Use LEFT and RIGHT arrow keys to navigate between flashcards;
Use UP and DOWN arrow keys to flip the card;
H to show hint;
A reads text to speech;
74 Cards in this Set
- Front
- Back
RCT
|
1) hypothesis
2) inclusion ↔ exclusion criteria 3) informed consent 4) DSMB 5) IRB 6) randomize accepted subject 7) asses outcomes |
|
DSMB
|
Data Safety Monitoring Board
under what conditions will we stop a trial? → overwhelming differences or futility |
|
IRB
|
Institutional Review Board
give Yes/No/Maybe review ethical issues of study |
|
Internal Validity Measures
|
1) subjects = entry criteria?
2) @ baseline, are study groups comparable? 3) loss to follow-up? 4) compliance assessed? 5) blinded? 6) agree w/ definition of outcome? 7) statistical power? |
|
External Validity Measures
|
Cost
Availability Believable? Clinically beneficial? |
|
RCT +/-
|
+ gold std.
+ equally distributes confounders (known and unknown) + incidence data & multiple outcomes - unethical (in cases)? - costly (time & money) - poor for rare outcomes |
|
observation ↔ intervention
|
obs: case report
case series cross sectional case control cohort int: RCT cohort |
|
Confounding
|
related to distribution of variables across study arms
|
|
Bias
|
related to errors in study design
|
|
ENHANCE
|
statins for ↓ cholesterol:
↑ clearance ↓ production are drugs better in combo? Chol. ↓ 27% w/ combo, but no change in atherosclerotic plaque |
|
Case Report
|
singular report on a case
to alert medical community no comparison group |
|
Case Series
|
plural report on a series of observed cases
to alert medical community no comparison group |
|
Cross Sectional Study
|
Prevalence
data mining for association and observations |
|
Case Control Study
|
subjects chosen by outcome
+ rare diseases - rare exposures + quickly study multiple exposures |
|
Cohort Study
|
subjects chosen by exposure status
prospective ↔ retrospective + rare exposures - rare outcomes + generate incidence data |
|
Nested Case Control
|
internal case control w/in context of a prospective study
|
|
Prevalence
|
total cases
---------------- = I x duration total population "snapshot in time," inc. old cases |
|
Incidence
|
Cumulative I. (total cases over a set period of time)
Incidence Density -- cases*personyears^-1 *New Cases |
|
Mortality
|
deaths
---------------- population |
|
Case Fatality
|
deaths
-------------------- those affected |
|
RR
|
Ie
---- Io |
|
AR
|
Ie - Io
|
|
OR
|
ad
----- bc |
|
z
|
x - mean
------------- SD |
|
variance
|
spread
|
|
SD
|
√v
|
|
Epidemic
|
occurrence of a disease in an area with a greater than usual expectancy
|
|
Endemic
|
constant presence of a disease in an area
|
|
Pandemic
|
worldwide epidemic
|
|
Mean
|
Average
|
|
Median
|
Middle value
(if even # of values, average 2 in middle) |
|
Mode
|
most common value
*poss. to be bi-modal |
|
Range
|
highest - lowest
|
|
Percentile
|
% of variables at or below value
|
|
Interquartile Range
|
middle 50% {exclude <25th & >75th percentile}
|
|
Coefficient of Variation
|
SD
--------- mean |
|
Continuous Data
|
like a wave on the ocean
data limited only be accuracy of measurement |
|
Discrete Data
|
are OR are not
categorical, nominal, ordinal |
|
Z-score
|
# of SD away from average
--> implies normal distribution |
|
Gaussian Distribution
|
1 SD = 68%
2 SD = 95% 3 SD = 98% if normal dist., mean=median=mode |
|
Problems of Internal Validity
|
Chance
Bias Confounding |
|
Chance
|
design inherently flawed
--> erroneous conclusions |
|
Information Bias
|
"garbage in --> garbage out"
|
|
Interviewer Bias
|
unique to case-control
*blind interviewers and use standard questions |
|
Selection Bias
|
way subjects are selected associated with disease
(ex. story of lupus alcohol link) only in case control / retrospective cohort *varied definitions |
|
Reporting Bias
|
social concerns → misreporting by subject
|
|
Recall Bias
|
notion that cases think harder about exposure status than controls and thu overestimate their status
unique to case control |
|
Surveillence / Detection Bias
|
avoid by blinding treating physicians
notion that cases may be looked at in more detail, and thus more incidences of disease found |
|
Loss-to-Follow Up Bias
|
10% ++
>20% -- Red Flag is it differential or even? |
|
Misclasification Bias
|
entries ↔ outcomes
*sometimes difficult to conclude if condition present, may put things in the wrong bucket |
|
Random / Non-differential
|
equal in both study arms
truth → null hyp. avoid w/ strict definitions of outcomes |
|
Differential / Non-random
|
can push in both directions
avoid by blinding everyone *more important w/ subjective outcomes |
|
Confounding
|
one side stacked for potential confounders
*association between result & exp. *independent risk factor cannot be from intermediate failure to fix → bias |
|
Interaction
|
:-)
external validity issue Are risks / benefits equal for all populations? → a good thing to know for treating various populations clinically |
|
Health Determinants
|
Fixed <----------------------> Modifiable
(genetics, time) env., lifestyle care, society |
|
Genomic Medicine
|
NOT Genetics
look at whole population distribution |
|
Pharmacogenomics
|
goal: to tailor drug therapies based on a patient's genetics
|
|
Aging
|
Lose adaptive capacity with age
theories: genetic -- pre-programmed on cellular level stochastic -- chance, accum. of env. factors. |
|
Health Determinants
|
Environment
Lifestyle Medical Care +Society **all hopelessly intertwined |
|
Social Capital
|
Ability to take advantage of people who care about you
-->getting involved is a strong + influence |
|
Stress
|
a normal physiological response
chronic response has long term (-) effects What do you do about it {clinically}? |
|
Common Health Behaviors
|
diet
exercise smoke Alcohol / drugs Sex Recklessness **Our ability to change them is ABYSMALL! |
|
Our impact as physicians?
|
↑ prolonging life
↑ reducing suffering ↓curing disease / healing ↓ prevention |
|
Limitations of Statistical Significance
|
Statistical Significance ≠ Patients' Benefit
|
|
P-value
|
Probability of getting a difference of x% or larger due to chance alone
|
|
α
|
Level of Significance
arbitrary "line in the sand" to help us make a decision freq = 0.05 |
|
P < or = α
|
chance is unlikely {but NOT impossible}
∴ reject null on statistical significance ∴ likely hood of getting result or better due to chance alone <5% |
|
P > α
|
fail to reject null
∴ conclude result is due to chance alone (not statistically significant) |
|
Power
|
1 - P(type II error)
ability to detect a difference of a certain magnitude or more if one exists |
|
Power (cont.)
|
opp. P value, calc. assuming alt. hyp. true
unlikely to find true differences w/ small sample ↑ power α ↑ sample size |
|
1 Tail <-----> 2 Tail
|
1) existing vs. {better} treatment
***not worried about negative outcomes 2) discerning differences btw A & B (regardless of direction) |
|
Type I Error
|
false +
conc. diff., but no diff. p < α, but still due to chance |
|
Type II Error
|
false -
conc. no diff., but truth is difference p > α |
|
α adjustments for multiple subgroups
|
"data dredging" will find something where p < α
∴ α'= α / # of analyses |