Evaluating the Role of Random Error Flashcards
If the observed result is false, what could be the 3 alternate explanations?
- Bias
- Confounding
- Random Error
Random Error
- aka chance
- arises from:
- measurement errors
- sampling variability
What are some ways to reduce random error?
- increase sample size
- repeat a measurement or entire study
- use an efficient study design
Hypothesis Testing
- determines if random error/chance causes association
- P value
Null Hypothesis
- no association b/w exposure and outcome
- RR=1, OR=1, RD=0
Alternate Hypothesis
- is an association b/w exposure and outcome
- RR≠1, OR≠1, RD≠0
Define P Value:
- given that H0 is true, the p-value=probability of observed and extreme results by chance only
- Range: 0-1
If p ≤ .05…
- significant
- reject H0, Accept HA
If p>0.05
- not significant
- do no reject H0
Problems with use of P-value in research
- Does not:
- imply medical, biological, or public health significance
- rule out bias or confounding explanations
- mean H0 is true
- P≤0.05 is arbitrary→ judgement errors
What does the confidence interval tell you
- range of hypotheses that are comparable to the data
What does a wide confidence interval indicate?
- smaller population size
- large amount of random error
what does a narrow confidence interval indicate?
- large sample size
- small amount of random error
If null hypothesis is within confidence interval range, the what would the p-value be?
- P-value> 0.05
- not significant
- H0 not rejected
Confounding
- mixing of effect b/w exposure, outcome, and confounder (3rd variable)
- exaggerates or minimizes true association
Criteria to be a confounder?
- Must be associated with:
- exposure
- outcome
- independent of exposure
- not intermediate step in cause pathway b/w exposure and disease
What are the effects of confounding?
- account for all or part of association
- cause overestimate or underestimate of association
What variables can be potential confounders?
- risk factors for disease
What are the different ways to control for confounding?
- Design Stage:
- Randomization (RCT)
- Restriction
- Matching
- Analysis Stage:
- Stratification
- Multivariat analysis
Randomization
- with sufficient sample size
- control for known and unknown confounders
- not guaranteed
Restriction
- restrict admission criteria for study
- limits individuals in specific category of confounder
- ex: Race limited by age or height
What is the goal for controlling confounding in design phase
- eliminate or reduce variation in confounding factor b/w compared groups
Restriction: Advantages and disadvantages
- advantages
- straight forward
- convenient
- inexpensive
- disadvantages
- limits generalization
Matching
- select subject so potential confounders are distributed in identical manner among:
- exposed and unexposed
- cohort study
- case and controls
- case control study
- exposed and unexposed
Stratification
- evaluate association within homogenous categories (strata) of confounding variable
Bias
- systematic error
- results in and incorrect measure of association:
- creates association when there is not one
- bias away from null
- Mask association when there Is one
- bias towards the null
- creates association when there is not one
- arise in all study types:
- experimental
- cohort
- retrospective=mostsusceptible
- case-control
- evaluated but not fixed in analysis design
What are the 2 main type of bias?
- selection
- observation
What type of studies are Selection Bias most likely to occur in? Why?
- most likely to occur in:
- case control
- retrospective
- bc Exposure and outcome occurred at time of study selection
What is the solution to selection bias?
- nothing can be done
Observation Bias
- error in the info on exposure or outcome from subjects
- affects groups unequally
- occurs after subject have entered the study
- results in incorrect classification of subjects:
- exposed vs unexposed
- diseased or not
What are the different types of observation bias?
- Recall
- Interviewer
- misclassification
recall bias
- remember or report exposures differently
- more or less accurate than those w/o disease
Interviewer bias
- difference in soliciting, recording or interpreting info
misclassification bias
- subjects exposure or disease status is classified wrong
- incorrectly reported or recorded
Loss of subjects to f/u
- Problem in cohort studies when related to exposure and disease
What diseases are appropriate for screening?
- serious disease w/severe consequences
- progressive disease & tx is more effective at earlier stage
- High Prevalence in DPCP
- Detectable pre-clinical phase
What are some characteristics of good screening test?
- economical
- convenient
- free of risk and discomfort
- acceptable to large population
- highly valid and reliable
test validity
- does it accurately identify those with and without preclinical disease
- sensitivity
- specificity
Test reliability
- same result each time
2x2 table for screening test outcomes:
Define Sensitivity
- probability that people with the disease will test positive
- True Positive
How do you calculate sensitivity?
a/(a+c)
Define Specificity
- probability that people without the disease have a negative test
- true negative
t
How do you calculate specificity?
d/(b+d)
define Positive Predictive value
- the likelihood that a positive test result indicates the existence of the disease
- how worried should the patient be?
How do you calculate positive predictive value?
a/(a+b)
define negative predictive value
- likelihood that a negative test result indicates the absence of the disease
- how reassured should they be?
How do you calculate negative predictive value?
d/(c+d)
What do predictive values depend on?
- sensitivity
- specificity
- prevalence
Randomized control Trials
- gold standard for study designs
- bc of greater internal validity
Steps to conduct RCT
- enroll subject
- allocate to tx groups (Tx vs placebo)
- randomization
- blinding
- F/u for relevant period
- study groups monitored for outcome
- compare rates of outcome in different groups
RCT Step 2: Allocate Tx or placebo by randomization. What are the benefits? Overall Goal?
- Benefits:
- unbiased assignment
- known and unknown confounders are balance
- leads to comparability
- minimizes selection bias and confounding
- Goal:
- Groups are identical-all determinants except tx
- any difference due to tx
- Groups are identical-all determinants except tx
RCT: What groups are subjects randomized into?
- Tx Group
- New tx
- other doses of same tx
- Other: Combo therapy
- Comparison group
- Placebo
- Other: usual care (current tx in use)
What is the primary objective for using a placebo?
- minimize bias
Uses of placebo?
- makes group as comparable as possible
- depends on subjectivity of outcome
- cannot be done in some situations
- pills vs surgery
- Allows study to be blinded
What is the primary objective of blinding in RCT?
- Prevent bias from affecting results
What are the different types of blinding?
- Single-Blind
- subjects unaware of treatment group
- Double-Blind
- Both subjects & investigators unaware of tx groups
- cannot be done in some situations (pills vs surgery)
- Both subjects & investigators unaware of tx groups
- Triple Blind:
- Subjects, investigators admin tx, investigators eval outcomes all unaware
How to establish cause
- Valid Association
- even if valid, not all associations are causal
- eliminated alternative explanations
Hill’s Criteria
- Causality is strengthened by:
- temporal sequence
- risk factor first
- strong association
- dose-response relationship
- consistent findings
- biologic credibility
- results of experiment
- temporal sequence
Temporal sequence
- a factor must come before outcome to cause it
What types of studies provide the best evidence that causes precedes the effect?
Prospective Cohort Studies
Randomized Clinical trials
Strength of Association
- Large RR supports cause relationship
- by itself, can’t conclude cause
- bias or confounding
- by itself, can’t conclude cause
- Small RR also possible
Presence of Dose-Response relationship
more exposure means greater risk
