EPIDEMIOLOGIC PERSPECTIVES ON
LIFE-STYLE MODIFICATION AND
HEALTH PROMOTION IN CANCER RESEARCH’
David D. Celentano2
Research, intervention, and evaluation of tion” is the epidemiologic method. As the
programs to reduce behavioral risks for inci- basic framework of public health, insights
dent cancers, promote early detection, alle- gathered from epidemiologic investigations
viate pain and dysfunction, and improve the on these issues offer avenues for addressing
quality of life of patients with cancer typi- factors of interest in behavioral and psy-
cally address these issues from the perspec- chosocial research.
tive of individual patients. This reflects the Under the rubric of life-style factors and
disciplinary bases of psychosocial oncology health promotion we include the following:
research (predominately the clinical, patient- (1) those factors that reflect societal position,
oriented focus of medicine and psychology) including social class/socioeconomic status
and their research methods. While providing and occupation; (2) behavioral risk factors as-
valuable insights into the course of disease sociated with elevated rates of cancer, prin-
and the impact of coping strategies, this ap- cipally smoking (and smoking cessation),
proach does not place the findings in a alcohol consumption, diet, and exposure to
broader context. The extent to which these sunlight; (3) screening behaviors, factors as-
experiences can be generalized cannot be de- sociated with adherence to early detection of
termined. premalignant disease; and (4) psychosocial
A complementary approach to assessing factors, including personality, coping, social
the variables we aggregate under the head- support, and related issues. Several recent re-
ings “life-style factors” and “health promo- views (1, 2) have addressed many of the is-
sues in considering life-style factors in
relation to cancer initiation and promotion.
Source: Cancer 1991; 67 (3, suppl. 1): 808-812. Reprinted
with permission from J. 8. Lippincott Company, Phila-
delphia, Pennsylvania., United States of America.
‘Presented at the American Cancer Society’s Second THE EPIDEMIOLOGIC PERSPECTIVE
Workshop on Methodology in Behavioral and Psychosocial
Cancer Research, Santa Monica, California, United States
of America, December 5-8,1989. Epidemiology is principally concerned
Qvision of Behavioral Sciences and Health Education, with describing the distribution of diseases
Department of Health Policy and Management, Johns
Hopkins School of Hygiene and Public Health, Baltimore, within defined populations and assessing the
Maryland, United States of America. determinants of disease (etiology) (3, 4).
Kleinbaum et al. (5) added two additional The unifying framework of the epidemio-
aims: to predict future cases and the health of logic approach is the host-agent-environment
the population and to control disease by pre- paradigm initially developed to explicate in-
vention, eradication, prolongation of life, and fectious disease cycles (7). Three influences
improving the quality of patients’ lives. An must simultaneously be addressed to deter-
understanding of the natural history of dis- mine the likelihood of disease: (1) the sus-
ease and how psychosocial factors affect it ceptibility of the host (an individual, a
suggests points for intervention to influence subgroup, or the entire population); (2) the
its course. agent, or that which is directly implicated as
Fletcher et al. (6) contrasted epidemiology the source of the problem; and (3) the envi-
from clinical approaches along the following ronment, which includes all other aspects of
domains: (1) epidemiology locates its obser- the situation, and must be viewed as dynamic
vations within specific groups (representa- in nature. The clinical perspective is focused
tiveness), (2) including all members, whether on decision-making oriented toward disease
or not they have come to the attention of the mechanisms of the basic sciences to alter the
medical care system; and (3) data are analyzed course of disease. These paradigm differences
on an aggregate level using (4) somewhat lead to separate pathways to conceptualiza-
crude categories of disease by clinical stan- tion and intervention.
dards. Epidemiology is somewhat more in- An example demonstrates these sources of
terested in determining (5) how often and in influence and differences in the public health
what circumstances an event occurs, rather approach to intervention from the clinical ap-
than exactly how it occurs and (6) analyzing proach. In viewing the influence of cigarette
data based on populations to determine the smoking on lung cancer, it is clear that al-
relative increases or chances that exposures though we have not yet uncovered the etiol-
lead to disease. Clearly, this approach differs ogy at the ultimate pathogenic, mechanistic
dramatically from individually based per- level, the epidemiologic evidence is clear.
