Below are the first 10 and last 10 pages of uncorrected machine-read text (when available) of this chapter, followed by the top 30 algorithmically extracted key phrases from the chapter as a whole.
Intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text on the opening pages of each chapter.
Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.
Do not use for reproduction, copying, pasting, or reading; exclusively for search engines.
OCR for page 30
3 Methodology
As might be judged from the preceding chapter's discussion of the
nature of cancer, it will not be easy to determine what causes cancer.
It is especially difficult to identify the connections between cancer
and what people eat, not only because of the complex nature of the dis-
ease, but also because of the complex nature of the food supply, the
variations in eating habits, and the limitations of scientific tools.
The classic diet-related disease is associated with a deficiency of
one or more nutrients. The discoveries of the causes and cures of dis-
eases such as scurvy (caused by a lack of ascorbic acid) and beriberi
(caused by a lack of thiamine) led to the development of a specific model
for nutrition research in which nutrient requirements were determined by
producing deficiencies in laboratory animals or volunteers.
The relationships between diet and chronic disease did not emerge
as a major interest to investigators until the causes of the princi-
pal deficiency diseases were identified. Just as it was once difficult
for investigators to recognize that a symptom complex could be caused
by the lack of a nutrient, so until recently has it been difficult for
scientists to recognize that certain pathological conditions might re-
sult from an abundant and apparently normal diet. Adverse effects on
health associated with nutrient excess in humans have long been recog-
nized. Obesity is the most noticeable among them. Other adverse effects
result, at least partly, from the availability (and overuse) of vitamin
and mineral supplements. Certain vitamins and most of the minerals are
known to be toxic above certain levels. But these known adverse (patho-
logic) effects of vitamin and mineral overdoses have, like the deficiency
diseases, a conspicuously direct relationship with the nutrients in ques-
tion. That is, the effects of denying or restoring a nutrient to an ex-
perimental subject, whether animal or human, are usually observable
within a short time--at most, months. The links between diet and meta-
bolic, degenerative, and malignant diseases are considerably less obvious.
However, because such conditions as atherosclerosis or cancer are probably
associated with dietary patterns that extend over a number of years, the
causative agents are difficult to identify.
The possible relationships between diet and cancer have been investi-
gated in studies of human populations and in laboratory experiments using
various in vitro systems (to check substances for their ability to mutate
bacteria and mutate or transform other cells) or animal models (to test
substances directly for carcinogenicity). This chapter provides a synop-
sis of the strengths and weaknesses that are inherent in the methods used
30
3-1
OCR for page 31
Methodology 31
to study these relationships. It also explains the approach adopted by
the committee in evaluating the epidemiological and experimental evidence.
EPIDEMIOLOGICAL METHODS
General Approaches
In epidemiological research on cancer and diet, investigators seek to
associate exposure to dietary risk factors with the occurrence of cancer
in defined population groups. The studies are largely observational, and
may be of several different types:
Descriptive Studies. These studies describe the patterns of disease
occurrence in one or more populations, in components of the same popula-
tion, or in a single population over time. The observed patterns may be
related to certain other environmental variables or characteristics of
the population, such as demographic factors, industrial pollution, or
diet. Data from descriptive studies are suggestive, rather than defini-
tive, and serve primarily to identify population groups at risk and to
generate hypotheses for further investigation.
Correlation Studies. These studies, based on aggregate exposure
data and observed outcomes, provide the next step in establishing mean-
ingful associations. The crudest of these studies are ecological studies
in which national per capita food intake is related to patterns of can-
cer incidence or mortality. This type of analysis is frequently able
to utilize existing data and is a valuable tool for generating new hy-
potheses. At a more refined level, interviews with carefully selected
individuals may be correlated either with group-specific cancer rates or
with regional differences in rates. In such analyses, the data on expo-
sure and those on outcome may be representative of exposure of differ-
ent groups in the population. Consequently, they often do not reflect
true individual associations and thus may be misleading. This is often
referred to as an ecological fallacy.
C~c--~nntrn1 Studies. Unlike descriptive and correlation studies,
case-control studies enable investigators to collect data for individuals
rather than for groups, and they are designed to control for confounding
variables. In these studies, exposure data (such as dietary intake) are
collected for cases with a specific type of cancer and are then compared
with similar exposure data for a suitably selected noncancer group,
usually referred to as "controls" or compeers. Differences in exposures
between the two groups that cannot be accounted for by chance occurrence
(random errors) or by known biases (systematic errors) represent true
associations between individual exposure and disease and may actually
reflect causal relationships (Ibrahim, 1979; MacMahon and Pugh, 1970~.
The strength of the association can be measured by an odds ratio
calculated from a 2 by 2 contingency table.
3 - 2
OCR for page 32
32 DIET, NUTRITION, AND CANCER
Cohort Studies. Similar to case-control studies, cohort studies
focus on individuals and control for confounding variables. Furthermore,
they are less susceptible to bias than case-control studies because the
exposure data are collected prior to the occurrence of the disease. In
the simplest cohort studies, occurrence rates of disease (e.g., cancer)
over time are compared between two groups of individuals with similar
characteristics but with different histories of exposure (e.g., none vs.
any; low vs. high) to the factors being studied. Higher or lower inci-
dence of disease in one group relative to the other implicates the expo-
sure variable as playing a role in the etiology of the disease. Cohort
studies are reported relatively infrequently because the low incidence of
the disease requires following large groups for long periods. This neces-
sitates considerable expenditures of both time and money. Furthermore,
even if a cohort study is prospective, it is limited in that the cohorts
were self-selected and were not randomly assigned as in true clinical
trials or intervention studies. However, dietary intake data from sev-
eral cohort studies of coronary heart disease have enabled investigators
to perform retrospective cohort analyses of diet and cancer (see Chapter
5~.
