Tip:
Highlight text to annotate it
X
Art Reingold: Good morning. Morning. So as I
said on Monday, today represents somewhat of a transition. So
for the last several sessions we've been talking about
experimental study designs in which the investigator randomly
assigns people to one exposure group or the other. Most
epidemiologists who are outside a clinical setting don't do
randomized control trials or field trials. And most
epidemiologic studies are observational in nature rather than
experimental in nature. And so for the next couple of weeks
we're going to be now focusing on what we refer to as
observational study designs and the quintessentially most
important difference between experimental studies and
observational studies is the fact that people end up in
different exposure groups by themselves rather than being
randomly put into one exposure group or another.
And as you might imagine, that poses challenges because
in the real world people who choose to smoke or not to smoke
are different people. They differ with regard to many, many
other potential exposures and factors. People who choose to
take oral contraceptive pills or not. People who choose to use
condoms or not. Whatever the exposure is in the real world
when people are making their own choices rather than been
randomly assigned. There is the very real possibility that the
people in the two different exposure groups will differ in very
important other ways. And this really gets to an issue called
confounding and we'll talk about confounding in a few weeks.
But confounding is really the key difference between randomized
studies and observational studies. So there are some of some
sort of interesting other study designs we'll talk about.
There are some sort of hybrid study designs. But when you
think about analytic study designs there are basically three
broad types: Cohort studies, cross sectional studies and case
control studies.
So if you basically think about there being an exposure
and an outcome, cohort studies are studies in which we start
out putting people into different exposure categories.
Ascertaining exposure and then determining who develops the
outcome and who doesn't in different exposure groups. So you
determine exposure first and then you determine outcome after
that.
In cross sectional studies you ascertain them both
simultaneously. And in case control studies you start out
ascertaining the outcome. You know the outcome status first.
And then you go back and retrospectively determine the
exposure. These are in theory three mutually exclusive
categories in terms of which are you determining first. And
it's remarkable to me how often even doctoral students who are
in very sophisticated programs seem not to be figure out given
a study which category it falls into. So what I'd like to
suggest you do, any time you see a study, ask yourself a
question, what is the exposure of interest? What is the
outcome of interest and which did they know first? Which did
they determine first. And then it should be immediately
apparent what the study design is.
Okay?
But for some reason a lot of students seem to get this
wrong. Okay. Cohort studies are what we're going to be
talking about today, Friday and next Monday before moving onto
cross sectional studies.
So cohort studies are also referred to as longitudinal
studies. They can be certainly be looked at, they can
certainly be used to look at exposures that actually increase
the risk of disease or exposures that decrease the risk of
disease. In theory there are fewer ethical concerns about
doing observational studies than experimental studies because
people are in these groups to begin with. Although you could
legitimately ask about the ethics of following people with some
exposure you think is dangerous to them. And simply watching
them in their exposure category and seeing what happens to
them. So, these studies are not free of ethical concerns.
Um, the cohort studies are the studies that most
closely mimic randomized controlled trials. Because in both
settings these cohort studies and in randomized trials we start
out knowing who is in different exposure groups. We follow
them to see who develops the outcome and who doesn't. Okay?
The difference is randomization takes place in the
trials, in the experiments and it doesn't take place in cohort
studies. So this absence of randomization, the likely unequal
distribution between exposure groups and potentially
confounding factors is the key difference between trials or
experiments and cohort studies.
The concept of a cohort study comes from the Roman army. And if you've
ever seen any of those movies, Ben-Hur, or whoever it is, and you have a
picture of Roman soldiers, the Roman soldiers in theory march forward
into battle together. And a cohort was in the Roman army one of ten
divisions of a legion, a company or a band especially of warriors so it
was a group of Roman soldiers and the idea is here a cohort is a
situation which people are marching forward together in time. Okay? So
different people in different exposure groups marching forward together
in time. Now in cohort studies in different exposure groups you can
calculate sometimes the rate, sometimes the risk. If you can calculate
rates and risk you can calculate rate ratios and risk ratios. Rate in
the exposed divided by rate in the unexposed. Risk in the exposed
divided by risk in the unexposed. Hazard rate in the exposed divided by
hazard rate in the unexposed.
