Biostats Flashcards
Post test probability
Pre test prob is 10%
liklihood ratio is 16
Post test probability
10% pre test porbablity = 1:9 odds
16 is lilihood ratio
16+9=25
16/25= post test proabiliy
Minors <18 yrs old excpetions to needing parental consent
- married
- preggo
- finacially indpendent
- riaisng kid
- living by theirself
- in the army
- OCPs or Pregnancy Tx
- STD
- Psych illness
*need parental consent for abortions
difference in comeptency vs capaacity
Competent= legal by court
-permanent is dementia and someone appointed b court ot make decisions, dleirum is temporary
capacity = medical by me
break of confidentiality (2)
- hurt themselves or others
2. Abuse
termnally ill patients make sure you prescribe
LOTS of opiods
Make sure to rpeort HIV aids, syph, measles all the good stuff
withdrawal of care anf amily disagrees with patient?
seek out ethics comittee
High sensitivity but low spec= False ____
High spec but low sens= False____
Positive
Negative
95% sens 95% spec Low prevalence of disease 1% 10% 50% 90%
1- not helpful
2- Kinda helpful, negative result reduces lieklihood and positve test says it needs more tests
3- Very helpful
4- not helpful- positive adds ntohign and negative is a false negative
Disease present Disease absent
+/exposure A B
-/no exp C D
Know it
Specificity Equation
D/ D+B
PPV equation
A/ A+B - liklihoood a positive result is positive
High prevelance== higher PPV\
Need cohort study not case control- incidence needs to reflect he population
NPV equation
D/ C+D - liekly hoood the negaitve result is disease free
Lower pevelance = higher NPV
Need cohort study not case control- incidence needs to reflect he population
Incidence define
new cases per year
prevalence define
total number of existing cases in a population at certin snap shot of time
Absolute risk define
PROBABILTY of n event occuring over a time frame
1% chnce of getting Disease X in 5 years
Relative risk define
COHORT porspective studies comapring groups exposed to a risk factor or not
RR<1 in event less liekly in exposed group
RR>1 exposed group more likely to have event
Odds ratio defined
CASE CONTROL studies, RETROSPECTIVE studies.
it compares rate of epxosure to those with and without disease. Less accruate than OR
Relative rate equation
a/a+b /// c/c+d
Odds ratio equation
AD/BC
Absolute risk reduction define and equation
Risk from exposure to something + Background of disease.
RANDOMIZED CONTROL TRIALS. BEST EQUATION AND WAY BETTER THAN RRR
Absolute risk (adverse events) in placebo group - absolute risk in treated patients
Absolute risk reduction define and equation
Risk from exposure to something + Background of disease.
RANDOMIZED CONTROL TRIALS. BEST EQUATION for benefit AND WAY BETTER THAN RRR
Absolute risk (adverse events) in placebo group - absolute risk in treated patients
RRR
RADNOMIZED CONTORL TRIALS
ratio between 2 risks
Ratio and can look deceptively large
event rate in control - event rate in experimental patients /// event rate in control
Or Unexposed-exposed///unexposed
Example of ARR AND RRR
reduction of adverse event is 0.01% to 0.004%
ARR 0.01-0.004 = 0.006%
RRR 0.01-0.004///0.01= 0.6 or 60%!!!
NNT define and equation
number of patients ot be treated to prevent 1 adverse event
1/ARR
p value defined
less than 5% chance that is was by chance or there is a5% chance the null hypothesis that there is no association holds true
CI defined
Like p value, is this real or is it by chance. It says that there is a 95% chance that the observed risk will fall between a certain range.
