require\ wme period 0f"re~ular" smoking for an individual to be clawificd as an t`ver
maker. 12% of?i:! individuals reported being neler smoker\. However. when assessed
concurrently ti ith another questionnaire in which regular smoking was not defined and
the respondent self-defined waking. 7 percent fewer subjects t II9 of 252) reported
being never smokers.
Thus, the use of more clearly defined questions. wch as specifying 100 cigarettes in
;1 lifetime. or 1 cigarette per day for I year. or 5 ci_rarettes per week for I year. Mill
reduce misclaGtIcatlon. However. some misclassification will still occur for thaw
individuals who hmohed for relatively brief periods during their lives but cannot
accurately remember hou long they smoked or accurately estimate the number of
cigarettes they smoked.
Attention also must be paid to defining current or former smokers. Some studies.
such as the Cancer Prevention Study I (CPS-1) (Hammond and Garfinkel 1969). define
current smokers as those who respond affirmatively to the question "Have you smoked
within the past year?" Other studies u$e smoking in the past 6 months as the guideline
for current smokers (Coultas et al. 1988). The criteria for questions identifying current
smoker\ can range from having smoked in the past year. to the past 6 months. to the
past week. or to an unspecified period. A few additional questions will enhance the
specificity of the definitions of current smokers and former smokers. These items. or
comparable ones. have been used in previous surveys. for example. the 198X Baseline
Prevalence Survey for the Community Intervention Trial for Smoking Cessation.
funded by the National Cancer Institute: `.At what age did you start smoking on a
regular basis?": "On the average. about hou many cigarettes did you smoke per day
during the lact I?. months you smohed?": and for former smokers. "When did you quit
smoking cigarettes'?" (recorded to exact date if possible). These item\ provide udd-
tional information for defining ever smohers. or stratifying by levels of exposure. and
for determining the period of abstinence.
The dynamic nature of smoking ce\\ation highllghts the importance of being aware
that any categorical definition of former smoker in relation to the health effects of
smohing cessation will include former woher\ who h:r\,e been abstinent for \,arying
period\ of time. Optimally. questions on smohin, ~7 historv should ascertain the duration
                                         _
of abstinence for former \moher\. and if possible. abstinence period\ should be treated
aj continuous or categorized vuriablc\ in an anal>si\. thus avoidins the problem ot
treating former smoher\ ;IS ;1 single group.  Howewr. benefit\ of ce\\ation are still
clearly observed in spite of the limitation\ of using categorical data.
The mo\t common minimum period\ ofabstinencc u$ed for defining former smoking
statu\ are 2-l hours. 7 days. and 30 da\\. The National Interagency Council on Smoking
and Health ( 1973) recommended using ;I minimum of 7 da> s ofab\tinence for defining
cessation. However. becuuw of the nature of mokin g. usin_r ;t short abstinence period
to define former smoher\ i\ not optimal in epidemiologic studies. The degree of
misclassification of former smoker\ M ill depend on the minimum duration of abstinence
u\ed to define former smokers and the criterion wed to consider determine relapse.
Many studie\ do not specify a minimum duration of abstinence for indi\,idual\
classified 3s former smohers at ;I particular point in time. Data from such \tudie\ on
the aswciation of smohin, 17 ce\\ation L+ ith health and disease outcome\ mu\t bc


interpreted cautiously. For example. in the reports of the Whitehall Civil Servants
Study (Rose and Hamilton 197X; Rose et al. 1982). the criterion used to define
abstinence is not indicated. The only information provided is that the smokers reported
that "they were then smoking no cigarettes at all" (Rose and Hamilton 1978).

Regardless of the criteria used to define abstinence. the methodology for assessing
smoking status, including questionnaire items. needs to be carefully described by
investigators. Optimally these items should enhance the process of obtaining informa-
tion regarding the duration of abstinence. making it possible to fully determine the
relationship of smoking cessation to health and disease outcomes. When reviewing
studies of the health effects f smoking, the definition of the former smoker must be
carefully assessed, and the effect of the definition on the findings must be carefully
examined.

Temporal and Frequency Issues

Studies vary according to whether smoking is assessed retrospectively or prospec-
tively and whether a single assessment or a series of assessments is used. The category
of never smokers can be assessed retrospectively. usually relying on a single assess-
ment. Requiring subjects to reconstruct more detailed smoking histories can be very
demanding. Nevertheless, simply classifying individuals as former smokers or current
srnl  i reveals very little about the amount of smoking exposure experienced. More
pen. .Lnt questions regarding exposure include "How) long have you been abstinent
from cigarettes`?`: "At what age did you start smoking`?": "How many cigarettes did
you smoke during different periods of your life'?": "How many times did you stop
smoking'?"; and "How long did you remain abstinent during each of these occasions'?"
A series of repeated assessments can result in inconsistencies such as some in-

dividuals reporting smoking at one assessment and later reporting that they never
smoked. In a followup study in England. for example, Britten (198X) found 1.296
participants aged 36 who claimed that they had never smoked. Of these. 232 ( IX.7
percent) previously had reported smoking less than I cigarette per day, and 102 (7.9
percent) previously had reported smoking at least I cigarette per day for at least I year.
Of the 102 who reported previously that they had been regular smokers, 93 percent
reported that the last time they had smoked was at least IO years prior to the survey.

If the Britten study had used only one retrospective assessment of the subjects at age
36.323 percent of the 1,296 subjects would have been classified as never smokers and
32.6 percent as former smokers. Assuming that reports at a young age were more
accurate because memory bias was less likely to occur, the serial assessment indicates
that a more accurate categorization would be 29. I percent for never smokers and 36.5
percent for former smokers. Britten (1988) estimated that misclassification of this
magnitude, when applied to a study by Friedman and colleagues ( 1979). would result
in only a S-percent increase from 2.41 to 2.53 in relative risks of death for former
smokers compared with never smokers.

Krall and colleagues ( 1989) found that of 87 middle-aged adults. X7 percent accurate-
ly recalled their smoking status of 20 years earlier. but only 71 percent accurately
recalled the amount that they had smoked. Furthermore. underestimation of the amount

-77


smoked was tu ice as common for 20 years earlier ( I7 vs. 9 percent) and six times more
common for 32 years previously (37 vs. 6 percent). Persson and Norell (1989) found
that in a random sample of 9.394 individuals in Sweden. retrospective information
obtained 6 years later resulted in a strong tendency to overestimate previous cigarette
consumption among individuals who had increased their smoking (69 percent over-
estimated) and to underestimate among individuals who had decreased their smoking
(39 percent underestimated). Subjects with unchanged cigarette consumption showed
the highest levels of agreement (X9 percent) between original and retrospective infor-
mation. Rather than reconstructing full smoking cessation histories that are subject to
biased reporting. many retrospective studies rely on more limited categorization such
as never. former, and current smokers.

Retrospective studies enable researchers to assess long periods of smoking abstinence
without the need to observe the subjects over a long period of time. as would be
necessary in prospective studies. Case+zontrol studies. for example. can compare cases
with smoking-related diseases with controls with histories of being abstinent for IO to
20 years: in a prospective study. it may be impractical or impossible to study health
consequences of cessation with more than IO to 20 years of abstinence (Chapter 2. Part
II).