You can also calculate odds ratios. And odds ratios
are a perfectly legitimate effect measure, measure of
association in a cohort study. But traditionally people tend
to prefer to calculate rates or risks and rate and risk ratios
rather than odds ratios and I will show you a cohort study in
which the investigators calculated odds ratios. You can also
calculate rate or risk differences and you can calculate the
attributable fractions either of exposed or in the population.
So, there are a number of things in terms of measures
of association and measures of potential impact that you can
calculate in a cohort study. So the basic notion here just
like in a randomized trial we put people into different
exposure groups, smokers, nonsmokers, whatever. We follow them
over time. We see who gets the disease and who doesn't. In
this instance I'm just showing you the calculation of
cumulative incidence. So it's just the number who developed
the disease in the exposed group over the total number exposed.
The number who developed the discuss in the unexposed group or
the total number of unexposed. And then you can divide one by
the other or subtract one from the other to calculate a risk
ratio or a risk difference.
You can do the same thing using person time in the
denominator -- so, sorry. The risk ratio then in this is
basically A over A plus B divided by C over C plus D. So risk
in the exposed divided by risk in the unexposed is the risk
ratio. You can do the same thing in many cohorts calculating
incidence density and therefore having in the denominator
person time instead of number of persons. And you can then
calculate incidence density rates and then calculate a rate
ratio. Rate in the exposed divided by rate in the unexposed.
Sorry, by the way I was told these slides didn't get posted in
advance. I apologize for that. They'll be posted right
afterwards. There was a screw up on this. In any event you
can sometimes calculate rates and rate ratios. And so this is
a graphic presentation of what's going on. We have some large
population, let's say the population of California or Oakland
or whatever it is. We choose a sample because we can't follow
everybody. If that sample is a cohort in the cohort we
determine who has the exposure and who doesn't or the risk
factor and who doesn't. We follow them over time to see who
gets disease and who doesn't. A pretty simple concept.
Okay. So just to illustrate this before we get into
some more details. I want you to be familiar with a number of
really important landmark cohort studies over time and this is
one of them. So this is the British doctor's study begun in
1951 in which Sir Richard Doll basically asked all physicians
licensed to practice or in England, in the United Kingdom,
sorry, in 1951 he got a list of every physician practicing in
the United Kingdom and tried to enroll them in a cohort study
and follow them over time. And this cohort in fact lasted over
50 years. He was smart enough to start this cohort when he was
young. He was wise enough to live to be very old and he
followed this cohort for over 50 years. Okay?
Um, and so but you can see in 1951 he got replies from
about 40000 physicians. You can see there were more male
physicians than female physicians in the UK and he continued to
follow them up at regular intervals. So what does it mean to
follow them up?
What would be the key thing you'd be interested in, in
following up this group of 40000 people?
>>>: (Inaudible).
Art Reingold: So the first question is what
outcome is he trying to study? In this case he's interested in
lung cancer mortality. Death from lung cancer. That was the
hypothesis, the hypothesis was that smokers would have a
greater risk of dying of lung cancer than nonsmokers. Okay?
So clearly he needs to particularly ascertain who is
still alive and who has died. Right? And particularly then
the cause of death in these people.
Okay? So how do you do that? Well at that point in
time at least for the first 40 years he would send out periodic
mailings. One of you'll see many, many cohorts of health care
providers. Why do people choose health care providers to put
into cohorts?
>>>: (Inaudible).
Art Reingold: They are very generally compliant.
They are generally easily to find. There tend to be lists of
licensed nurses or physicians or veterinarians or dentists or
whatever it is. And they are pretty easy to follow.
So you have relatively little in the way of loss to
follow up. So he would send out periodic mail questionnaires.
He could monitor deaths through various obituaries and
basically the office that collected data on everyone who died
in Britain. And then he would also write to people living at
the last known address if he didn't get a response from a
letter. Now he achieved extremely high levels of follow up.
He basically could determine pretty much for everyone if they
died or if they were still alive.
Even over decades. And so here the basic notion is
being able to ascertain who is still alive and who isn't and if
they died of what did they die. Okay?
And so here are the key data. These are just among the
male physicians age 35 and over. So this is just the one
analysis of these data. So you can see this is person time.
So he's calculating incidence density and the first thing to
observe here is a lot more person time in smokers than
nonsmokers. Why is that?
>>>: (Inaudible).