It uses OR and RR
If CI includeds 1 it is not cliniclaly significant
if it is 1.9 but CI is .8-3 it is not significant
RR is 2 and CI is 1.2-3.5 then there is OBSERVED 2x risk for smoking to cause cancer but in reality the ACUTAL risk is somewhere between 1.2-3.5 X the risk if you were to smoke
Prospective studies
future outcomes
control for biases
retrospective
past events
less reliable
cohort study
population observed overtime grouped on basis of exposure to a factor and watching for a psecifci outcome *NOT good for rare conditions **USE RR to intepret results either pro or retro can work
Case ocntrol study
Retorspective study
look at ppl with a disease and without the disease and look fro exposure to risk factors
*Good for rare diseases
**USE OR to intepret resutls
Randomized contorl trials
prospective study to randomly assign ppl to treatment or placebo group to see if tx made a difference
DBL blindi s gold standard
Crude mortality rate
Deaths/total population
Cause specifc mortalty rate
Disease related deaths/total population
Case fatality
Deaths/#ppl with disease
Standard mortality ratio
Observed deaths over expected detahs. If it is 2 then the partcualr group is twice is high as the genreal population
Crude birth rate
Live births/Total population size
Maternal mortlaity rate
Deaths/live births (not fetal losses)
DO suicidie contracts work?
No get guns out of the home
Alcohol abuse, anemia, old due, MCV low, no Abd pain- first step colonsicpopy or EGD?
Colonoscopy- cancer first then EGD if negative
WHat is an Anylsis of variance used for>?
MEan comparison between 3 groups. Itgivee F statisitc for variation between the groups
Mcnemar test
difference between 2 paired proportions - the patients serve as their own control (success/failure before and after the treatment in the same subjects)
pearson chi square
association between categorical variables
Success/failure in MAle/feamle
fischers exact test
small sample size for pearson chi squared
success/failure in <20 patients etc
paired t test between 2 means- patients as own control
Ex- mean BP before and after treatment
Correlation coefficiant
-1 to 1 and 0 is null
NO causality
Weak association the closer it gets to 0
Difference between ARR and RRR and RR
RRR is proportion or percentage
ARR is difference
RR is the increased likelihood of the risk in (un)exposed
.0018 and .0013
ARR is .5
RR is 1.38 or 38% more liekly
RRR is 28%
smaller sample size will make what type of error?
type 2 error
what makes a type 1 error
significane of the testing? large the alpha the mor eliekly a type 1 error
when does erly termination happen in a study
overhwlimgin positive or negative result
what is outocme musclassifcation
if the end point is to determine if you have hcancer but you cant accurately test for that cancer
If you need a higher power for a study do you need a larger or smaller sampe size?
Larger
If you have an alpha of .01 and you rise it .05 what does that do to sample size need?
nothing, just accpeting more chance becuase alpaha = p value
If you expect a intervention to cause a 50 effect on glucose but really it oly effects it by 30 will you need more or less sampe size?
more, smaller effect takes more ppl to accurately detect it
difference in 1 tail vs 2 tail test
1 tailed test have more power and fewer subjects needed bc it only olooks in one direction, did the intervention increase or decresae it but not both
Prospective study to ID risk of getting a disease in a group in 1 year- RR or OR
Retrospective to see if a Pill caused a disease- OR or RR
RR
OR
If there is a -.8 correlation factor what is the coeffiecient of deteremination to tell you the variablity of the outcome factor?
.64
-.8 *- .8
Correlation
Scattered but in one direction= weak
straight line but not up or down= 0 correlation
30 out of 60 (.5) smokers vs 10/40 (.25) non smokers
Attributable risk
PErctange of attributable risk
Percatage of entire population from smoking?
.5-.25= 0.25
0.5 - .25///.5 = 0.5
40/100-10/40= .15»_space;».15/.4-0-.375
NNT
NEw drug survivial is 25/50 vs old drug 25/100
50% mortality and 75% mortality
.75-.5-.25
1/.25= 4!!!
Need to look at bimodal distrubtion of spe and sens curves
False positive an dpalse negative values is the little area around the cutoff point
90& sensitivity means how many false negatives
10%
true positive and false negative for all that have the disease
RR of 2 w/ 95% Confidence interval 1.5-2.5
this means that we are 95% confident that our true value lies between these value ranges and the relative risk is 2x greater for the intervention group.
p would then be <0.05
if .4-1.2 range this includes 1 and means there is a greater than 0.05 risk for this to happen and not statisically significant
if 0.74 then a decresaed risk for RR
CI of 0.5-3.8 means
Sample size is too small
95% CI equation
Mean * 1.96*SD//// Sq Rt of N
CI will narrow and SEM will narrow with a alrger sample size
Look at the sens and spec curves
the little area over the cutoff line is the sens and spec values, tighter = better. Left is sensitive usually
Liklihood ratio vs predictive values- test perfromance values
Predictvie values change based on prevalence
liklihood of 9 for a test means 9X more liekly to have a psotive result than that who do not have the disease.