Intervention Studies. In these studies, which are sometimes called
experimental studies, the investigator randomly assigns the subjects to
two (or more) groups, which are then exposed (or not exposed) to differ-
ent levels of the substance being studied. Although such studies are
ideal for establishing true causal relationships, opportunities for con-
ducting this type of study are rare. In the past, intervention studies
have most often been undertaken to test the effectiveness of vaccination
programs or new treatments for disease. Their use in future research on
diet and cancer will be discussed in a second report to be prepared by
this committee.
Methods For Determining Dietary Intake
Several standard methods with markedly different levels of precision
are used to determine what people eat. Some of these methods are based
on government production statistics; others use information obtained from
individuals about what they have purchased, prepared, or eaten.
Group Dietary Data. Comparisons of diets for different population
groups are generally based on one or two types of data: national per
capita food intakes (also called food disappearance data) or house-
hold food inventories.
Most cross-national studies of cancer incidence comparing national
per capita "intake" of various foods or nutrients are based on figures
derived from food balance sheets. The intakes are calculated by adding
the total quantity of food produced in a country to the quantity of food
imported, and then subtracting the sum of food exported, fed to live-
stock, put to nonfood uses, and lost in storage. These estimates are
3-3
OCR for page 33
Methodology 33
then divided by the total population to yield per capita intakes.
Comparisons of cancer rates at various sites with national per capita
intakes of, for example, fat, fiber, and animal protein are derived from
data such as these. Although national per capita intakes have been very
useful in providing leads for further research, they are inaccurate as
measures of food that has actually been eaten. They really only measure
food that has "disappeared" into the food supply--which is why they are
sometimes called "food disappearance data." They do not account for food
produced by individuals, for waste in stores, restaurants, or homes, or
for differences in consumption within a country by different age and sex
groups.
In this report, the term "per capita intake" is used synonymously
with "food disappearance data."
Household food inventories are used in epidemiological studies
.
to obtain data on the eating patterns of groups of persons who dif-
fer geographically, socioeconomically, ethnically, or in other ways.
Food intake over a fixed period, usually 1 week, is estimated either
by trained workers who visit individual homes or by the person in the
household responsible for food preparation who is asked either to re-
cord purchases and menus or to recall household food use. Average per
capita intakes of food and nutrients are calculated by dividing the
total household intake by the number of persons in the family. A major
limitation of this method is that it assumes uniform food distribution
for members of the individual household.
Individual Dietary Data. Of necessity, individual food consumption
data must be provided by individual assessments--usually reports from
the subjects themselves, but occasionally reports from family members
who share their living quarters. Such information is obtained from three
basic sources: recent (e.g., 24-hour) recall, food records, or diet
history.
The recent recall is used most frequently to measure individual
consumption. In this method, subjects are asked what foods they con-
sumed over a recent specified time--usually 1 to 7 days. The 1-day (or
24-hour) recall only requires that each person estimate the amounts of
specific food items consumed during the preceding 24 hours. However,
since the foods consumed may vary considerably from one day to the
next, 24-hour recalls are more reliable as a source of group data than
as a source of individual data, i.e., the average for an entire group
is probably reasonably representative of the eating pattern for that
group. A 24-hour recall may be recorded by the subject or, more often,
by a trained interviewer. He or she may be asked to recall all items
or only certain foods eaten during the specified period. One sampling
problem is inherent in the 24-hour recall: diets during the weekend may
differ greatly from those consumed during the week. To increase the
representativeness of the 24-hour recall, this method is often combined
with a consumption frequency questionnaire in which subjects are asked
how often they eat selected groups of foods.
3 - 4
OCR for page 34
34 DIET, NUTRITION, AND CANCER
In studies based on food records, participants are asked to main-
tain an accurate diary of all foods consumed during a specified period
(e.g., 1 week). The subjects must estimate the quantity or weigh or
measure each food item eaten at home, allow for inedible portions and
plate waste, and note and measure all ingredients in recipes. They
must also record estimated amounts of foods consumed away from home.
Although the weighed diet record was long viewed as the ideal standard
in estimating dietary intake, it requires, at a minimum, a great deal
of interest and cooperation on the part of the subjects and, hence,
selects for certain types of people. Moreover, this method is likely
to cause subjects to modify their eating patterns to some extent, if
only for purposes of reducing their work load (e.g., by eating fewer
mixed dishes). The accuracy of this method is also compromised in
developed countries, where much of the food eaten is neither prepared
nor consumed in the home. Finally, this method is unsuitable for very
large-scale surveys or studies because of the time and effort involved
in providing detailed instructions to the subjects, in making frequent
follow-up contacts, and in coding the unstructured information from the
records. Despite these limitations, the food record has been used to
validate other methods used for collecting dietary intake data in the
same study population.
Unlike the recent recall, the diet history method does not seek in-
formation on intake during a specified day or week but, rather, attempts
to determine the average pattern of consumption during a particular per-
iod of the subject's life, e.g., just before the onset of an illness.
The intake of selected items or the usual dietary pattern for total in-
take is obtained through interviews or, less often, by self-administered
questionnaire. The information is recorded as frequencies of consumption
or, preferably, as estimated total amounts for the period of study. The
method requires very thorough training of interviewers (or subjects, if
self-administered), careful standardization of the questionnaire, ade-
quate allowances for differences in food preparation, and the provision
of suitable food models to facilitate quantification.
Each of the methods for estimating individual intake has its strengths
and weaknesses, but they share certain limitations. People vary in their
abilities to estimate exactly how much of something they have eaten, and
may sometimes fail to notice (or forget to report) their consumption of
certain foods (e.g., side dishes at meals, peanuts taken from a readily
available supply). Respondents may also know nothing about the ingredi-
ents of the dishes set before them. Furthermore, as mentioned above,
they may alter their eating habits when asked to record their intake. In
case-control studies, there is an additional problem: subjects who are
ill (i.e., cases and sometimes controls) may have altered their diets as
a result of their illness. Although patients are generally asked to re-
call what they ate before the onset of their illness, they may not be
completely successful in this effort.