spectives. Table 1 outlines a comparison of Rather than a simple cause-effect relation-
clinical and epidemiologic approaches to ad- ship, environmental factors must also be ad-
dressing psychosocial and behavioral factors dressed, including other exposures that might
in cancer research. It serves as a heuristic for elevate the risks for lung cancer (e.g., the syn-
defining key epidemiologic parameters dif- ergistic relationship between cigarette smok-
ferentiating these perspectives. ing and asbestos exposure on lung cancer
TABLE 1. Comparison of Clinical and Epidemiologic Approaches to the Investigation of life-style
Behaviors in Cancer Research
Issue Clinical Approach Epidemiologic Approach
Purpose Determine occurrence in individual patients Determined relative frequency in population
Focus Biologic, personality traits Interaction of host, agent, and environment
Source of subjects Patient series (referrals, admissions) Patients, community, or population survey
Designs Experimental, quasi-experimental Observational, analytic
Treatment of time Change, survival Retrospective and prospective, survival
Statistical methods Comparison of groups (means, ANOVA) Analysis of risk, assessment of
or to published norms dose-response effects
Reliability Test-retest, internal consistency Internal consistency, record verification,
Validity Criteria (physiologic and/or “expert”) Selection bias, confounding
Ceneralizability ? To population from which derived
Inferences Prognosis, recovery, personality Relative risk, etiology, social context
Prevention focus Individual behavior change, group norms Community and cultural change, public policy
234 Epidemiologic Perspectives
incidence). Other workplace exposures might tant consideration. Selection bias is the amount
prove equally important in either initiating or to which we overestimate or underestimate the
promoting the carcinogenic process. The clin- effect (risk) resulting from how subjects are se-
ical approach would also include a similar lected for the study. The principal sources of
line of questioning, although the aim would selection bias are the following: (1) comparison
be to determine the biologic significance for groups, common to almost all research de-
the patient in question. For example, in ap- signs, (2) sampling frame, (3) incomplete fol-
proaching a smoking history, equivalent in- low-up and nonresponse, and (4) selective
formation would be assessed, although the survival (8). Of particular concern is the re-
use of these data would differ. For the epi- liance upon hospital-based comparison
demiologist, the rate of smoking in the sub- groups (referred to as Berksonian bias f91).
population from which the case resides
would be of paramount interest, as well as the
comparison of this rate to a larger defined STUDY DESIGN
population and its rates of lung cancer occur-
rence. Furthermore, the fact that smoking is Design differences also are apparent in the
more common among blacks and persons of comparison of the clinical and the epide-
lower socioeconomic status would be ana- miologic/public health approach. Clinically
lyzed to determine if excess deaths were as- based investigations typically use experimen-
sociated with these factors, simultaneously tal protocols or quasi-experimental designs.