Sensitivity/ (1-specificity)
look up ROC curves
you want them to be close to 1 area under the curve so it to go up to the top and then over
- The higher you go on the Y axis the more sensitive it is, so the cut off value is lower to get more +s
- do if you get a negative reuslt it is mroe liekly to be negative
Predictive value equations
Used for post test probability
A/(A+B) PPV
D/(C+D) NPV
80% have positive tests that have the disease 10% have positve test that dotn have the disease. Prevalence is 10%. what is PPV?
sens=.8 spec=.9
(sensitivityXprevalenc)x[sensXprevelence)+(1-spec)(1-prev)]
0.80.1 / (0.80.1) + (0.10.9)= 0.47 PPV
ORRRRRR- plug in numbers for 100 ppl
out of 100 ppl, 10 have the disease and 90 dont. Of the 10, 8 will be +s and of the 90 who dont 9 will be false negative (10% or what is leftover on specficity, if specifcity is 70 than it would be 30% of 90)
8/(8+9)
As specifcity goes up, the false positives go down and the PPV is stronger.
(even if snesitivity is low)
length time bias vs elad tiem bias
Lead time: you detect a disease after its course is already set and no intervantions chnages the outcome but it appears like survival is longer
Length time: you catch a less aggressive form of the disease so it appears like better surivival
Selection bias
Berkson fallacy- hospitlaized patient
Referral-sampled from specifc medical centes
loss to follow up in ochort studies
non respoense bias if refuals isrealted othealth
prevalence bias or Neyman-incidence is based on prevalnce. Dm2 of MI vs no DM2 and MI. the DM2s who died with MI cant report their incidence.
suspecitpbility bias- Severity of the patients tocndition interfers with treatmnet they are offered to skew results
Hazard ratio
similar to relative risk.
Null is 1 or baseline risk.
It ooks at meaure of eefecr in surival or tiem-toevent analysis.
Example- MI risk following surgery.
Confounders
Must link it to the exposure! and! the outcome! of itnerest
Use statify analysis to tkae out the confounder
ways to reduce confoudning
randomize
or match- Use race, age, geogaphicla location to match u the groups as best you can
you can restrict - or set inucsuoin cirteria but it reduces generalizbailtiy to the population
effect modification
Basically is confoudning but it is a natural phenomena that can tbe corrected in the study design.
the exposure of interest is modfiied by another variable
no flaws in design or anayals
OCPs and breast cancer - family hx of breast cancer
Asbestos and lung cancer- smoking
estrogens and DVT- smoking
Randomization
- minimizines confounding
- simialr distributions of gorups
- only the interbention is being looked at
blinding
Lowers placebo and observer bias
Intention to treat odel
Preserves randomization and prevent non complaince bias
COntorl gorup starts taking the med later or the interbention gorup stops taking the med- this model will look at thier treatment staus at the momen when randomized
Deviations
68% within 1 stanrard deviation each way
95% (2.5% on each side!)
99%
Mean, median and mode are very close to eachother
Z score
Subtract the mean from all values and divide by SD
“how many SD a given value is from the mean”
Skewed distributions
Tail to RIght is postive skew
Mode=peak
Median= to the right of mode
Mean= to the very right because it is affected by hihg values
Routine labs
CBC BMP UA Lipid EKG
low salt low calorie/fat exercise alcohol smoking
6 months of this then HTN meds
If group A has a relative risk 0.4 compared to B, what is the increased that gorup B has?
2.5 increased risk
1/0.4
Correlation
-1 to 1 range
r=.99 vs .3 shows a big difference in .99 has a greater positive preidctive facotr
You need to look at your fav photos for types of tests
Chi squared is fro cateogrized proportions - like high or low with CRP or No CRP.
paired t test uses same individual like BMI now, take adrug, and BMI after in that person (even tho sample size is larger than 1
difference between smiths and colles fracture?