It is especially difficult to relate diet to a disease like cancer,
which has a long time course, because we need to learn not only what
3 - 5
OCR for page 35
Methodology 35
people ate yesterday or during the previous week, but also what they
consumed in the more distant past. (The length of time between expo-
sure and onset of disease depends partly on whether the dietary com-
ponent being studied is an initiator or promoter.) The notion that
subjects can accurately report not only what they usually eat but also
what they usually ate is, for the most part, untested, although limited
data suggest that "recall" of a diet consumed 20 or more years ago may
more closely reflect present food choices than past ones (Garland and
Ibrahim, 1981~.
There is considerable potential for variation in the technique used
by interviewers and the introduction of bias during dietary interviews,
especially when very detailed information is required as in studies of
cancer. Depending on the hypothesis being tested, the interviewer may
need to elicit careful descriptions of food preparation methods, of the
fats and oils used for frying, of usual portion sizes, of seasonal vari-
ations in intake, etc. Eliciting such information requires considerable
probing on the part of the interviewer. During this process, subjec-
tivity may be introduced in the recording of responses. For these rea-
sons, researchers active in this field spend considerable time training
interviewers and developing effective instruments and aids (for example,
see Morgan et al., 1978~.
Asking subjects for the same information in two or more different
ways by using several methods in conjunction with one another may also
help to overcome some of these problems. Estimates of quantity can be
improved by using realistic or abstract food models (Morgan et al.,
1978), photographs of graded portion sizes (Hankin et al., 1975), and
similar devices. The strengths and limitations of the major epidemio-
logical methods to study effects of diet have been discussed extensively
in a number of reports (Beaton et al., 1979; Graham and Mettlin, 1979;
Graham _ al., 1967; Hankin et al., 1975; Marr, 1973; Mettlin and Graham,
1978; Morgan et al., 1978; National Academy of Sciences, 1981; Nichols
_ al., 1976; Nomura et al., 1976; Reshef and Epstein, 1972~.
Biological markers are also used to obtain indirect estimates of
individual intakes. This method has the appeal of objectivity, since it
entails the direct measurement of substances in serum, tissues, or body
wastes as a reflection of actual dietary exposures. Apart from the
difficulty in collecting such data from healthy controls, there are other
reasons why this method has not been widely used in epidemiological stud-
ies of diet and cancer. Foremost is the difficulty of identifying an
appropriate indicator of past intake. For example, serum levels of some
dietary components, such as cholesterol, do not correlate with informa-
tion on intake and may reflect homeostatic balances or long-term patterns
of consumption (Pearson, 1967; Underwood et al., 1970~. However, recent
reports on vitamin A serum levels suggest that some such measurements
may nevertheless be useful in predicting cancer risk in cohort studies
(Cambien et al., 1980; Kark et al., 1980; Wald et al., 1980~. A particu-
larly troublesome aspect of case-control studies using biological markers
3 - 6
OCR for page 36
36 DIET, NUTRITION, AND CANCER
(e.g., the relationship of fecal steroids to colon cancer) is that the
markers may themselves reflect consequences rather than antecedents of
the disease.
Analysis of Dietary Data. Regardless of the method used to collect
food intake data, the reported foods must be grouped into categories be-
fore they can be analyzed. Before this can be done9 some decision must
be made concerning the kinds of variables that should be compared with
data on the occurrence of cancer. The very first level of decision may
be whether to classify the data in terms of foods or nutrients--e.".,
whether the variable of interest is vitamin C or citrus fruits, carotene
or grams of dark green and deep yellow vegetables. In principle, the
at the outset of the
and forrn~t of the
important analytic variables should be idlest; f ; "A ~
study, since that decision will determine the nature and format of the
data that are collected. For example, if the variable of interest is
total calories from fat rather than the characteristics of specific fats,
which may differ according to their sources and processing, then the
interviewer need not help the respondent differentiate between animal
fats and vegetable oils or between liquid and hydrogenated shortenings.
If vitamin C is considered to be the relevant variable rather than fresh
citrus fruit, then no effort need be made to sort out the various forms
in which oranges might be consumed (e.g., as freshly squeezed or frozen
juice, or as whole orange segments). Thus, the nature of the hypothesis
determines the nature of the classification used for data collection.
This explains much of the discrepant data from different investigations
of the same cancer site, although the source of the discrepancy may not
be immediately apparent from even the most careful perusal of the pub-
lished reports.
Since much of the research on the relationship between diet and
cancer has been based on hypotheses regarding the effects of nutrients,
the raw data on foods consumed has most often been translated into nu-
trients, such as grams of protein, animal protein, total fat, satu-
rated and unsaturated fat, cholesterol, and complex carbohydrates. The
quantitative estimates are usually based on food composition tables,
such as those developed by the U.S. Department of Agriculture. (For an
example of these estimates, see Morgan et al., 1978.) Unfortunately,
the mean values recorded in such sources as USDA Handbook _. 456 (U.S.
Department of Agriculture, 1975) may not reflect the specific composition
of the foods eaten by subjects in a particular study. For example, wide
variations in the nutrient content of unprocessed and processed foods
can result from modifications in processing procedures (e.g., the addi-
tion or removal of nutrients) over time. However, such inaccuracies will
merely tend to weaken any detected association rather than introduce a
spurious association.
Analyses based on individual foods or food groups are not encumbered
by the need to estimate nutrient intake, but are often difficult to in-
terpret because of the multiple comparisons involved. In such analyses,
the specific substances responsible for an effect may be difficult to
identify.
3 - 7
OCR for page 37
Methodology 37
Overall Assessment of Epidemiological Approaches
The major strength of epidemiological studies is that their focus on
human populations circumvents two important limitations of laboratory
research. First, since humans are observed directly, the results do not
have to be extrapolated from one species to another. Second, since the
levels and patterns of exposure studied are those that actually occur
among people, interpolation to low doses from the artificially high ex-
posure levels frequently required in laboratory research can also be
avoided. In addition, since the varieties of human experience produce
wide range of exposures to a given risk factor, epidemiological invest)
gations are often able to examine directly the effects of different
levels of exposure (i.e., dose-response).
