EPIDEMIOLOGIC RESEARCH:

1

• EPIDEMIOLOGIC RESEARCH

Things to Think About
Including some examples from the wonderful world
of diet and cancer
2

• James R. Hebert, Sc.D., Professor, Arnold School
  of Public Health Department of Epidemiology and
  Biostatistics

Director, South Carolina Statewide Cancer
Prevention Control Program
3
Observations

• We are generally driven by Outcome yield (as
  indicated by high rates of disease) go to SC
  because CA rates are very high here
• But Information yield (as indicated by
  optimizing the likelihood of observing a true
  relationship between the risk factor under study
  and the outcome RR or OR) is far more
  important

4
Fat Consumption (proportion EI) Comparing US
individuals to 154 Country Means
5
Effect of Truncating Tobacco Exposure Hebert JR,
Kabat GC. J Natl Cancer Inst 199183872-4.
Males
6
Effect of Truncating Tobacco Exposure Hebert JR,
Kabat GC. J Natl Cancer Inst 199183872-4.
Females
7
Observations

• Timing of exposure in relation to outcome (and
  therefore measurement of things such as cancer
  initiators and promoters) is crucial (etiologic
  relevance)
• Non-linearity is the norm for biological
  processes, perhaps more so for carcinogenesis
  than for more mundane processes (e.g., cancers
  often exhibit log-normal growth)

8
Examples

• Uterine thalidomide exposure and phocomelia
• Calcium intake and bone mineralization
• Tobacco and a variety of cancers/ precancers
  (e.g., oral leukoplakia)
• Diet BrCA
• Physical activity BrCA
• BrCA timing of surgical interventions
• Etc., etc., etc., etc., etc., etc., etc., etc.,
  etc.,

9
Observations

• Outcomes (e.g., cancers of a particular anatomic
  site) may represent 2 sometimes very
  different diseases
• Susceptibility may vary drastically across
  individuals
• Conventional views of reality are conditioned by
  world view of researchers and accessibility/compli
  ance of subjects

10
Esophageal Cancer

• In Western countries, 95 of the variability in
  incidence is attributable to tobacco and alcohol
  use.
• Within the U.S., rates for Blacks are about 2-3
  times those of Whites (and they were 3-4 times
  just 15 years ago!) yet the use rates of tobacco
  and alcohol are about equal. Also see
• 1. Hebert JR. Differences in biological responses
  to cigarette smoking remain unexplained. Am J Pub
  Health 1991 811679-1680.
• 2. Hebert JR, Kabat GC. Menthol cigarette smoking
  and Oesophageal cancer Results of a case-control
  study. Int J Epidemiol 1989 1837-44.

11
Esophageal Cancer

• Right now in the U.S. esophageal cancer incidence
  rates are
• ?205 in Black men compared with Whites
• ?240 in Black women compared with Whites

12
Esophageal Cancer

• Incidence rates of squamous cell carcinoma, the
  common form in African Americans, are falling
  (slightly)
• Rates of adenocarcinoma are rising (rapidly), and
  this change is seen almost exclusively in Whites

13
Esophageal Cancer Incidence, by Age Sex
14
Esophageal Cancer

• Mortality rate differences are even larger than
  incidence rate differences
• Black men in SC have an esophageal cancer
  mortality rate that is 3.83 times higher than
  that of their White counterparts
• The mortality rate for Black women is 2.53 times
  higher than that of their White counterparts

15
Esophageal Cancer Mortality, by Age Sex
16
(No Transcript)
17
(No Transcript)
18
Esophageal Cancer

• Non-Western populations (in China, South American
  and in the Caspian Littoral) have rates higher
  than Western populations and much lower exposures
  to tobacco and alcohol
• Interactions evident in Western populations do
  not seem to be fully operative and there appears
  to be at least one other main effect or some
  other important interaction

19
State of South Carolina
Total Area 31,113 mi2 Total Population
4.2m Proportion AA 31 gt40 of rural
population is AA
20
State of South Carolina
21
Esophageal Cancers in SC 1996-2000
Squamous
Adenocarcinoma
Squamous
Squamous
22
Menthol Cigarette Sales and Age-Adjusted
Esophageal Cancer Rates in Blacks
23
How Could Menthol Explain these Differences?

