Public Health Sciences Flashcards
Cross sectional study
Frequency of disease and risk factors both assessed in the present
Measured by disease prevalence
Case control study
Group of people with a disease compared to a group without disease, look at ODDS of prior exposure or risk factor makes a difference
Measured by OR
Cohort study
Looks at a group with a given exposure/risk and a group without and assesses risk factor association with disease development later on
Measured by RR
Can be prospective (who will develop) or retrospective (who developed)
Phase I drug trials
Assesses safety, toxicity, pharmacokinetics/dynamics in small # of healthy volunteers
Phase II drug trials
Assesses if it works – treatment efficacy, optimal dosing, adverse effects in small # pts with disease
Phase III drug trials
Compares tx to standard of care or placebo to see if its an improvement in a large number of randomly assigned patients with disease
Phase IV drug trials
Postmarketing surveillance – if rare/long-term adverse effects may be withdrawn from market
Sensitivity
TP/(TP+FN) or 1-FN
Def: when disease present, how many test positive
Highly sensitive rules OUT disease (i.e. low false negative) – best for screening
Specificity
TN/(FP+TN) or 1-FP
Def: when disease not present, how many test negative
Highly specific rules IN disease (i.e. low false positive) – best for confirmation after screening
PPV
TP/(TP+FP)
Def: Proportion of positives that are true positives
Person with a positive test actually has disease
Varies with pretest probability (higher pretest prob –> higher PPV)
NPV
TN/(TN+FN)
Def: Proportion of negatives that are true negatives
Person with a negative actually doesn’t have disease
Varies w/ pretest probability (higher pretest prob –> lower NPV)
LR+
Sense/(1-spec) = TP/FP
>10 useful diagnostic test
LR-
(1-sens)/spec = TN/FN
<0.1 useful diagnostic test
(- –> negative sign on top!)
Odds ratio
OR=(a/b)/(c/d) or ad/bc
Used in case control studies to depict odd of event given an exposure vs odds of it happening w/o exposure
Relative risk
=[a/(a+b)]/[c/(c+d)]
Used in cohort studies to determine risk of developing disease in exposure group divided by risk in unexposed group
-For rare disease (i.e. low prevalence) – approximates RR
-If 1 –> no relationship between exposure/disease
-If >1 –> positive association between disease and exposure
-If <1 –> negative association between disease and exposure
Attributable risk
Difference in risk between exposed and unexposed groups – proportion of disease attributable to exposure
AR=[a/(a+b)]-[c/(c+d)]
Relative risk reduction
Proportion of risk reduction attributable to an intervention vs control
RRR=1-RR
Absolute risk reduction
Difference in risk (not proportion) attributable to intervention vs control
ARR = [c/(c+d)]-[a/(a+b)]
ABCD on table!
disease
+ -
risk factor + a b
- c d
NNT
Number needed to be treated for 1 pt to benefit (lower is better)
=1/ARR
NNH
Number needed to be exposed to risk factor for 1 patient to be harmed (higher number is better)
=1/AR
Precision
aka reliability Reproducibility – increased=lower SD, higher statistical power
Accuracy
aka validity Trueness of test measurements – absence of systematic error/bias in a test
Selection bias
Non random sampling or treatment allocation so that population in study is not representative (usually a sampling bias
Berkson bias
Study pop from hospital – less healthy than general pop
Healthy worker effect
Study populatio in healthier than general pop
Non-response bias
Participating subjects differ from those who do not respond
To reduce selection bias…
Randomization, ensure choice of right comparison/reference group
Recall bias
Awareness of disorder alters recall (esp in retrospective studies) – recall exposure upon hearing about similar cases
To reduce recall bias…
Less time from exposure to followup
Measurement bias
Information gathered in a distorted manner
Hawthorne effect
A measurement bias – participants change behavior in response to be observed
To reduce measurement bias…
Use objective, establish testing methods for data collection, utilize a placebo group
Procedure bias
Subjects in diff groups not treat the same
To reduce procedure bias
Use blinding and placebos
Observer-expectancy bias
Researchers belief in efficacy of a treatment changes it’s outcome (Pygmalion effect, self-fulfilling prophecy)
To reduce observer expectancy bias…
Blind and use placebos
Confounding bias
Factor is related to exposure and outcome but not causal – can distort/confuse effect of exposure on outcome
To reduce confounding bias…
Multiple/repeated studies, crossover studies (patients are their own control), matching (patients similar in both groups), restriction, randomisation
Lead-time bias
Early detection is not the same as longer survival
To reduce lead-time bias…
Measure “back end” survivial (adjust survival according to severity of disease at time of diagnosis)
Variance
=SD^2
Standard error
Estimate of how much variability exists in a theoretical set of sample means around the true population mean
=SD/[sqr of n]
Positive skew
Mean>med>mode (tail to right)
Negative skew
Mode>med>mean (tail to left)
Type I error (alpha)
Stating there is a difference when none exists (accusing an innocent man) – incorrectly reject Ho (false pos)
alpha – probability of making a type I error – p is judged against alpha level of significance
Type II error (beta)
Stating that there is not a difference when there is one (blindly let the guilty go), incorrectly accept Ho (false neg)
beta – probability of making a type II error – related to statistical power
Power
=1-beta Increased power (lower beta): -bigger sample -larger expected effect size - increased precision of measurment
Confidence interval
Range of values within which the true mean is expected to fall w/ a specified probability
For population = mean +Z(SE)
95%, Z=1.96
99% Z=2.58
If a CI includes 0 – dont reject Ho
If CI for 2 groups overlap, no significant diff
If they dont – significant dif