Art Reingold: Because at that time virtually all
doctors smoked. Okay. Pretty simple. Smoking was an
extremely common phenomenon, particularly among males and most
British physicians smoked back in the 1950s and 60s. So you
can see in 15000-person years among nonsmokers one lung cancer
death. So the incidence density rate here, seven per hundred
thousand. Among smokers 98000-person years, 80 lung cancer
cancer deaths, rate of 81. Well you can calculate a rate ratio
it's 81 divided by seven. Right? So the rate ratio is pretty
close to 12. Right? So the interpretation of that is smokers
are 12 times more likely to die of lung cancer than nonsmokers.
Right? Pretty simple.
You can also and then if you're interested in the rate
difference you just take the rate in the exposed and subtract
the rate in the unexposed so the rate difference is just 74 per
hundred thousand for lung cancer mortality. And you can
calculate the attributable fraction exposed. So the
attributable -- and I understand we did post the formula for
those who wanted to know the derivation of the formula. The
relative risk minus one over relative risk or incidence in the
exposed minus incidence in the unexposed over incidence in the
exposed. Either one basically gives you 91 percent. And the
interpretation of that is that among smokers, 91 percent of
lung cancer deaths are attributable to smoking. And in theory
preventable if we could get everyone to stop smoking.
Right? Pretty simple.
Even more details. So he divided people into levels of
smoking, light, moderate and heavy smokers. And that's defined
by numbers of grams rather than the current approach which is
number of cigarettes or packs smoked. And here you can see
basically the exact same approach. Here are the rates. Light
smokers, moderate smokers, heavy smokers. Obviously all of
them higher than nonsmokers and increasing within increasing
amounts of smoking. So anybody know how we refer to this
observation that the more you smoke the greater your risk of
dying of lung cancer? That's evidence of what?
>>>: (Inaudible).
Art Reingold: A dose response effect. And when
we talk about causal inference in about six weeks that's one of
the criteria we use to see whether we believe that there's a
cause-effect relationship. Is there a dose response effect?
The higher the dose the greater the likelihood of the outcome.
So here you can see evidence of a dose response effect and he
could calculate rate ratios, light smokers versus nonsmokers,
moderate smokers versus nonsmokers. Heavy smokers. And you
can see this increasing rate ratio with increasing level amount
of smoking. So the more you smoke the greater your chances of
dying of lung cancer. Pretty simple.
Again, here you can see this was by 1990. This was
almost 40 years later. And this is now looking in fact at
overall survival among the physicians. Smokers versus
nonsmokers and basically showing because they are all basically
followed that there's in essence about a shortening of life by
about seven and a half years among smokers compared to
nonsmokers. So this is not lung cancer mortality, this is all
cause mortality.
So smoking shortened lives by about seven and a half
years on average. Okay? And that's one of the beauties of a
cohort study is you can look at things like this. He could
also show in fact that again the more you smoked the greater
the shortening of your life.
So here you can see basically evidence of a dose
response effect that basically the heaviest smokers had the
greatest reduction in duration of survival.
And here he could also then, as you might expect, some
people stopped smoking. And that raises issues in a cohort
study, what happens when people's exposure status changes? So
in this case what happens when they stop smoking?
And here you can see basically people who stopped
smoking before the age of 35, so they've been smokers but
stopped before the age of 35 basically have no shortening of
their life compared to people who continued to smoke.
Okay? So if you quit before the age of 35, basically
no shortening of your life.
And this is another way of showing these data. These
are from a different study, but basically showing the same
thing. So here is the relative risk of all cause mortality in
smokers and basically showing time since quitting and here are
never smokers over here. So basically you continue to have an
excess mortality, um, up until about 10 to 15 years after you
quit. And if you can make it until about 10 to 15 years after
quitting you basically will have then a normal life span after
that. Okay?
And there's Sir Richard who fortunately when he was
alive used to make regular visits to Berkeley and we had the
opportunity to have him speak a number of times, but
unfortunately he himself died.
Now another cohort which has looked at this we'll talk
about that you'll hear a lot about is the nurse's health study.
The nurse's health study is a cohort study by people done at
Harvard following over a hundred thousand nurses and basically
very, very similar prospective cohort study design. And here
you can see these are female smokers in this case. About
105000 women in the nurse's health study. The baseline 1980
and here you can again see that among these nurses enrolled in
1980 that over 50 percent of them were either current or prior
smokers.
