - 18
-------
the unexposed group can be shifted to the exposed group before the p-value exceeds
0.05. If one could identify the current and/or ever-smokers in a study, then they would
be removed from the analysis rather than reclassified. Statistical significance should be
less sensitive to removing subjects than to reclassifying them, as done in this example.
i .
I
I
Reclassification would be appropriate for nonsmokers who are incorrectly classified
according to exposure. :
The main concerns raised in the literature regarding misclassification concerns
cases.
As discussed above, however, misclassifications could occur in cases or controls,
and could occur differentially. The net reclassification (in the direction of reducing sta-
tistical significance) that could be distributed between controls and cases without.
exceeding the arbitrarily set significance level of 0.05 would likely fall between the 7 and
11 percent values found for cases and controls, respectively.
2.6. COHORT STUDIES: BACKGROUND
At this point we shift from review and discussion of the epidemiologic evidence
in case control studies to the three cohort studies that have been conducted: Garfinkel
et al. (1981); Gillis et al. (1984); and Hirayama (1981a, 1984); to be abbreviated as
GARF(Coh), GILL(Coh), and HIRA(Coh), respectively. The use of
"Coh" in
parentheses is used to distinguish a cohort study from a case control study. The three
studies are included in most of the references cited for summary descriptions and
comparisons of case control studies in Section 2.1. The Surgeon General's report (SG,
1986) sketches the basic features of the cohort studies and the salient topics of
controversy and discussion that appeared in the literature up to its time of preparation.
The Scottish study, GILL(Coh), which observed only a very small number of lung
cancer deaths (6 men and 8 women), is not discussed further in this report.
2 - 19
-------
Unlike the case control studies, several of which have appeared since the NRC,
IARC, and SG reports of 1986-87, the two major cohort studies, GARF(Coh) and
HIRA(Coh), first appeared in 1981. Consequently, most of the issues regarding these
two studies and their apparently discrepant results surfaced well before the three major
reports were prepared. Critical scrutiny of the Hirayama study had already appeared
and had been adequately addressed by Hirayama, as described in the SG and NRC
reports.
Judging from the roundtable discussion at the symposium
"Medical
Perspectives on Passive Smoking" (Lehnert, 1984), previous challenges to Hirayama's
work
regarding data analysis and other issues appear to have been resolved, aside
perhaps from the omnipresent issue of potential misclassification. Without relinquishing
the misclassification banner, even Mr. Lee offered a (qualified) concession to the
strength of the statistical evidence in the Hirayama study: "It is ... clear in Dr.
Hirayama's data that if one takes the age of the husband or wife into account and . does
the analysis correctly, there is a statistically significant association in lung cancer risk,
but the significance is not nearly as marked as in the incorrect analysis."
Paradoxically, the study by Garfinkel and colleagues at the American Cancer
Society (ACS) has undergone much less questioning and critical examination, although
the problems experienced in conducting the study and the potential for error in the
results has not gone unnoticed. The difference in outcomes in HIRA(Coh) and
GARF(Coh) has been a source of concern to many, but to our,knowledge no one has
conducted a statistical review of GARF(Coh), as has been the case with HIRA(Coh), or
compared the statistical methodology in the two studies. Those topics are addressed in
the following section.
2 - 20
-------
2.7. SOME COMPARATIVE ASPECTS OF TWO COHORT STUDIES:
HIR(Coh) and GARF(Coh)
2.7.1. Overview
An increase in risk of lung cancer from ETS was observed in both cohort studies,
with statistical significance (P<0.05), achieved in HIRA(Coh) but not in GARF(Coh).
In the former study, the observed risk is nicely patterned, consistently increasing with
higher levels of exposure from spouse smoking. Data from the American study estimate
a higher risk at the lower of two exposure categories (spouse smokes <20 cig./day) than
at the higher one (spouse smokes 20+ cig./day). Some researchers have interpreted this
outcome as evidence that there is not a "dose-response" relationship in the American
study, or more strongly, that the results demonstrate that there is no increased risk of
lung cancer from ETS exposure. We find the statistical evidence inconclusive regarding
a possible association between lung cancer incidence and ETS exposure, and consistent
with what one would expect to observe in either the presence or absence of a true dose-
response relationship. This conclusion follows from the 95% confidence intervals for the
lung cancer mortality ratio at the low «20 cig.jday) and high (20+ cig./day)
exposures, equal to (0.85, 1.89) and (0.77, 1.61) respectively. Technically these values
could be adjusted for making multiple comparisons and for the use of a confidence
interval corresponding to a two-sided test of significance rather than a one-sided test.
The basic conclusion that the data are consistent with a wide range of possibilities,
however, would remain. To illustrate this point with the confidence intervals given, the
value 1.0 (corresponding to no increase in lung cancer mortality) is in both confidence
intervals. Values corresponding to a substantial dose-response relationship are also in
the intervals, e.g., 1.25 and 1.50 at the low and high exposures, respectively.
2 - 21
-------
The American cohort study appears to contain more statistical uncertainty than
the Japanese study, either due to real differences in risk associated with ETS, the
presence of other factors that contribute to uncertainty in general, or both. Some of the
general factors contributing to uncertainty in study data are related to sample size,
variability in the population sampled, sample design and protocol, treatment of missing
or incomplete data, accuracy and reliability of collecting and reporting data, and
methods of statistical analysis. When the data produce a clear pattern such as
HIRA(Coh), with a consistent upward trend across exposure categories and age groups
that cannot be ascribed to chance alone, one has some assurance that the sources of
variability ("noise") are sufficiently under control relative to the strength of an effect
("signal") in the data. This does not preclude possible bias, however, which is related to
the tendency for estimates to be over- or understated on the average, and is not a
component of variability per se.
In the following section the Japanese study (HIRA(Coh)) anq. the American
study (GARF(Coh)) are reviewed, with an emphasis on the cultural differences in the
populations sampled and the differences in study design, ex~cution, and analysis of data
that may help to compare outcomes of the two studies. In the final section, data
comparisons are made for the two studies to evaluate if there is widespread differences
across all age-exposure group combinations, or just specific ones.
2.7.2. Comparative Review and Discussion of Two Cohort Studies
HIR(Coh) is a census-population based study of adults aged 40 or above, begun
III 1965 in 29 Health Center Districts in Japan. A total of 200 cases of lung cancer
occurred among the 91,540 nonsmoking married women that were followed. In the whole
study, 265,118 subjects were enrolled (122,261 males and 142,857 females) including
unmarried women, accounting for 94.8% of the total census in the study area. Subjects
2 - 22
-------
were tracked by establishing a record linkage system between the risk factor records and
death certificates (Hirayama, 1983b, 1984). Blind interviews were conducted on the
study subjects (NRC, 1986).
In the Japanese study, relative risks of 1.42, 1.58, and 1.91 were observed for
nonsmoking wives with husbands who smoked 1-14, 15-19, and 20+ cigarettes per day,
respectively. The corresponding value for women whose husbands were ex-smokers is
1.36, between the values for nonsmoking and light smoking husbands but closer to the
latter (Hirayama, 1984). The observed increase in risk across the exposure categories,
with ex-smokers classified as exposed between nonsmokers and the group for 1-14
cig./day, is statistically significant by the Mantel-Haenszel test (one-tailed P<0.002).
Also, an increasing trend in risk related to husbands' smoking is reported when data are
analyzed by age group (ten year periods), using either husbands' or wives' ages, or by
occupational group or duration of exposure. No other characteristic of husbands or wives
was found to be associated with the risk of lung cancer in nonsmoking women
(Hirayama, 1983a).
The risk differential from exposure to ETS in the Japanese study was observed
to decline with age, for all exposure groups. It is reported that although the risk
relationship persists, there is an observable difference in the strength of the relationship
for older and younger women. A clear-cut relationship was observed in younger women
while the difference was rather small in older women (Lehnert, 1984). Some recent work
showing an effect of age on lung cancer risk in active smokers may have some bearing
on this issue. Moolgavkar et al. '(1989) conclude that age likely influences lung cancer
risk among smokers independently of duration of smoking. If this is true for passive
smokers, then age of a nonsmoker exposed to ETS may be a factor aside from duration
of exposure. Another possibility is suggested by Wells (1988b) who studied adult
mortality from passive smoking. He comments that in passive smoking deaths we are
2 - 23
-------
dealing with only the very most susceptible people. It may be that the more susceptible
people constitute a relatively smaller proportion of the aged population at risk. This
would follow if a subpopulation relatively more susceptible to lung cancer from exposure
to ETS is also more susceptible to other health risks as well, reducing the rate of
survival to old age. Future res arch may shed some light on this issue.
GARF(Coh), the American Cancer Society Prevention Study I, began in 1959
when a pyramidal structure of 68,000 volunteers in 25 states enrolled more than one
million men and women for long-term follow-up. Volunteers were instructed to recruit
people they knew well. Subject participation was fairly evenly divided across large
cities, small cities and suburbs, small towns, and rural areas. Overall, about 3% of the
population over .the age of 45 in 1,121 counties was recruited. Enrollment included all
family members of age 30 or above, provided at least one member of the househ.old was
at least 45. In the general plan of the study, the volunteers were to report annually on
the status (alive or dead) of the subjects they enrolled, but some volunteers moved
. away or died in subsequent years, many of the subjects moved, and some volunteers did
not follow instructions.
Each year, for six years, the volunteers were asked to report the vital status of
the persons contacted (alive or dead). For subjects who had died, death certificates were
obtained from state departments of health to determine whether death was due to lung
cancer or not. Additionally, for the first six years, physicians who certified the cancer
deaths were contacted and asked to supply information to verify the primary sites of the
cancers. In the first six years, information was received confirming the primary site of
cancer in 78% of the cases and microscopic confirmation was obtained in 69% of the'
cases. It was found that death certificates overstated the lung cancer rates by 11.8%
(Garfinkel, 1981, 1984, 1985). The study was essentially terminated after six years, as
originally planned in 1965, until it was decided to conduct a second follow-up in
2 - 24
-------
1971-72. It is reported that the follow-up was successful for 98.4% of the subjects. It is
also reported, however, that the follow-up was terminated because tracing became
increasingly difficult due to death or movement of the volunteers and their substitutes
(Garfinkel, 1985). Apparently death certificates did not continue to be followed-up by a
medical report after the first six years. For lung cancer cases in all women, married or
not, 203 out of a total of 564 (36%) reported by death certificates were accompanied by
a medical report.
The American stud¥ is not conclusive regarding a possible association of lung
cancer with ETS exposure. The ratio of observed to expected lung cancer deaths,
referred to as the mortality ratio by Garfinkel (1981), is 1.27 and 1.10 for nonsmoking
women with husbands who smoked <20 cig./day and 20+ cig./day, respectively.
Neither value is reported to be statistically significant. When data from the Hirayama
study are grouped according to the same exposure levels «20 cig./day and 20+
cig./day), the observed relative risks are 1.45 and 1.91, with one-tailed p-values of 0.03
and 0.001 respectively (Hirayama, 1984).