On the other hand, epidemiological studies present some special
difficulties. To begin with, such research is limited by its need to
rely primarily on observational data, because it is difficult and often
unethical to conduct experiments (i.e., intervention studies) on groups
of humans. Furthermore, observational epidemiological studies are open
to errors or bias. For example, persons who agree to participate in such
studies or who are selected as participants by the investigator (e.g.,
hospitalized patients) may not comprise truly representative groups of
subjects and may yield misleading findings.
Unlike studies of cancer among smokers and nonsmokers, dietary stud-
ies are confronted with the inherent difficulty of determining reasonably
precise exposures. For example, the degree to which cases have been
exposed to a particular dietary component may not be sufficiently
different from that of controls to demonstrate any effect. Furthermore,
it is often difficult to determine the specific dietary constituents to
which study participants have been exposed.
Another difficulty inherent in epidemiological studies of diet and
cancer is the long latency period between first exposure and overt man-
ifestation of illness. In case-control studies, this delayed onset makes
it necessary for investigators to learn what the subject ate during some
period beginning long before the study began, or to assume that recent
intakes reflect past exposures. In prospective cohort studies, the in-
vestigator must collect current dietary data and then either wait (for
up to 20 to 30 years) for the disease to appear or identify sufficiently
large groups of subjects for whom there are adequate retrospective
dietary data.
Accuracy in the measurement of both the exposure and the outcome
variables is especially difficult to attain in the studies of diet and
cancer. For example, the frequent dependence on recall data from inter-
viewed subjects virtually guarantees imprecise measurement of dietary
3 - 8
OCR for page 38
38 DIET, NUTRITION, AND CANCER
exposure, which might mask small but real differences between cases and
controls. Correlation studies may suffer from differences among coun-
tries such as completeness of cancer reporting, diagnostic practices, and
terminology. Furthermore, because cancer incidence (occurrence) data are
frequently not available for such studies, reliance must be placed on
mortality data instead. Since mortality reflects survival as well as in-
cidence, it is not an ideal measure for cancer etiology, particularly for
sites where survival rates are high and have notable international vari-
ation. These and other considerations make it especially difficult to
identify subtleties in the relationship between the degree of exposure
and risk of disease.
Most of these deficiencies in epidemiological studies of diet and
cancer are likely to result in nondifferential misclassification, thereby
reducing the likelihood that a given study will be able to demonstrate
true differences that exist between the groups compared. Therefore, the
results of epidemiological studies may often be assumed to represent
conservative estimates of the true risk for cancer associated with the
dietary exposures of interest.
LABORATORY METHODS
As interest in the possible relationship between diet and cancer has
grown in recent years, increasing attention has been paid to the chemical
carcinogens in our diet. The foods that we eat contain a vast number of
separate chemical entities: several thousand as additives and many times
this number as natural constituents. Most of these chemicals are present
in relatively low concentrations, but even small amounts of some potent
carcinogens might be important if they are present in commonly consumed
foods.
There are three major laboratory methods for detecting and identify-
ing chemical carcinogens: analysis of molecular structure, short-term
tests, and long-term bioassays in animals. The first two methods provide
information about potential carcinogenicity, whereas the third provides
direct evidence of carcinogenicity in laboratory animals.
Analysis of Molecular Structure
In a review of the large body of evidence pertaining to the role of
structure-activity relationships in predicting carcinogenic activity,
Miller (1970) suggested that most, if not all, chemical carcinogens are
ultimately electron-deficient reactants (Miller, 1970~. Carcinogens have
been identified in more than a dozen chemical classes, which share no
common structural features (Miller and Miller, 1971, 1979~. Furthermore,
even within classes, closely related chemicals may differ with respect to
carcinogenicity--e.g., 2-acetylaminofluorene (2-AAF) is a well-known car-
cinogen in several species of animals, whereas its close relative 4-AAF
is not carcinogenic (Office of Technology Assessment, 1981~. The major
3 - 9
OCR for page 39
Methodology 39
utility of the analysis of molecular structure is to screen a variety of
chemicals quickly and to treat the results as warnings rather than as
definitive indicators of carcinogenic activity.
Short-Term Tests
Interest in establishing short-term, relatively quick and inexpensive
procedures for the identification of chemical carcinogens has increased
during the past several years as a result of the realization that the
list of potential chemical carcinogens is growing faster than our capac-
ity to test the materials (Bridges, 1976~. Therefore, greater numbers
of potentially hazardous compounds must be screened and placed into a
priority system for further testing. This appears to be the primary role
of short-term tests.
Since these tests can be conducted quickly (often in only a day or
two) and inexpensively, they are useful for screening substances for
potential carcinogenicity. For these tests to be useful, they must not
only be faster, easier to interpret, more sensitive, and less expensive
than the standard feeding studies, but they must also be reliable and
relevant to the in viva assay upon which they are modeled.
A number of validated short-term tests can be used to examine the
capacity of a substance to cause mutations, other genetic alterations,
or neoplastic transformation. These tests can be used with a variety of
biological systems such as bacteria, yeast, mammalian cells, and intact
animals.
To date, the most widely used method appears to be the Salmonella/
microsome assay (also called the Ames test), which utilizes several spe-
cifically constructed Salmonella typhimurium strains to detect various
kinds of mutations and genetic damage (Ames et al., 1975~. It is gen-
erally agreed, but not without considerable controversy, that there is a
high degree of correlation between the mutagenicity of compounds in the
Salmonella/microsome assay and their carcinogenicity in laboratory ani-
mals (McCann and Ames, 1976; Purchase et al., 1978; Sugimura et al.,
1976~. However, recent studies show that this correlation is dependent
upon the class of chemical being investigated. For most aromatic amines,
polycyclic aromatic hydrocarbons, and direct alkylating agents, there
appears to be a high degree of correlation. On the other hand, chlori-
nated hydrocarbons are difficult to identify as mutagens in the Ames
test, although they are known to be carcinogenic. In vitro mutagenicity
tests have one major drawback: although they may provide a good indica-
tion of whether or not an agent is carcinogenic, they produce very little
information on its relative carcinogenic potency.
