Stat Flashcards
odds ratio<1
decreased risk
odds ratio=1
equal risk
odds ratio>1
increased risk
odds ratio= relative ratio
outcome is rare
to strengthen the argument for causality, consider (7)
consistency, plausibility, dose-response, temporality, strength of relationship, reversibility, lack of alternative explanations
types of descriptive studies (3)
detail one observation
ecologic study, case reports, case series
case reports
○ One or few patients
○ Link clinical medicine and public health
○ Publications and rounds
○ Rare disease/cases
case series
More than a few patients Good details CONS Small, highly selected group No hypothesis No comparison
Ecologic Study
Ecologic Study
○ Goal: comparing disease rates between population groups
○ Exposure (predictor or risk factor) —> disease (outcome or response)
○ “ecological correlation” or “aggregate risk” = exposure-outcome relationship
○ Suggests a link associated with a group
○ Ex) countries with higher fat diets = higher breast cancer rates
Ecologic Study: pros
Etiological hypothesis
Use to set research priorities
Low cost
Study large population
Studies hard to study environmental health questions
Ecologic Study: cons
No individual data
“ecological fallacy”
One could infer inappropriate individual relationship
**be careful not to over-interpret results
Analytical Study types (3)
cohort, case-control, cross-sectional study
Case-control study
*rare diseases
outcome –> exposure
○ Moves backwards in time
○ Find those with disease and look back at their exposure
○ Controls: from at risk population (had opportunity for exposure/disease), but free of disease at time
○ Odds ratio: estimates risk
Case-control study: pros
Study rare or long latency diseases
Requires few subjects
Faster, Less time
Evaluate multiple exposures (risks) as potential causes of disease
Case-control study: cons
Relies on subject’s recall for past exposures; biases
Difficult to select appropriate control group
Odds ratio only estimates relative risk
Cannot calculate incidence rates
Cross-sectional study
*quick measure
exposure and outcome at same time
○ “prevalence study”
○ Ex) who is more dissatisfied with weight: male or females?
Prevalence ratio
Cross-sectional study: pros
Good measure of disease prevalence What to expect in clinical setting Evaluate screening and diagnostic tests Help plan health services Quick- ask one question Easy Inexpensive
Cross-sectional study: cons
Measure disease/exposure at same time Cannot determine causality Cannot determine temporal relationship of exposure and disease Limited: study prevalence only Cannot determine disease incidence
Cohort study
*rare exposures
exposure—> outcome
○ Moves forward in time
○ Follow patients over time to see if they develop disease
○ Compare incidence of new development of disease
○ Ex)is physical fitness related to respiratory illness risk
assesses relative risk, attributable risk
Cohort study: pros
An evaluate multiple outcomes
Provide actual measure of risk of outcome
Can extract incidence and relative risk
Approximates Random control design
Cohort study: cons
Potential loss for follow-up
Needs large number of subjects
Takes a long time- not efficient to wait for outcome
Expensive, lots of staff
Randomized clinical trial
*best evidence Experimental Study • Randomly assign participants to one or two treatments • Produce comparable, similar study groups (equal known/unknown risk factors) • Removes investigator bias by allocating participants randomly • Valid statistical tests • Comparison groups: ○ No intervention ○ Observation- Hawthorne effect ○ Placebo ○ Usual care • Blinding/ masking
define: correlation
measures strength of association btwn 2 variables
define: regression
method for relating predictor to outcome
Q: “does an association exist”
“quantify the strength of the association”
find correlation
Q: “use the relationship to predict”
“does the observed relationship agree with this theory”
“estimate the parameters of this model
find regression
how to measure linear association
correlation coefficient
steps in using correlation coefficient (4)
1 observe (x,y) variables for random sample
2 plot pairs of points in scatter plot
3 find pattern of association
4 estimate population correlation coefficient (p)
correlation coefficient: range
-1 < r < 1
correlation coefficient: r=0
no linear association
loose clustering
correlation coefficient: r= 1
perfect positive linear association
tight cluster
correlation coefficient: r= -1
perfect negative linear association
tight cluster
when to use standard pearson correlation
random sample
normal distribution
when to use spearman rank correlation
decrease influence of outliers
ranks variables low to high and recalculates
“least squares” regression line
minimizes the sum of squared differences from best fit line
y= A+ Bx y= (intercept) + (slope)x
when to use multiple linear regression
adjust for cofounders
when outcome is continuous
simple logistic regression: use
estimate odds ratio
when dependent variable is categorical (binary)
multiple logistic regression: use
estimate “adjusted” odds ratio
for multiple predictors, binary outcome
to assess association between 2 continuous variables use…..
correlation or linear regression
to assess:
- association between continuous (or categorical) predictor variables
- estimate odds ratios for categorical (binary) outcome
- odds ratios after adjusting for other variables
logistic regression
ex) BMI & High BP
ex) age & anemia
estimate a regression line for a curve shaped scatter diagram?
linear relationship is unlikely
correlation coefficient ~0
computation of simple linear regression is contraindicated
What kind of study is appropriate for an outcome that is rare?
Case-control
Stratification
divide total sample into subgroups to deteremine odds ratios
Matching
manipulate study to directly compare factors you think are biasing
e.g., exposed old ppl to control old ppl
Adjustment
use regression models (odds ratios)
