Week 5 Flashcards
EBM levels of evidence pyraamid
meta analysis systemic review RCT cohort study case control study case series and case reports animal studies/laboratory studies
What is a RCT used for
treatment questions and diagnosis questions
Cohort studies
- type Q
- when use
- answer questions of prognosis and etiology/harm
- use when no RCT
Case control studies
- what Q
- when use
- answer questions of prognosis, etiology/harm
- used when no cohort studies
case series and case reports
- what Q
- when use
- answer questions of prognosis or etiology/harm
- when no case control studies
4 classifications of research
Nature of research
time frame of research
investigator approach
type of data
what are the 2 natures of research
descriptive and explanatory
what are the 3 time frames of research
prospective, retrospective, cross sectional
what are the 2 types of investigator approach
observational and experimental
what are the types of data
qualitative and quantitative
Observational investigator approach
investigator not influence what subjects exposed to
-track natural course/progression
Experimental investigator approach
-investigator controls exposures that may influence outcome of interest
4 components of true experimental design
1) manipulation
- investigator controls what happens during study
2) control
- presence of group that does not receive intervention being studied
- accounts for outside factors that may affect study
3) random assignment
- study subjects randomly allocated to intervention or control
4) random selection
- study subjects randomly chosen from total eligible population
are experimental studies randomized wrt selection of control/intervention subjects from population
yes
are observational studies randomized wrt selection of control/intervention subjects from population
no
why is non-random allocation to exposure/control a problem in observational studies
ie. what is the problem with non-randomized studies
-variable may be due to selection
+ex. physician chose more sick people for exposure rather than healthier
-characteristics heavily influence outcome
Characteristics of observational studies (3)
exposure –> outcome
less rigid than controlled studies
ASSOCIATIONS NOT CAUSE AND EFFECT
why observational studies?
not possible have RCT to support every intervention
-still accept some interventions in spite of no RCT
+parachute example
necessity of observational studies (3)
- extent of disease (distribution/epidemiology)
- etiology of diseases (risk factors)
- evaluate interventions (such as medications) in large populations to detect rare outcomes
Exposure event rate
EER - proportion of subjects in exposure group experiencing the event
EER = a/(a+b) ; a= outcome, b=no outcome where a and b are in the exposure group
Control event rate (CER)
CER - proportion of subjects in control group experiencing event
CER = c/(c+d) ; c= outcome and d = no outcome where c and d occur in control group
Relative risk
risk of developing disease or adverse event in participants EXPOSED to specific variable compared to those not
RER = EER/CER
RR (relative risk) = 1
no association
RR (relative risk) < 1
negative association
-exposure/intervention decreases likelyhood of outcome
RR (relative risk) > 1
positive association
-exposure/intervention increases likelihood of outcome
When dealing with observational what wording must be used to identify a relationship in variables
Associated
-(NOT “CAUSED”)
three types of observational studies
cohort studies
case control studies
cross-sectional studies
Cohort
-group of people who share one or more characteristic (treatment, disease, residence etc)
Cohort studies
- who is the population
- how population allocated
- the population is a cohort
-there is NO RANDOM ALLOCATION
+self selected
How collect data for cohort studies
1) retrospective
2) prospective
+either or
-
what is the risk of prospectively collecting data
-may induce additional bias
+people know they are being studied, change behaviour
Characteristics of cohort study
- can measure incidence (new diagnosis) and prevalence (new and old diagnosis) among population
- can assess ASSOCIATION between exposure and outcomes
Measuring the effect size of an intervention involves what calculations (3)
relative risk reduction (RRR)
Absolute risk reduction (ARR)
Number needed to treat (NNT)
Relative Risk Reduction
RRR = (CER-EER)/CER
ex interpretation:
handwashing is associated with a relative reduction of childbed fever mortality of 56% compared to no handwashing
Absolute Risk Reduction (ARR)
ARR = CER - EER
ex interpretation
-handwashing is associated with an absolute reduction of childbed fever mortality by 9%
Number Need to Treat (NNT)
Applies meaning to ARR
NNT = 1/ARR
* if outcome is unwanted, NNH, number need to harm
ex. 1 MD washing hands saves 0.09 life (ARR)
xMD wash saves 1 life
11MD save one life
largest study ever recorded
- type
- name/what about
- prospective cohort study (1948)
-Framingham Heart Study (FHS)
+ID risk factors for heart disease
FHS ID’d risk factors
Age, Physcal activity, body weight, diabetes mellitus, hypertension, cholesterol
smoking indivs dev CHD at larger rate
does association mean causation
no
-SO STUPID
can we establish causation without being able to manipulate exposure
- If we cannot manipulate exposure (randomization to exposure or control), several rules can establish causation
- causality requires assess multiple pieces of evidence, NOT JUST STATISTICAL ASSOCIATION
Bradford Hill
-guideline for establishing causation
- Temporal relation (ex. smoke first then get disease)
- Biological plausibility
- consistency
- strength of the association (RR, OR)
- Dose-response relationship
- study design
- judging the evidence
Can you establish causation from observational studies
-YES, with guidelines for causation such as bradford hill
Case-Control study
-5 key points
-start collecting CASES and assess exposure among them
-Collect CONTROLS from the same (or similar) “original” population
+must be very similar to the cases
-MUST BE RETROSPECTIVE
-cannot assess incidence or prevalence
+we dont know the OG population
-Can still assess association
What do you do when the outcome trying to be monitored in a study is rare
start small with a case control study
-ex. regional cancer centre
Measuring association in a case-control study
ODDS RATIO
-what are the odds of cancer amoung smoking patients
odds = (probability of success)/(Pfailure)
ex.
- roll 6 on dice
- odds= 1/5
Odds Ratio
NUMERATOR
Poutcome among intervention)/Pno outcome among intervention
DENOMINATOR
Poutcome among control/Pno outcome among control
OR=num/denom
Interpreting odds ratio
OR = 0.63
non-smokers have 0.63 lower odds of developing cancer
OR
NOT smoking reduces the odds of cancer by 37%
How is OR an approximate estimation of RR
when a disease is rare a
OR vs RR
- what does OR estimate
- overestimate or underestimate
- when is degree of estimate greater
- OR is an estimation of RR
- OR always overestimates RR
- this overestimation is greater when RR>1
-
how do you interpred the graph given %incidence (x) and odds ratio (y)
ALWAYS compare the ‘overestimation’ through the analysis of relative risk reduction
example
OR is 0.73 and RR is 0.75
The OR is considered an UNDERESTIMATION because:
given OR=0.73, RRR= 27%
given RR=0.75, RRR= 25%
SO, based on RRR, OR is an overestimate
Case Control study: choosing controls
- does it matter how many controls you chose
- avoid
- NO, it does not matter how many controls you choose
- avoid biased controls such as people hospitalized due to accidents
Case-control study: bad control examples
- using pediatrics for smoking case-control studies
- having more smokers than non-smokers in controls than cancer cases grp etc
Bias types (4) -what do they do
selection bias
information bias
data analysis bias
survival bias
distort relationship between exposure and outcome
Confidence interval (95%) -when is there statistical significance
if the confidence interval does not cross 1
ex. 1.27 - 1.69
NOT STATISTICALLY SIG EXAMPLE
-0.78 - 1.23
Confounding variables
things that influence the outcome that are not related to the exposure
-such as the biases
ex. post-menopausal women and estrogen
- doctors prescribed the estrogen to healthier women
- selection bias - confounding variable
- result is due to selection bias NOT exposure
LOOK AT SECOND TO LAST SLIDE TABLE
EEEEEEE UNGA BUNGA