Again, at least when I was in the clinical world back a
long time ago most nurses smoked. Very, very common
phenomenon, both nurses and respiratory therapists seemed to
think smoking was a good idea, but if yo go into the health
care settings today it's unusual to see doctors or nurses
smoking. So they've certainly gotten the message and this
basically shows again this is now in women instead of men.
Women in the United States basically showing the same
phenomena. Here's total mortality, cardiovascular disease
mortality, respiratory disease mortality, lung cancer
mortality. Basically among people who quit and these are
mostly people who quit after the age of 35. So they've been
smoking for a long time. In general your mortality for many
things, risk does not go down to that of a nonsmoker for
anywhere from 10 to 15 years after you quit.
Um, for vascular disease basically heart attacks the
risk goes down to that of a nonsmoker relatively quickly, but
for things like lung cancer mortality it really takes a long
time for you to get back to the risk of a nonsmoker.
Okay.
Okay. So when we talk about cohort studies, this is
where one of these areas where the terminology gets sloppy and
problematic. Okay? So the words prospective and retrospective
get used to mean two completely different things. Okay?
So in one sense all cohort studies are prospective.
They are all prospective in the sense that you start out
knowing the exposure, you follow people over time to see who
gets the disease and who doesn't.
In that sense all cohort studies are prospective.
Unfortunately there's another use of the term prospective.
Which as we'll talk about that, that doctor's study, the
nurse's study are prospective in the sense you enroll people
today. You follow them into the future. Right?
And you live a long time you can follow them for a long
time. But in that sense there's also a retrospective cohort
study. And Craig Steinmaus will talk about retrospective or
historical cohort studies next Monday. But those are cohort
studies where you go into the past and assemble a cohort in the
past and follow them up until today.
Okay? And he will explain how that's done. It's very
commonly done in occupational settings. And in that sense some
cohort studies are prospective starting today following people
into the future. Some are retrospective going into the past
and following them up until today.
And, in fact, some are ambidirectional going into the
past, following them up until today and continuing to follow
them into the future. Okay?
So, unfortunately this word prospective means two
completely different things and it can be a little confusing.
Now another thing that can be confusing is this issue of open
or closed or fixed cohorts. And the terminology I'm about to
show you on a slide is slightly different from the textbook
because I haven't changed it the way I did in the other slides.
We'll come back to that in a minute. Um, in again in cohort
studies you can calculate risks or rates or odds. You can
calculate risk ratios or rate ratios or odds ratios and you can
calculate all of these various measures.
Okay? Okay. Well I've already said this so I'm going
to skip that. Okay. So here this is the terminology as I've
taken it from Ken Rothman's book rather than your book,
Aschengrau, so this is slightly off from what's in the
textbook. I apologize.
So in your textbook they talk about closed and fixed
cohorts and open or dynamic cohorts. And what I've actually
written here according to your textbook is a fixed cohort. A
closed cohort is a subcategory of a fixed cohort. So you can
see that the textbooks are not all in complete agreement about
this. A fixed cohort is one in which the members are basically
defined. No new members can be added. And by definition the
size of the cohort will decline over time as people die.
So, a famous cohort where all the survivors of the
atomic bomb blast in Nagasaki. That was a closed -- excuse me,
a fixed cohort. By definition it was limited to people who
were in Nagasaki when the atomic bomb went out of and survived
the initial bomb blast. So nobody can ever be added to that
cohort. It could only shrink as people over time died. Okay.
Now your textbook says a closed cohort is a
subcategory, a fixed cohort in which in essence there's no loss
to follow up.
And so that, the typical example would be in a food
borne disease outbreak, people get sick eating at a picnic or
at a wedding. You can determine a hundred percent of the
people who were at that wedding. And the follow up period is
so short that there is no loss to follow up. You find
everybody and follow them throughout their risk period. Okay.
But in Ken Rothman's book he basically refers to these as
slightly differently. Open cohorts are cohorts in which new
members can be added. So the nurse's health study is an open
cohort. They could add additional nurses when they wanted to.
Right? And so I suggest we stick with the terminology
as it is in the textbook in terms of this issue of open and
closed cohorts.
Um, cohorts can be selected to be representative of the
general population. So another famous cohort is the Framingham
study. Anyone here from Massachusetts? Anyone here run the
Boston marathon? So if you've run the Boston marathon as I
have, you know that you run through Framingham when you're
running the Boston marathon. So it's a town about 15 miles
west of Boston. Okay. Well that was intended to be
representative of the population of the United States.