As discussed in Section 2.7.1., apparent differences between outcomes of the two
.
studies could be due to one or more sources: a real difference in risk in the populatbns
studied; differences in the way the studies were designed, conducted, or interpreted;
chance occurrence alone. There is some suggestive evidence for the first two
alternatives. Subjects in the American study were followed for 12 years in comparison to
16 years in the Japanese study (Hirayama, 1984), so the proportion of subjects with
lung cancer would be expected to be less in the American study, other factors aside. As
reviewed in the Surgeon General's report (SG, 1986), the relatively high risks observed
for nonsmokers whose husbands smoked led to speculation that Japanese women may
report themselves. as nonsmokers when they actually smoke (also see Lehnert, 1984),
part of the general concern over misclassification in both case control and cohort studies
2 - 25
-------
of ETS. Some reassurance of the validity of self-reported information from Japanese
women, in general, came from the case control study of Akiba et al. (1986) (designated
AKIB in this report). That study found strong concordance between smoking status
reported by the women themselves and the reports from next of kin.
Hirayama has emphasized the importance of properly defining passive smoking,
which he classifies as direct passive smoking, when the exposed subject is within a
proximity of about 1-1.5 meters from the source of exposure, and indirect passive
smoking, which applies to exposure from a greater distance (Hirayama, 1984; Lehnert,
1984). Direct passive smoking is of much greater concern than indirect passive smoking
(Lehnert, 1984). Japanese wives may experience more direct passive smoking if they
tend to be in closer proximity to their smoking husbands than American wives. Related
factors that may contribute to a net increase in exposure of Japanese wives relative to
their American counterparts include house sizes, the number of smokers per volume of
air, climate and ventilation, proximity of nonsmoking spouse's sleeping area to spouse's
smoking area, and the amount of time a nonsmoking spouse is in the home. Hirayama
(1981b) notes additional differences between Japan and America that may influence
exposure, such as a higher percentage of office workers among females in the United
States than in Japan, a higher divorce rate in the United States, and the custom in
Japan of smoking without first requesting consent.
There are also some differences in the methods of analyzing and interpreting data
in the two cohort studies, as described in the next section.
2.7.3. Comparative Data Analysis of Two Cohort Studies
The measures of risk reported in HIRA(Coh) and GARF(Coh), the odds ratio
and the mortality ratio, respectively, are not equivalent. Neither are the statistical
methods for controlling for age the same. The method applied by Hirayama (Hirayama,
2 - 26
-------
1984) is the Mantel-Haenszel procedure, commonly used to standardize for a,ge and
other factors that may have an influence. To control for age by this method, for
example, study observations are grouped by time intervals. Comparisons between
exposure groups are made at each time interval and then the results are combined
I -
across intervals to test for a. difference between exposure groups (the extended M-H
procedure). The analysis described by Garfinkel (1981), used in previous analyses of the
I
same cohort study ~oncerned with topics other than passive smoking (Hammond et aI.,
1976; Hammond, et al., 1975), is of a somewhat different nature. To adjust for age, each
age group is weighted by a factor according to person-years with a smoking husband.
Then the data are treated as quantal response data, i.e., there is one large data set with
observations weighted by person-years. Expected deaths are based on the lung cancer
rates by five-year age groups in women with nonsmoking husbands applied to the
person-years of women with smoking husbands.
To adjust the analysis for several variables at once, in order to take into account
potential confounding factors, the method previously applied to other data in the
American study is used (Hammond et al.; 1975, 1976). Matched groups are formed from.
the data, with the matching on age, race, highest educational status of husband or wife,
residence, and whether or not husband is occupationally exposed to dust, fumes, or
vapor (Garfinkel, 1981). The ratios of the number of adjusted lung cancer deaths in the
low «20 cig.jday) and high (20+ cig.jday) exposure categories to the control group,
i.e., the nonsmoking women with nonsmoking husbands, are reported to be 1.37 and
1.04, neither of which is statistically signficant.
Based on the description in the two references to Hammond et al. cited by
Garfinkel, under this method of adjusted analysis the data for subjects who cannot be
fully matched would be discarded. This appears to have occurred in the analysis of
passive smoking, judging from the estimates of the number of adjusted lung cancer
2 - 27
-------
deaths in exposure groups (a range of 25 to 36, Table 5, Garfinkel, 1981). Provided we
understand the procedure correctly, a downward adjustment would be expected due to
the limitation of simultaneously matching on several variables. It would be of interest to
compare the results obtained by a survival analysis approach that adjusts for covariates
simultaneously while using all the data. Such an approach would also have some
limitations, but it is likely that the power to detect an effect, if there is one, would be
improved.
Using data for the American study that includes age at time of death, duration
on study, and whether death was due to lung cancer or another cause (supplied by L.
Garfinkel), the Mantel- Haenszel method was applied with age and duration on study
controlled. The results were not close to statistical significance.
To see if we could gain further insight into the outcome in the American, study,
the descriptive data from GARF(Coh) corresponding to the age groups and exposure
classifications of data published for HIRA(Coh) (Hirayama, 1984, Table 1) were placed
side-by-side for visual comparison (Table 2-8). Relative to the general pattern of
response in the Japanese study, the American data appear to be at greatest variance
from what might be expected in the two subgroups at highest exposure (20-|- cig./day)
with age classifications 40-49 and 50-59. Further review of those data for completeness,
possible sources of bias, or unanticipated anomalies may be useful.
2.8. FROM HAZARD IDENTIFICATION TO RISK ASSESSMENT
In this chapter, the epidemiologic evidence of an association of ETS exposure
and lung cancer incidence has been considered. The question addressed has been
simply: Is there evidence of increased lung cancers among persons chronically exposed
to ETS after adjusting for sampling variability, i.e., beyond the laws of chance? Chance
2-28
-------
---_J
occurrence alone is extremely unlikely, based on the consistent outcomes of the several
analytical approaches described in this chapter.
Whether ETS exposure causes an increase in lung cancer, only contributes to it,
or acts as a surrogate for another cause(s) correlated with ETS exposure cannot be ascer-
tained from data analysis alone. Most studies attempted to account for other factors
that may contribute to lung carcinogenesis or otherwise confound the study interpreta-
tion. Particular attention was given to adjusted analyses and trends in Sections 2.3 and
2.4 above.
If ETS is not implicated as a causal factor at all, the "joker" (to quote.
Hirayama (Lehnert, 1986)) is elusive indeed.
If ETS is a risk factor for lung cancer, to what extent can exposure levels be
quantitatively related to risk estimates, i.e., can a meaningful. quantitative dose-
response relationship be determined? Published risk assessments of ETS have largely
relied on extrapolation of the overall relative risk estimate to a population of interest,
an approach for which some reservations are implied by our Section 2.2, or extrapola-
. tion to an estimated dose-response relationship for active smokers based on a. presump-
tion of equivalence of passive smoking at some specified ETS exposure level to active
smoking of some number of cigarettes, i.e., a "cigarette equivalence" for ETS exposure.
That approach may depend heavily on assumptions related to the biokinetics of active
and passive smoking, and the distribution of carcinogens between the particulate and
vapor phases in mainstream smoke and sidestream smoke. That topic is addressed in
the next chapter, followed by a review and discussion of dose surrogates and risk assess-
ments for passive smoking in Chapter 4.
Acknowledgment.
The authors are grateful to numerous researchers for helpful sug-
gestions and discussions, and for communication of recent research work and study data.
In particular, we would like to thank (in alphabetical order):
R.C. Brownson, K.M.
2 - 29
-------
. Cummings, R. Everson, L. Garfinkel, W. Hofmann, D. Hoffmann, W.K. Lam, J. Lew-
tas, T. Martonen, J. Repace, E. Riboli, D.L. Swift, S.R. Tannenbaum, and A.J. ';Yells.
2 - 30
-------
TABLE 2-1. CASE-CONTROL STUDIES - CHARACTERISTICS
Matched Final sample Adjusted statistical
Study Location variables matched for ETS? analysis?
AKIB Japan Age, sex, resi- Yes No
(Akiba et al. (Hiroshima, dence, RERF
1986) Nagasaki) participant
BROW1 USA Age, sex No2 Yes
(Brownson et (Colorado)
aI., 1987)
BUFF USA Age, sex No2 No
(Buffler et al., (Texas)
1984)
CHAN Hong Kong Matched but vari- No2 No
(Chan and Fung, abIes unspecified
1982)
CORR3 USA Age (~5), sex, No2 No
(Correa et al., (Louisiana) race
1983)
GAO China Age (~5) No2 Yes
(Gao et al., 1987) (Shanghai)
GARF USA Age (~ 5) Yes Yes
(Garfinkel et al.,
1985)
GENG China Age (~2), sex, No2 No
(Geng et al.' 1986) (Tianjin) race, marital status
HUMB USA Age (~ 10), sex No2 Yes
(Humble et aI, 1987) (N ew Mexico) ethnicity
INOU Japan Age, year of No2 Yes
(Inoue and (Kanagawa, death (~2.5),
Miraijama, 1988) Miura) district
KABA USA Age (~5), sex, Yes No
(Kabat and (New York) race, hospital
Wynder, 1984)
KOO Hong Kong Age (~5), resi- No2 Yes
(Koo et al.' 1987) dence, housing
-------
TABLE 2-1. Continued
Study
Matched
variables
Final study
matched for ETS?
Location
Includes an adjusted.
statistical analysis?
LAMT Hong Kong Age (x 5), resi- N02 No
(Lam et aI., 1987) dence.
LAMW Hong Kong Age, socioeconomic N02 Yes
(Lam, 1985) status, residence5
LEE England Age, sex, hospital No2,4 No
(Lee et aI., 1986) location, time of
interview
PERS Sweden Age (x 1), sex Yes Yes
(Pershagen et
aI., 1987)
SHIM Japan Age (x 1), hospital, Yes Yes
(Shimizu et aI., (Nagoya) admission date
1988)
SVEN Sweden Age N02 Yes
(Svensson et aI., (Stockholm)
1988)
TRIC Greece Age, socio- N02 No .
(Trichopoulos (Athens) economic status6
et aI., 1981)
VARE USA Age, sex, county, Yes Yes
(Varela, 1987) (New York) smoking history
WU USA Age (x 5), sex, N02 Yes
(Los Angeles) race
1 Adenocarcinoma only.
2Not matched on smoking status (smoker/non-smoker).
3Bronchioa.lveolar cancer excluded.
40ngoing study modified for passive smoking with follow-up.
5"Similar" in age, SES, and residence.
6"Similar" in age and SES.
