Epidemiology and Biostatics
observative study
cohort study: compare group with exposure or risk factor to group without
can be prospective or retrospective
relative risk (RR)
adoption study: compare siblings raised by biological vs adoptive parents
twin concordance study: compare the frequency with which monozygotic twin vs dizygotic twin develop same disease
cross-sectional study (横断研究): frequency of disease and risk-related factors are assesed in present
case-control study (症例対象研究): compare people with disease and without disease
odds=p/(1-p), odds ratio (OR)={p/(1-p)}/{q/(1-q)}
Clinical study
phase1
small number of healthy volunteers or patients with disease of interest
safety, toxicity, pharmacokinetics, pharmacodynamics
phase3
phase2
phase4
moderate number
treatment efficacy, optimal dosing and adverse effect
large
compare to current standard of care (or placebo)
postmarketing
long-term
evaluation of diagnostic test
TP, FP, FN, TN
sensitivity and specificity are fixed property, PPV and NPV vary depending on disease prevalence
negative predictive value (NPV)
positive predictive value (PPV)
sensitivity
specificity
TP/(TP+FN)
probability that when the disease is present, the test is positive
for ruling out disease and low false-negative rate
prevalence=(TP+FN)/(TP+TN+FN+FP)
screening in disease with low prevalence
TN/(TN+FP)
probability that when disease is absent, the test is negative
for ruling in disease and low false-positive rate
confirmation after a positive screening test
probability that a person who has positive test result actually has the disease
TP/(TP+FP)
probability that a person with negative test result actually does not have disease
TN/(TN+FN)
likelihood ratio
compare likelihood with disorder to that without disorder
LR⁺>10 and LR⁻<0.1 indicate a very useful diagnostic test
LR⁻=(1-sensitivity)/specificity
LR⁺=sensitivity/(1-specificity)
Quantifying test
number neede to harm
number needed to treat
a:disease+,risk+; b:disease-, risk+; c:disease+, risk-; d:disease-, risk-
relative risk reduction
atributable test
odds ratio
relative risk
avsolute risk reduction
odds of ratio with disease vs without disease
OR=(a/c)/(b/d)
in case-control study
in cohort study
risk of developing disease in exposed group vs in unexposed group
RR={a/(a+b)}/{c/(c+d)}
RR=1→no association
RR>1→exposure associated with ↑disease occurrence
RR<1→exposure associated with ↓disease occurence
difference between exposed and unexposed group
AR=a/(a+b)-c/(c+d)
intervention as compared to control
RRR=1-RR
ARR=c/(c+d)-a/(a+b)
the difference in risk attributable to the intervention as compared to a control
lower number=better
NTT=1/ARR
number of patient who need to be treated for 1 patient to benefit
number of patient who need to be exposed to a risk factor for 1 patient to be harmed
higher number=safer
NNH=1/AR
incidence vs prevalence
incidence
prevalence
during a specified time period
newcase/people at risk
at a point of time
↑prevalence→↑PPV and ↓NPV
existing cases/total people
prevalence = incidence for short duration disease
prevalence > incidence for chronic disease
precision vs accuracy
precision(reliability)
accuracy
absence of random variation
consistency and reproducibility
trueness of test measurement
absence of systematic error or bias
↑→↓standard deviation,↑statistical power
bias and study error
recruiting
interpreting bias
performing
selective bias
berkson bias: from hospital
non-response bias: participating subject differ from nonrespondent
nonrandam sampling or inappropriate treatment allocation of subject
strategy: randomization, ensure the choice of the right comparison/ reference group
recall bias
awareness of disorder alters recall by subjects
strategy: decrease time from exposure
measurement bias
distorted manner
Hawthorne effect: participant(血圧計) change behavior upon awareness of being observed
strategy: objective, standardized, previously tested method, use placebo
procedure bias
not same treatment
strategy: blinding and use placebo
observer-expectancy bias
strategy: blinding and use placebo
researcher's belief change outcome
confounding bias
two or more complex causal pathway
strategy: multiple or repeated study, cross order study
not on causal pathway, it distort effect
lead-time bias
early detection is confused with ↑ survival
strategy: measure"back-end" survival
length-time bias
latency period affect result
strategy: randomized controlled trial
statistical distribution
normal distribution: Gaussian, bell-shaped, mean=mode=median
nonnormal distribution
measure of central tendency
measure of dispersion
median=middle value of a list of data sorted from least to greatest
mode=most common value
mean=(sum)/(total number)
standard deviation(標準偏差): how much variability exists, around mean
standard error: an estimate of how much variability exists in a set of sample means around the true population mean
varianc=(SD)², SE=SD/√n
positive skew: mean>median>mode
negative skew: mean<median<mode
bimodal(二峰性): eg. Hodgikin lymphoma, suicide
statistical hypothesis
outcome
Alternative(H₁): hypothesis of some difference or relationship
Null(H₀): hypothesis of no difference and relationship
correct result
incorrect result
stating that there is an effect or difference when one exist, or stating that there is no deffect or difference when none exists
type 1 error
type 2 error
stating that there is an effect or difference when none exist
α is probability
preset α level is 0.05
stating that there is no effect or difference when one exist
β is probability
↓β by: ↑sample size, expected effect size, precision of measurement
confidence interval
95% is often used (95%→Z=1.96, 99%→Z=2.58
if mean of variable include 0, or odds ratio or relative risk include 1, H₀ is not rejected
CI=x±SE
if CI overlap→no significant difference
range of value within which true mean of the population is expected to fall
Meta-analysis
improve strength of evidence and generalizability of study findings
pools summary date from multiple studies for more precise estimate ofnthe size of effect
Pearson correlation coefficient
common test
ANOVA: 3 or more group
Chi-square(χ²): check difference between 2 or more percentage or proportions og categorical outcome
t-test: check difference between means of 2 groups
the closer r is to -1, the stronger the linear negative correlation
coefficient of determination = r²
the closer r is to 1, the stronger the linear positive correlation
-1≦r≦1