EBM Exam 1 Flashcards
Descriptive statistics vs Inferential statistics
Descriptive: describe and summarize data
Inferential: make inferences to larger pop beyond the data collected
Simple random sample
each person has equal prob of being selected (prob sample)
Stratified random sample
Divide into M and F and select 10% of each gender – ensure that both men and women are represented equally (prob sample)
Cluster sample
select 10 clinics in NE OH then select 50 pt from each clinic (prob sample)
Systematic sample
select every pt that walks through the door at the clinic
prob sample
Convenience sample
advertise over internet, newspapers
approach people in waiting room
Nominal
Ordinal
Interval
Ratio
Nominal: cannot be ordered (gender, race)
Ordinal: can be ordered (likert scale)
Interval: meaningful intervals (temp)
Ratio: absolute zero, ratios are possible (age)
Discrete vs. continuous
Discrete: counts, no fractions (ex. number of pts)
Continuous: infinite number of values (age)
Match test w/ scale of data for dependent variable
Differences in proportion
Chi square (nominal)
Match test w/ scale of data for dependent variable
One or 2 means
t-test (interval or ratio)
Match test w/ scale of data for dependent variable
More than 2 means
Wilcoxon rank sum test (ordinal)
ANOVA w/ F-tests (interval or ratio)
Match test w/ scale of data for dependent variable
Differences in variances
F-test (interval or ratio)
Match test w/ scale of data for dependent variable
Association b/w 2 variables
Spearman rho (ordinal) Pearson r (interval or ratio)
Match test w/ scale of data for dependent variable
Predicting the value of a variable
Logistic regression (nominal) OLS regression (interval or ratio)
Match test w/ scale of data for dependent variable
Predicting the value of a censored variable
Cox proportional hazards analysis (nominal)
Mode
value that occurs most often
nominal and ordinal
Median
value in middle of distribution, 50th percentile
ordinal or interval/ratio
Mean
average
(population and sample means)
(interval/ratio)
Normal distribution
mean, median, and mode have same value – at top of bell curve
Range
difference b/w lowest and highest scores
Variance
mean of the squares of all the deviation scores in the distribution (the mean square)
What percentage of the area under the curve falls w/in 1, 2, and 3 SD from the mean?
1 SD from mean: 68%
2 SD: 95%
3 SD: 99.7%
Prevalence vs incidence
prevalence: number of people w/ disease at given time (chronic)
incidence: number of NEW cases of a disease w/in a certain time period (acute)
Prevalence is affected by…
incidence (high incidence inc prevalence)
recovery (high recovery rate dec prevalence)
mortality (high mortality dec prevalence)
Maternal mortality
death of woman while pregnant or w/in 42 days of termination of pregnancy from any cause related to or aggravated by the pregnancy or its management
Neonatal mortality
rate of infant death during first 28 days after live birth
Infant mortality
number of infant deaths in first yr of life for every 1,000 live births
Under-5 mortality (child mortality)
probability per 1,000 that a newborn baby will die b/f reaching age 5
Life expectancy
how long a person is expected to live, based on yr of birth, current age, and other factors
Health-adjusted life expectancy
number of healthy yrs a person is expected to live at birth by subtracting the yrs of ill health
Yrs of potential life lost
estimating the avg time a person would have lived had he or she not died prematurely
Quality-adjusted life yrs
measure of the value of health outcomes
Disability-adjusted life yrs
sum of the years of life lose due to premature mortality in the pop and the yrs lost due to disability
Why use relative risks?
stable across populations with different baseline risks and
are, for instance, useful when combining the results of
different trials in a meta-analysis
When are relative risks used vs odds ratios?
Relative risks: when prospective cohort studies or RCTs are conducted
Odds ratios: used for case-control studies b/c we do not know the true incidence of a disease/outcome
Validity
IS THERE BIAS? did the study measure what it claimed to test?
how accurate is the study?
is there bias (systematic error)?
Internal validity
are the results of the study valid for the pop studied?
External validity
are the results of the study valid for the larger pop? are they generalizable?
Reliability
HOW PRECISE ARE THE RESULTS?
do you get similar results if you measure more than once?
is the study precise in measurement?