Other short-term _ vitro and in viva tests in use include assays
for the induction of DNA damage and repair or mutagenesis in bacteria,
in yeast, in Drosophila melanogaster, or in mammalian cells in culture.
.
Whole mammals can be used in the dominant lethal test, mouse spot test,
3-10
OCR for page 40
40 DIET, NUTRITION, AND CANCER
tests for heritable translocations, and tests for chromosome aberrations.
These mammalian mutagenesis bioassays offer promise as prescreening tools
since they seem to provide both qualitative as well as quantitative data,
but they are more expensive to perform and require more time than the
other assays. The _ vitro transformation systems are potentially useful
for screening carcinogens, but they are also expensive and time-consuming.
Moreover, the reliability of early markers of oncogenic transformation is
unknown. If the _ vitro transformation tests have to be carried out to
the point of injecting presumably transformed cells into a syngeneic ani-
mal to determine if the cells develop into a tumor, then the expense and
time involved are the same or possibly greater than required for some in
viva carcinogenicity test systems.
In general, short-term tests have a number of drawbacks:
~ Carcinogens or modifiers of carcinogenesis may operate by mecha-
nisms not involving DNA damage and repair. Thus, some agents, e.g.,
tumor promoters, which are particularly relevant when one considers diet,
are not likely to be detected by these tests.
O The effects of absorption, transport, activation, detoxification,
and excretion are not taken into account.
~ Quantitative risk assessment cannot be made easily.
O Despite positive results for mutagenicity in a battery of such
tests, many scientists do not accept such evidence alone as an indica-
tion of carcinogenicity. Long-tenm bioassays in whole animals are still
necessary to make this determination (International Agency for Research
on Cancer, 1980~.
Long-Term Bioassays
These tests, which are conducted in animals, have been the most
widely accepted methods for determining the carcinogenic effect of
substances. In the absence of data on humans, all substances demon-
strated to be carcinogenic in animals are regarded as potential car-
cinogens for humans, and the empirical evidence overwhelmingly supports
this hypothesis.
The standard procedure in long-term bioassays is to feed substances
at levels that are just below the maximum tolerated dose for a major
portion of the lifespan of the animal (usually rodents, which have a
lifespan of 2 to 3 years). The rationale for feeding very high doses of
a substance in chronic bioassays is that the number of animals that de-
velop cancer increases as the dose of the test substance is increased.
To conduct a valid experiment at high doses, only a small number of ani-
mals (a few hundred) is required. An important variable that determines
the outcome in these tests is the potency of the carcinogen: the greater
3-11
OCR for page 55
The Relationship Between Nutrients and Cancer 55
There are some data indicating the magnitude of the changes in per
capita intake of certain food items and constituents; however, many of
the changes are not adequately documented. For example, there are no
data on the consumption of whole wheat flour, commercial baby food, or
home-produced vegetables. Moreover, there is no indication whether
fresh vegetables eaten in the Northeast were grown in that region or
whether they were shipped by train from California or by air from
Mexico (Brewster and Jacobson, 1978~. A variety of differences in
their chemical composition (e.g., in their vitamin and mineral content)
can result from differences in the way in which these vegetables were
grown, transported, and stored.
Between 1940 and 1977, per capita intake of food color additives
increased tenfold. Soft drink consumption increased 1.5 times in
just 16 years--between 1960 and 1976 (Brewster and Jacobson, 1978~.
Although total intake of fruits and vegetables increased slightly
between 1909 and 1976 (Table A-2), the intake of fresh fruits and
vegetables] actually declined (Table A-3), a major portion of that
decline having occurred after 1948. Changes in the per capita intake
of certain individual commodities are especially striking. For example,
the intake of fresh potatoes is more than two-thirds lower than it was
at the turn of the century and more than one-half lower than it was 30
years ago, whereas the intake of processed potatoes has increased by a
factor of 44 during the same 30 years. The per capita intake of pro-
cessed citrus fruit juice--which accounts for much of the increase in
overall fruit intake--increased dramatically from an average of less
than one 4-oz (120-ml) serving per person annually in 1948 to 117 4-oz
servings per person in 1976 (Brewster and Jacobson, 1978~. Similarly,
the intake of canned or bottled tomato products (e.g., paste, sauce,
catsup, and chili sauce) increased from 2.25 kg per capita in 1920 to
10.1 kg per capita in 1976 (Brewster and Jacobson, 1978~. All of these
changes reflect the proliferation of food products on the market--from
less than 1,000 at the end of World War II to well over 10,000 at
present (Molitor, 1980~.
The term "fresh" applied to fruits and vegetables commonly refers to
produce that has been, at most, washed, trimmed, and chilled. The term
"processed" has many meanings; for example, preservation by canning and
freezing, which result in some chemical but little structural change;
extraction and dehydration such as the preparation of orange juice con-
centrate, which produce significant structural and possibly major
chemical changes including nutrient loss; and processes that involve
extensive separation of foods into components, or the fabrication of
new foods such as "chips" made from molded rehydrated potato flakes,
which result in marked structural changes that may have equally marked
effects on the chemical composition of foods.
A-5
OCR for page 56
56 DIET, NUTRITION, AND CANCER
TABLE A-3
Annual Per Capita Intake of Fresh and Processed Fruits,
Potatoes, and Other Vegetables in the United Statesa
Annual Consumption, kg/Person
Vegetables
Fruits (Excluding Potatoes) Potatoes
Year Fresh Processed Fresh Processed Fresh Processed
1909 75.6 3.61 83.7 7.7 81.9 0.2
1927 76.5 8.1 85.5 11.7 63.9 0.2
1948 73.8 19.8 82.8 21.15 50.0 0.2
1965 4 7. 3 27.5 63.5 29.7 30.6 5.4
1976 52.7 36.5 65.3 31.1 24.3 9.9
aAdapted from Page and Friend, 1978.