-------
TABLE 2-2. CASE-CONTROL STUDIES - CHARACTERISTICS
Percent Percen t
Proxy Female2 Source Number Female
Response1 Age of Female Controls
Study Ca Co Ca Co Controls Controls "Exposed,,3
AKlB 90 88 70.2 . Atomic bomb 270 70
35-95 . survivors
BROW 69 39 66.3 68.2 Cancer cases4 47 155
BUFF 82 76 30- 79 30- 79 Cancer cases6 196 847
CHAN . . 39- 70+ 39-79+ Orthopaedic 139 47
patien ts
CORR . . . . Hospital 133 46
patients 8
GAO 0 . 35-69 35-69 General 375 74
population
GARF . . ~40 ~40 Cancer cases9 402 61
GENG . . ~65 ~65 . 93 44
HUMB . . ~85 ~85 General 162 56
population
INOU 100 100 . . Cerebrovascular 64 .
disease (deaths)
KABA . . 61.6 53.9 Patients10 25 60
KOO . . . . "Healthy" 11 136 49
LAMT . . . . "Healthy" 12 335 45
LAMW . . 67.5 66 Hospitalized 144 44
orthopedic
patients
LEE 3813 38 35- 74 35- 74 Patients14 66 68
PERS .15 . .16 . .17 347 43
-------
TABLE 2-2. Continued
Study
Percent
Proxy
Response1
Ca Co
Co
Female
Age
Ca
SHIM . . 59 58
35-81 35-81
SVEN 0 0 66.3
TRIC . . 62.8 62.3
VARE
3319
68.119
3319
67.119
WU
<76
<76
.
.
Source
of
Controls
Pet'cen t
Female
Controls
"Exposed,,3
Number
Female
Controls
Patients18 . .
General 174 66
population
Hospitalized 190 43
orthopedic
patien ts
New York State 21820 .
Dept. of Motor
Vehicles
Neighborhood 12 52 63
1"ca" and "co" stand for "cases" and "controls", respectively.
2Single values are the average or median. Paired values are the range.
3Definition of "exposed" differs between studies. See Table 2-4 and Table 2-5.
4Persons' with cancers of bone marrow or colon in Colorado Control Cancer Registry,
5"Exposed" if husband smoked.
6Population-based and decedent comparison subjects selected from state and federal records.
7 "Unexposed" includes up to 32 total years of living with a household member who smoked.
8 Assorted ailments.
9 Colo- rectal cancer.
10Diseases not related to tobacco.
llSelected from a healthy population.
12Living in neighborhood of matched case.
13 Applies only to the 143 patients in the follow-up study.
14Excluding lung cancer, chronic bronchitis, ischemic heart disease, and stroke.
15No overall percentages given.
16Two control groups: 15-65 and 35-85 for both cases and controls in groups 1 and 2
respectively.
17 Two controls groups were randomly chosen from the cohort under study.
18patients in the same or adjacent wards with other diseases.
19Includes males and females. .
20Includes 69 former smokers.
-------
TABLE 2-3. CASE-CONTROL STUDIES - ETS SOURCES
Adulthood Child hood
Spouse(s) Others at A way from Exposure From
Study 1 >1 Home Home Mother / Father
AKIB X X
BROW X X X
BUFF X
CHANI X
CORR X X
GAO X X X X
GARF X X X
GENG X X2 X
HUMB X
INOU X
KABA X X X
KOO X X
LAMT X
LAMW X X X
LEE X X3
PERS X X
SHIM X X X
SVEN X X X X
TRIC X
VARE X X X4
WU X X X X
1 As reported in Chan, 1982. In Chan, 1979, exposure is described as at home or
at work.
2Exposure to mother's or father's smoking is presumed to mean in adulthood.
3Separate for workplace, travel, leisure.
4Separate for workplace and social circumstances.
-------
TABLE 2-4. MEASURES OF ETS EXPOSURE
AND EXPOSURE TO POTENTIALLY RELATED SUBSTANCES
ETS Exposure Measures Related Exposures
Cigarettes/ Total Total Cooking/ Work/
Study Day Years Cigarettes Other Heating Environ.
AKIB X
BROW hrs/day X
BUFF X X
CRAN Xl X
CORR pack-yrs
GAO X X X
GARF X2 hrs/day
GENG X X
HUMB X X
INOU X3
KABA no units X
KOO X X X4
LAMT X5
LAMW no units
LEE X6 X
PERS no units X
SHIM X X X
SVEN X7
TRIC X8 X
VARE X X X person-yrs
-------
TABLE 2-4. Continued
Study
Exposure Measures
Cigarettesj Total Total
Day Years Cigarettes
Other
Related Exposures
Cookingj Workj
Heating Environ.
wu
X9
x
x
1Exposedjunexposed determined from a single question, "Are you exposed to the
tobacco smoke of others at home or at work?" (Lam T.H. et al., 1987).
2Cigjday smoked by husband at home.
3Smoker at home defined as ~ 5 cigj day.
40thers include total hours of exposure and mean hrsjday.
5 A woman was considered exposed to her husband's smoke if they had lived together
continuously for at least one year.
6Exposure designated as 0 (unexposed), 1,2,3.
7 Exposure is "yes" or "no" for each source.
8Exposed within last 5 years.
9Exposed if husband smoked.
-------
TABLE 2-5. CASE-CONTROL STUDIES - "UNEXPOSED" vs. "EXPOSED"
FEMALES FROM RAW DATA
No. No. P~
Study Exposure Cases Controls Rl C.l.l SI
AKIB
Female 0 21 82 1.52 (0.88.2.64) 1.48 0.07
(Cig/ day) ~1 73 188 1.5 2
(1.0,2.5)
Male 0 16 101
( Cig/ day) ~1 3 19 1.8 (0.5,5.6)
B ROW3
Female unexposed 15 40 1.52 (0.39.5.99) 0.61 0.27
exposed4 4 7 1.5 .
Male 0-3 2 11
(Hrs/ day) ~4 2 8 1.38 .
BUFF
Female 0:"32 8 32 0.81 (0.34.1.90) -0.49 0.69
(Tot. yrs) ~33 33 164 0.8 (0.3,1.8)
Male 0-32 6 34
(Tot. yrs) ~33 5 56 0.5 (0.2,1.7)
CHAN
Female unexposed 50 73 0.75 (0.43.1.30) -1.02 0.85
exposed 5 34 66 0.8 (0.4,1.3)
CORR
Female6 0 8 72 2.07 (0.82.5.20) 1.52 0.06
(Pack-yrs) ~1 14 61 2.07 (0.8,5.0)
Male 0 6 ~.o
(Pack-yrs) ~1 2 .
-------
TABLE 2-5. Continued
Study
GAO
Exposure
No.
Cases
Sl
p~
Female
(Tot. yrs)
GARF
Female
(Cig/ day)
:=GE~
Female
(Cig/day) .
HUMB8
Female
(Cig/day)
INOU
Female
C (Cig/d~)
KABA
Female
, .
KOO
Female
(Cig/day)
No.
Controls
R1
1
C.!.
0-19
~20
(0.82.1.73)
0.18
0.91
99
276
1.19
57
189
.
.
1.31 (0.87.1.98) 1.29
1.31 (0.99,1.73)
lAJA/!- ~ t.~
6'- 'if +- ~ /. 8' to - A:4!.. (f' '1
~ (1.09.4.28) 2.19 0.01 '-7
2. (~,4.5tl,. '.
. (A~~Y('
2.34 (0.83. 6.61)~.57 0.06
1.8 (0.6,5.4) '-..
~.
- f'-~ . ¥?- ~~~
~;49 1;. 3~ GZt J7~'0'9;,1.1O?~~~'\ -:1 1-' 'c
-) ~;~~:(~ - ~~ ~~~ 9";fQ
. . , ~. r~ l:in. 71~\ ,\
unexposed .11 10 0.79 (0.25.2.48) -0.41 0.66 t">-~ .
exposed10 13 15 .. . ~ <)
o
~17
0.10
44
90
157
245
o
~1
20
34
52
41
o
~1
5
15
71
91
o
exposed 11
35
115
70
66
1.55
1.55
(0.90.2.67)
(0.94,3.08)
0.06
1.56
-------
TABLE 2-5. Continued
No. No. P~
Study Exposure Cases Controls R1 1 Sl
C.!.
LAMT fi~ s7#/~~
. , I f 1-~
J ' "
Female 0 84 183 ,.t'! ~ 'r-,.
~ (1.16.2.35) 2.77 0.003 ~
(Cig/day) ~1 115 152 1.69". (1.16,2.35) ~ 4../
LAMW .~(
Female unexposed 23 80 2.01 (1.09.3.71) 2.22 0.01
exposed 11 37 64 2.01 .
LEE
Female unexposed 12 10 21 1.03 (0.412.56) 0.06 0.48
exposed 13 22 45 1.0014 ( 14
0.37,2.71)
PERS
Female unexposed 15 34 197 1.2814 (0.76.2.15) 0.91 0.18
exposed 16 33 150 14 (
1.28 0.75,2.15) 1)
~ ~\(I Sll (M t1.1' / /,'... ~ /L) !~~, I
SVEN /
( JJ
Female unexposed 10 60 1.26 (0.57.2.81) 0.57 0.28
exposed 16 24 114 . .
TRIC
Female 0 24 109 2.13 (1.19.3.81) 2.53 0.0063
(Cig/day) ~1 38 si . .
1-1.,,41.( ill. ,;;
WU l./ ~/,h /c t-
Female17 unexposed 9 19 1.12 (0.46.3.24) 0.39 0.35
exposed4 19 33 . .
"----,
~.
,
l.-J(
~ I
>.L
)
-------
Footnotes for Table 2.5.
1Values of R, C.I., S, and Ps on the first row of an entry (boldface) are our calculations for
Mantel-Haenszel odds ratio. Values in the second row are from the study. S is the square
root of the Mantel-Haenszel statistic with sign of (-) if R < 1 and (+) if R> 1. PSis one-
tailed from normal tables, and equals one-half the corresponding two-sided P value for the M-
H chi-squared statistic. Confidence intervals are 95% unless noted otherwise.
290% C.I.
3Data communicated from R.C. Brownson.
4Exposed if husband smoked.
5Exposure based on single question, "Are you exposed to the tobacco smoke of others at
or at work?" (Lam, T.H., et a1., 1987a).
6Data partially from Table 12-4, NRC (1986).
7 Cigar or pipe smoking by husband while at home is included in category of ~ 1 cigjday.
8Data communicated from C.G. Humble.
9Husbands who smoke less' than 5 cigfday are presumed not to smoke at home.
10Based on spouse's current or past smoking habits.
llExposed if husband ever smoked in presence of spouse.
120nly the controls in the follow-up study.
13Exposed if husband ever smoked during marriage.
14Standardized for age.
15Data for controls from Saracci and Riboli (1989).
16No measure of exposure given.
17 Data from Blot and Fraumeni (1986).
home
-------
TABLE 2-6. CASE CONTROL STUDIES - "UNEXPOSED" vs. "EXPOSED"
FEMALES FROM ADJUSTED STATISTICAL ANALYSES
Study
BROW
GAO
GARF
HUMB
>INOIL
KOO
J^AMW3
LEE5
PERS1
SHIM2
SVEN
VARE4
WU
Exposure
<3 vs. >4 (hrs/day)
0-19 vs. >40 (yrs. with
smoking husband)
0 vs. 20 (cig/day)
0 vs. >21 (cig/day)
-^<4 vs. >20 (cig/day)
0 vs. >21 (cig/day)
Exposed by husband
Exposed by husband
0 vs. >16 (cig/day)
Exposed by husband
Exposed in both childhood f
and adulthood vs. exposed U^-
in neither ^"p*..