Prospective studies have the potential for more reliable and valid measures of
smoking status over time. especially when using a series of assessments, than do
retrospective studies. In intervention trials, for example. all subjects enter the trial as
current smokers. Following intensive intervention. subjects are identified as continuing
smokers or former smokers (abstinent). By assessing subjects at specified intervals
such as every 1 or 6 month\ over a series of years. especially when paired with
biochemical verification (Chapter _. ' see section on Biochemical Markers). researchers
can reduce the measurement bias and he more confident in the reliability and validity
of measures classifying continuing and former smokers and specifying length of
abstinence for former smohers. In MRFIT (Ockene et al. 1990) for example. a series
of4month followups over 6 year\ enabled researchers toclassify participants into three
categories: persistent quitters (continuous abstainers since the initial intervention).
intermittent quitters (abstinent for periods of time since the initial intervention). and
continuous smoherx (not abstinent during any of the followup periods). Such precision
in measurement is generally not possible or necessary in epidemiologic studies.

Prospective stud& may use 3 single assessment to categorize current. former. and
never smokers. These studies then prospectively, examine the categories to detect
differential rates of morbidity~ and mortality.. As discussed above. the assumption that
individuals vvill not change their smoking status maybe a tlavv, with \uch single
as\es\ments.

Improving Self-Report Measures

Ideally. assessments of smoking statu\ need to include standardized questions to
determine smoking status. that is never. current. and former smokers. For example. to
be categorized as a never smoker. the necessary response bould be "no" to a standard
question such as. "Have you ever \mohed at least I cigarette per day for at least I year?"

28


Whenever possible. questions should be used that allow continuous rather than
dichotomous scales for rejpon$e. A question such as "Do you smoke regularly?" results
in a dichotomous response scale. This scale provides much less information than does
a continuous scale. such as the question. "On the average. how many cigarettes do you
smoke per day`?" which can range from 0 to 20. 40. 60. or more. Multiple questions
such as. `* Have you smoked even a puff of a cigarette in the past 7 days?": "How many
cigarette% do you typically smoke each da),`?": and "How many cigarettes do you
typically smohc each weeh'l" can be used to refine a category such as current smokers.
Inclusion of other indices. such as biochemical markers of smoking (e.g.. sali\,acotinine
levels). can also be used to describe smoking statu\.

In a followup study. measures of smoking status optimally should be repeated over
multiple occasions. especially for dynamic categories lihe current smokers and former
smokers. which are open to change over time. Repeated measure\ over a series of
occasions provide further reliability and validity for assessments and alw provide
greater statistical power for detectin g differences betueen groups.  Nevertheless.
studies with only a single or a few assessments of smohing behavior have been
extremely informative.

Alternative BehaGral Measures

As a measure of smoking, self-report by questionnaires and interviews is the most
common. the least expensive. the easiest to use. and the most feasible in epidemiologic
studies (Frederiksen. Martin. Webster lY7Y: Pechacek. Fox et al. 1983). However.
other behavioral measures have also been used in clinical studies. Because these
measures are generally not used in large-scale epidemiologic studies. they w*ill be
presented only briefly m this Chapter.

Self-monitoring by the smoker. a measure of smoking commonly used in intervention
studies. involves recording by paper. pencil. and mechanical counters each cigarette as
it is smoked. The monitoring itself may be a reactive measure and alter the behavior.
depending on the nature of the monitored behavior and motivation (Abrams and Wilson
1979: Frederiksen. Martin. Webster 1979; Lipinski et al. 1973: McFall 1978: Orlean\
and Shipley 1982). It is an intrusive measure that is normally restricted to small \tudie\
of high intensity. Other behavioral measures, such as direct observation. collecting and
counting cigarette butts (McFall lY78). and measurin f their length (Auger. Wright.
Simpson 1979). are even more costly and intrusive and less appropriate for
epidemiologic and large intervention studies.

Alternative types of behavioral reports for validation of smoking status include
verification by an informant (Shipley I981 J. by self-report measure\ tising multiple
questions about smoking behavior or status as part of the same interview or question-
naire (see above). and by samplin 2 on multiple occasion\. Examples of the latter
usually involve long periods of time and often rc\ult in multiple sources of di\-
crepancy. (See Lee I9XX for summary.)


Surrogate Assessments

In some circumstances researchers may need to obtain information from sources other
than the index subjects. With some study designs, for example a case<ontrol study of
lung cancer, some subjects are unavailable to answ'er questions because of illness or
death. In cohort studies. or intervention studies with mortality endpoints, surrogate
interviews are sometimes required to assess smoking during the interval preceding
death.
Failure to obtain surrogate reports can cause considerable bias in some instances. In
a case+ontrol study of oral cancer. Greenberg and coworkers ( 19X6) obtained inter-
views with I I2 cases (67.9 percent) and surrogate reports for 23 cases ( 13.9 percent).
Cases needing surrogate report3 had more advanced stages of disease at the time of
diagnosis and were more likely to be black and less educated than cases interviewed in
person. Cigarette smoking and drinking hard liquor were more common among these
cases. Therefore, failure to include surrogate reports would have resulted in under-
estimates of the strength of association between cigarette exposure and hard liquor and
the risk of oropharyngeat cancer.
Pickle. Brown and Blot (1983) found that siblings of index subjects provided the
most complete data about smoking in the subject's family of origin and early life events.
Spouses and offspring supplied the most complete data about smoking history during
adult life. Incomplete data generally increased with the amount of detail requested. so
that there were considerably higher nonresponse rate\ for a detailed smoking history
(approximately SO percent) than for the history of a broad smoking status. such as never
smoker (approximately IS percent). Surrogates beyond a spouse or close relative
prov,ided much higher nonresponse rates for almost all questions in all statuses.
McLaughlin and colleagues ( 19X7) examined the reliability of retrospective surrogate
reports obtained IO years after initial reports and compared these with retrospective
self-reports using data from the NHANES-I (Cornoni-Huntley et al. 19X3). Correct
identification of previous smoking status was generally provided by most types of
surrogates. except siblings of male decedents. The combined level of agreement for all
surrogates ranged from X5 to 95 percent and was remarhably similar to that from
self-reports of living subjects. Thirty-five percent of the surrogates could not provide
data on when smoking began compared with I percent in self-reports. Surrogates who
responded tended to provide a later age for starting. Surrogates did. however, provide
estimates of years smoked that vvere comparable to the original reports. In this study.
siblings and other surrogates provided less reliable report\ than spouses. offspring. or
parents of subjects.
Lerchen and Samet (19X6) interviewed widow\ of lung cancer patients who had
supplied theirow>n smoking histories vvhile alive. They found that of 77 uiv,es of current
smokers, all supplied information about the cases' cigarette smoking status (ever/never)
that was in perfect agreement with the information supplied by the cases themselves.
Sixty-six (X6 percent) w'ere able to supply complete responses about their husbands'
smoking behavior. For those who responded. however. mean values reported by cases
and their wives were not significantly different for age at which cases started smoking.
years smoked. or average number of cigarettes smoked per day. Wives tended to report


20 cigarettes smoked daily even when their husbands smoked substantially more or
fess. Pershagen and Axelson (1982) also reported perfect agreement regarding
smoker/nonsmoker status when information was obtained from a close relative (parent.
wife. or child) for I4 lung cancer cases compared with information that had previously
been obtained from the cases by the physician. Blot. Akiba. and Kato (19X4) also
interviewed next of kin in a case<ontrol study of lung cancer among atomic bomb
survivors who had previously provided information regarding their own smoking
behavior while they were alive.  The investigators found that only I percent of
surrogates reported that a subject had never been a smoker while the subject reported
that he or she had smoked. suggesting that the identification of never smokers by next
of kin is very accurate. There was poorer agreement regarding those who smoked. with
I3 percent of surrogates indicating that a subject had smoked while the sub.ject had
reported never smoking.

