EPIDEMIOLOGY and STATISTICS
Screening
Tests
Risk Reduction Statistics
- the
Types of Study *Bias
- note that errors can either be random or systematic
- bias = systematic error
- random error = type 1 and types 2 errors
--> type 1 error = alpha = wrong rejection
--> type 2 error = beta = wrong accepting of the null
*Recall Bias
- common with interview style studies
STATS
Standard Error
Notes:
- note that sample error is inversely related to sample size
Case presentation:
EPI Principles and Terms
Incidence
Incidence vs Prevalence
Notes:
- this question simply tries to trick you on incidence vs prevalence
-incidence tells you nothing about the amount of people in population currently who have the disease - prevalence = incidence x survival duration
--> survival duration = 1 / fatality
Health Promotion and Disease Prevention
Primary Health Prevention
- no disease is currently present
- but there are evident risks that you can counsel to lower
- promotion of health
Secondary Health Prevention
- disease could be present, but assymptomatic
- mainly screening for cancer, blood, hypercholesteremia, etc.
Tertiary Health Prevention
- disease is present and there are symptoms as well
- trying to lower any extra symptoms
- cure the disease or slow its progress
Pernicious Anemia Case
Notes:
- note that health promotion is the main primary health prevention strategy because you can see very clear risks in a patient that you can cunsel them to lower
- but there is no disease present yet and no symptoms
- you are trying to lower disease later down the road
Clinical Case
Types of Statistical Tests
Types of *Studies
Stats / EPI you should know off-hand
- general cancer stats and risks
α = alpha , β = beta , Type 1 / 2 Errors
Type 2 Error = Beta
and Statistical Power
- power = 1 - β
- failure to detect a difference since the n number not big enough
- this is not forgiveable since all need to do is find more people
--> where alpha errors exist no matter what
Cancer Incidence
Women :
- Incidence = breast, lung, colon
- Mortality = lung, breast, colon
Men
- Incidence = prostate, lung, colon
- Mortality = lung, prostate, colon
note that lung is second for both in incidence, but switches for the sex cancer for mortality as the number 1
Accumulation Risk Effect
- for both risks and protective factors
- some risk factors or protective factors take longer periods to give their risk
- SMOKING for example accumulates over years and gets worse and worse with more use
Accumulation Risk Effect example
Notes:
- accumulation effect simply means a risk or protection accumulates over time
- this is not true for all risks
- classic example is smoking
--> reason why we take a pk year history since it has the accumulation effect - same true here for antioxidant use over a lifetime
Example:
Negative and Positive Predictive Values
- NPV and PPV
Accumulative Incidence
Accumulative Incidence
Example:
Notes:
- note that cummulative incidence is over a specified time period like a year in this example
- it does NOT count the people in the population who already have the disease so you have to subtract them from the total population first
- you don't subtract anything else, even deaths since these people may have gotten the disease in the time period
*Attributable Risk
- What percentage of risk can be attributed to smoking?
--> attributed risk % = risk difference / (total risk RR)
--> attributed risk % = (RR -1) / RR - Subtract then divide
Attributable Riske example
Example:
Notes:
- Attributable risk = RR - 1/ RR
Infectious Diseases
Diarrhea Outbreak Example
Notes:
- NTDs have different severity
- severity of a neural tube defect can range from moderate, such as spina bifida, to severe and non-life-compatible, such as anencephaly
- the high AchE is because in NTDs it leaks out into the amniotic sac from the CSF of the open neural tube
Example:
Smoking Risks
- smoking is one of the biggest risk factors for most diseases
- biggest mortality risk reducer in MI (even more than aspirin, BP etc.)
Smoking = biggest mortality risk for MI, expecially for Diabetics
Notes:
- note in the graph to the left smoking is the largest factor for mortality
- it is even more pronounced in diabetics for getting an MI
- you would think that aspirin would be directly related to CAD and reduce mortality more, but smoking is still bigger, even more in diabetics
Case example
Example:
Prevalence
- prevalence = incidence x time period
Point Prevalence Example
- point prevalence same as normal prevalence but measured at a specific time
Notes:
Example:
*Odds Ratio
- prob of event happening vs not happening
Odds ratio example
Example:
Basics of the Punnet Square
- make sure to put either all numbers or ALL percents
- NEVER mix them up!