O.vs. 20 (cig/day) , . '
, f
Exposed by husband
f&i>
R
1.68
1.7
1.70
^
1.2
3.09
1.19
2.64
1.00
2.4
1.1
't™ S>
(S
0.94
fO
j?s>
95% C.I.
(0.39, 2.97)
(1.0, 2.9)
(0.98, 2.94)
—
(0.26, 5.5)
(1.04, ll.Slf
fi
(0.46, 3.03)
•
(0.37, 2.71)
(0.6, 8.7)
•
(0.2, 3.7)
•& if'S" •
<)rtr> affu. -4
(0.76, 1.17)
(0.6, 2.5)
S
1.78
1.95
1.90
0^23
1
f 1.65
*£> 0.36
0.00 "
1.33
•
1.89
-0.54
0.49
PS
0.04
0.03
0.03
1T41
0.05
0.36 '
•
0.50
0.09
a
0.03
0.70
0.31
1See footnotes 15-17 of Table 2-2.
9
Higher R values associated with adult exposure to smoking by mother or by father's
husband. Insufficient information to calculate the S statistic.
3No units of exposure. R = 2.64 with p = 0.02, and R=1.61 with P = 0.19, for periph-
eral and central lung adenocarcinoma, respectively.
4From Table 4 of Varela, 1987.
See footnotes 12-14 of Table 2-5. Study is included with adjusted statistical analyses
in this table since analysis was standardized for age.
-------
TABLE 2-7. CASE-CONTROL STUDIES - EXPOSURE-RESPONSE
TRENDS FOR FEMALES
Analysis
Study Exposure R 1 P- Trend U nadj usted Adjusted
C.I.
AKIB 0 1.0 0.06 X
(Cigj day) 1-19 1.3 2
(0.7,2.3)
2
20- 29 1.5 (0.8, 2.8)
~30 2.1 2
(0.7, 2.5)
CORR 0 1.0 . X
(Pack-yrs) 1-40 1.18 .
~41 3.52 .
GAO 0-19 1.0 . X
(Tot. yrs)3 20-29 1.1 (0.7,1.8)
30-39 1.3 (0.8, 2.1)
~40 1.7 (1.0, 2.9)
GARF 0 1.0 < 0.025 X
(Cigj day) 1-9 1.15 (0.8, 1.6)
10-19 1.08 (0.8, 1.5)
~20 2.11 (1.1,4.0)
GENG 0 1.0 . X
( Cigj day) 1-9 1.40 (1.1, 1.8)
10-19 1.97 (1.4,2.7)
~20 2.76 (1.9,4.1)
HUMB 0 1.0 . X
1-20 1.8 2
(0.6, 5.6)
~21 1.2 2
(0.3, 5.2)
INOU 0-4 1.0 <0.05 X
(Cigj day) 5-19 2.58 2
(0.4,5.7)
2
~20 3.09 (1.0, 11.8)
KOO 0 1.0 . X
(Cigfday)4 1-10 2.33 (0.9, 5.9)
11-20 1.74 (0.8, 3.8)
~21 1.19 (0.5, 3.0)
-------
TABLE 2-7. Continued
Analysis
Study Exposure R 1 P- Trend Unadjusted Adjusted
C.!.
LAMT5 0 1.0 <0.01 X
(Cig/day) 1-10 2.18 (1.14,4.15)
11-20 1.85 (1.19, 2.87)
~21 2.07 . (1.07,4.03)
LEE6 0 1.0 . X6
Low 0.92 .
High 0.81 .
PERS7 0 1.0 . X8
(Cig/day) 1-15 1.8 (0.6, 5.3)
~16 6.4 (1.1, 34.7)
TRIC9 0 1.0 . X
(Cig/day) 1-20 1.95 .
~21 2.55 .
VARE10 0 1.0 . X
(Cig/day) 1-20 0.79 (0.6, 1.1)
21-40 0.91 (0.6, 1.3)
41-60 1.23 (0.6, 2.4)
61-80 0.42 (0.1, 2.3)
80+ 2.86 (0.3,27.7)
WUll 0 1.0 . X
(Yrs. exposed 1-30 1.2 .
as adult) ~31 2.0 .
1Confidence intervals are 95% unless noted otherwise.
290% C.!.
3Years lived with a smoking husband.
4Cig/day smoked by husband.
5 All histologies.
6Exposure at home only. Standardized for age, spouse smoking, and whether
marriage was ongoing or ended.
7 Small cell carcinoma only. Observed risk was lower for other histologies combined.
8Standardized for age.
9Data from Trichopoulos et al. (1983).
10From Table 2 of Varela (1987) for spouse smoking, presumably including males.
11 Adenocarcinomas only.
-------
TABLE 2-8. TWO COHORT STUDIES - FEMALE LUNG CANCER
DATA FOR SIMILAR AGE AND EXPOSURE GROUPS1
Age3
Husband's Smoking Habit2
Study4 Nonsmoker 1-19 20+
G 9/23,743 6/11,791 12/26,918
(3.8) (5.1) (4.5)
H 4/6,229 14/13,779 16/10,764
(6.4) (10.2) (14.9)
G 31/25,108 25/13,528 21/24,184
(12.3) (18.4) (8.7)
H 10/7,791 28/13,720 24/9,820
(12.8) (20.4) (24.4)
G 23/15,138 16/6,884 20/7,299
(15.2) (23.2) (27.4)
H 18/7,120 37/9,756 23/4,651
(25.3) (37.9) (49.4)
40-49
50-59
60-69
1 Entries are (number of lung cancer deaths)/(number at risk). Values (x104) are in
parentheses. Data for age 70-79 are omitted because of small sample sizes and small
number of lung cancers observed. Data for "G" were supplied by L. Gariinkel.
Data for "H" are in Hirayama (1984).
2 Cigarettes/day.
3 Women's age for Gj Husband's age for H.
4 G: GARF (Coh) H: HIRA (Coh).
-------
BROW
TRIC
PERS
GENG
LAMW
LAMT
CORR
CHAM
Koo
HUMS
KABA
GARF
wu
SVEN
LEE
AKIB
GAO
BUFF
PERCENT
******.
15
**.*.****.*.*..***.**
4&]
*.....***.**..*.*****.
43
.*..*.******.********.
4&4
*..*.**...*..**....*..
44
*.*.**..***..*****.....
45
****.***..*..****......
46
**...****..******.*...*.
47
***.*****.*.*..*.*......
49
**.*.***.**..***.*.....*....
!)6
***********..***...*...**...*.
GO
.*****.*.**.*******.**.....**.
.51
**.......***..***...*.**.****...
63
.**.*...**.**.***..*.*.**.**.....
66
.*..*****.***.*.*......*...***.**.
68
.*........*........................
10
*.....*...*.***...*.*..*.****.*..****
74
.........*....****.*.*.*..................
84
----------+---------+---------+---------+--
20
40
60
80
PERCENT
Figure 2-1.
Percentage of controls exposed to ETS by study.
-------
S statistic
CHAN
BUFF
**********
-1. 02
*****
-0.49
RASA
****
-0.41
LEE
WU
*
0.06
****
SVEN
******
0.39
0.57
BROW
******
0.61
GAO
*********
0.91
PERS
*********
0.91
GARF
*************
1.29
AXIB
***************
1.48
CORR
***************
1.52
Koo
****************
1.56
HUMB
****************
1.57
GENG
**********************
2.19
LAMW
**********************
2.22
TRIC
*************************
2.53
LAMT
****************************
2.77
~+-----~---+---------+---------+--------
-1.00
0.00
1.00
S Statistic
2.00
Figure 2-2. Ordered values of the S statistic from raw
data of studies in Table 2-5.
-------
S Statistic
VARE
***********
-0.54
LEE
0.00
HUMB
*****
0.23
Koo
*******
0.36
wu
**********
0.49
PERS
***************************
1.33
INOU
*********************************
1. 65
BROW
************************************
1. 78
SVEN
**************************************
1.89
GARF
**************************************
1.90
GAO
***************************************
1.95
----+~------+-------+-------+-------+-------+-------
-0.40
0.00
0.40
0.80
1.20
1.60
S statistic
Figure 2-3. Ordered values of the S statistic from adjusted
analyses of studies in Table 2-6.
-------
Figure
studies
I .....
.....
. . .....
I .....
.....
.....
1.' . .....
I ..... J.t . .....
..... I .....
..... .....
1.' . ..... .....
I ..... .....
..... ..... . . .....
..... ..... I .....
1.' . ..... ..... .....
I ..... ..... .....
..... ..... ..... .....
..... ..... ..... a.1 . .....
1.. . ..... ..... ..... I .....
I ..... ..... ..... .....
..... ..... ..... .....
..... ..... ..... .....
1 . ..... ..... ..... ..... . . .....
I ..... ..... ..... ..... ./ .....
..... ..... ..... ..... .....
..... ..... ..... ..... .....
0.' . ..... ..... ..... ..... .....
I ..... ..... ..... ..... 1.1 . .....
..... ..... ..... I .....
..... ..... ..... ..... .....
0.1 . ..... ..... ..... ..... ..... .....
I ..... ..... ..... ..... ..... .....
..... ..... ..... 1 . ..... ..... .....
..... ..... ..... ..... I ..... ..... .....
0.. . ..... ..... ..... ..... ..... ..... .....
I ..... ..... ..... ..... ..... ..... .....
..... ..... ..... ..... ..... ..... .....
..... ..... ..... ..... 0.' . ..... ..... .....
0.1 . ..... ..... ..... ..... I ..... ..... .....
I ..... ..... ..... ..... ..... ..... .....
..... ..... ..... ..... ..... ..... .....
..... ..... ..... ..... ..... ..... .....
--------------------
1-1' .O-It "-30 ~-.o -u
AKIB CORR
I .....
.....
. . .....
I .....
.....
.....
1.' . .....
I .....
I ..... .....
..... .....
1.1 . ..... 1.1. .....
I ..... I .....
..... .....
..... .....
1.. . ..... 1.4 . .....
I ..... I .....
..... ..... .....
..... ..... .....
1.2 . ..... ..... 1.' . .....
I ..... ..... I ..... .....
..... ..... ..... ..... ..... .....
..... ..... ..... ..... ..... .....
1 . ..... ..... ..... ..... 1 . ..... ..... ..... .....
I ..... ..... ..... ..... I ..... ..... ..... .....
..... ..... ..... ..... ..... ..... ..... .....
..... ..... ..... ..... ..... ..... ..... .....
0.' . ..... ..... ..... ..... 0.1 . ..... ..... ..... .....