Sandler and Shore (1986) examined the quality of data provided by adult offspring
on parents' smoking and drinking. The data were from 5 IX cancer cases and 5 IX
healthy controls aged IS to 59. When possible, mothers provided data on their own
smoking and their husbands' smoking. Of 9X2 subjects who had lived with their natural
mother, 97 percent provided data on their mothers' smoking status. Of those whose
mothers reported never having smoked cigarettes , 2.7 percent were reported as ever
having smoked by the adult child. Of those mothers who reported ever having smoked.
8.8 percent were reported as never smokers. Of those fathers reported by the mother
as never smokers, 17.2 percent were reported by subject5 as ever smokers. Of those
fathers reported as ever having smoked cigarette\ . 2 I. I percent were reported as never
smokers by their adult children. Even with the quantity of cigarettes collapsed into
categories to include answers of less than I pack, I pack. and more than I pack. the
proportion of mothers and subjects whose responses exactly agreed was X2.0 percent
for mothers and 49.2 percent for fathers.

Humble, Samet. and Skipper (1984) interviewed 46 subject-spouse pairs, waith 2
people in each of 38 of these pairs acting as the subject and as a surrogate for his/her
spouse, thus producing X4 total subject-surrogate pair\. For the 30 current or previous
cigarette smokers whose spouses gave complete smoking data regarding the subjects.
the subjects reported a mean use of 17.8 cigarettes per day compared wjith 14.3 reported
by their spouses. The difference was not significant.

Investigations indicate that useful information on smoking can be obtained in
epidemiologic investi&ations that must rely on surrogate information (McLaughlin et
al. 1987). Although greater misclassification occurs wshen surrogate reports are used
compared with self-reports. consideration of variables \uch as the relationship of the
informant, length of time he or she had known the case. the topic of the questions. and
complexity of the data gathered from the informant can add to the validity of the data
(Roget and Reid I975 ).

Nonbehavioral Measures

Methods other than self-report have been used to assess smohing status. Some
researchers have cxprecsed concern that sell`-report Ls hen used alone can bc an 111.


accurate measure that underestimates the amount of cigarettes smoked (Haley and
Hoffmann 1985: Marsh et al. 19Xx; Warner 197X) because subjects often underreport
levels of cigarette consumption or misrepresent themselves as former smokers (Luepker
et al. 19X9: Murray and Perry 19X7: Windsor and Orleans 19X6; Russell 19X2: Stookey
et al. 1987). Underreporting also has been linked to "digit bias." that is. subjects tend
to report in terms of multiples of ten and underestimate actual consumption (Pechacek.
Fox et al. 19X3; Vogt 1977: US DHHS 1989).

Between 1974 and 1985. estimates of U.S. cigarette consumption based on \elf-report
accounted for only about 70 percent of consumption estimates based on cigarettes taxed
and sold (Hatziandreu et al. 1989). This ratio has remained relatively stable. Most of
this discrepancy is lihely to be due to underreporting or a "rounding down" to the nearest
multiple of a half-pack of daily cigarette consumption (Kozlowski 19X6). although
misreporting of smoking status may play a role as well.

Validation of self-reports with measures such as biochemical assessments represents
a possible means of decreasing misclassification due to misreporting (Luepker et al.
1989: Windsor and Orleans 1986). However. some researchers note that biochemical
validation techniques present different problems that also cause misclassification. thus
favoring the use of self-report (Assaf et al. 1989: Crossen. Dougher. Belew 1983:
Hansen, Malotte. Fielding 1985; Hatziandreu et al. 19X9: Kornitzer et al. 19X3: Petitti.
Friedman. Kahn 19X I ). As noted above. sensitivity and specificity of the biochemical
measures are not perfect. In addition. the procurement of biochemical measures from
a large majority of self-reported quitters is not as feasible in large-scale interv,ention
trials or observational studies as it is in smoking studies of a smaller scale and a more
clinical nature. Subjects in the population samples do not have the same commitment
to studies that volunteers have to clinical studies. and the former are more likely to leave
the study area. which makes validation difbcult (Ockene et al. 1989). Validation aI40
requires more personal contact than is generally employed in observational or large-
scale field studies, and the additional contact may not be acceptable to the subjects or
feasible in the context of the study.

The section below on physiologic measures discusses methods other than behavioral
measures that have been used to a\ses'r cigarette smoke exposure. These measures are
then contrasted with self-report. and the varying needs for biochemical measurement
among different populations are considered.

Physiologic Measures

Smohing behavior has been assessed by measuring physiologic changes that result
from smoking (Pechacek. Fox et al. 19X-4). Smohing and smohe exposure are reflected
in a variety of acute and chronic physiologic measure\ primaril) because of the strong
pharmacologic effects of nicotine. These effects include changes in heart rate. blood
pressure. hand tremor. and skin temperature.  Each of these measures has a wide
variability under normal conditions and is affected hy man\ factor\ other than smohin~.
thu\ limiting usefulness ;I\ a measure ofsmohing (Pechaceh. Fo\ ct al. 19X1).

32


Biochemical Markers

Cigarette smoke i\ a complex mixture of chemicals. some of w hich are present in the
tobacco leaf and some of v. hich result from chemical reactions during either the curing
procec\ or smoking (US DHEW 1979: US DHHS 1986. 1989). Three chemical
constituents of tobacco smohe, carbon monoxide (CO). hydrogen qanide (HCN). and
nicotine. pass through cigarette filters and are pre\ent in inhaled tobacco smoke in
concentrations high enough to be absorbed and detected in persons who smoke. These
chemicals are measurable as intact compounds or as metabolic products.

Exposure to CO can be assessed in the blood a\ carbo\yhemoglobin (COHb) or as
CO in expired alveolar air. Method\ are a\,ailable for measuring cotinine. the primary
metabolite of nicotine, and SCN-. a metabolite of HCN. in urine. blood, and za1ib.a.
Other measures. such as skin+urface sampling for nicotine (Naliji and Lawrence 19X8)
are not as well established.