Okay?
You can take cohorts that are representative of geographically defined
areas. So Warren Winkelstein's San Francisco men's health study was a
random sample of men living in certain areas of San Francisco and
intended to be representative of men in those zip codes. But certainly
not intended to be representative of all men in the United States.
Okay? So intended to be representative of geographically defined area.
Again people frequently enroll health providers because they are easy to
find and easy to follow. And as particularly Craig will talk about in
the occupational epidemiology area, people frequently will enroll
cohorts of highly exposed individuals.
So, for example, people with a high exposure to
asbestos because of working a job exposure where in the general
population you have trouble finding people with high levels of
exposure.
So, sometimes cohorts are chosen to be representative
of highly exposed groups. And within the cohorts sometimes
it's a comparison within the cohort between those who are
exposed and those who are not exposed such as smoking doctors
versus nonsmoking doctors or smoking nurses versus nonsmoking
nurses. Sometimes the rates and risks in your cohort are
compared to some second external cohort. And sometimes in fact
they are compared to the general population. But the basic
notion is you're going calculate the rate or the risk or the
odds of disease in the exposed group and compare it to some
comparison unexposed group.
Now I'll point out that in there may be different types
of outcomes you might want to study. So one way of, one type
of outcome is an unrepeatable outcome. You can only have it
once. Okay? So you can only have your first heart attack
once. You can have more than one heart attack, but you can
only have your first heart attack once.
Right? So a first heart attack would be an
unrepeatable outcome. Dying from lung cancer. That can only
happen to you once as far as I know. So that can't be
repeated. Or you can have outcomes that can occur to you more
than once. So having an episode of diarrhea. You can have
multiple episodes of diarrhea. Or an exacerbation of your
asthma. You can multiple episodes of asthma.
And these have to be treated very differently from an
analytic point of view. If you can have the outcome more than
once. Okay. Yes.
>>>: So if someone has cancer and then they go
into remission and then the cancer, you have cancer again. Say
you have breast cancer around you get it again.
Art Reingold: So if you are talking about
recurrence of the breast cancer. The breast cancer is treated.
The breast cancer recurs, no, that would not generally be
included in any of these studies as a second episode of cancer.
Now somebody with breast cancer can then go onto develop lung
cancer or leukemia or some other completely separate cancer and
in theory I suppose there could be studies where that might be
a relevant outcome.
But certainly not a recurrence of the same cancer.
Because basically these cohort studies are primarily done to
look at the etiologic relationship between some exposure and
some outcome.
And once you've developed breast cancer, that's really,
you contributed the information that we need in terms of
looking at that relationship. And so the outcome can be either
repeatable or unrepeatable. It can be clinically apparent or
clinically silent.
If the outcome is clinically silent such as ***
infection, many *** infections are clinically inapparent, you
may not know when the person actually acquired their infection.
You tested them on January 1st and they were negative.
You tested them on July 1st and they were positive. You have
no idea when during that six-month interval they became
infected. Right?
So, we pretty much arbitrarily assign their date of
infection to the midpoint between those two observation time.
That's the standard approach. So you can outcomes that are
clinically silent or clinically apparent. And of course we can
be studying things where there is a long latency or induction
period or is short induction period. And that really is
critical to think about in terms of designing these studies.
So if you are interested in the relationship between
radiation or asbestos and cancer that may be a 20-year time
period between when the exposure occurs and when the disease
occurs.
So, studying a cohort for six months or for a year
doesn't make a lot of sense. Right?
As opposed to things with a very short latency period
of hours or days where you could potentially learn what you
need to learn with a cohort followed for a much shorter period
of time.
I'll also point out that sometimes the outcomes can be
dichotomous. You get lung cancer or you don't get lung cancer.
You get meningitis or you don't get meningitis. You get *** or
you don't. A dichotomous outcome. You either get it or you
don't.
Other outcomes can be continuous. So, something like
blood sugar level or level of blood pressure. Those are
continuous outcomes.
In general and in fact invariably if the outcome is
continuous and all you want to see is whether there's a
difference in the mean between the exposed and the unexposed
group, the mean blood pressure, the mean blood sugar level, the
sample sizes needed for cohorts to look at continuous variables
are much smaller than the sample sizes needed to look at
dichotomous outcomes.