I ..... ..... ..... ..... I ..... ..... ..... .....
..... ..... ..... ..... ..... ..... ..... .....
..... ..... ..... ..... ..... ..... ..... .....
0.1 .. ..... ..... ..... ..... 0.1 . ..... ..... ..... .....
I ..... ..... ..... ..... I ..... ..... ..... .....
..... ..... ..... ..... ..... ..... ..... .....
..... ..... ..... ..... ..... ..... ..... .....
0.' . ..... ..... ..... ..... 0.. . ..... ..... ..... .....
I ..... ..... ..... ..... I ..... ..... ..... .....
..... ..... ..... ..... ..... ..... ..... .....
..... ..... ..... ..... ..... ..... ..... .....
O.J . ..... ..... ..... ..... 0.' . ..... ..... ..... .....
I ..... ..... ..... ..... I ..... ..... ..... .....
..... ..... ..... ..... ..... ..... ..... .....
..... ..... ..... ..... ..... ..... ..... .....
----------- --------------
0-1. 10-11 la-It _0 1-' 10-11 ,.-10
GAO GARF
2-4.
in
Plots of
Table 2-7.
relative
risk
against
exposure
for
-------
Figure
I .....
.....
.....
'.1 . .....
I .....
.....
.....
.....
, . ..... .....
I ..... .....
..... .....
..... .....
..... .....
1.S . ..... .....
I ..... ..... .....
..... ..... .....
..... ..... .....
..... ..... .....
I . ..... ..... ..... .....
I ..... ..... ..... .....
..... ..... ..... .....
..... ..... ..... .....
..... ..... ..... .....
0.$ . ..... ..... ..... .....
I ..... ..... ..... .....
..... ..... ..... .....
..... ..... ..... .....
..... ..... ..... .....
1.' . .....
I .....
.....
.....
1.' . .....
I .....
.....
.....
1.. . .....
I .....
.....
.....
1.' . ..... .....
I ..... ......
..... ......
..... .....
1 . ..... ..... ...."
I ..... ..... ...."
8l1li... ..... ....<\
..... ..... ....11
0.1 . ..... .....
I ..... .....
..... ..... ......
..... ..... ......
0.' .. ..... .....
I ..... ..... .....,
..... ..... .....
..... ..... .....
O.t . ..... ..... ....,
I ..... .....
..... ..... .....
..... ..... .....
0.' . ..... ..... .....
I ..... ..... .....
..... ..... .....
.....
---------_._-------
1-'0 -u
HUMB
-------------------------------------------------
1-'
10-11
.-20
GENG
I .....
.....
.....
2.2 . .....
I .....
.....
.....
, . .....
I .....
.....
.....
1.. . .....
I ..... .....
..... ......
..... .....'
I.. .. ..... ......
I ..... I ..... .....
, . ..... ..... .....'
I ..... ..... .....
..... l.t . ..... .....
..... I ..... .....
.-. ..... ..... .....
1.1. -.. ..... ..... .....
I ..... ..... 1.' . ..... .....
..... ..... I ..... .....
..... ..... ..... .....
..... ..... ..... ..... .....
, . ..... ..... 1 . ..... ..... ..... .....
I ..... ..... I ..... ..... ..... .....
..... ..... ..... ..... ..... .....
..... ..... ..... ..... ..... .....
..... ..... 0.1 . ..... ..... ..... .....
1.' . ..... ..... I ..... ..... ..... .....
I ..... ..... ..... ..... .....
..... ..... ..... ..... .....
..... ..... 0.' .. ..... ..... ..... .....
..... ..... I ..... ..... .....
1 . ..... ..... ..... ..... ..... .....
I ..... ..... ..... ..... ..... ..... .....
..... ..... ..... O.t . ..... ..... ..... .....
..... ..... ..... I ..... ..... ..... .....
..... ..... ..... ..... ..... ..... .....
0.1.. ..... ..... ..... ..... ..... ..... .....
I ..... ..... ..... 0.' . ..... ..... .....
..... ..... ..... I ..... ..... ..... .....
..... ..... ..... ..... ..... ..... .....
..... ..... ..... ..... ..... .....
---....-------- --------------------------------..
... I-II -'0 1"10 11-20 >-21
INOU KOO
2-4.
Continued.
-------
Figure
1.1 i
I i
1.' i
1.' i
I..' .
I
1.1 i
1 i
..1 i
..1 i
... i
..1 i
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
1 .
I
0.. .
I
a.1 .
I
a.. .
I
0.2 .
I
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
------------..--...",
-
IIGII
LEE
I
. i
Ii
. i
. .
I
I i
1 i
--------..-...--..............
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
-.....------------..----.
I
I.. i
'.1 i
I i
1.' i
1.' .
I
1.' i
,.. i
, i
0.8 .
I
..1 i
... i
..1 i
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
LAMT
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
1-10
11-10
..u
I
'.1 i
I i
1.1 .
I
, i
..1 i
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
..e..
..u
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
. .....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
1-11
PERS
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.....
-----------
--------------------------------.......----...
I..
2-4.
1-10
TRIC
Continued.
>-11
.
1-30
21-40
4,-..0
VARE
.. -I.
-------
REFERENCES FOR CHAPTER 2 AND APPENDIX
Akiba, S.; Kato, H.; Blot, W.J. (1986) Passive smoking and lung cancer among Japanese
women. Cancer Research 46:4804-4807.
Balter, N.J.; Schwartz, S.L.; Kilpatrick, S.J.; Witorsch, P. (1986) Causal relationship
between environmental tobacco smoke and lung cancer in non-smokers: a critical review of
the literature. Proceedings of the Air Pollution Control Association, 79th annual meeting.
Blot, W.J.; Fraumeni, J.F. (1986) Passive smoking and lung cancer. J Natl Cancer Inst
77(5):993-999.
Blot, W.J.; McLaughlin, J.K. (1985) "Practical issues in the design and conduct of case-
control studies: use of next-of-kin interviews." In: Statistical Issues in Cancer Epidem-
iology. (Blot, W.J., Hirayama, T., Huel, O.G., eds.) Hiroshima: Sanei, pp. 46-62.
Breslow, N.E.; Day, N.E. (1980)
Scientific Publications No. 32.
Statistical methods in cancer research.
Lyon: IARC
Brownson, R.C.; Reif, J.S.; Keefe, T.J.; Ferguson, S.W.; Pritzl, J.A. (1987) Risk factors
for adenocarcinoma of the lung. Am J EpidemioI125:25-34.
Buffler, P.A.; Pickle, L.W.; Mason, T.J.; and Contant, C. (1984) The causes of lung
cancer in Texas. In: Lung Cancer: Causes and Prevention. (Mizell, M. and Correa, P.,
eds.) New York: Verlag Chemie International, pp. 83-99.
Chan, W.C.; Fung, S.C. (1982) "Lung cancer in non-smokers in Hong Kong." In: Cancer
Campaign, Vol. 6, Cancer Epidemiology. (Grundmann, E., ed.) Stuttgart: Gustav Fischer
Verlag, pp. 199-202.
Copeland, K.T.; Checkoway, H.; McMichael, A.J.; Holbrook, R.H. (1977) Bias due to
misclassification in the estimation of relative risk. Am J Epidemiol105(5):488-95.
Correa, P. (1986) "The passive smoking-cancer controversy." In: Mechanisms in Tobacco
Carcinogenseis. Banbury Report 23:343- 360.
Correa, P.; Fontham, E.; Pickle, L.; Lin, Y.; Haenszel, W. (1983) Passive smoking and
lung cancer. Lancet 2:595-597.
Cummings, K.M.; Markello, S.J.; Mahoney, M.; Bhargava, A.K.; McElroy, P.D.; Mar-
shall, J.R. (1986) Measurement of current exposure to environmental tobacco smoke. Am
J EpidemioI130(1):122-132.
Dalager, N.A.; Pickle, L.W.; Mason, T.J.; Correa, P.; Fontham, E.; Sternhagen, A.;
Buffler, P.A.; Ziegler, R.G.; Fraumeni, J.F. (1986) The relation of passive smoking to
lung cancer. Cancer Research 46:4808-4811.
Doll, R.; Peto, J. (1986) Passive smoking. Br J Cancer 54:1019- 1021.
Eriksen, M.P.; LeMaistre, C.A.; Newell, G.R. (1988) Health hazards of passive smoking.
Ann Rev Public Health 9:47-70.
-------
Gao, Y.; Blot, W.J.; Zheng, W.;Ershow, A.G.; Hsu, C.W.; Levin, L.1.; Zhang, R.;
Fraumeni, J.F. (1987) Lung cancer among Chinese women. Int J Cancer 40:604-609.
Garfinkel, L. (1981) Time trends in lung cancer mortality among nonsmokers and a note
on passive smoking. J Natl Cancer Inst 66:1061-1066.
Garfinkel, L. (1984) Passive smoking and cancer: American experience. Prev !vIed 13:691-
697.
Garfinkel, L. (1985) Selection, follow-up, and analysis in the American Cancer Society
Prospective Studies. Presented at a Workshop on the Selection, Follow-up, and Analysis
in Prospective Studies held at the Waldorf-Astoria Hotel, New Yor, N.Y., October 3-5,
1983.
Garfinkel, L.; Auerbach, 0.; Joubert, L. (1985) Involuntary smoking and lung cancer: a
case-control study. J Natl Cancer Inst 75:463-469.
Geng, G.; Liang, Z.H.; Zhang, G.1. (1988) "On the relationship between smoking and
female lung cancer." In: Smoking and Health, Elsevier Science Publishers, pp. 483-86.
Gillis, C.R.; Hole, D.J.; Hawthorne, V.M.; and Boyle, P. (1984) The effect of
environmental tobacco smoke in two urban communities in the west of Scotland. Eur J
Respir Dis 133(Supp1.):121-126.
Hammond, E.C.j Garfinkel, 1. (1975) Aspirin and coronary heart disease: findings of a
prospective study. Br Med J 2:269- 271.
Hammond, E.C.; Garfinkel, L.; Seidman, H.; Lew, E.A. (1976) 'Tar' and nicotine content
of cigarette smoke in relation to death rates. Environmental Research 12:263-274.
Hirayama, T. (1981a) Non-smoking wives of heavy smokers have a higher risk of lung
cancer: a study from Japan. Br Med J 282: 183-185.
Hirayama, T. (1981b)' (Letter) Br Med J 283:916-917.
Hirayama, T. (1983a)
Lancet 2:1425-1426.
Passive smoking and lung cancer: consistency of association.
Hirayama, T. (1983b) Passive smoking and lung cancer. Presented at the 5th World Con-
ference on Smoking and Health, Winnipeg, Canada.
Hirayama, T. (1984) Cancer mortality in nonsmoking women with smoking husbands
based on a large-scale cohort study in Japan. Prev Med 13:680-690.
Hollander, M.; Wolfe, D.A. (1973) Nonparametric statistical methods. New York: John
Wiley and Sons.