Extensive review\ of the literature on the use of biochemical markers as measures of
smoking status are provided by Bcnowit/ ( 1983). Haley and colleague\ ( 1986). Lee
( 198X). Pechacek, Fox. and colleagues ( 1981). and Windsor and Orlean\ ( 19X6).
Cummings and Richard ( 198X) supplied a review ofoptimal cutoffs for the biochemical
measures discussed here. This Section i\ not intended to provide an indepth review of
the variability and biochemical rationale for these meajure5 and will only provide an
overview of the use of biochemical assessments for smoking status.

Terminology

Sensitivity and specificity. characteristics of a test such as a biochemical assessment.
are measures of validity, the extent to which the test measures truth (Fletcher. Fletcher.
Wagner 1987). Typically. sensitivity and specificity are determined by comparing the
test results against a reference or "gold" standard. For smoking. self-reported status
has most often been used as the standard for assessing biochemical markers. The
sensitivity of a biochemical test for smoking exposure is the proportion oftrue smokers
who are classified as smokers by the biochemical test. The specificity of a biochemical
test for smoking exposure is the proportion of true nonsmokers who are classified as
nonsmokers by the biochemical test. A test of 100-percent sensitivity and lo(-percent
specificity would perfectly discriminate true smokers from true nonsmokers. However.
this degree of validity is not reached by any presently available biochemical marker.
In addition, the standard to which biochemical measures are compared. typically
self-reported smoking statu\. may be of limited validity. and thereby cause apparent
sensitivity and specificity to be reduced.

When continuous measures are used to test for smoking status. a cutpoint must be
chosen such that those individuals whose test value exceed\ the cutpoint are classified
as smokers and those with values below the cutpoint are classified as nonsmokers
(Cummings and Richard 198X). The level at which the cutpoint is set determines the
sensitivity and specificity of the test. Lowering the cutpoint improves the sensitivity
at the expense of specificity. Raising it will improve specificity at the expense of
sensitivity (Cole and Morrison 1980; Browner. NewJman. Cummings 198X). Selecting


a cutpoint depends on the relative importance of mislabeling an actual smoker as a
nonsmoker with a very insensitive but specific test versus mislabeling an actual
nonsmoker as a smoker with a very sensitive but nonspecific test. This tradeoff between
sensitivity and specificity is discussed in more detail elsewhere (Fletcher. Fletcher.
Wagner 1987).

An important contextual issue concerns the validity with which the biochemical
measure classifies individuals. When the test is applied to a population of smokers and
nonsmokers. the proportion of the persons who test positive. that is. above the specified
cutpoint. who are actually smokers becomes an important concern. This issue. distinct
from the question of w/hat proportion of smokers are above the cutpoint. is the crucial
measure of how much misclassification occurs. This proportion. the positive predictiv{e
value of a test, depends not only on specificity and sensitivity but also on the prevalence
of the condition in the population being tested (smoking in this example). The less
prevalent smoking is in the screened population the lower the positive predictive value
of a test (Browner. Newman. Cummings 198X).

The relative misclassification rates for smoker\ and nonsmokers. determined in part
by the estimated prevalence of smoking in the population to which the cutpoints are
applied, are particularly important in studies which use biochemical tests to verify
self-reported smoking cessation (Cummings and Richard 1988; Ruth and Neaton. in
press). For example. the pressure to quit smoking that is present in formal smoking
cessation programs may result in a high proportion of continuing smokers who report
not smoking. The use of cotinine validation in such circumstances (high prevalence of
false reporting) result\ in a high positive predictiv,e v*alue, as opposed to the lower
positive predictive value when the $ame test is applied to self-reported former smokers
identified in a population-based survey (low prev)alence of false reporting).

In biochemical validation studies. such as those reported in a subsequent section of
this Chapter. after optimal cutpoints are set using \elf-report in one population as the
gold standard. the biochemical marker then becomes the gold standard against which
self-reported smoking status is measured in another population.

Carbon Monoxide

High concentrations of CO are present in cigarette smoke (US DHEW 1979: US
DHHS 1986. 1989). Absorbed rapidly into the bloodstream during smoke inhalation.
CO has a half-life of-t to 5 hours in sedentary adults (Stewart 1975). Direct measure-
ments of CO can be taken from exhaled alveolar air or estimated by measuring the
percentage of hemoglobin combined with CO (COHb) (Stewart 1975).

Sensitivity of exhaled CO for classifying active smoking is generally in the range of
80 to 85 percent but can be affected by diurnal variability as well as other factors
(Benowitz 1983). Given the short half-life of CO. levels are influenced by time of day
and time elapsed since last cigarette. Measurements tahen late in the day, standardized
from time since last cigarette. are likely to give the best estimates of CO levels
(Frederiksen and Martin 1979: Horan, Hackett, Linberg 1978: Hughes. Frederiksen.
Frazier 1976). Using self-report of recency of smoking can increase sensitivity
(Bauman. Koch. Bryan 1983). Sensitivity is poor for light smokers (Fortmann et al.

34


1984: Vogt 1982). and specificity can be reduced by exposure to CO present in the
environment as a result of industrial and automobile pollution, environmental tobacco
smoke, indoor combustion sources. and use of products such as marijuana (Biglan et
al. 1985: Frederiksen and Martin 1979; Stewart 197.5). In spite of this. only 2 to 5
percent of nonsmokers in general populations will exceed I percent COHb (Janzon et
al. I98 I ; Kahn et al. 1974). tising COHb levels from a national probability sample.
the Radford and Drizd (1982) reported the 95th percentile for COHb to be I .77 percent
in nonsmokers, aged I2 to 74. If a 2-percent cutpoint is applied to this sample. 3.6
percent of nonsmokers would be incorrectly classified as smokers.

Thiocyanate

High concentrations of HCN. a toxic gas. are present in cigarette smoke. HowevJer.
HCN is very active chemically and is rapidly detoxified by the liver into SCN-(Langer
and Greer 1977: Boxer and Rickards 1952). Because SCN-accumulates in body fluids.
such as saliva. urine, and blood, it is used as a biochemical measure of exposure to
tobacco smoke. The biologic half-life of SCN- has been found to vary quite a bit (Bliss
and O'Connell 1984) although the length of time usually noted is between IO and I4
days (Langer and Greer 1977; Vesey 1981). Salivary SCN- can be measured most
reliably in parotid gland secretions (Shannon. Suddick. Dowd 1974): however. parotid
gland secretions show some seasonal and diurnal variability (Shannon. Suddick. Dowd
1974). When serum and saliva samples are compared, the levels are IS to 20 times
higher in saliva than serum (Langerand Greer 1977; Pechaceh et al. 1979: Vesey I98 I ).
However, saliva levels are more variable (Pechacek et al. 1979).