Crytptogenic stroke and ASD and PFO
Notes:
- I put in the 75% into the square by accident
Clinical Case
*Special Types of Bias
Pygmalian Bias = smart Pigtail Gretchen IQ bias
- think of smart kid like Gretchen with Pigtails
- type of observer bias where they already have a concieved notion of an outcome
- comes from study of students where their IQs given to the teacher made the teacher think they were smarter
--> smart IQ kids = pigtails = Pygmalian bias
Notes:
- note that
Clinical Case
Hawthorne Effect Bias
- think Pierce Hawthorne FAKING heart attacks because he KNOWS people are WATCHING and STUDYING him
- Hawthorne effect is where people being study become aware and fake their behaviour
Notes:
- note that int his case the people being studied = doctors
- Hawthorne effect is they realize this and fake their behaviour like pierce faking a heart attack
Clinical Case
Berkson's Bias
- think of Pete Burke at the hospital
- Berkson bias is where CONTROL patients are chosen from the hospital
- more likely to be sick so are not good controls
Notes:
- note that
Clinical Case
*Hardy-Weinberg Genetics
- prevalence of alleles in the population
- assumes there is no evolution, or changes in the population
Notes:
- p = normal allele frequency % in population
- q = mutant allele frequency %
- all add to 1
- q2 = phenotype of disease
- 2pq = carrier of disease
Clinical Cases
Clinical Case
Notes:
- note that
Clinical Case
Confounding variables vs Effect modification
- note these are both external variables that have an effect on the exposure and the disease
- difference is in stratified analysis
--> confounding variables show the RR is about the same when stratified by the outside variable
--> effect modification is where there is a large effect in the RR when you do stratification by the new variable
Confounding Variables Bias
- something not accounted for in the study as the cause of an outcome
- study finds association with significant p value, but doing stratification (= dividing up the people by age stratification) with an outside variable shows no difference
- every study STOPS confounding variables by getting MATCHED groups by baseline when comparing controls to exposed
CASES
Clinical Case
Notes:
- note that
Clinical Case
Effect modification
- opposite of confounding
- effect modification is where there is a large effect in the RR when you do stratification by the new variable
- can be positive or negative
*NPV = Negative Predictive Value
NPV = Negative predictive Value
example:
Notes:
- note that in the 2x2 table the TP = sensitivity and TN = specificity
--> whenever you are given a sensitivity and specificity fills these in right away
*PPV = Positive Predictive Value
- Positive Predictive --> depends on PREVALENCE
Clinical Cases
Clinical Case
Notes:
- note that
Clinical Case
PPV = Positive Predictive Value example
example:
Notes:
- note that same for PPV and NPV
- for PPV you take the True positives and divide by all positive results
PPV = Positive Predictive Value example 2
example:
Notes:
- note that same for PPV and NPV
- for PPV it depends on PREVALENCE
Sensitivity and Specificity
- think of graph and cutoff point fr a certain marker
- right graph = higher marker being screened for
--> positive screen test
--> disease present - left graph = lower marker being screened for
--> negative screen test
--> NO disease
*Sensitivity
- sensitivity = TP/FN
--> % of people with the disease picked up by the test
--> trying to get as many people WITH disease to test positive
Clinical Cases
Clinical Case
Notes:
- note that
Clinical Case
*Specificity
- Specificity= TN//FP
--> % of people correctly ruled out as not having the disease - False positive rate = FP is related to specificity
- FP = 1 - specificity
Clinical Cases
Clinical Case
Notes:
- note that
Clinical Case
Specificity and
FPR = False
Positive Rates
Notes:
- Specifity and FPR = false positive rates do not change and are independent of the disease prevalence in different populations
- this means that you find the specificity, then FPR = 1 - specificity
--> you can then apply this FPR to another population with a different prevalence of the disease - specificity = % of healthy people identified by a screening test
--> specificity = 1 - FPR - note when filling out the chart, always fill:
--> total population
--> prevalence - this also gives the non-disease = healthy
--> FP = % of healthy people = 1 - specificity
FPR example:
Sensitivity
Notes:
- note for extremely rare disease you want a high sensitivity
- for very common diseases, you want a higher specificity
example:
Clinical Cases
Clinical Case
Notes:
- note that
Clinical Case
Type 2 Error = Beta
and Statistical Power
Notes:
- α = prob of type 1 error
- type 1 error = α = prob of rejecting the null when it is actually true
--> think this is the usual error we use as α = 0.05 meaning that for a stat test we assume there is a 5% chance we make a type 1 error and reject the null as false, when there is a small chance it may still be true - β = prob of type 2 error
- type 2 error = β = prob of not rejecting the null when it is actually false
--> type 2 error is directly related to statistical power because if you don't have enough power in your study or large enough sample, you may have a type 2 error and not reject the null when it is actually false - statistical power = (1 - β )
Case presentation:
Statistical Power calculation from B error example
Case presentation:
*Type 1 Error = Alpha and 95% confidence level
- confidence level = 1 - α
- type 1 error is where you detect a difference, but incorrectly
--> actually no difference
--> only 95% sure there is a difference - think alpha errors are understood to happen= built into study since you know there is an alpha = 5% chance you have a type 1 error
*NNT NNH = Number
Needed to Treat / Harm