Okay. So if the outcome is mean blood pressure as
opposed to hypertensive versus not hypertensive it will take a
much smaller sample size.
So, some of the challenges in analyzing cohort studies.
Well first of all it's ascertaining exposure status and taking
into account someone's status changes.
And then classifying them correctly and changing their
exposure status can certainly give rise to challenges in doing
these types of studies. Again the outcome can be clinically
silent, its exact timing unknown. And there can be a
substantial induction period. So all of these are factors that
need to be taken into account when thinking about cohort
studies, this is simply meant to show an example here.
Thinking about smokers in the study, basically they may change
the amount they smoke. They may quit smoking. They may begin
smoking. Although that's very uncommon. Cigarette companies
know if they don't get you by about the age of 20 they are not
going to get you. People don't generally start smoking after
their teenage years, but many people do quit.
So they then have to be classified correctly in terms
of their status and the time that's accumulated relative to
their exposure status needs to be thought about very carefully
in such studies.
Okay. So I want to illustrate some of these points now
with some of these studies we'll talk about some of them this
morning and we'll finish up with the rest of them on Friday to
illustrate some points.
So, again the Framingham study is one of the most
famous cohort studies in epidemiologic history. And here the
basic question was do certain factors increase your risk of
having heart disease. Cardiovascular atherosclerotic heart
disease and there were basically three hypotheses about what
might be risk factors. So what were those risk factors?
>>>: (Inaudible).
Art Reingold: Anybody know?
>>>: (Inaudible).
Art Reingold: Blood pressure. High blood
pressure was one.
>>>: (Inaudible).
Art Reingold: Smoking was two. Cholesterol was
three. Okay? So people basically had done studies suggesting
that those were risk factors for having a heart attack or for
cardiovascular disease so they wanted to look at that in a
cohort study.
So, they went, they decided, they chose Framingham
Massachusetts because they thought it was in some way
representative of the population of the United States and they
enrolled a cohort of individuals in Framingham.
So, another important point is that in general when we
assemble a cohort to study over time we have to screen them at
the beginning and throw out anybody who already has the disease
we're interested in.
If they already have the outcome we're interested in
then they can't be in our cohort. Right? They already have
lung cancer then they can't be in our cohort if the outcome of
interest is lung cancer.
So, there's invariably screening that takes place to
make sure that people are disease free with regard to the
outcome of interest at the beginning of the study.
So, you take a random sample of adult residents of
Framingham they screen them and get rid of anybody with
prevalent cardiovascular disease and then they follow them over
time. So just to give you a sense here you can see that they
here have a random sample actually in Framingham they decided
to both take a random sample and then to add volunteers.
Rather odd study design but nevertheless that's what they did.
So it's a mix of a true random sample of the population and a
call for volunteers.
Um, but basically you can see they've got the sample
then they get the people who respond to a request to
participate. Then they throw out everybody who already has
prevalent heart disease. They end up with here a total
population of around 5000 adults. Okay?
About a little more in the way of women then men, but
basically about half men and half women. And free of
cardiovascular disease and now they're going to follow them
over time to see what predicts cardiovascular disease.
So, you can imagine there are reams of data. There are
hundreds of publications, many, many different follow ups of
this population.
In fact, children and grandchildren of Framingham are
being followed today. Here you can see one example of this.
This is looking at cholesterol. You take people at the
baseline and divide them into three levels of their serum
cholesterol level, low, medium and high. Although many of
these people today would be considered to have high cholesterol
because our standards have changed. But nevertheless three
levels of cholesterol. Here are the number of people. Here
are the number of cases of basically ischemic heart disease or
heart attack. Here is the risk per thousand. These are
cumulative incidence. Then you can calculate a risk ratio and
show that with increasing serum cholesterol you have a greater
risk of having a heart attack. Right? Pretty simple.
You can also calculate the risk difference or the
excess risk. It's just the risk in the highest level compared
to the risk in the lowest level. So that would be the excess
risk. You can calculate the attributable fraction exposed for
high cholesterol versus low cholesterol and again you can
pretty simple calculations.
Okay. And you can show that having a high cholesterol
versus a low cholesterol gives you an increased relative risk
and then you've got an excess risk that accounts for pretty
substantial fraction of ischemic heart disease in the high
cholesterol group.