These striking changes in the food supply need to be taken into
account when one examines the relationship between diet and cancer.
On the one hand, any change in cancer incidence resulting from major
changes in food processing that occurred before 1900 (e.g., roller
milling of grain) or up to 40 years ago (e.g., flour enrichment) would
probably have been observed long before now. Conversely, because of
the long latent period between exposure and manifestation of cancer,
effects from changes introduced less than 10 years ago might not yet be
evident. If, as is often the case, changes in food-processing methods
are poorly monitored, the extent of exposure to substances resulting
from those processes will not be known. In such cases, it will be
difficult to make any associations between those substances and cancer
incidence.
A more difficult problem is encountered in case-control studies:
here one must determine what foods were consumed by subjects one or
more decades in the past. It is necessary to make one of two assump-
tions when collecting such information: that people can accurately
remember their typical dietary patterns of 10 or more years ago, or
that present diets adequately reflect diets consumed in the past. Both
of these assumptions are most subject to inaccuracies when there have
been continual shifts in the numbers, types, and varieties of available
foodstuffs. Even when the kinds and amounts of foods consumed in the
past can be accurately determined, their chemical composition remains
- ~ ~ ~ For
example, a frozen pizza made with imitation cheese, tomato extender,
and soy-protein pepperoni is composed of a very different collection
of chemicals than the apparently similar product made 10 years earlier
with mozzarella cheese, tomato paste, and meat sausage.
A-6
unknown and may have chanced significantly over the decades.
~ . .
OCR for page 57
The Relationship Between Nutrients and Cancer 57
The proportion of manufactured products in the average diet has
been increasing, especially in developed countries, but the detailed
composition of many of these products is not known. Manufacturers
often consider it proprietary information. Figures on the production
of ascorbic acid illustrate both the scale of the potential effects
of processing and the difficulty of monitoring such effects (Table A-4).
During the past 20 years, there has been a sixfold increase in the
tonnage of ascorbic acid produced. But in standard food composition
tables prepared by the U.S. Department of Agriculture, only the amount
of ascorbic acid used for food fortification is recorded. The disposi-
tion of the remainder is unknown. Most of the imbalance is probably
consumed in the form of vitamin supplements. Nonetheless, the fact
remains that large amounts of ascorbic acid, as well as other nutrients,
are added to foods for "technical" reasons, e.g., for their antioxidant
~' ' ~ "nutritional" reasons. These amounts do not
, . ,
show up on tables of nutritional value, although vitamins are monitored
nore carefully than most other components of the food supply. Most non-
nutritive substances are not monitored at all, and as a consequence
almost nothing is known about their presence or fluctuations over tine.
Saccharin, for example, is a nonnutritive substance intentionally added
' ~ ~~~ i-- ~~ ~ themselves--and, to a
Properties. as opposed to
~ - ,
to food--by manufacturers or by the consumers
very large extent, it is knowingly consumed. Yet, in epidetniological
studies it has proved very difficult to obtain reliable data on indi-
vidual saccharin intake. Obviously, it is even snore difficult to
obtain consumption data for substances that are neither monitored, as
are the nutrients, nor consumed intentionally, as is saccharin.
TABLE A-4
Production of Ascorbic Acid in the United Statesa
Year
1960
1965
1970
1974
1982
Amount Produced
(Metric Tons)
2,392
3~914
5,470
1O3O54
14~800
(estimated)
aData from U.S. International Trade Commission, 1980.
A-7
OCR for page 58
OCR for page 60
OCR for page 61
OCR for page 62
OCR for page 63
OCR for page 64
OCR for page 65
Representative terms from entire chapter:
food supply
58 DIET, NUTRITION, AND CANCER
Because of this paucity of information, it is possible to make
only the crudest assessments of factors that may affect the composition
of foods. For example, one can determine whether fruits are available
fresh, frozen, or canned, whether potatoes are available fresh or dried,
but not whether macaroni contains soy flour or whether tomato paste
contains modified starch, 6-carotene (for color), and added vitamin
C--among other things.
It is not clear whether all the changes in the food supply have
increased, decreased, or had no effect on the incidence of cancer.
Overall U.S. cancer rates at most sites other than lung and stomach
have remained relatively stable for several decades. This might
suggest that the food supply has contained an unchanging cluster of
cancer-causing or protective substances throughout much of this per-
iod, despite the extensive changes in the composition and quantity
of many of the foods consumed. It is also possible that any changes
capable of affecting cancer rates (positively or negatively) have
occurred too recently to be reflected in cancer statistics. But even
if cancer rates rise or fall in the future, it may prove very diffi-
cult to identify which, if any, specific compositional modifications
are involved, since so many different changes are going on simultane-
ously. This is illustrated in Figures A-2 and A-3, which show the
changing sources of fat in the U.S. diet. These figures reveal that
the relatively stable consumption of "total table spreads" and "total
cooking fats" masks a dramatic shift in the sources and, hence, the
composition of the fats involved. The use of butter and lard has
decreased sharply, whereas margarine and shortening (usually based on
vegetable oil) have come into much wider use (Brewster and Jacobson,
1978).
20
-
, 1 5
Q
A)
Q
En
~ 10
o
-
y
5
o
:\
The Relationship Between Nutrients and Cancer 59
25
20
Total Cooking Fats
o
In
o, 15
=
-
cn
`' 10
o
J
5 _
O , 1 , 1 , 1 , 1 , 1 , 1 ,. 1
1910 1930 1950 1970 ~
1 976
i_
.
, _
Shortening /
~_~ i./
_ ~
i
\
~ Lard
1 Q30 1 950
YEAR
FIGURE A-3. Intake of cooking fat. From Brewster and Jacobson, 1978.
At any particular time, cancer rates probably reflect the sum of
many changes, some producing positive and others negative effects. For
example, the introduction and subsequent wide use of refrigeration and
the increased use of mold-inhibitors and antioxidant s have probably had
~ ~ Together, these changes have markedly decreased the
positive effects.
population's exposure to rancid and/or moldy foods and to foods
preserved by salting, smoking, or drying.