The increment of SCN- in light smoker5 is low, and there is much overlap of SCN-
levels in light smokers compared with nonsmokers (Fortmann et al. 1983: Neaton et al.
I98 1; Vesey et al. I98 I ). However. detection of light smoking in adults using SCN-
levels is better than in adolescents (Windsor et al. 1985). This is likely to be related to
the fact that adolescents are often in the process of learning how to smoke and inhale.
and they may not have an established pattern of smoking (Pechaceh. Murray et al. 19X-t).
For example. among younger adolescents only one-third or less could be identified on
a single assessment (Hunter, Webber, Berenson 1980: Luepker et al. 1989: Pechacek.
Murray et al. 1984). Specificity represents a more severe problem than sensitivity. A
large number of food products are sources of either cyanogenic giycosides (e.g..
almonds, bamboo shoots, sugar cane) or naturally occurring SCN- (e.g.. caulitlow~er.
broccoli, beer) and can produce levels of SCN- in saliva equivalent to the average levnels
of smokers (Langer and Greer 1977: Neaton et al. 19X I; Pechacek et al. 1979: Swan et
al. 1985).

The relatively low specificity and sensitivity of SCN- testing compared with cotinine
and CO make SCN- a less useful outcome measure for smoking cessation studies
(Gillies et al. 1982; Fortmann et al. 1984) unless adjustments are made using carefully
collected dietary and environmental exposure data. A prime advantage of using SCN-
for biochemical validation of smoking abstinence is its long half-life compared with
other biochemical measures (Fortmann et al. 1984; Steinman 1985; Murray et al. 1987;


Pechacek. Fox et al. 19X1). \vhich i\ of particular interest in population surveys where
longer term ub5tinence i5 of concern.

Cotinine

Cotinine. a metabolic byproduct of nicotine. i\ distributed throughout extracellular
fluid and is excreted through the kidnqs and salivary gland\ (Benowitz 19X?). About
15 to 20 percent is eliminated in the urine unchunsed. and the re\t is metabolized
(Benowitz 1983). The half-life estimates ofcotinine are variable and range from IS to
30 hours (Carey and Abram\ IYXX: Knight et al. 19X5: Greenberg et al. 19X-l: Haley
and Hoffmann 19X5: Haley et al. 19X7: Scpkovic. Haley. Hoffmann 19X6). The
differences in estimated half-life for cotinine reflect not only individual difference\ in
metabolism but also difference5 between \mohers and nonsmohers (Haley. Sepkovic.
Hoffmann 19X9: Sepkovic. Haley, Hoffmann I YX6: Hale} et al. I Y87). Cotinine le\ cls
vary with the diurnal cycle and are best asse\\ed late in the day (Benowitz 19X.3).
Methods are available for measuring cotinine in saliva. urine. and blood. Urinary le\,els
have been suggested to be too variable (Pechacek. Fok et al. 1981). and plasma or herum
levels appear to be the most hpable (Benou itz I983 ). However. sampling saliva because
of ease of procurement and accuracy in classifying smoker\ and nonamohers ha been
recommended as a useful. noninvasive method that can be applied to large-zcale
intervention trials (Abrams et al. 19x7).

Because nicotine is unique to tobacco. cotinine is a highly valid marker for almost
any tobacco use (Haley. Axelrad. Tilton 19X3: Rus~ttll et al. 19x1: Wald et al. 19X1:
Zeidenberg et al. 1977). Although nicotine has been aaessed in home studies. it is
recommended that cotinine be used because it ha\ a more enduring and stable blood
level (Langone. Gjika. Van Vunaki\ IY73). Detecting regular smoher\ by analysis of
cotinine in blood. urine. or saliva ih almost certain. and even light smokers and
intermittent smoker5 are caily detected (Benowitf lYX3: Haley. Axelrad. Tilton 19X3:
Paxton and Bernacca 1979: Zcidenberg et al. 1977: Carey and Ahrams IYXX: Williams
et al. 1979). In one investigation. Y5 percent ofadole~cent e\er smokers were detected
by cotinine (William\ et al. lY7Y). Specificity i\ al\o high: regular smokers typicall!
have blood cotinine levels of 200 to 100 ng/mL. light smoker\ have -IO to 50 ngiml.
and nonsmokers are typicsll>, helo\{ IO ng/mL. When nonmohers are aaeshed. the\
rarely have any detectable cotininc' (Benowit~ IYXi: Hale!. Axelrad. Tilton IYX3:
Sepkovic and Hale) IYX.5: Zeidenbers et al. 19771.

In comparative studieb of different biochemical measures of smoking. cotinine ha3
emerged a$ the measure of choice (Abram\ et al. lYX7: Hale). .4xelrad. Tilton 19X3:
Jarvis et al. IYX-I. 19X7: Knight et al. 19X5: Pojer et al. 1YX-l) because of itz superior
senGtivity and specificit!.  However. it i\ more expensive and more analytically
complex than the other biochemical measure\.

The value of biochemical meaures is limited to short-term abstinence and cannot be
used to document continuous abstinence in long-term \tudie\. CO. with a half-life of
3 to 5 hours. can validate self-reports of not having smoked in the pact 23 to 3X hour>
(Benowitz 19x3). Cotinine. with a half-life of I5 to 40 hours. would have limited
application for validation beyond a few day\. SCN-. ivith a half-life of 10 to l-1 day\.

36


has been used to validate self-reports of not having smoked in the past 7 days and may
be useful to validate up to 3 to 4 weeks. However. specificity of this measure is low
compared with cotinine and CO.

Bogus Pipeline

The bogus pipeline, an assertion to subjects that biochemical assessments will be used
to assess smoking status when they will actually only be collected but not evaluated. is
used mostly in research with adolescents. One of the reasons given by researchers for
continuing to use biochemical verification for at least some proportion of the total
subjects is the assertion that if the subjects believ,e biochemical validation will occur.
they will be more likely to provide valid responses to self-report measures. This "bogus
pipeline effect" was first presented by Evans. Hansen. and Mittelmark ( 1977) from the
work of Jones and Sigall ( I97 I ) concerning smoking among adolescents. It is believed
that there is great pressure among adolescents to misreport smoking activities, Murray
and coworkers ( 1987) provided an estensiv#e review of this aspect.

Murray and Perry (1987) attempted to determine the conditions under which a bogus
pipeline will be effective by manipulating conditions ofanonymity. They demonstrated
that a bogus pipeline for adolescents is more likely to have an effect if there is an
expectation that subjects would otherwise perceive large amounts of pressure to report
not smoking and there is a credible pipeline message. However, their findings suggest
that an effective procedure to ensure anonymity can reduce this pressure and likewise
reduce the need for the pipeline.

Contextual Issues Affecting Biochemical Assessment

The accuracy of self-report measures, the desirability for behavioral or biochemical
validation of self-report. and the type of assessment needed are issues that need to be
considered in the context of the type of study. the nature and size of the study sample.
and possible refusal problems.

The nature of the subject sample can affect the likelihood of misreporting and
therefore the desirability of validation by biochemical assessment. In Table I. studies
demonstrating misreporting rates for individuals who report cessation but who are
assessed to be smokers by cotinine or nicotine measurement are classified into three
types of subjects: untreated volunteer samples. intervention samples, and high-risk for
disease and/or medical patients. Table 2 presents a similar classification of studies
demonstrating misreporting with CO validation. The tables are adapted from Lee's
work (1988) with the inclusion of additional studies. In cases where multiple cutoff
criteria are recorded, the values closest to the optimal cutoff are reported. Several
studies should be viewed as outliers and are noted in the tables,. These studies reported
unusually high rates of individuals who reported not smoking but were above the
cutpoint and also employed cutoff criteria far below optimum cutpoints (Cummings
and Richard 198X).