• Pyrollized menthol could exert a direct effect or
  its mild anesthetic properties could lead to
  changes in diet e.g., allowing smokers to
  consume beverages at a higher temperature
• Menthol may modify specific nutrient effects

24
In Our 1989 Study Of Menthol Cigarettes
Esophageal Cancer, we Looked at

• Current Cigarette Smokers
• Cases With Esophageal Cancer
• 216 Males 96 Females
• Controls With Non-Tobacco-Related Diseases
  Matched on Age (?5years) and Sex
• 305 Males 157 Females
• Participants from 20 Hospitals in 9 U.S. Cities

Hebert JR, Kabat GC. Menthol cigarette smoking
and Oesophageal cancer results of a case-control
study. Int J Epidemiol 19891837-44.
25
Statistical Methods

• Simple Univariate Statistics
• Exploratory Analyses to assess relationships
  among the variables
• Two-Stage Logistic Regression (to account for
  intercorrelations, to allow for increased degrees
  of freedom, and to provide a conservative
  estimate of the effect of menthol cigarettes on
  risk of esophageal cancer)

26
PERCENTAGE IN OUR STUDY GROUP EVER SMOKING
MENTHOL BRANDS UNAMBIGUOUS
WHITES WHITES WHITES NON-WHITES NON-WHITES
Cases Cases Controls Cases Controls
MALE MALE 10 15 10 15
FEMALE FEMALE 22 13 32 30
27
CRUDE ODDS RATIOS - FEMALES
O.R. 95 CI
EVER MENTHOL 2.05 (1.09-3.87)
gt 10 YRS MENTHOL 2.03 (0.91-4.56)
gt 15 YRS MENTHOL 2.83 (0.77-10.36)
Cases 96 Controls 157 Cases 96 Controls 157 Cases 96 Controls 157 Cases 96 Controls 157

N
28
RESULTS OF FIRST STAGE LOGISTIC MODELS - FEMALES
Odds Ratio 95 Confidence Interval P-value

Education (gtHS) 0.72 0.37-1.38 0.33
Religion (Jewish) 0.16 0.05-0.49 0.001
Alcohol (gt 1oz/d) 2.79 1.55-5.03 0.0006
Race (Black) 4.60 2.11-10.04 0.0001
Cigarettes (20/d) 1.52 0.79-2.79 0.22

Non-Menthol Smoking (years) 1.02 0.99-1.05 0.15
29
RESULTS OF SECOND STAGE FITTING - FEMALES
Odds Ratio 95 Confidence Interval P-value
Stage IIa Menthol Smoking (lt10 y) 1.50 0.54-4.17 0.44
Menthol Smoking (?10 y) 2.30 0.93-5.72 0.07
Stage IIb
Menthol Smoking (years) 1.05 0.75-4.17 0.09
30
THESE RESULTS DO NOT RESOLVE THE ISSUE
DEFINITIVELY BECAUSE OF

• The lack of representativeness of Blacks in our
  study population
• The inability to define menthol cigarette
  exposure unambiguously
• The lack of data on potential effect-modifiers,
  such as dietary factors (including temperature of
  beverages consumed)

31
New Hypothesis

• Menthol, alone or in combination with EtOH, may
  modify permeability and solubility of tobacco
  carcinogens
• Azzi C, Zhang J, Purdon CH, Chapman JM, Nitcheva
  D, Hebert JR, Smith EW. Permeation and reservoir
  formation of 4-(methylnitrosamino)-1-(3-pyridyl)-1
  -butanone (NNK) and benzoapyrene (BAP) across
  porcine esophageal tissue in the presence of
  ethanol and menthol. Carcinogenesis
  200627(1)137-45.