,
-
The effect of other changes is less clear. Although there has been
relatively little change in the overall percentage of calories derived
from fat, protein, and carbohydrate, there have been marked shifts in
consumption patterns from vegetable to animal protein, from complex to
simple carbohydrates, and, as already noted, from animal to vegetable
fats. The increase in per capita intake of fat from meat has compen-
sated for a decline in the intake of dairy fats. In addition, there
has been a marked increase in the intake of separated vegetable oils
that have been structurally altered by hydrogenation and other
treatments.
A-9
60 DIET, NUTRITION, AND CANCER
There have also been changes in the nature of the fat-soluble
contaminants present in the diet. In federal inspections for pesti-
cide residues, contaminants have been found most frequently in meats
and fats (U.S. Food and Drug Administration, 1980~. The Comptroller
General (1979) reported that of 143 drugs and pesticides likely to
leave residues in raw meat and poultry, 42 were known to cause or
suspected of causing cancer. Twenty years ago, fats were much less
likely to carry such residues since the use of both drugs in animals
and pesticides has increased markedly in the interim (Smith, 1980~.
A fivefold increase in the per capita intake of french-fried
potatoes is part of a trend toward a much greater consumption of
products crisped by exposure to heated fat or to extreme dry heat.
Such products include potato chips, fried snacks, crackers, and
ready-to-eat breakfast cereals. Many products of such browning
reactions have proved to be mutagenic in laboratory tests as have
the by-products resulting from the frying and broiling of meat and
fish (see Chapter 13~. Hence, products in this category must be
regarded as potential contributors to carcinogenesis.
Several other changes may also be important, but their effects on
carcinogenesis are not known. For example, there has been a documented
decline in the consumption of certain types of vegetables, especially
in their fresh state. The effect, if any, of the marked increase in
the consumption of cooked (and often burned) tomatoes is also unclear
as is the effect of the documented decline in the consumption of fresh
cabbage, since the total long-term consumption of other cruciferous
vegetables (e.g., broccoli, cabbage, and kale) is impossible to cal-
culate given the lack of accurate data on home production. However,
the documented decrease in the annual per capita intake of sweet
potatoes, from 11.1 kg per person during 1976 to 2.4 kg during 1980,
combined with the declining consumption of fresh dark green and deep
yellow vegetables, has very likely decreased the intake of dietary
fiber and naturally occurring 6-carotene, which recently have been
studied for their possibly protective roles in carcinogenesis (Chapters
8 and 9~.
Despite (or perhaps because of) the paucity of information pertain-
ing to the composition of our contemporary food supply, foods have been
most often used in epidemiological studies as indicators of the presence
of particular nutrients or they have been grouped for analysis accord-
ing to certain nutrients they have in common. Given the multitude of
other chemicals present in the diet, it is notable that epidemiological
studies have found significant relationships between the occurrence of
cancer and estimated intakes of such nutrients as fat, vitamins A and
C, or protein (see Chapters 5, 6, and 9~. This would seem to indicate
either that these nutrients must play a role in the development of
cancer or that they serve as indicators of other substances that do.
Epidemiological associations between cancer and nutrients are
often based on the presence in the diet of certain foods. For
example, citrus fruits have sometimes been used as indicators of
A-10
The Relationship Between Nutrients and, Cancer 61
the presence of vitamin C, although they obviously have much more in
common than ascorbic acid. They contain, among other substances,
flavonoids (Chapters 13 and 15~. The dietary presence of vitamin A has
often been based on green and yellow vegetable consumption (Chapter 9),
although the active agent in those foods may not actually be vitamin
A. Peto _ al. (1981) suggested that carcinogenesis may be inhibited
by 6-carotene (the plant constituent that can be converted to vitamin A
in the body), rather than by the vitamin itself. Their report suggests
that, when examining naturally occurring compounds in foods, we should
not limit our attention to those already identified as having a nutri-
tional role.
Until fairly recently, fiber was also overlooked as a possible
protective factor in carcinogenes~s. For many years, fiber was re-
garded as a collection of inert substances in foods, even though it was
known to be present in relatively large amounts, compared to vitamins
and minerals. These substances were even regarded as a nuisance factor
that might interfere with the absorption of minerals in unrefined diets
Since most traditional diets contain large amounts of such indigestible
residues, fiber came to scientific attention as a result of observa-
tions that peoples consuming "primitive" diets high in complex carbohy-
ates (including fiber) appear to be spared a number of maladies, includ-
ing bowel cancer, that are common to populations consuming more refined
diets.
These simple observations have led to ongoing investigations con-
cerning which components of carbohydrate should "count" as fiber, which
of them might play a role in carcinogenesis, and how (or whether) fiber
affects the incidence of certain diseases or whether it acts merely by
displacing other dietary substances that are either carcinogens or
promoters of carcinogenesis.
The recent findings concerning fiber remind us again that sub-
stances in food other than those presently classified as nutrients may
be instrumental in the development of cancer. Milk is one major food
that is difficult to classify in cancer studies. As a source of
vitamin A (Mettlin and Graham, 1979), whole milk may be a beneficial
component of the diet; but as a source of fat (Blair and Fraumeni, 1978
Howell, 1974), it may have deleterious consequences. The category
"dairy products" or "milk products" may combine milk products such as
butter, cheese, cream, yogurt, low-fat milk, and cottage cheese, some
of which are very different from each other in composition. In a case-
control study conducted by Phillips (1975), dairy products other than
milk were associated with breast cancer. Hirayama (1977) reported that
the ingestion of two glasses of milk daily was associated with a lower
risk of gastric cancer in a large cohort.