Okay? Now here are just some of the data from
Framingham. These are now the relative risk for developing a
heart attack. Men and women, three different ages of entry and
here is for blood pressure, cholesterol and smoking. Blood
pressure, cholesterol and smoking. And these are the relative
risks and this is an issue we're going to come back to later in
the semester. But just to give you an example here if you look
at blood pressure, high blood pressure is associated with about
a three and a half fold increased risk of having a heart attack
if you're young. And with no increased risk of a heart attack
if you are older. Okay? So the relative risk varies depending
on your age.
And this is a phenomenon we refer to as effect measure
modification. That is to say, in this case the relative risk
is different depending on the level of some third variable. In
this case the third variable is age.
So blood pressure is a more potent predictor of your
risk of a heart attack when you are young than when you are
old. And we'll come back to this later in the semester. Okay?
Similarly you can see that smoking among women and
among men is a much more potent predictor of having a heart
attack if you are in your thirties than if you are in your 50s
or 60s. Okay?
By the way, while the relative risks for smoking are
much higher for lung cancer, are -- the relative risks
associated with smoking are much higher for cancer than they
are for heart disease. Cigarette smoking kills far more people
through heart disease than it does through lung cancer.
Because heart disease is a much, much, much more common cause
of death. So much more mortality from cigarette smoking is due
to heart attacks than due to lung cancer.
Okay. Now some of you may be familiar with this.
People have now actually gone back and looked at Framingham
study in order to test the hypothesis that obesity might be
spread socially. It might be contagious, if you will. Not
through an infectious agent, although that is a separate
hypothesis that obesity is caused by a virus. But this
hypothesis is that basically obesity is spread through social
networks.
Okay. This is a famous study done by people at Harvard
a couple of years ago, a network analysis going back and
looking at the Framingham data, looking at the development of
obesity and concluding that basically obesity is spread through
a social network phenomena. I think some people don't believe
this study, but I'm not sure where I stand on it. Nevertheless
the Framingham study is still being used to look at all kinds
of things and as I've said the offspring and the grandchildren
of Framingham are still being studied to look at a variety of
things. So if you are interested in that article, the spread
of obesity in large social network over 32 years looking at the
development of obesity in this cohort in terms of the social
network and suggesting that who your friends are and their
levels of obesity have a huge impact on your likelihood of
becoming obese.
Okay? And if you are interested you can read this
article. Okay. This was the newspaper account of this that
your friends make you fat. (Laughter).
Okay. Now again just to point out this nurse's health
study, a very famous study I won't go into details, but over a
hundred thousand nurses assembled, followed over time and then
they basically measure every conceivable exposure and every
conceivable outcome at every conceivable point in time and
there's an almost infinite number of publications from the
nurse's health study. You name the exposure and the outcome
and they've studied it in the nurses. So I won't get into
details here. But when you've got a cohort like that, you
know, you can basically, there's some good news to the nurse's
health study. Here's the relative risk of a heart attack by
caffeine intake basically showing that you don't have to worry
about how much caffeine you take in, at least in terms of your
risk of a heart attack if you are a female nurse.
Okay. So, um, we'll cover one more cohort here before
we break. And so this was something I noticed in the newspaper
a number of years ago. An article suggesting that the more
orgasms you have, the longer you will live. And being an
epidemiologist I was interested in the study as you might
imagine. So I had to dig it out. It wasn't one I was familiar
with. This is a cohort called the Caerphilly study and
Caerphilly is town in Wales in the United Kingdom.
Done by a famous British epidemiologists named George
Davey Smith. So I tracked down this study. The Caerphilly
study attempted to screen all men 45 to 59 years of age living
in the town of Caerphilly and five adjacent villages. So this
is back in the late 70s, early 80s. A total of 2500 men were
seen. They did at the baseline a history, a physical exam
serum cholesterol, echocardiogram and the first 1200 men were
asked about their frequency of intercourse. 918 provided
responses. But the question was abandoned after discussion
with local practitioners because of a possible effect on the
overall response rate. In other words, a number of men were
apparently were put off by being asked this question and were
refusing to participate. So they stopped asking this question,
but they had information about this for the first 1200 men or
so. And then they're following these men over time and looking
at who lives for how long and who dies. Okay?
So, um, responses to *** intercourse were classified
into categories ranging from never to daily. But then were
basically compressed into three categories less than monthly,
twice a week or more and anything in between.