For untreated volunteer samples. the mode for individuals classified as smokers by
biochemical assessment who reported not smoking is zero, and no sample exceeds 5

37


TABLE I.-Measures of false reports of not smoking from studies using nicotine and cotinine as a marker

Wllll;nrl\ 1'1 d
I lY7Y 1

H;tlcy. AxclwJ.
TlllWl ( IYX3)

WaltI c, aI , I YXJ)

Nm )l und
l.a&ncc ( I'JXX)

I'icrce cl ill. ( I YX7 J

0 (O/27)

2 (2/0X)



0 (O/IX)



0.`) (2/Z I )



I .3 ty2.32,
2.1 (S/2.32,


2.2 (33/1,?60)



2.5 (20/X0X,
1.2 (34/X()X)

0 (O/43,


1.0(3/h?.!)


TABLE I.--Continued

Kwwll et al
c I')X7?

Stoohq 111 al.
,19X7)

Saltvary nIcoIine           7.1 (l/13)

Ilrtnary nlcotme           II=?. N<hO

3x.x ,57/l-17,


TABLE l.--Continued

No
No
No
No
No
No

tladdwv et al. (IYX7)    IJS pr~&!n;ltlt w,11111'11               No

:! pg/l(H) mL urinary nicotine or
10 pg/loO tnL urinary cotininc

13.7 ng/tnl. wum cot inine
14.2 ng/mL. uliv;tt-y cotitliw
397 ng/mL urinary cotinint'
21 .X ng/mL allvary ntcotine
7.3 ng/niL plnw~;i nicotine
5X.6 ng/tnL urtnary nlcotiw
IO ng/tnL wrum


TABLE 2.-Measures of false reports from studies using CO as a marker

No

NV

NU       IO ppm CO (cxpirctl atr)
No        l.7fh CO(llh)

No

6 ppm CO (eupiraI air)
X ppm CO (c\pved air)

x ppn1

YLY

No

YC\

x ppm CO

x ppn1

2'4 COHh
-lo; (`OHh
W/r (`Otlh

% (n/N)
False report\

Comments

1.x ( I/Z1 )

Ih(lY/l2l)
lXCl2/l'l,

0
3

0 (O/?O)

4.2 (37/XYO,

7o.ht22/1(17)
(J.3 t lO/lO7~
1.7 t51107r


TABLE 2--Continued

M:tlct,ln1 1`1 a.
( 19x0)" h

Ra\r et al. ( 19X0)

YC\

O.X'd (`OHh

I% (`OHh

(`0

I .h'i (`OHh

(`0 or (`OHh

(`0

YL%        (`0 0, (`()I Ill

x.x ( v3.l I

0(0/33,

I3ct \* cc`,,
I .-I ( 117-l) ;md
4 7 I l/Z-l,

0 (O/X)


TABLE 2.--Continued

    Popul;ttion

1JK \moker\ nttendmg fetwrd
practitioners

US worhde 5moktng control study

Jamrwth. Fan ler
et al. (Ic)X?)"

Clwel et al.
(I'w.5)



Land0 and
McGovern ( 19X.5 1

Richmond and
Wch\tt`r ( 10x5)

UK wwker\ in trial of nicotine gum

French trial of acupuncture and
nlcotinc gum


IJS whjcct\ undergotnp variou\
treatmcnta for rliminattn~ wlohinf

Au\traltan wioker\ in a general
practtce: mndomi/ed tnitl ot
rttect\ of advice to five up

Ahram\ et al.
lI9X7)

Worh\ite cessation

(;l>nn. Gruder.       Chicago Lung Awxxttion
Jqxr\hl ( I'MI       ce\\i111011 wiy

Part 111. titFh-rt\h/mc~lic;tl patient\

Told to      Criterion for titlw
pivz up    reports of not vllohirtg


Some group\    7 ppm co



  YC\       IO ppm (`0



  Ye\       I 2 ppm <Y 1



  Ye5       5 pptn(`0

Q (n/N)
False wport\


About 37



0 (O/J)


2x.0 (7/35)



0 (O/14,

Ye\

Te\t group

Ye\



Yt3

IO ppm CO (exp~retl ;ur)    15.ht7/45,

3-mo to I -yr followup

6.tno ti~llowup

Sample ot \tuJy
parttclp:tnt\tN=23):
I -yr followup

C!p to 2.mo fdlowup

Criteria not \tnted:
h-t110 follow~up


TABLE 2.--Continued


percent for either cotinine or CO. For intervention studies, values are typically 2 to 5
percent for cotinine and 0 to IO percent for CO. High risk/medical samples appear to
have the highest rates of misclassification of former smokers with the rates exceeding
20 percent. For example. as shown in Table I, Jarvis and colleagues ( lYX7) reported
very low rates ( I percent) of false reporting in vascular patients who were not advised
to quit compared with the rate in high-rish patients uho here advised to quit ( I7
percent). It is likely that the pressure to stop smoking influenced the accuracy of patient
reporting.
Observation studies in which no intervention occurs. or intervention studies in which
there is minimal intervention or interaction M ith smokers. are less lihely to prompt false
reports of smoking cessation than studies in which intensive inter\.ention does occur.
In the former types of studies. in which no or low-intensity inter\,ention occurred. there
was a much lower prevalence of subjects reporting a 2-l-hour quit attempt during the
past 6 months or current abstinence (Prochasha et al. lYX5) than in intensiLe interven-
tion studies. making misreporting less likely. A greater tendency to misreport in no or
low-intensity intervention studies might occur with adoleccents. for whom pressures to
report not smoking may be omnipresent (Pechsceh. Murray et al. 19X-I: Chapter 2. see
section on Bogus Pipeline). A similar pressure might occur in some other instances.
such as worksites in which a ban has been placed on smoking. where no intervention
occurs but there may still be pressure on individuals to misreport. However. no studies
have looked at the possibility of misreporting in such instances. The context in which
the study tahes place is likely to influence the degree of misreporting. Data currently
being collected from smoking cessation programs in a wide variety of contexts may
help to clarify this issue.
Clinic interventions and intensive interventions. on the other hand. typically ask
participants to set a quit date. Close relationships are developed with the counselors,
and self-reports of quitting are often given initially in a peer group. Under these higher
demand conditions. biochemical verification may be needed to decrease the mis-
reporting of current smokers as former smokers. For example. in MRFIT, special
intervention subjects claiming to be former smokers at followup examinations had mean
SCN- levels between those of never smokers and continuing smokers (Ockene et al.
1982). Similar discrepancies between reported and validated cessation rates did not
occur for the usual care men who had not received intensive intervention.
The use of biochemical tests for validating self-reports in epidemiologic studies has
a number of limitations. The tests do not have perfect sensitivity and specificity: their
half-lives do not necessarily fit the timeframe to be covered: and not all subjects are
willing to provide the necessary samples for assessment. A very sensitive test may
misclassify subjects as smohers if they have heavy passive smoke exposure (DiGuisto
and Eckhard 19%: Haddow. Palomaki. Knight lYX6: Haley et al. IYXY: Jarvis et al.
1985). smoke occasionally (i.e., I or 2 cigarettes on isolated occasions) (Williams et al.
1979). and/or use nicotine in some other form. such as nicotine polacrilex gum or
smoheless tobacco (Cohen et al. IYXX: Slattery et al. IYXY). Biochemical marhers are
also limited because they assess relatively short-term cessation (less than 2 weeks). and
in studies concerned with the impact of cessation on health. there is more interest in
evaluating consequences of long-term cessation.