3 measures of reliability
test/retest reliability
repeatability and reproducability
precision of measure
Types of bias in external validity
Sample size too small
Volunteers used
Inclusion and exclusion criteria too select
Efficacy vs effectiveness
Efficacy: determine whether an intervention is successful under IDEAL circumstances
Effectiveness: determine whether an intervention is successful under REAL WORLD clinical settings
Types of bias in internal validity
Measurement or info bias --recall bias --ascertainment bias Intervention bias Attrition bias
Types of bias in internal validity
Measurement or info bias
+ 2 types
were the predictor and outcome variables measured accurately?
- -recall bias: participants may not remember past events
- -ascertainment bias: researchers or participants have knowledge of who is receiving the intervention, lack of blinding
Types of bias in internal validity
Intervention bias
did the authors select an unusually high dose for the comparison drug?
Types of bias in internal validity
Attrition bias
loss to follow-up b/c too many people drop out, lack of intention-to-treat analysis
Ecological fallacy
conclusions about indv are based only on analyses of group data
Hawthorne effect
people who know they are being studied may modify their behavior or feelings
Confounding varaibles
true effect is due to an unmeasured variable that affects the results, lack of randomization to intervention and control groups
Case control
Cohort study
Cross-sectional
*these are all types of what kind of research?
Case control: outcome –> exposure
Cohort: exposure –> outcome
Cross-sectional: snapshot at one time period AKA prevalence/frequency survey
*all are analytic (observational, primary)
Case control study: cannot calculate what measures?
Why?
What type of study allows for calculation of these?
Cannot calculate relative risks or attributable risks b/c no pop denominator
*cohort study: can calculate incidence rates, relative risks, attributable risks
PICOT stands for…
P: pop I: intervention C: comparison O: outcome T: time
Health disparities
Health disparities are preventable differences in the burden
of disease, injury, violence, or opportunities to achieve
optimal health that are experienced by socially
disadvantaged populations
Peer review
Peer review
is the critical assessment of manuscripts
submitted to journals by experts who are not
part of the editorial staff
FRISBEE stands for
why is it important?
Follow-up Randomizaiton Intention-to-treat Similar baseline Blinding Equal treatment Equivalence to your pt **validity of research
FINER stands for
Feasible Interesting Novel Ethical Relevant
PPICO stands for
Problem Patient/population Intervention Comparison Outcomes ***clear clinical question in systematic review
3 models for Meta-Analysis
Fixed effects model: any difference found among study results due to chance
Random effects model: difference b/w study results due to chance and other effects
–popular when interventions thought to be more variable
Bayesian Meta-Analysis
Forest plot
Quickly visualize the results of individual
studies and possibly for pooled data
L’Abbe plot
Quickly shows the amount of contribution of individual studies to the outcome sample size is proportional to circle size
Funnel plot
bias unlikely vs likely
bias unlikely: dots to left and right of zero
bias likely: dots to right of zero
Causes of publication bias
reporting bias
true heterogeneity
data irregularities
chance
Qualitative sources of heterogenetiy among studies
Patients in studies (differences in gender, age range, disease state)
Interventions (drug vs. placebo, drug A vs drug B)
Outcomes (death, inc chance of MI)
Clinical research design of indiv studies
Tests for heterogeneity
Mantel-Haentszel Chi-Square test
Breslow-Day Test
Cochran’s Q Test
I^2 statistic
Positively skewed
Normal
Negatively skewed
Pos: mean and median to the RIGHT of the mode
Normal: mean, median, mode are the same
Neg: mean and median to the LEFT of the mode
Standard normal or Z distribution has a mean of ___ and a SD of ____
mean = 0 SD = 1
For normal distribution to be fully defined, what 2 measures must be known?