Humble, C.G.; Samet, J.M.; Pathak, D.R. (1987) Marraige to a smoker and lung cancer
risk. Am J Public Health 77:598-602.
Inoue, R.; Hirayama, T. (1988) "Passive smoking and lung cancer in women."
Smoking and Health. Elsevier Science Publishers, pp. 2~3-285.
In:
-------
International Agency for Research on Cancer (IARC). (1987) Environmental carcinogens
methods of analysis and exposure measirement. Volume 9: passive smoking. (O'Neill, I.
K., Brunnemann, K. D., Dodet, B. and Hoffmann, D., eds.) Lyon: IARC Scientific
Publications No. 81.
Kabat, G. C.; Wynder, E. L. (1984) Lung cancer in nonsmokers. Cancer 53:1214-1221.
Kilpatrick, S.J. (1987) Misclassification ot environmental tobacco smoke exposure: its
potential influence on studies of environmental tobacco smoke and lung cancer.
Toxicology Letters 35:163-168.
Knoth, A.; Bohn, H.; Schmidt, F. (1983) Passive smoking as a causal factor of bronchial
carcinoma in female nonsmokers. Medizinisch Klin 78:66-69. .
Koo, L.C.; Ho, J.H.; Saw, D.; Ho, C.Y. (1987) Measurements of passive smoking and esti-
mates of lung cancer risk among non-smoking Chinese females. Int J Cancer 39:162-169.
Kuller, L.H.; Garfinkel, L.; Correa, P.; Haley, N.; Hoffmann, D.; Preston-Martin, S.;
Sandler, D. (1986) Contribution of passive smoking to respiratory cancer. Environ Health
Persp 70:57-69.
Lam, T. H.; Kung, LT. M.; Wong, C.M.; Lam, W.K.; Kleevens, J.W.L.; Saw, D.; Hsu, C.;
Seneviratne, S.; Lam, S.Y.; Lo, K.K.; Chan, W.C. (1987) Smoking, passive smoKing and
histological types in lung cancer in Hong Kong Chinese women. Br J Cancer 56:673-378.
Lam, W.K. (1985) A clinical and epidemiological study of carcinoma of lung in Hong
Kong. Thesis submitted to the University of Hong Kong for the degree of Doctor of
Medicine.
Leaderer, B.P. (1988) Measuring exposure to environmental tobacco smoke.
prepared for the U. S. Environmental Progection Agency.
Lehnert, G. (1984) Roundtable discussion from Symposiom "Medical Perspectives on Pas-
sive Smoking." Prev Med 13:732- 733.
Report
Lee, P.N. (1987a) Lung cancer and passive smoking: association an artefact due to mis-
classification of smoking habits? Toxicology Letters 35:157-162.
Lee, P.N. (1987b) Passive smoking and lung cancer association: a result of bias? Human
Toxico16:517-524.
Lee, P.N. (1988)
Springer.
Lee, P.N.; Chamberlain, J.; Alderson, M.R. (1986) Relationship of passive smoking to
risk of lung cancer and other smoking- associated diseases. Br J Cancer 54:97-105.
Misclassification 01 smoking habits and pass~ve smoking.
Berlin:
Mantel, N. (1963) Chi-square tests with one degree of freedom: extensions of the Mantel-
Haenszel procedure. JAm Statist Assoc 58:690- 700.
Mantel, N. (1987a) Lung cancer and passive smoking. (Letter) Br Med J 294:440-441.
Mantel, N. (1987b) Re: Passive smoking and lung cancer in. Swedish women (Letter) Am
J Epidemio1126(6):1214-1216.
-------
Mantel, N.; Haenszel, W. (1959) Statistical aspects of the analysis of data from retrospec-
tive studies of disease. J Natl Cancer Inst 22:719-748.
Meier, P. (1987) Commentary. Statistics in Medicine 6:329-331.
Miller, G.H. (1984) Cancer, passive smoking and nonemployed and ~mployed WIves.
Western J Med 140(4):632-635. .
Moolgavkar, S.H.; Dewanji, A.j Luebeck, G. (1989) Cigarette smoking and lung cancer:
reanalysis of the British doctors data. J Natl Cancer Inst 81(6):415-420.
National Research Council (NRC) (1986). Environmental tobacco smoke.
DC: National Academy Press. .
Washington,
Pershagen, G.; Hrubec, Z.; Svensson, C. (1987)
Swedish women. Am J EpidemioI125(1):17-24.
Remmer, H. (1987) Passively inhaled tobacco smoke: a challenge to toxicology and pre-
ventive medicine. Arch ToxicoI61:89-104.
Passive smoking and lung cancer in
Riboli, E. (1987) Passive smoking and lung cancer: epidemiological evidence and ongoing
international collaborative studies. Toxicology Letters 35:19-27.
Sackett, D.L. (1979) Bias in analytic research. J Chronic Dis 32(1-2):51-63.
Samet, J.M. (1988a) Involuntary exposure to tobacco smoke. Annals of Sports Medicine
4(1):1-15. .
Samet, J.M. (1988b) Environmental tobacco smoke and cancer. Report prepared for the
U.S. Environmental Protection Agency.
Sandler, D.P.; Everson, R.B.; Wilcox, A.J. (1985)
cancer risk. Am J EpidemioI121(1)37-48.
Saracci, R. (1989) Passive smoking and cancer risk. IRAC report of panel of experts.
Prepared at the request of the European School of Oncology through the Europe Against
Cancer program of the European Economic Community.
Passive smoking in adulthood and
Saracci, R.; Riboli, E. (1989) Passive smoking and lung cancer: current evidence and
ongoing studies at the International Agency for Research on Cancer. Mutation Research
222:117-127.
Schlesselman, J.J. (1982) "Case-control studies: design, conduct, analysis." In:
Monograpns in Epidemiology and Biostatistics. New York: Oxford University Press, p.
190. .
Shimizu, H.j Morishita, M.; Mizuno, K.; Masuda, T.; Ogura, Y.; Santo, M.; Nishimura,
M.j Kunishima, K.; Karasawa, K.; Nishiwaki, K.j Yamamoto, M.; Hisamichi, S.;
Tominaga, S. A case-control study of lung cancer in nonsmoking women. (1988) Toboku
J Exp. Med 154:389-397.
Svensson, C.; Pershagen, G.j Klominek, J. (1988) Smoking and passive smoking in rela-
tion to lung cancer in women. Department of Epidemiology, National Institute of Environ-
mental Medicine, Stockholm, Sweden.
-------
Trichopoulos, D. (1988) Passive smoking and lung cancer. Scand J Soc Med 16:75-79.
Trichopoulos, D.; Kalandidi, A.; Sparros, 1. (1983) Lung cancer and passive smoking: con-
clusion of Greek study. (Letter) Lancet 667-668.
UberIa, K. (1987) Lung cancer from passive smoking: hypothesis or convincing evidence?
Int Arch Occup Environ Health 59:421- 437.
U.S. Surgeon General (SG) (1986) The health consequences of involuntary smoking. U.S.
Department of Health and Human Services, Public Health Service.
Vandenbroucke, J.P. (1988) Passive smoking and lung cancer: a publication bias? Br
Med J 296:391-392.
Varela, L.R (1987) Assessment of the association between passive smoking and lung
cancer. Dissertation to Yale University in candidacy for the degree of Doctor of
Philosophy.
Wald, N.J.; Nanchahal, K.; Thompson, S.G.; Cuckle, H.S. (1986) Does breathing other
people's tobacco smoke cause lung cancer? Br Med J 293:1217- 1222.
Weiss, S.T. (1986) Passive smoking and lung cancer: what is the risk? (Editorial) Am
Rev Respir Dis 133:1-3. .
Weiss, S.T.; Tager, LB.; Schenker, M.; Speizer, F.E.
involuntary smoking. Am Rev Respir Dis 128:933-942.
'VeIls, A.J. (1988a) Re: Passive smoking and lung cancer: a publication bias? (Letter)
Br Med J 296:1128.
(1983) The health effects of
Wells, A.J. (1988b) An estimate of adult mortality in the United States from passive
smoking. Environment International 14:249-265.
Wells, A.J. (1988c) Misclassification as a factor in passive smoking risk. (Letter) Lancet
638.
Woolf, B. (1955) On estimating the relationship between blood group and disease. Ann
Human Genet 19:251-253.
Wu, A. H.; Henderson, B. E.; Pike, M. D.; Yu, M. C. (1985) Smoking and other risk
factors for lung cancer in women. J Natl Cancer Inst 74(4):747-751.
Yusuf, S.; Peto, S. R; Lewis, J.; Collins, R; Sleight, P. (1985) Beta blockade during and
after myocardial infarction: an overview of the randominized trials. J Prog Cardiovasc
Dis 27:335-371.
-------
APPENDIX
SUMMARY DESCRIPTIONS OF ELEVEN CASE CONTROL STUDIES
BROW. The case control study of risk factors for adenocarcinoma by Brownson et al.
(1987) includes 23 never smoker cases (19 females) among the 102 cases interviewed.
All subjects were white, had microscopically confirmed cancers incident from 1979 to
1982, and were identified through the Colorado Central Cancer Registry which covers
the five county Denver metropolitan area. In the study as a whole, interviewed cases
represented 68.5% of the 149 cases meeting eligibility criteria.
Controls were chosen
from persons with cancer at sites unassociated with cigarette smoking and were matched
to the cases on age and sex. Of the 169 eligible controls, 131 (77.5%) were interviewed.
Surrogate respondents were required for 69% of cases and 39% of controls.
Passive smoke exposure was analyzed both as a dichotomous variable based on
the smoking status of the spouse and as a stratified variable based on the hours per day
that the subject was in the presence of persons smoking.
Other variables pertain to
previous smoking, education, income, occupation, and residence history as an indirect
measure of exposure to total suspended particulates.
The relative risk for adenocarcinoma among female never smokers exposed four
or more hours per day relative to a lower exposure was 1.68 (95%CI= 0.39-2.97) after
adjustment for age, income, and occupation. Similar nonsignificant risk estimates were
shown when smoking by the spouse was considered as a dichotomous variable. The high
proportion of surrogate source data led the authors to conduct parallel analyses limited
to self-reported data. Results from those analyses were described as highly comparable
and indicated possibly higher risks than those reported based for all respondents"
A-I
-------
Note: The number (19 females) of never smokers in this study is much too small
to make even a large observed relative risk (1.68) statistically significant. Further,
combining ever smokers and never smokers (possibly to increase the sample size) makes
the results difficult to compare with previous findings. After further a.djusting the
analysis for education and income, in addition to previous smoking habits and age, there
was no indication of a potential association of lung cancer with ETS exposure.
GAO. Gao et al. (1987) report the results of a large (1407 subjects) population-based
case control study of lung cancer etiology in Shanghai China, where lung cancer rates
for women are among the highest in the world.