In large-scale studies. use of biochemical assessments is generally not feasible; thus.
mandatory use of such assessments and subsequent classification of refusers as smokers
(as suggested by home investigators involved in clinical intervention studies e.g..
Windsor and Orleans 1986) would result in an unacceptable distortion of the outcome
data. In addition, some subjects may drop out if validation is required. The effect of
lost subjects on study results may be difficult to estimate. In contexts other than
intensive intervention trials. self-reported smoking status at the time of measurement
and concurrent biochemical assessment have been demonstrated to be highly concor-
dant (Fortmann et al. 1984: Petitti. Friedman, Kahn I98 I) (Tables I and 2). This high
concordance supports the use of self-report as a valid measure of smoking status in
observation studies of the health effects of smoking cessation.

PART 11. ASSESSING THE CONSEQUENCES OF SMOKING CESSATION

Study Designs Used to Assess the Consequences of Cessation

Overview of Study Design

Most evidence on the health benefits of smoking cessation derives from studies of
human populations and not from animal studies or other types of research. Research
on humans can be classified as experimental (the investigator assigns subjects to be
exposed or not exposed to the risk factors or preventive factors of interest) or observa-
tional (the investigator does not determine whether subjects are exposed or not exposed
to the factors of interest; exposure reflects the subjects' choices or some other process).
Intervention studies include randomized or nonrandomized community-based inves-
tigations and clinical trials. The clinical trial. involving randomization of subjects to
be exposed or not exposed to an intervention, has been used to investigate the effects
of smoking cessation in patient groups and in populations. The observational designs
include the ecologic study, the cross-sectional study. the cohort study. and the case-
control study.

The biases potentially affecting these studies can be broadly classified as selection
bias. information bias. and confounding bias (Table 3) (Kleinbaum. Kupper. Mor-
genstern IYE). Selection bias refers to distortion of an exposuredisease relationship
by the mechanism through which subjects are selected. Information bias arises from
the incorrect categorization of subjects as exposed or not exposed or as diseased or not
diseased. The resulting misclassification of subjects on exposure or disease status may
occur in a random or nonrandom fashion (Chapter 2. Part I ). Confounding bias refers
to the distortion of the apparent effect of an exposure on risk caused by association with
other factors that affect outcome (Last 1988). In the subsequent review of the study
designs used to assess the benefits of smoking cessation. sources of bias most relevant
to each design are highlighted.

46


TABLE 3.-Examples of potential methodologic problems in investigating the
     health consequences of smoking cessation

Con\equenceh

Current smoker\ developing symptom\ of
disease qutt smoking

Apparent benefn\ ofcr\\ation are reduced

Self-reported former smoker\ are actually
mohtng (information bia\)

Apparent benefit\ of ce\%ttw are reduced

Former \moher\ tend to have smoked less
than per\tstent smokers (confounding htah)

Fatlure to xxount for the dlfferencr ma)
exaggerate the appsrent benefit\ of
ceaation

Former smoker\ tend to have a healthier         Failure to account for the dlfferrnce ma!,
lIfestyle than persistent \moher> tconfoundmg      exaggerate the apparent benefit\ 01
hia\)                               ceaatmn

Smoking practtce\ and the presence of
smoking-related direases affect panicipntton
in btudtes (selection hias)

.4pperent benefit\ of cr\wtlon ma) hr
mcreawd or decrrawd

Small number of wbject\ in a stud)

A heneficlal effect ofcrsttion may not
reach sttisttcal stgntficancr

Ecologic Studies

Ecologic studies represent a descriptive approach for examining the relation between
risk factors and disease. Groups, rather than individuals, are the unit of analysis in
ecologic studies. For example, changes in lung cancer mortality rates for selected
countries have been examined for correlation with changes in measures of smoking for
those countries. such as the percentage of smokers or per capita cigarette consumption
(US PHS 1964; Cairns 1975: Cummings 1984; Doll and Peto I98 I ). Ecologic studies
often have the advantage of being performed inexpensively and feasibly by using
already available data. This design has well-described limitations related to the
estimation of exposure and control of confounding, and may yield seriously biased data
on exposuredisease relationships (Kleinbaum, Kupper, Morgenstern 1982: Rothman
1986).

Cross-Sectional Studies

In a cross-sectional or prevalence study, exposure and outcome are assessed at the
same point in time among individuals in a population. Because cross-sectional studies
measure exposure and outcome variables simultaneously. the true temporal relation
between exposure and disease may be obscured (Rothman 1986). However. cross-
sectional studies can be readily performed and have supplied much of the evidence on
smoking cessation and nonmalignant respiratory diseases (Chapter 7).

47


Cross-sectional studies may be affected by selection hia\. Because cigarette smoking
is a strong cause of disease and death. groups studied cross-sectionally may not
accurately reflect the natural history of smoking. smoking cessation. and the develop-
ment of smoking-related illness. The proportion of heav ier smokers and more suscep-
tible smokers may be reduced compared with the original birth cohorts giving rise to
the cross-sectional study population (McLaughlin et al. 1987). Former smokers who
stopped because ofthe development ofdisease may be underrepresented. whereas those
who stopped to reduce the rish of illness may be overrepresented.

Information bias is also of potential importance in cross-sectional studies. Pre-
existing conditions in survey participants may affect recall of past smoking or may alter
the approach used by interviewers to gather smoking information.  However. as
summarized in Tables I and 7. cross-sectional surveys generally demonstrate low rates
of misreporting of smoking status when compared with cotinine and CO levels.

As mentioned previously. a single observation on smohing behavior may lead to
misclassification of smokers because of the dynamic nature of smoking behavior.
Former smokers are typically a heterogeneous group with periods of abstinence ranging
from days to years. For example, in the 1986 Adult Use of Tobacco Survey (US DHHS
1989). the subjects' responses were classified in IO categories. -l of which included
former smokers. Of the former smokers. 12.5 percent had quit within the past 3 months.
7.X percent had quit in the past 3 to 12 months . 77.3 percent had quit in the past I to 5
years. and 57.4 percent had quit 5 or more y'ears earlier.