- NNT ARRRRR... important
- NNT = 1/ARR
- NNT for reduction in absolute risk
*Odds Ratio vs. RR
- the OR always over estimates the RR
- the OR tends towards the RR when the prevalence of a disease is low, or the prevalence is close to the incidence
--> you can see this in the picture to the right as the factor prev./inc tends to --> 0 in both the exposed and unexposed groups,
--> the OR tends --> RR
Clinical Cases
Clinical Case
Notes:
- note that
Clinical Case
NNT = Number
Needed to Treat
Notes:
- NNT is the inverse of the ARR = absolute relative risk
Example:
Solution:
NNT example 2
Notes:
- NNT is the inverse of the ARR = absolute relative risk
Example:
*OR= Odds ratio
- odds looks backwards
--> splits up Disease vs no disease
--> looks at odds of exposure vs no exposure - OR = ADDS BATCIO = ad / bc
*RR = Relative Risk
- RR looks at exposure vs NO exposure
--> risk of getting the disease - RR = Exp+Disease / total exposed
--> divided by (NOexpose+Disease / total NOexposed)
Clinical Cases
Clinical Case
Notes:
- note that
Clinical Case
RR = Relative Risk
Solution:
Example:
Notes:
- important to know the difference between relative risk, absolute risk, and odds ratios
- sometimes RR = relative risk can overestimate the risk of n exposure
--> example: RR of UTI from not getting circumcised is high, but the actual AR = absolute risk difference is low because UTI in boys are so extremely rare
*Normal Distribution
- 68 95 99.7 Rule
--> 1,2,3 SD from the mean
--> % of population in those SDs
Clinical Cases
Clinical Case
Notes:
- note that
Clinical Case
*Normal Distribution
- 68 95 99.7 Rule
--> 1,2,3 SD from the mean
--> % of population in those SDs - actual 95% population = 1.96 SD
- actual 99% population = 2.58 SD
*Lead-time Bias
Clinical Cases
Clinical Case
Notes:
- note that
Clinical Case
Lead-time Bias
Example:
Notes:
- note that lead time bias happens when there appears to be a longer survival rate for times right after the screening process
- if a screening test survival rate doesn't hold up at later times that means there is lead time bias
*Observer Bias
Clinical Cases
Clinical Case
Notes:
- note that
Clinical Case
Observer Bias
Example:
Notes:
*Observer Bias case 2*
Example:
*Selection Bias
- attrition bias = loss of follow up
--> special kind of selection bias
--> loss of one specific group to follow up that skews the results
*RCTs = Randomized Control Trials
- gold standard of studies
*Observational Studies
*Case Control Studies
- "CASE is first, risks are 2nd"
- good for rare diseases
--> since need to find SACES of the disease first, then test for risks after - only study type that goes in reverse
- pick CASES = have disease
--> regardless of exposure = find this after - pick controls as = NON CASES of diseases
--> regardless of exposure = find this after
CASE control studies CONTROL for the CASE of DISEASE 1st
--> figure out the risks exposure after
Clinical Cases
Clinical Case
Notes:
- note that
Clinical Case
Case example
Notes:
- note that CASE control studies are unique because they are the only one where they divide patients into CASES and NON cases of a disease
--> they then look back to see if they have risk + exposure or NOT - this is the opposite of retrospective cohort studies, observational studies, etc.
--> they instead divide into risk exposure vs non risk exposure --> then find whether they get a disease or NOT
Clinical Case
Case example 2
Notes:
- note that CASE control studies controls are NON DISEASE people
- for CASE controls, think you control for the CASE of DISEASE
- then figure the risks after
Clinical Case
*Ecological Studies
- "eco" = environemnt
- think of the whole species in the environment
- eco studies are at population level, NOT individual
Clinical Cases
Clinical Case
Notes:
- note that
Clinical Case
Case example
Notes:
- note that ecological studies are eco so it is the whole species and the environment they live in
- thus eco studies are at the population level, NOT the individual level
- ecological fallacy = trying to predict individual outcomes based on eco studies or the entire population
Clinical Case
*Prospective Cohort Studies
- find a cohort of people who are either exposed or not exposed to a risk and see if they develop a disease
Prospective cohort example
Clinical Case
Notes:
*Crossover study
- crossover from one treatment to another in the SAME patient to compare the 2 treatments
- randomly group patients to AB group or BA group
- AB = treatment A / washout period / treatment B
- BA group is the oppositie
- pros for crossovers = patients are their own controls
- cons for crossovers = need washout period long enough to make sure no effect on second treatment
P Value
- the prob of getting a finding just given to chance, assuming that the null hypothesis is true
T-test
- for comparing means of 2 separate samples
--> note that the t test is a special case of an F test ANOVA
T test vs Chi square
Notes:
- note
Clinical Case
Chi-Squared Test
- same independent variable (continuous with mean and sd)
- comparing two nominal groups (classic example is gender men vs women and comparing a certain variable)
Chi square test example
Notes:
- note that Chi square test is for 2 nominal variables, often comparing men and women as two categories
Clinical Case
*ANOVA = Analysis of Variance
- for comparing means of more than 2 sample populations
- note that the t test compares means of 2 sample populations
--> the t test is a special case of an F test ANOVA
*Skewed Normal Distributions
- SKEW = means the MEAN is skewed by a TAIL in either the Positive or NEgative direction
- "MEAN, Median, Mode" = MEAN is always skewed the most by the tail
--> then Median
--> then MODE - makes sense as the MODE hass to be near the HUMP of the Curve
Attributable Risk
- Subtract then divide
- AR = (RR -1) / RR
- AR = (RRe - RRue / RRe
*Home