being aware that access to medical care, pre- Although control groups are commonly used,
ventive intervention services, and counseling their source and generalizability are often open
strategies are less available to high-risk sub- to question. More commonly, patients are ran-
groups. Prevention strategies would go be- domized to various treatments, and group
yond advising smoking cessation and include means are then compared over time. With re-
policy level strategies to decrease smoking by spect to life-style behavior modification, indi-
advocating tax code changes and reducing vidualized approaches are compared (e.g.,
availability of tobacco products for particu- individual counseling versus group counsel-
larly vulnerable populations. ing versus physician advice) and relative quit
rates monitored over time. Similar interven-
tions are seen for other factors, such as dietary
SOURCES OF SUBJECTS modifications, controlling alcohol use, and pro-
moting aerobic exercise. Public health ap-
The source of subjects demonstrates a major proaches tend to focus more on altering the
difference in the two methods. Traditionally, community and cultural context within which
clinical (and psychologic) research has relied the individual lives (20). To address these is-
upon patients presenting for treatment at a sues, data are collected using strategies that
given facility If there are no barriers to health provide information that can be transcribed
care utilization and if the factor under consid- into comparisons of rates among those with
eration is independent of care seeking, patient and without specific risk factors (and combi-
series are an acceptable method of subject ac- nations thereof). Typically, when the disease is
crual. Much of the psychologic and behavioral rare, case-control epidemiologic studies are uti-
literature on life-style risk factors and cancer lized (II), in which retrospective recall of ex-
utilizes just such samples. If, however, there are posures to risks is ascertained for defined cases
selection criteria operable that differentially and selected controls (e.g., cases with invasive
triage patients to treatment on the basis of other cervical cancer matched with a random sam-
characteristics associated with the factor under ple of women from the same area who are of
investigation, then bias emerges as an impor- the same age and race fZ21). Cohort studies
prospectively observe a population enrolled at variety of chi-square methods). In particular,
baseline to determine directly the risks associ- there is a heavy reliance in epidemiology for
ated with incident disease. The latter design using multivariate statistical analysis, often
avoids many of the biases associated with case- logistic regression methods, reflecting the un-
control studies, but the time and expense re- derlying multifactorial framework. In clinical
quired must be considered (13). Clinical trials studies the use of ANOVA and similar tech-
(both nonrandom and randomized) can be niques predominates, reflecting small sample
viewed as subsets of cohort studies, allowing sizes but the general approach as well.
a more informed assessment of association and
TREATMENT OF TIME Test-retest methods of assessing reliability
(or precision) predominate in clinical frame-
Clinical approaches to the analysis of time works, although there is concern with mea-
are predominately focused on assessing sures of internal consistency for multi-item
change in physiologic functioning as well as scales. Epidemiology relies more heavily upon
patient illness status; recall of history or record verification for the detection of infor-
events is also manifest. Issues of survival are mation bias, distortion in the estimation of im-
essential in clinical research. Epidemiologic pact (or risk) due to measurement error, or
approaches to the subject of time take similar subject misclassification (5). This emerges
perspectives, although the research designs from faulty measurement of the exposure con-
treat time somewhat differently. Retrospec- dition (poorly worded questionnaire, inter-
tive recall of exposures in cases and controls view procedure, or indicator) or the disease
is commonly assessed, whether in the case- condition (any inaccurate diagnostic proce-
control design or in prospective cohorts, to dure) .
elucidate risk. Survivorship, whether based
on proportional hazards models or simpler
life table enumeration, is a common way of VALIDITY
Assessments of validity are most commonly
appeals to criteria in clinical research. Either
STATISTICAL METHODS consensus (or prevailing wisdom) exists on
cutting points for the determination of impor-
The analytic methods selected for making tant treatment decisions or critical values are
comparisons directly follow the design and accepted. For example, investigations of psy-
purpose of the investigation. Clinical reports chologic functioning might be verified by a
frequently compare groups (treated versus psychiatrist’s rating or by exceeding standard
not treated; smokers versus nonsmokers) and values. In epidemiologic research two issues
report on differences in mean values or com- are of greatest relevance: ruling out selection
pare the results of a patient series to published bias and confounding. Kleinbaum et al. (8)
normative data. Epidemiology has its own set demonstrated how selection factors can be cor-
of statistics (odds ratio, relative risk, attribut- rected or avoided, principally through design
able risk) that have been developed to assess considerations or through analysis. In either
different types of risk assessments; in addi- case a determination can be made when selec-
tion, the concern with dose-response rela- tion bias exists and its impact on both the mag-
tionships also calls for alternative forms of nitude and direction of the bias. Confounding
determining differences in groups (hence the results can be achieved by a risk factor being
236 Epidemiologic Perspectives
affected by other extraneous (and perhaps this goal differs, however. Individual behav-
causally linked) factors. In general, compar- ior change is generally the strategy employed
isons of crude versus adjusted effects are made clinically. Physicians provide advice (smoking
to determine if distortions are present. cessation, alcohol consumption reduction,
weight and cholesterol control, increased ex-
ercise) or make referrals to formal programs
GENERALIZABILITY (e.g., SmokeEnders, AA, Weight-Watchers),
which also have the individual as the focal
The extent to which one may generalize point. Family members or peers may be in-
from clinical investigations is dependent volved (e.g., in cardiac rehabilitation exercise
upon the factor being addressed. Certainly, if programs) but primarily as a source of sup-
measures of interest are independent of pa- port to maintain adherence to the recommen-
tient selection factors, there are no limits to dation. The public health approach seeks to
generalizability. However, if the investigation alter individual behavior through interven-
includes factors that might be affected by pa- tions focused upon the community at large.