Cohort Studies

In a cohort study. the \ubjrcts are selected on the basis of exposure status (e.g..
smoking behavior) and observed for de\.elopment of disease. Observation may be
forward in time (prospective). backward in time (historical or retrospective). or both.
Correct conclusions can usually be made about the temporal relation between exposure
(smoking cessation) and outcome (reduction of morbidity, or mortality). With the
cohort design. multiple health outcomes can be considered simultaneously. For ex-
ample, the CPS-I and CPS-II conducted by the American Cancer Society (ACS)
examined the effect of smohing bells\ ior on total mortality and specific causes of death.

In a study of \mohing cessation. selection bias could affect the findings of cohort
studies if subjects lost to observation were more or less lihely to benefit from smoking
cessation than subjects remaining under observation (Greenland 1977). For inten,en-
tion studies and cohort studies. the rate of sub,ject loss provides an index of the potential
selection bias.

In a cohort study of smohin, 0 ccs\ation. some Ini\cla\zification of exposure may be
introduced if the classification of smoking status is based on a single assessment.
Although the categorization of smohing status may' be correct at the time the informa-
tion is collected. inevitably some former smoker\ will resume smoking and some
current smokers will stop. The extent of the resulting error will increase with the
duration of followup. The resulting misclassification will tend to underestimate the
effects of quitting because those who relapse to become current smoker\ would not be
expected to experience beneficial effects attributable to quitting.

48


For example. in ACS CPS-I involving nearly I million people. Hammond and
Garfinkel ( 1969) studied changes in smoking status over a Z-year period. Male former
cigarette smokers in 1959-60 who reported that they were smoking in 196142 varied
according to duration of prolonged abstinence reported in the lYS9-60 survey. For
respondents abstinent le5s than I year in 1959-W. 37.3 percent reported smoking 2
years later; of those reporting abstinence for I to 2 years. 19.1 percent were smohing :!
years later; and of those reporting abstinence of more than 2 years. 1.6 percent were
smoking 3 years later. For all males who were former smohers in 1959~60. I I .3
percent reported smoking 2 years later. For all female former smoker\ in 195Y-60. 6
percent reported smoking 2 vears later. In the U.S. Veterans Study (Roget and Murq
IYXO: Kahn 1966). male veterans itt a cohort of 23X.X16 were classified based on
responses to questionnaires administered in 1954 or in 1957 (if the 1951 questionnaire
was not returned) and then folloued for 16 years to determine the relationship betbeen
tobacco use and mortality. Undoubtedly, many of the original current smokers became
former smokers as a result of the strong trend of smoking cessation among U.S. males
durin_g the followup period (US DHHS lYX9).

Repeated assessment of smoking status in a cohort stud) can mitigate misclassifica-
tion due tochanges in smoking status over time (Chapter 2. Part I). Repeated measures
are often feasibly made in cohort studies to minimiLe the effects of misclassification.
Alternatively. validation substudies can be conducted within the cohort to quantify
misclassification errors (Greenland I9XX).

Case-Control Studies

Casexontrol studies involve selection of study suqjects based on the presence (cases)
or absence (controls) of a disease. Exposure and other attributes of cases and controls
(e.g.. smoking status or lifetime cigarette consumption) are then measured. The groups
are compared with respect to the proportion having the attribute of interest to calculate
the exposure odds ratio. which estimates the relative risk associated with exposure.
Case-control studies can generally be conducted in les\ time than cohort studies or
intervention studies and are less expensive to perform. Case+ontrol studies are well
suited for evaluation of disease\ with low incidence rates.

Case+zontrol analyses may be affected by information bias and selection bias.
Case+ontrol studies are prone to information bias if lifetime exposure histories are
collected by interview (Schlesselntan 19x21. Retrospective lifetime histories of smoh-
ing or other exposures obtained from ill or elderly sub.jecth may introduce misclassifica-
tion. SimilarI\.. studies that rel\, on reports from surrogate\ to assess smohing ma).
misclassify exposure. If individuals classified as cases recall more accurately or less
accurately than those classified as controls, differential misclassification result\ (Gordix
1982). Differential misclassification may also be introduced ifre\pondent~ deliberateI>
falsify answers or if interviewers differentialI> gather information from cases and
controls (interviewer bias): interviewer\ not blinded to case-control \tatu\ may probe
more intensely for a putative causal exposure in cases than in controls (Sachett I Y79).
Blinding is often not feasible. and meticulous attention must be directed to training
interviewers and to designing questionnaire\ to rcmovc the po\\ibilit!, of intervieuer


bias. Although selection bias may affect any case-control study that is not population
based. it is unlikely to be of particular importance in most casexontrol studies of
smoking cessation.

Intervention Trials

Intervention trials are designed to test a hypothesized cause-effect relationship or the
benefits of a preventive program by modifying the putative causal or preventive factor
and measuring the effect on relevant outcome measures. Intervention trials may be
directed at individuals or groups. such as communities. Regardless of the unit of
observation. the trials may be conducted wsith (e.g.. a clinical trial) or without ran-
domization to the intervention.

Clinical trials are most commonly used to assess therapeutic interventions. but this
design has also been used to evaluate preventive interv,entions. such as smoking
cessation. A clinical trial includes one or more comparison groups in which subjects
receive the control intervention: subjects are randomly assigned to the treatment and
comparison groups to ensure that the groups are comparable with respect to charuc-
teristics potentially affecting the outcomes of interest. Individuals or groups such as
communities can be the units of randomization. Within the limits of chance. random
assignment makes the intervention and control groups similar at the onset of study.

Although widely used to test smoking cessation methods. clinical trials have been
used infrequently to assess the health benefits of smokin, 0 cessation. In comparison
with observation studies. the clinical trial design offers the potential for eliminating or
more tightly controlling bias from the selection of subjects and from confounding.
However. for many health outcomes, both a large sample size and a lengthy followup
period may be needed to have sufficient statistical pow'er. Moreover. in a study of
smoking cessation. the power of the trial also depends on the extent of the reduction in
smoking in the intervention group. in comparison with the control group. In the
reported smoking intervention trials. only ;I minority of participants attained continuous
or prolonged abstinence following most cessation interventions (Hunt. Barnett. Branch
1971: Hunt and Bespalec lY73: Ockene et al. 1990). Even with intensiv,e. prolonged
inten entions. as in MRFIT. only 42 percent of smokers within the special intervention
group were not sntohing at h-scar follow up. and only 76 percent of baseline smobers
                     2
had been continuously abstinent from cigarettes over this prolonged period (Ockene et
al. IYYO).

Only a few clinical trials provide information relevant to the health benefits of
cessation (Chapter 3). In the Whitehall Civil Servants Study, (Rose et al. 19821. the
investigators randomly intervened in smoking with advice from a phy,sician in a group
of men at high rish for cardiopulmonary disease. In MRFIT. smoking intervention w'as
one component of the rish factor intervention program directed at the special interven-
tion group (MRFIT Research Group IYX3).

In tnost clinical trials that assess the effect of cessation on disease outcomes. such as
the Whitehall Civil Servants Study (Rose et al. 1982). the tn\,estigators did not monitor
longitudinally the persistence of quitting or levels of biochemical markers. The only
clinical trial that has provided these measures is MRFIT (Ochene et al. lY90). Although

SO