Potential cases included all female
patients with newly diagnosed primary lung cancer incident between February, 1984,
and February, 1986, who were 35 to 69 years of age at the time of diagnosis and were
residents of urban Shanghai. After exclusion of 93 patients who died, all remaining 672
cases were interviewed. Eighty-one percent were diagnosed by tissue biopsy or cytology
and 19 percent by repeated x-ray. Adenocarcinoma was the predominant (61 %) diag-
nosIs.
Controls were frequency matched within five-year age strata and randomly
selected from the general population of the Shanghai urban area. Of the totaJ of 735
controls interviewed, only 9.7% were secondary controls, chosen mainly because the first
selected control had moved from the Shanghai urban area or was found to be outside
the eligible age range.
The study includes 246 cases and 375 controls who were
nonsmokers (presumably had never smoked cigarettes). Logistic models were used to
estimate relative risks of disease" adjusted for other study factors.
Among all subjects no significant increase in risk was observed for overall ETS
exposure during childhood (OR [odds ratio]=1.1, 95% CI=O.7-1.7) or adult life
(OR=O.9, 95% CI=O.6-1.4). For these calculations, exposure was said to have occurred
if the subject had ever lived with a smoker. However, when exposure was defined in .
A-2
-------
terms of husbands' smoking and the analysis was limited to nonsmoking women, lung
cancer risks tended to increase with the number of years of exposure, with the highest
observed risk (OR=1.7, 95%CI = 1.0-2.9) occurring in the comparison of those with 40+
years of exposure to those with 20- years exposure, after adjustment for age and
education (p. 605, Table II) The relative risk in this comparison was higher (OR=2.9,
95% CI=1.0-8.9) for squamous-and-oat cell carcinoma alone.
No test for trend over
levels of ETS exposure was reported.
In the discussion of the results, the authors note the upward trend in risk
associated with increasing years of exposure among nonsmoking women married to.
smokers. They conclude that ETS may be a contributing causative agent, but that
other factors need to be considered as well, e.g., pre-existing lung disease, hormonal
conditions, and especially exposure to cooking oil vapors.
Note: The study was not undertaken specifically to look at ETS lung association.
Despite the large number of nonsmokers, it was not possible (or the authors chose not)
to use women married to nonsmokers as a comparison group in their Table II. That
may have been necessitated by the high prevalence of cigarette smoking among Chinese
males.
GENG. In a brief article describing work similar in design and purpose to Gao et al,
Geng et al. (1987) report the results of their study of lung cancer risk factors among
women living in Tianjin, where the rates of lung cancer mortality are the highest in
China. All 157 female cases were resident in Tianjin for at least ten years and were
pair-matched to 157 controls for sex, race, age (within 2 years) and marital status.
Diagnosis was predominantly by histologic or cytologic review (84.7%), although com-
puterized tomography (10.8%) and clinical or x-ray (4.5%) methods were also used to
identify cases. The authors describe the case group as representative of Tianjin female
A - 3
-------
lung cancer patients in terms of age and distribution of residents. They further state
that the prevalence of smoking among the controls (40.8%) is similar to that seen
among the Tianjin adult female population. The participation rates for cases and
controls is not given, but other studies from China have reported very high response
rates.
The study report available in the literature is fairly brief. Neither the method for
assigning ETS exposure nor information about personal smoking status are discussed.
Both multiple conditional regression and stratified analytic techniques were used to
calculate reported risk estimates, but the authors do not stipulate which variables were
controlled for the analyses.
The authors report that among the odds ratios of paSSIve smoking from
husbands, fathers, mothers, and colleagues, only that from husbands is significant.
However, it is not clear whether this amount applies to smokers and nonsmokers
combined in the same analysis or whether the analyses of ETS were restricted to
nonsmokers only. The authors do explicitly state in Table 5 that the odds ratio for lung
cancer in nonsmoking women married to smokers is 2.16 (95%CI=1.05-4.53), but it is
not clear why this estimate differs from the odds ratio of 1.86 for nonsmoking wives
.~
with smoking husbands in Table 7. The odds ratios for lung cancer increase with the
number of cigarettes smoked per day by the husband and the duration of exposure to
the husband's smoking (Table 6).
No tests for trend are provided, however, and
whether these findings apply to all subjects as a group or only to the nonsmokers is not
clear.
One interesting finding in Table 7 of this brief report is the similarity of esti-
mated effects associated with ETS exposure from a husband only (OR=1.86, 95% CI=
1.04-3.5) and active smoking by the wife only (OR=2.61, 95% CI=1.4-4.6). Further,
A-4
-------
these independent risks can be seen to interact on a multiplicative scale among smoking
women married to smoking husbands (OR=4.9, 95% CI=1.8-9.5).
Whether these
estimates include adjustment for other factors is not stated.
HUMB.
The study by Humble and colleagues (Humble, et aI, 1987) includes 28
incident cases described by interview to be lifelong nonsmokers (8 men, 20 women).
Cases were identified through the population-based New Mexico Tumor Registry while
controls. (130 men, 162 women) were chosen through randomly generated phone
numbers and Health Care Financing Administration rosters of Medicare participants.
Controls were frequency-matched to cases by ten-year age groups and by sex. Subjects
were the nonsmoking subset in a larger study of lung cancer risk factors in which 88.5%
of cases and 83.1% of controls eligible for interview had participated. Of the 28 lung
cancers among nonsmokers, 24 had a histologic diagnosis in the Tumor Registry record.
However, in a separate review of histologic materials for 17 of these cases, only eight cell
types concurred with the Registry.
Subjects or their proxies were interviewed regarding their personal smoking
habits, smoking by their spouses(s), and their occupational exposures. Surrogate inter-
views (usually with the spouse) were necessary for 19 of the 28 cases, but for only 13 of
the 292 controls. No effect of information source was noted when analyses were run sep-
arately for self-reported and surrogate-reported cases using self-reported controls as the
comparison group.
Small numbers precluded a separate reporting of the odds ratio
(OR) for males.
An elevated risk of lung cancer was reported for all subjects combined and for
females separately. In logistic models, including adjustment for age and ethnicity and
sex when appropriate, the ORs are 2.6 (90% CI = 1.2-5.6) for all subjects and 2.2
(CI=0.9-5.5) for females.
~sk increased with the duration of spousal smoking (chi-
A - 5
-------
..4..J':" f - V'~ tJ<" ""~ r U Jfd.
squared statistic for trend equals 2.01 for all subjects and 1.23 for females alone) in
cross-tabular analyses, but not in results from multiple logistic models. No trend was
seen over the average number of cigarettes smoked per day by the spouse. Separate
analyses for current and former smokers r.evealed no increased risk associated with
marriage to a smoker.
Cell-line specific analyses were precluded by the small number of cases with
histologic confirmation of their diagnosis, the poor concordance of histologic designa-
tions in the Registry file, and the special review. The high proportion of cases with surro-
gate respondents may actually have improved the quality of data regarding exposure to
a spouse's cigarette smoking, as spouses were the principal source of surrogate data.
Exclusion of four former smokers (by information from other sources) did not alter the
results. Size of the case series allowed only crude stratification of duration and amount
when testing for trends, and may explain the marginal significance of findings reported
separately for women.
INOU. In a large case control study of smoking and lung cancer in two Japanese cities,
Inoue and Hirayama (1987) identified 37 women who died with lung cancer. Twenty-
eight of these women (75.7%) were nonsmokers (definition not given).. Cases were
matched for age, year of death (within 2.5 year~), and residential district to 74 controls
who had died of cerebrovascular disease. Sixty-two (83.8%) of the controls were nonsmo-
. . . jf-4 ~]3P2
kers. Husbands' smoking status was available for 29 of the 37 cases and 54 of the 74
controls.
Interviews were used to gather dat~ for analysis, but the authors do not
describe the characteristics or degree of relatedness of the surrogate respondents.
Neither do they describe the degree of cooperation among the study subjects.
Mantel-Haenzsel odds ratios were used to estimate the relative risks of disease
associated with ETS, adjusted for age alone and for age and residential district
A - 6
-------
(necessary given the different socio-economic natures of the two areas). With stratifi-
cation for both age and district, the odds ratios are 2.58 (90% CI=0.44-5.7) when
husbands smoked less than 19 cigarettes a day, and 3.09 (90% CI=1.04-11.81) when hus-
bands smoked 20 or more cigarettes a day. The chi-squared test for trend is significiant
"\
(p<0.05).
LAMT. The large case control study by T.H. Lam and colleagues (Lam et al" 1987) was
undertaken to assess the respective roles of active and passive smoking in lung cancer
etiology among women living in Hong Kong.
Only patients with a pathologist's
confirmation (98% by histological or cytological review) were included. Those with rare
tumors, e.g., carcinoids, were excluded. Women were interviewed in hospital and then
age-matched to healthy female controls selected from within their own neighborhoods.
Interviews took place between 1983 and 1986 and approximately 99% of all eligible
subjects responded.
Never smoker status for both subjects and their husbands was defined as having
never smoked as much as one cigarette a day, or its equivalent in other tobacco
products, for at least one year.
A woman was 'considered exposed to her husband's
tobacco smoke if she had lived with her smoking husband in the same household
continuously for at least one year. If the husband was an ever- smoker, information on
the type of tobacco and amount usually smoked per day by the husband and the
duration of exposure was obtained. Never-married women were included as nonexposed
to ETS. The authors describe the results of separate analyses on cigarettes only, and on
all forms of tobacco, as similar and only report the latter. Relative risk (R) and 95%
confidence intervals were calculated for each level of ETS exposure. Fisher's exact test
(two-sided) was used to check whether the relative risk was significantly different from
unity. Multivariate methods do no appear to have been used.
A - 7
-------
Among the total of 444 cases and 443 controls were 199 cases and 335 controls
who had never smoked and for whom data on husbands' smoking were available. For
never smokers the relative risk for lung cancer of all types from ETS exposure Ms 1.165
(95% CI=1.16-2.35); for adenocarcinoma the relative risk is 2.12 (95% CI=1.32-3.39)7
The risks for small and large cell carcinomas are 3.00 and 3.11, respectively, but these
estimates are not statistically significant. Trends in relative risk for cancer at all sites,
and for adenocarcinoma by the amount of tobacco smoked daily by the husband, are
both significant with P<0.001. The authors discount the possibility that misclassifica-
tion bias could have lead to the observed results, given the low prevalence of smoking
(4.1%) among women in Hong Kong and the strength of the findings in the present
study.
LAMW. The dissertation of Lam (Lam,W.K., 1985) was the third case control study of
risk factors for lung cancer among females in Hong Kong. The nonsmoker cases, all^
with histologic or cytologic confirmation of adenocarcinoma, were part of a larger case,.
series of 161 interviewed Chinese female lung cancer patients diagnosed at a large,
regional general hospital with disease incident between January 1981 and April 1984.
Fifteen cases with three other lung cancer histologies, as well as any patients with^
metastatic disease, were not included. Nonsmoking controls (n=144) were part of a'
larger series of 185 Chinese, mostly lower income female patients admitted to the
^
orthopedic wards between 1982 and 1984. Cooperation of potential subjects exceeded
99%.