32
Observations

• Humans are not simple systems (so interactions
  occur on multiple levels)
• We have done a poor job of anticipating, let
  alone helping to create scientific innovation,
  though there are occasional flashes of brilliance

33
Observations

• What people know (or believe that they know) is
  not always obvious because we epidemiologists
  rely heavily on self-assessment, this can be a
  real problem
• We need to anticipate as much as possible about
  peoples motivations and beliefs

34
Observations

• We need to realize that what cannot be controlled
  by design may be controlled analytically
• This kind of control may be both necessary and
  preferable - the RCT is often impractical, it is
  not true to life, and it is almost always
  expensive

35
Yet More Observations

• Ecological studies produce results that are
  often very different from Analytic studies
  e.g., dietary factors explain 80-90 of the
  variability in hormone-sensitive CA rates in the
  former (at least in cross-national studies) and
  around 10 in the latter
• Correlatedness among cancers changes with level
  of aggregation (internationally r 1.0 for BrCA,
  PrCA CRCA), but not within populations (an
  example of ecological inversion)

36
Now, Some Questions

• If such large differences exist in rates, why are
  estimates for most risk factors (OR or RR) lt3.0?
• Why are results so different according to level
  of aggregation (e.g., cross-nationally vs.
  between states in the U.S.)?
• Why are results derived in studies of individuals
  so different from those obtained in ecological
  studies?

37
How About Some Answers?

• Diet assessment methods are inadequate to
  estimate true exposure with reasonably sufficient
  accuracy and precision, especially over long
  periods
• Based on our work, resulting biases could
  easily distort epidemiologic effect estimates
• Use of retrospective diet assessment methods in
  case-control study designs introduces additional
  information bias
• And this also could distort estimates of effect

38
From the Conventional Epidemiologic Perspective,
there is

• Type I Error the probability of accepting the
  alternative hypothesis of an effect on the
  assumption that the null hypothesis is true
  (often worded as the error of rejecting a true
  null hypothesis declaring a difference when
  one does not exist) Test statistic exists
  usually a 0.05 (expressed as the p value)

39
And

• Type II Error the probability of accepting
  the null hypothesis on the assumption that the
  alternative hypothesis is true (often worded as
  the error of failing to reject a false null
  hypothesis i.e., declaring that a difference
  does not exist when in fact it does) Test
  statistic exists usually 1-b 0.80 (100
  power 80)

40
There is a More Insidious Category

• Type III Error incorrect inference resulting
  from a faulty conception of how the world works,
  or selection of a study design that produces an
  answer even if correct to the wrong question
  Test statistic does not exist

41
Another Answer (more specifically to the Type III
Error Problem)

• Within-study-group contrasts, in both relevant
  exposures and cancer-related outcomes, are often
  lacking
• this reflects the Information Yield vs.
  Outcome Yield problem

Also see Hebert JR. Epidemiologic studies of
diet and cancer The case for international
collaboration (Commissioned as part of the
Eminent Scientist of the Year 2004 Award by the
World Science Forum). Austro-Asian J Cancer
20045(Special Issue - Recent Advances and
Research Updates)140-53. Hebert JR. Invited
commentary menthol cigarettes and risk of lung
cancer. Am J Epidemiol 2003158617-20.
42
So, what can we do about this?

• Study disease-risk factor relationships in
  situations where we expect to obtain the largest
  information yield
• Create large study-group contrasts in relevant
  exposures by designing and implementing effective
  interventions
• Engage other basic, clinical, and behavioral
  scientists in considering what is really
  important and in paying attention to important
  clues that are not readily discernable from their
  vantage points (thereby lowering type III error
  rates)
• Be open-minded and inquisitive be willing to say
  yes

43
Thanks to the Many People and Institutions that
have influenced my thinking, especially

• University of Washington
• Cole P. Dodge (UNICEF)
• Ross Prentice (FHCRC)
• Harvard University
• Glorian Sorensen (DFCC)
• Karen Peterson
• Mohamed el Lozy
• Walter Willett
• Larry Kushi (Kaiser P.)
• Bombay University Healis
• Prakash C. Gupta
• Boston University Bedford VA
• Donald Miller

• American Health Foundation
• Ernst Wynder (deceased)
• Geoffrey Kabat
• University of Massachusetts
• Ira Ockene (Prev Cardiology)
• Judy Ockene
• Jon Kabat-Zinn (emeritus)
• University of South Carolina
• Tom Hurley (EPID-BIOS)
• Bill Hrushesky (Dorn VA)
• Frank Berger (Biol CAS)
• Jane Teas (SCCC)
• Harris Pastides (EPID-BIOS)