There have been surprisingly few studies linking specific foods
with either increases or decreases in cancer rates. Where there have
been such studies, e.g., those on cruciferous vegetables, the data
A-11
62 DIET, NUTRITION, AND CANCER
underscore the fact that it will be difficult for epidemiologists to
sort out the specific chemicals of concern. For example, the consti-
tuents of cruciferae responsible for their apparent effect on the
occurrence of cancer may be, as Chapter 15 suggests, indoles, isothio-
cyanates, or other nonnutritive substances demonstrated to affect car-
cinogenesis in the laboratory. But it is not yet possible to attribute
the epidemiological associations to any such substances simply because
of the simultaneous presence in these vegetables of such other consti-
tuents as fiber, 6-carotene, ascorbic acid, or calcium.
Moreover, the identification of these associations is complicated
not only by the composite nature of single foods but also by the in-
terrelated variations in the intakes of a number of foods in any given
diet. By eating more broccoli, one ordinarily eats less of something
else. More broadly, those who increase their consumption of vegetable
products must, of necessity, simultaneously reduce their consumption of
animal products since these are the only two classes of substances
(other than table salt and water) ordinarily consumed by humans. A
reduced intake of animal products will normally result in a decreased
consumption of nutrients such as animal fat, animal protein, heme iron,
preformed vitamin A, and zinc; of mutagens formed during the cooking of
meat; and of such fat-soluble contaminants as pesticides and drugs used
for animals. This tendency for certain nutrients and other substances
to occur together in certain types of foods accounts for the strong
direct correlations among such dietary variables as beef, all meats,
animal fat, and animal protein in epidemiological studies. Moreover, a
reduced intake of animal products is necessarily accompanied by an
increased intake of substances such as starches, fibers, and certain
vitamins and minerals that are present in the substituted vegetable
foods. Since all these dietary constituents increase and decrease
simultaneously, it is difficult to determine which ones, if any, are
involved when, for example, consumption of animal products and cancer
rates decrease simultaneously or when control subjects consume more
animal products than do cancer cases.
Individual diets are not composed of isolated substances or even
isolated foods but, rather, they contain thousands of unique combina-
tions of nutrients and other compounds that comprise the individual
food items. From the standpoint of public education and public health,
therefore, it is considerably less important to identify isolated com-
pounds that cause or protect against certain cancers than it is to
identify dietary patterns that enhance or minimize overall risk. The
conclusions and recommendations contained in Chapter 1 reflect this
committee's assessment of the evidence regarding some components of
these patterns.
SUMMARY AND CONCLUSIONS
Since the turn of the century, there have been extensive changes
in foodstuffs consumed by the U.S. population. Only a few of these
A-12
The Relationship Between Nutrients and Cancer 63
changes have been measured, and then only crudely. Levels of nutrient
intake that have been monitored have remained relatively constant
between 1909 and the present, but data indicate that this constancy
obscures major unmeasured changes in intake of other substances result-
ing from the declining consumption of certain commodities; changes in
the forms in which foods are consumed; or the introduction of entirely
new products and substances.
The relationship between these changes in the food supply and the
incidence of cancer is not yet clear. The fact that the food supply
has undergone major changes while the rates of cancer at most sites
have been relatively constant may suggest that none of the changes has
an effect on cancer incidence, that the changes have occurred too
recently to produce an effect, or, more likely, that some changes have
had a positive and some a negative impact. Data reviewed in this re-
port indicate that a number of substances in food other than nutrients
may play a role in the causation or the prevention of cancer. Thus, it
may be important in epidemiolgical studies to consider a variety of
food classifications and to monitor changes in the food supply in
addition to those that affect nutrients.
A-13
64 DIET, NUTRITION, AND CANCER
REFERENCES
Blair, A., and J. F. Fraumeni, Jr. 1978. Geographic patterns of
prostate cancer in the United States. J. Natl. Cancer Inst.
61:1379-1384.
Brewster, L. M., and M. Jacobson. 1978. The Changing American Diet.
Center for Science in the Public Interest, Washington, D.C. 80 pp.
Comptroller General. 1979. Problems in Preventing the Marketing of
Raw Meat and Poultry Containing Potentially Harmful Residues.
Comptroller General's Report to the Congress of the United States,
No. HRD-79-10, April 17, 1979. General Accounting Office,
Washington, D.C. 87 pp.
Hirayama, T. 1977. Changing patterns of cancer in Japan with special
reference to the decrease in stomach cancer mortality. Pp. 55-75 in
H. H. Hiatt, J. D. Watson, and J. A. Winsten, eds. Origins of Human
Cancer, Book A: Incidence of Cancer in Humans. Cold Spring Harbor
Laboratory, Cold Spring Harbor, N.Y.
Howell, M. A. 1974. Factor analysis of international cancer
mortality data and per capita food consumption. Br. J. Cancer
29:328-336.
Mettlin, C., and S. Graham. 1979. Dietary risk factors in human
bladder cancer. Am. J. Epidemiol. 110:255-263.
Molitor, G. T. T. 1980. The food system in the 1980s. J. Nutr.
Educ. 12 (Suppl. 1~:103-111.
Page, L., and B. Friend. 1978. The changing United States diet.
Bioscience 28 :192-197.
Peto, R., R. Doll, J. D. Buckley, and M. B. Sporn. 1981. Can
dietary carotene materially reduce human cancer rates? Nature
290:201-208.
Phillips, R. L. 1975. Role of life-style and dietary habits in risk
of cancer among Seventh-Day Adventists. Cancer Res. 35:3513-3522.
Smith D. T. 1980. Antibiotic additives: The prospect of doing without.
Farmline 1~9~:14-15.
U.S. Food and Drug Administration. 1980. Compliance Program Report
of Findings. FY 77 Total Diet Studies--Adult (7320.73~. Bureau of
A-14
~e R^~~ Be- -~ ~d ~~ ~
Foods, food and Drug Administration, U.S. Department of Health,
Education, and Welfare, Washington, D.C e [ 33] pp.
Ue S" International Trade Co _ fission. 1980. Synthetic Organic
Chemicals. United States Production and Sales, 1980 e USITC
Publication 1183e Office of Industries, OeSe International Trade
Co _ ission, Washington, D eC e 327 pp e
A-15