mean
SD
Test statistic
measures the degree to which observation varies from predicted
Type of data used with these tests:
Student’s T test
Chi-squared test
ANOVA
Student’s T test: continuous data
Chi-squared test: categorical data and proportions
ANOVA: comparing means of 2 or more pops
Null hypothesis
the intervention being studies has NO EFFECT
Type I error
incorrectly concluding that an effect exists when it does not (errouneous REJECTION OF NULL HYPOTHESIS)
**worry more about type I errors
Type II error
failing to recognize an effect that truly exists (erroneous ACCEPTANCE OF THE NULL HYPOTHESIS)
Alpha
prob of making a type I error
Beta
prob of making a type II error
To reduce alpha and beta, need to…
inc sample size
P value
prob of obtaining the observed results if the null hypothesis is true
Power
prob of finding an effect if it truly exists (1-beta)
Parameter estimation
determining the plausible range of values for a parameter of interest in a pop or experimental group
- -point estimate
- -CI
To inc the degree of confidence, how do you have to change CI
wider CI to inc degree of confidence
The mean of sample means should be ___ the mean of the whole pop
The SD of the sample means will be ___ the SD of the pop
Mean of sample means should be the SAME as the mean of the whole pop
SD of the sample means should be LESS THAN the SD of the pop
Odds ratio
measures the degree to which exposure to a risk factor or a treatment changes the odds of experiencing an outcome
Non-inferiority margin
maximum acceptable loss of efficacy
Prob of type I error ___ with number of independent hypotheses tested
increases
Bonferroni correction
adjust p values when doing multiple hypothesis testing
alpha = 0.05/# hypotheses tested rather than alpha = 0.05
Sensitivity
how reliably does the test pick up disease when present?
to calculate, only consider pts known to have disease
neg result on highly sensitie test rules out disease
Specificity
does this test avoid false-pos?
to calculate, only consider pts who do not have the disease
pos result on a highly specific test rules in disease
Pre-specified hypotheses
based on prior research, clinical/biochemical reasoning, or other first principles
precisely defined in study protocol
Post-hoc
even if highly statistically significant, not proof of association or causality
–to be convincing, findings must be both improbable and have a plausible explanatory mechanism
ANOVA
more than 2 groups F statistic (variance b/w and within groups) If have more variance b/w groups than within the groups, you will get a bigger F statistic which will be more likely to be stat sig
Repeated measure ANOVA
when longitudinal data collected
–each person has repeated measures of the dependent variable
Therapy vs prognosis studies
Therapy: involve experimental intervention by researcher
–compare groups based on treatment
Prognosis: observational
–look for associations b/w variables
Parametric tests
normally distributed dependent variable ratio or interval level data measures are independent usually used to examine means (t tests, ANOVA) --sometimes difference in variance
Nonparametric tests
make NO assumptions about normal distribution
more conservative than parametric
used for analysis of medians and proportions
used for ranked data
used for nominal or ordinal data
Wilcoxon rank sum test AKA Mann-Whitney U test
nonparamentric test
when assumptions for t test don’t hold
ranks scores from lowest to highest
ranks analyzed as though they were original observations
null hypothesis: means of the ranks are equal
Independent vs dependent variable
independent: predictor
dependent: outcome
Dose-response relationship
in drug studies, what is the largest, most effective dose w/o serious side effects?
Types of regression ordinary least squares poisson (explain) logistic (explain) cox proportional hazards analysis (explain) hierarchical linear modeling
Poisson (count dependent variable)
Logistic (dichotomous dependent variable)
Cox proportional hazards analysis (survival analysis)
Logistic regression
predicting a binary (dichotomous outcome)
predicts the probability of the outcome variable
regression coefficients can be transformed into odds ratios w/ CI
used for multivariate analysis
First order interactions
the relationship b/w an independent variable and the dependent variable is conditional upon a second independent variable
Kaplan-Meier Curve
used with survival analysis
x axis: exposures
y axis: percent of event-free subjects
Relationship b/w CI width and sample size
CI width inc as sample size dec
–seen in survival curve as people die off
Survival analysis
censored observations
Kaplan-Meier curves plot timing of events
Cox proportional hazard analysis for multivariate analysis
based on regression analysis
What tests use multilevel modeling?
logistic regression
OLS regression
Prevalence and Diagnostic tests
Prevalence: tells dr the prior prob of the disease
Diagnostic test: alters the disease prob estimate
How are these measures affected by inc in prevalence? sensitivity specificity PPV NPV
sensitivity and specificity remain constant PPV inc NPV dec **inc true pos **inc false neg
ROC curves
method to find best cut point
used to compare tests
–test is best when has greater area under the curve
–best cut point is at the top of the curve
Downfall of screening tests
increases number of false pos
CAGE screening questionaire
Cut down
Annoyed (others)
Guilty
Eye opening