tient characteristics also associated with their Whether via public policy (banning of smok-
referral status (e.g., patients treated at a com- ing in public areas, on airplanes, or in offices)
prehensive cancer center), then inferences or through mass media and other community-
must be closely held. One of the features of the based approaches, the intent is to alter com-
epidemiologic investigation is that one may munity norms. Environmental alteration
generalize directly to the population from (eliminating or reducing risk or exposure) is
which cases and controls, or the prevalent co- the preferable route to promote behavior
hort, are drawn. change, thereby removing decision-making or
the need for behavior change by individuals.
The principal aim of the clinical approach
is to predict patient prognosis and to deter-
This discussion is limited in the depth of the
mine factors that may impede optimal func-
comparisons being drawn. In some respects
tioning or recovery. The focus is on the
we have used a heuristic to demonstrate the
individual patient or the treatment. The epi-
utility of epidemiologic methods to under-
demiologic aim is to ferret out presumed risk
stand life-style behavior factors in cancer re-
factors impinging upon the population at
search, such as tobacco use and alcohol
large and to compare these risks within the
consumption, and issues such as personality
dynamic of the environment. In some respect
and coping style. An appreciation of the epi-
the community focus of epidemiology is
demiologic method and the public health
sometimes forgotten but underlies the generic
paradigm demonstrates the remarkable dif-
ference in approach from the traditional clin-
ical approach. Both approaches are essential
to understand the dynamics of patient behav-
PREVENTION FOCUS ior and risks for cancer. In addition, these per-
spectives must be simultaneously integrated
Altering individual behavior to reduce risk with psychosocial (stress) paradigms in which
or to improve outcome is the ultimate goal of personal and social resources are considered.
both the clinician and the public health It is clear that a great deal has been written
practitioner. The route by which each reaches concerning life-style risk factors and their im-
pact on cancer risks. If we review the evidence research is the integration of a variety of re-
outlined by Doll and Peto (24), the majority of search traditions and paradigms, including
risks for cancer deaths are self-imposed, prin- basic medical sciences, health psychology,
cipally reflecting tobacco use and dietary fac- medical sociology, anthropology, and epi-
tors. Other factors, such as alcohol use, sexual demiology. This integration stretches our
behavior, and occupational exposures, surely contemporary knowledge and requires new
should not be ignored, but they contribute lit- thinking and perspectives on how people
tle in comparison to smoking and dietary fat confront and cope with cancer.
(although they have unique and important
public health relevance for other problems).
In reviewing the evidence for smoking and di- REFERENCES
etary interventions, it is clear that we have
been relatively unsuccessful to date in modi- 1. Levy SM. Host differences in neoplastic risk:
fying cancer risk behaviors. Whether this re- Behavioral and social contributors to disease. Health
Psycho1 1983; 22144.
flects inappropriate paradigms, naive views
2. Holland JC. Behavioral and psychosocial risk fact-
of the intervention process, or a lack of aware- ors in cancer. Human studies. In: Holland JC.
ness of the behaviors themselves must be ad- Rowland JH, eds. Handbook of Psychooncology:
dressed (25). Psychological Care of the Patient with Cancer. New
Social epidemiology, the study of the psy- York: Oxford University Press. 1989; 705-726.