There was little difference in the ages, occupations, years of schooling, or recent
residences of the 161 cases and 185 controls, so the author deemed it unnecessary to
control for (stratify on) these variables in the analysis of the 60 nonsmoking cases with
adenocarcinoma and 144 nonsmoking controls. Exposure to ETS was categorized
A-8
-------
separately for husbands and other sources, e.g., cohabitating relatives or coworkers. Sub-
jects were also queried regarding exposure to smoke from kerosene stoves and incense.
The author interviewed all cases and, with a single research assistant, all controls. Thus,
one may assume that interviews were not "blind".
The strongest and most statistically significant associations of ETS were with
peripheral adenocarcinoma, with the highest odds ratio (2.64) occurring when exposure
was based solely on husbands' smoking behavior.
Relative risks of 1.6 and 1.7 were
found for centrally located tumors when ETS was based on the husband's habits and
total exposure to passive smoking, respectively. When data from Table 7.5 of the study
are summed over sites, relative risks of approximately 2.0 are obtained with P<0.05,
regardless of exposure classification scheme. All odds ratios appear to be unadjusted for
any other study factors. No statistically significant risks from kerosene or incense were
found. The author concludes that the small sample size and use of only a single hospital
source for subjects are limitations.
SHIM. Shimizu and his colleagues (Shimizu et al., 1988) use a hospital-based case-
control study of lung cancer in women to examine the effect of involuntary exposure to
tobacco smoke from a variety of sources. Among 118 female patients with histologically
confirmed lung cancer, 90 reported having never smoked cigarettes.
Cases were
matched on hospital, age (within 1 year), and date of admission to patients being seen
for conditions generally unrelated to tobacco.
All subjects were asked to complete a
questionnaire about occupational history, kinds of fuels used for cooking and heating,
and smoking habits, including number of cigarettes smoked daily by parents, siblings,
the husband, and the husbands' parents in the home, as well as the amount of time
spent in the same room with the husband, and the duration of marriage. ETS exposure
at work was simply categorized by presence or absence of smokers.
A-9
-------
No association was observed between risk of lung cancer and smoking by
husbands, fathers, siblings, or coworkers. However, increased odds ratios were seen for
smoking by subjects' mothers (OR=4.0, P<0.05) and by their husbands' fathers
(OR=3.2, P<0.005). Dose-response relationships were not apparent for exposure by the
mother or the husband's father, but the authors suggest that subjects may ha.ve been
unable to recall the exact number of cigarettes in some cases (especially in childhood).
It is not clear whether variables such as occupational exposure to iron and other
metals, or type of heating fuel, were assessed. Neither is there mention of cooperation
rates by cases and controls. Adjustment of odds ratios for smoking by mother, smoking
by husbands' father, and occupational exposures to iron and other metals, caused
modest reductions in the point estimates, although smoking by husband's father in the
home, (adjusted OR=3.2) is still significant with P<0.005. The authors describe this
association as plausible since a high proportion of Japanese wives live with their in-laws
after marriage and their father-in-law may have already retired.
SVEN. The study of lung cancer etiology in women by Svenson et al. (1988) includes
34 cases with microscopically confirmed non-carcinoid cancer who had never been
regular smokers. Cases were patients referred to one of four clinical departments that
diagnose or treat lung cancer in Stockholm county, Sweden. Only patients who would
not benefit from specialist care, or who were not in physical or mental condition to
allow an interview, were excluded from eligibility. Cases were matched on age using
random selection from the population register in Stockholm County.
For the whole
study only seven subjects refused to be interviewed, resulting in a sample of 210 cases
and 209 controls.
Cooperation of nonsmoking cases and their matched controls was
presumably high as well.
A - 10
-------
Four physicians completed all interviews usmg a structured questionnaire that
included ETS exposure during childhood, as well as domestic and work environment
exposure during adulthood. Other questions concerned the consumption of foods rich in
vitamins A and C, and information about the dwellings where a subject had lived for
more than two years. No surrogate sources of information were used and squamous/
small cell carcinomas constituted, respectively, 57.9% and 20.6% of the case histologies.
Women who lived with a smoking mother as children (R=3.3), or were exposed
to ETS both at home and at work (R=2.1), or were exposed both as children and as
adults (R=1.9), showed the highest risks.
However, all estimates had very wide
confidence intervals owing to the small sample size, and tests of association between
ETS exposure and lung cancer incidence and tests for trend were all nonsignificant.
The authors describe the results for ETS as inconclusive, but note that most
estimates of relative risk are greater than unity. The statistical power. to detect an
increased risk of 50% from exposure to ETS was only about 0.1. The author suggests
that information bias may have precluded the identification of statistically significant
small increases in risk. Specifically, no information on the duration or intensity of ETS
.;
exposure was obtained in the study, so it was difficult to assess the relative importance
of domestic and workplace exposures.
V ARE. The case control study described by Varela in his 1987 dissertation (Varela,
1987) is based on 439 histologically confirmed primary lung cancer cases incident in
nonsmokers over an 18 month period in upstate New York. Sample size requirements
were set large enough that detection of a relative risk of the size reported by Hirayama
and Trichopoulos would be likely. However, to reach the calculated requirement of 450
matched case control pairs, it was necessary to include former smokers (55% of sample)
in addition to never smokers. Cases were identified through a special rapid reporting
A-ll
-------
system in all participating hospitals and through periodic review of the New York State
Cancer Registry. Controls were matched to cases on residence, age (within 5 years),
sex, smoking history, and whether the interview was with the subject (67%) or with a
surrogate for the subject (33%). Standardized interviews were conducted to collect data
describing exposure to a spouse's cigarette smoke in terms of cigarettes/day, total years
of smoke exposure, and total cigarettes smoked during the marriage. Information was
also collected on total exposure from all smokers in the household, from coworkers on
the job, and from exposure in social circumstances. The potentially confounding varia-
bles considered in the analysis include religion, income, marital status, other
occupational exposures, and number of cigarettes smoked/day for former smokers. The
study's total of 439 cases represents a cooperation rate of 84% among those selected for
interviews. Approximately two potential controls had to be called per case to obtain
enough study controls.
The author provides a systematic and exhaustive analysis based on linear logistic
models for pairwise matched data. These data were collected as continuous values to
allow analysis by source of exposure, e.g., spouse, other household smokers, coworkers,
and social encounters, using methods for both continuous data and for categorical data.
Analysis of household exposure was further complicated by the use of two alternative
assumptions regarding missing data for exposure at previous residences.
After extensive analyses no index of exposure to spouse's tobacco or smoking by
coworkers was associated with an increased risk for lung cancer. However, person-years
of total exposure from all smoking household members showed a statistically significant
linear trend. When exposure was fitted as a continuous variable, the unadjusted odds
ratio associated with 150 person-years of exposure was 1.86 (95%CI=1.22-2.83). Adjust-
ment for the potentially confounding variables listed above reduced the estimated OR
for 150 person-years of exposure to 1.56 (95% CI=1.00-2.41). Exposure to passive smoke
A-12
-------
In
social situations showed an anomalous protective effect in both adjusted and
unadjusted models (Tables 20,21,22 and Figures 25-28).
Note: The study contains extensive statistical analyses of which only a small part
have been described here. When a large number of tests are made, the likelihood that
one or more statistically significant result will occurr by chance alone increases. This
can cause results to be interpreted as more significant than may be justified.
The author suggests that his own finding of no effect from exposure to spouses'
smoke is understandable because the smoking habits of a spouse may not accurately
describe true exposure to passive smoke. By contrast, the household exposure variable.
which was designed to more fully capture exposure in the home was the only index that
was associated with increased risk of disease in this study. The greater association of
household exposures with epidermoid and small cell histologies (Tables 12, 13, 15, 16) is
not inconsistent with the apparent specificity of effect observed in PERS and GARF.
One difficulty with comparing the Varela study with other case control studies is the
inclusion of either males with females, or ex-smokers with never smokers, in' the
reported results. Although the analysis is very comprehensive, no reports for the risk of
female never smokers alone were found. The author suggests that differences in past
smoking habits of cases and controls may have a confounding effect. Although identical
proportions of cases and controls were former smokers, cases had smoked a, larger
number of cigarettes per day (28.9 vs. 23.8, p=O.0002).' Former smokers were not
included, however, unless they had stopped smoking at least ten years prior to the
interview. The author questions the validity of an apparent significant protective
association from ETS in social circumstances, suggesting the possiblities of biased
reporting and questionnaire artifacts as alternative explanations for this finding.
A - 13
-------
WU. Wu and her coauthors (Wu et. at., 1985) report the effects of ETS exposure as
. part of a larger study of determinants of lung cancer among white women living in Los
Angeles County. Eligible cases included only patients with microscopically diagnosed
primary adenocarcinoma (ADC) or small cell carcinoma (SCC) of the lung, i.ncident
between April 1, 1981 and August 31, 1982. Subjects also had to be English-speaking
residents and less than 76 years old at the time of diagnosis. One neighborhood control
was individually matched to each interviewed case using date of birth (within five
years) .
From a total of 490 eligible cases, 190 were dead or too ill to participate, eight
could not be located and 44 refused to be interviewed, leaving 220 (44.9%) as the
interviewed case group. After replacement of 85 potential controls who refused t.o partic-
ipate, 220 controls were also interviewed. Surrogate respondents were not used because
they were thought to be an unreliable source of information for ETS exposures and
dietary practices in childhood.
Cases and controls were interviewed by telephone regarding personal smoking
habits, exposure to ETS, history of lung diseases, dietary intake of vitamin A, types of
heating and cooking fuels used, and reproductive history. Information obtained about
childhood exposure to ETS included the amount and years of smoking by fathers,
mothers, and other household members. Questions on exposure in adulthood pertained
to smoking habits of spouses and other household members.
Study data were adjusted for potential confounding variables by application of
logistic regresssion. Estimates for the relative risk of ADC are provided separately for
nonsmokers, ex-smokers, and current smokers, but a small number of occurrences
precluded the corresponding calculations for SCC. For ADC and sce among smokers
and nonsmokers combined, after adjustment for personal smoking habits, no
A -14
-------
significantly increased risks were observed due to smoking by the subject's mother,
father, spouse, or coworkers. For the 29 nonsmoking ADC cases, no significant elevated
risk was associated with ETS exposure from either parent (R=O.6, 95% CI=O.2-1.7),
from spouses(s) (R=1.2, CI=O.5-3.3), or from the workplace (R=1.3, CI=O.5-3.3). The
observed relative risk for ADC increases with the number of years of adult ETS expo-
sure from spouse( s) and cowor,kers, but a test for trend is not statistically significant.
The authors attribute the ambiguous nature of their results to the lesser etiologic
role of ETS for ADC compared to SCC. Further, 12 (41%) of the 29 ADC cases are
bronchoalveolar cell carcinomas, which Correa et al. (1983) found to have a relatively
weaker association with passive smoking.
A - 15
------- |