Vous êtes sur la page 1sur 32

STAT 2 Lecture 3: Does smoking kill?

Review

The best way to test hypotheses is by carrying out experiments Compare a treatment group to a control group; groups should be identical except for the treatment Avoid confounding, bias

Review

Ideally: Use contemporaneous controls andomise groups !ouble blind

Can we always do experiments?

"ypothesis: smo#ing causes lung cancer Can$t force people to smo#e or not smo#e Can$t set up a control %o viable placebo

Today

!oes smo#ing #ill& 'hy did it ta#e the scientific community so long to decide whether smo#ing was ha(ardous or not&

Today

!ifferent designs for observational studies 'hat advantages and disadvantages do these different designs have&

I: Does smoking kill?

Lung cancer

are disease before modern times Incidence rose mar#edly in the th early )* century Cigarette smo#ing became th fashionable in the late +, century Could there be a connection&

1912: Isaac Adler

-bserved lung cancer more common among men than women

.the irritations and traumatisms which must be fre/uent in the more unprotected life of men 0the abuse of tobacco and alcohol, the many trades and vocations which are accompanied by irritations of the respiratory organs, etc12111 The entire sub3ect is not yet ready for final 3udgment14

192 : Lom!ard and "oering

5uestioned )+6 cancer patients and )+6 controls; controlled for gender and age Cancer group: 768 heavy smo#ers; control group: )*8 heavy smo#ers Cancer of lung9lip9chee#93aw: :7 out out :; heavy smo#ers

192 : Lom!ard and "oering

<mall sample <election of control group somewhat arbitrary -nly shows association =any confounding factors

19#$: %c&airer and %c&oniger

<urveyed families of lung cancer victims and of stomach cancer victims, as well as a control group In both control group and stomach cancer group, about >78 were smo#ers In lung cancer group, ,68 were smo#ers

19 ': "oll and (ill

Interviewed lung cancer patients newly admitted to hospitals Interviewed other randomly selected hospital patients

Control Cancer Smokers 1296 1350 Nonsmokers 61 7

19 #: "oll and (ill

<urveyed the smo#ing habits of :**** ?ritish doctors in +,;+ Trac#ed deaths for +* years

Cigarettes per day 0 1-14 15-24 25+ % of sample 10 40 30 20 % of lung cancer deaths 2 16 40 42

19 ): R*A* +is&er

Incipient cancer, or a pre@ cancerous condition, may induce smo#ing <mo#ing does not cause cancer, but both are caused by genetics

199#: ,raun- Caporaso- .age(oover

Axamined B>+ pairs of male twins where at least one died of lung cancer After controlling for other factors, identical twin of lung cancer victim no more li#ely to develop lung cancer than fraternal twin

199#: ,raun- Caporaso- .age(oover

!oesn$t mean there no genetic effect, only that it$s small compared to other effects <mo#ing causes lung cancer

II

Observational study designs

Cross/sectional studies

Axamine a population data set =ight only have data on groups, not individuals Coo# at incidence of disease and possible cause among different groups elate disease and cause fre/uencies

Cross/sectional studies

-ften re/uire no new data Can cover very large populations "uge amounts of confounding Impossible to determine cause and effect

Cross/sectional studies

Alcohol and heart disease "eavy drin#ers have a higher incidence of heart disease than non@drin#ers =oderate drin#ers have a lower incidence of heart disease than non@drin#ers

Case/control studies

Consider a group of people who have a certain condition Also consider a group of people who don$t have the condition Coo# for characteristics of the +st nd group that are different from ) Can loo# at individual differences

Case/control studies

Dood for trac#ing rare events Comparatively cheap and /uic# <till a high danger of confounding etrospective: loo#ing for patterns after the fact

Case/control studies

Seat Belt Studies Case@control studies of road accident victims seemed to show a huge decrease in serious in3ury if you wore a seat beat <eat beat laws have not reduced serious in3uries by as much as those studies predicted

Co&ort studies

Cohort: group of people who share a common characteristic Compare to a control cohort who differ in only one respect =atch for gender, race, economic status, other health, etc1 Trac# groups for a long time

Co&ort studies

=ost li#ely confounders can be controlled for Can almost isolate variable in /uestion !oesn$t rule out uncontrolled confounding factors Eery difficult to establish causality

Co&ort studies

Nurses' Health Study !esigned to assess ris# factors for cancer and heart disease "as trac#ed +)**** nurses since +,6B =any ma3or health findings have followed

%trengt& o0 evidence

+1 andomised experiments )1 Unrandomised experiments :1 Cohort studies 71 Case@control studies ;1 Cross@sectional studies BF1 Axpert opinion
%ote: exact ran#ing will depend on /uality of study

Recap

Recap

Cross@sectional studies loo# at incidences and possible cause at a particular point in time Case@control studies loo# for uncontrolled differences between a case group and a control group

Recap

Cohort studies trac# two groups that differ in a certain respect over an extended period of time Though none of these is as definitive as a experiment, if we eliminate all other plausible causes, we can ma#e strong statements

1ext time:

Turning numbers into pictures

Vous aimerez peut-être aussi