3. McMahon B. Pugh TF. Epidemiology: Principles and
chosocial determinants of physical health
Methods. Boston: Little, Brown and Company, 1970.
status, recognizes that social and cultural 4. Lilienfeld AM, Lilienfeld DE. Foundations of
factors are important in the etiology of Epidemiology. New York: Oxford University Press,
chronic disease (26-18). Kasl(Z9) suggested 1980.
that the contribution of this approach lies in 5. Kleinbaum DC, Kupper LL. Morgenstem H.
Epidemiologic Research. Belmont, CA: Lifetime
the level of aggregation and abstraction of
psychosocial/behavioral factors compared 6. Fletcher RH, Fletcher SW, Wagner EH. Clinical
with biologic factors. For example, associat- Epidemiology: The Essentials. Baltimore: Williams &
ing trace elements or specific dietary con- Wilkins, 1982.
stituents with a specific physiologic response 7. Sartwell PE, Last JM. Epidemiology. In: Last JM. ed.
Public Health and Preventive Medicine. New York:
may be demonstrated but would certainly re-
Appleton-Century-Crofts. 1980; 9-85.
main masked were we to reduce our focus to 8. Kleinbaum DC, Morgenstem H, Kupper LL.
“dietary intake.” Many of our measures of be- Selection bias in epidemiologic studies. Am J
havioral risk factors [How many cigarettes do Epidemiol 1981; 113: 452463.
you smoke per day? How often do you exer- 9. Be&son J. Limitations of the application of fourfold
table analysis to hospital data. Biometrics 1946; 2:47-53.
cise? What did you eat for the last 24 hours?]
10. Kasl SV. Social and psychological factors affecting the
are at this aggregated level. The new mea- course of disease: An epidemiological perspective.
surement methods developed in the Stanford In: Mechanic D, ed. Handbook of Health, Health
Five-Cities Project for exercise may serve as Care, and the Health Care Professions. New York:
a model for the developmental activities re- Free Press, 1983.
11. Breslow NE, Day NE. Statistical Methods in Cancer
quired for other behaviors (20).
Research. vol. 1. The Analysis of Case-Control
Clearly, we face a challenge in devising re- Studies. Lyon, France: International Agency for
liable, valid, and simple measures of human Research on Cancer, 1980.
behavior that lead to risks for cancer. Only _-.
13 Celentano DD, Klassen AC, Weisman CS, Rosen-
through links with medicine can we develop shein NB. Cervical cancer screening practices among
older women: Results from the Maryland Cervical
such measures and approaches, but it will
Cancer Case-Control Study. ] Clin Epidemiol 1988;
also require education on the part of the so- 41:531-541.
cial sciences. Furthermore, a challenge con- 13. Breslow NE, Day NE. Statistical Methods in Cancer
fronting the field of psychosocial cancer Research. vol. 2. The Design and Analysis of Cohort
238 Epidemiologic Perspectives
Studies. Lyon, France: International Agency for diseases. In: Freeman HE, Levine S, Reeder LG, eds.
Research on Cancer, 1987. Handbook of Medical Sociology. Englewood Cliffs,
14. Doll R, Peto R. The Causes of Cancer. New York: NJ: Prentice-Hall. 1979; 71-96.
Oxford University Press, 1981. 18. Sexton MM. Behavioral epidemiology. In: Pomerleau
15. Leventhal H, Cleary I’. The smoking problem: A re- OF. Brady JP, eds. Behavioral Medicine: Theory and
view of the research and theory in behavioral risk Practice. Baltimore: Williams & Wilkins, 1979.
modification. Psychol Bull 1980; 88:370-405. 19. Kasl SV. Stress and health. Annu Rev Public Health
16. Cassell J. An epidemiologic perspective of psy- 1984; 5:319-341.
chosocial factors in disease etiology. Am J Public 20. Sallis JF, Haskell WL, Wood PD et al. Physical activ-
Health 1974; 64:1040-1043. ity assessment methodology in the Five-City Project.
17. Graham S, Reeder LG. Social epidemiology of chronic Am 1 Epidemiol 1985; 121:91-106.