Colles FOOSH- radius shaft is volar to hand
Smith- hyperlfexion fall on it- radisu shaft is dorsal to hand (opposite mechanism)
Difference in confoudning and effect modifere
Confounding will effect both exposure and outcome but when you stratify the groups there is no association now
(shoe size and IQ are linked until you stratify or age ane now there is no link)
Effect modifier- positive or negative impact from another variable, when you strafity you will see association between one group and not the other
Survival analysis for months after chemo starting probability of making it 3 months?
take each interval times each other. 10% died the first monthm then 8%, then 6%
.9.92.94=
2 year mortality is the same for tx and placebo group but the treatment gorup is said to be effective, why?
End poitn is too long- time to event analysis
median lifespan is still 6 months longer for tx group
latency period
Talks abotu chornic diseases (time from getting it to outcome is longer so no difference and then a diffrence after years is due to latency)
systematic error is not effected by…
sample size
if someone else came in and did the exact same study they would achieve the same error results
compormises the vlaidity of the study
2/3 docotrs diagnose PNA, which means ____ is affected
reliability
Type 1 vs type 2 error
type 1- wrongful association- Alpha or p value
pvalue of .03 menas still a 3% chance no associaiton
type 2- Syaing no association when there is - Beta or 1-B is th epower of the study (proability of detecting an aosscaiton if it exists)
How to increase power of study (3)
power=odds to see treatment effect if there is one
- increase sample size
- difference in outcomes is large and nto subtle
- lower the alpha lower the power
standard mortality ratio
interst population/standard population
if 1.75 than 75% hihger than expected
class 2 3 4 shcok
1 no change
2 tachy over 100
3 blood pressure drops
4 rr and all go up, 140 Tahcy
Whatis Q and I2?
Used ofr heteorgenity, MIld vs Moderate alzheimers to see if differences between them
Q=think of it as P value. Greater than .05 then no heterogeneity
and I2 of 0% is no heterogenity, 25% is low amoun t, 50% mod and 75% high amount of heterogneity
heterogenity- just looking to see if there is a difference btween the two groups (used in a big anaylsysi )
Quality adjsuted years vs disabiltiy years
Quality is individual or little poutilaion
living 5 years like this is liek my revious living 1 year
Dislability is global burden- iving in perfect helthu up to stsndard life expectsncey age. Current situiotn vs ideal situaiotn
WHat is multiple linear regression used for?
1 Quant dep and mutliple dep vriables while ocntorlling for many other variables
multiple logisitc regression
1 dichotomus dpednent and muntliple independent
Diabetes dependent and obesity, age, BP
2 means
>2 means
2 sample t test
1 way analysis of variance
use median when…
skewed numbers
Standard incidence ratio equaiton
observed number of cases/expected number of cases
RRR when they give you cases and person years
1.8-1.3/1.8
354/196785 x1000 for 1.8 person eyars
lenght time vs lead time bias
length time is rapid vs slowly progressing disease
lead time bias is a test that diagnoses the same disease earlier
hemophilia A is X linked
recessive
intetnion to treat vs as treated analysis
as treated is analyzing them on what they actually got vs what thye were supposed to be randomized to. Randomization is lost
Intetnion to treat= you analyze the subject basedo n whiat gorup they were randomzied to despite if they dropped out or switched groups bc it can be treated as a result form the intervention. preserves the trial and is ocnservtive to results
1 2 3 SD Mean %s
68
95
99.7
primoridal primary sec tert quart preventions
primordial- stopping it before you even get a risk factor - lifestyle as a kid
primary- you have a risk facotr now take a statin BEFORE MI happens
sec- stopping progressions of disease (Stress test positvie for MI)
Tertiary- Cardiac caths
Quats- Sharing electronic medical records so repeat caths dont cause adverse effects
A 25%
B20%
what is NNT for A
25-20=.05
1/.05
20 NNT
one study showed good Ps and findings and the other study was rndomized and did not shows good ps… why?
Residucal confonfounding
can alrge population studies shed lgiht to indiivdual results/
no
OR
Can just divide th etwo numbers and get close or AD?BC
NNT and NNH is just minus big from small and then divide by 1
A risk o.4 compared to B risk sooo
40% risk of the risk of B gorup
1/.4 or the B risk has 2.5 fold higher risk
Odds ratio
AD/BC
case cotnrol stidueis
looking specifalclly at one hting and comapring it to contorl gorup