And then they found a correlation between frequency of
*** inversely correlated with age but not correlated with
all of these other risk factors.
So here you can see the data in terms of all cause
mortality, coronary heart disease mortality and see that in
fact there is a correlation here between frequency of ***
and the risk of dying or dying of a heart attack. And this is
controlling for social class, blood pressure, smoking,
cholesterol and a variety of other factors.
>>>: This is *** intercourse with other
people. Right?
Art Reingold: (Laughing) *** intercourse
generally occurs with other people, yes. In my experience,
yes.
>>>: But they are not talking about ***.
Art Reingold: So one of the problems with study,
there seems to be a conflation as you're pointing out between
*** and *** intercourse. As you're pointing out there
may be a disconnect between the two. Right? And I've never
been able to quite figure out how they got these two confused
because they do go back and forth between these two different,
somewhat different outcomes.
Now this is what they reported. Basically more orgasms
are a good way to lengthen your life and there are actually
similar studies in women, by the way.
So, I actually know Davey Smith. Here you can see the
conclusion if you want to read the article. It's in the
British medical journal. And despite having over 600
publications to his name this is the one that people ask you
about the most frequently for some reason.
Now when I actually dug out the study they also
actually looked at frequency of shaving. And I was interested
in that because I haven't shaved in over 30 years. And so here
you can see they classified men by their frequency of shaving.
Eventually put them into twice daily to once daily, every other
day, less than every other day. So they then break the men
into once or twice per day and less often than that.
And then they look at the question does shaving
correlate with your risk of dying? And basically I'll skip to
this. Here you can basically see that people who shave less
often, men who shave less often in fact have a greater risk of
all cause mortality. A greater risk or cardiovascular disease
mortality. And a variety, strokes and all other bad things
happen to you if you don't shave regularly.
Um, so any thoughts about this? Yes.
>>>: (Inaudible) personal hygiene (laughter).
Art Reingold: These guys aren't going to like
that. It's a hygiene hypothesis. Okay. Yes.
>>>: (Inaudible).
Art Reingold: So maybe it's in fact not really
how often you shave, but something related to hormone levels.
Yes.
>>>: Well, so (Inaudible).
Art Reingold: Sorry, feels what way?
>>>: (Inaudible) some people if they don't shave
for a while it's because they (Inaudible) other people choose
not to shave for other reasons (Inaudible).
Art Reingold: I think what's he's trying to
capture did people basically, some people have a heavier beard.
Some people have a lighter beard and have to shave more or less
often because of that. And most likely it is in fact related
to hormone levels. But it isn't necessarily shaving itself
that increases your life span. Right? But if you simply look
at these data you might say shaving is protective against all
cause mortality. Right? But I think it probably we can see
here that they basically think that it might have something to
do, a certain amount of confounding we'll come back to, but
basically it might be a hormonal effect. But as somebody who
hadn't shaved in 30 years I was sort of interested in this
because I was worried that perhaps if I'm not shaving I was
increasing my risk of dying. Here you can see this is from a
previous election period talking about politicians and facial
hair, facial hair on men. This is Erica Jong. Nonconformity,
masculinity and unruliness is what facial hair is all about.
Clinton has it all without facial hair. George W. Bush has
some smirk that some women find cute. I don't know exactly
what all that means, but nevertheless, um (laughter) as I said,
I haven't shaved in over 30 years. So I was worried that
perhaps I was increasing my mortality. That's what I look like
when I worked for the CDC in Atlanta in my better days and one
day I happened to be on Good Morning America talking about a
disease. And I looked like that at the time and it generated
this letter from someone in the Nebraska. Dr. Arthur Korngold
(laughter), I cannot believe in a god but your parents' genes
gave you quite a handsome face. Idiotic to hide your face with
facial hair. No federal employee should be kept who insists on
wearing a beard. If a man has difficulty proving his
masculinity without these hairy adjuncts I think that there
must be other ways to prove one's masculinity. Incidentally I
wish you people would find a cure for 65-year-old men who are
plagued with impotence. You would be making a real
contribution (laughter).
Um, so for the men in the audience who have beards I
don't know exactly what advice to give you. But this was by
the way the return envelope from Mr. Lawbaugh in Nebraska. So
I'll let you decide whether shaving might increase your
duration of survival or not. So we'll stop here on this upbeat
note and we're going to continue talking about prospective
cohorts.