Study Desgin And Preventing Bias Flashcards
3 keys to reproducibility
Use a study design that will answer the question of interest
Prevent bias in anyway possible
Exercise intellectual honesty, rigor and completeness
Cross section studies
Observational study that analyzes data at a specific point in time
Process
- draw a sample for a population at any given point
- identify those with and without the diseases
- identify exposures of interest
- examine association between exposure and outcome
Advantage:
- hypothesis generation is easy and is cheap
- can show association between something
- good for evaluating screenings
Limitations:
- only know about the population at a single point in time
- limited to collecting prevalence (who has the disease now) data
Cohort study
Longitudinal study of people who share a defining characteristic
- define some group of people prior to the onset of the disease and follow them overtime
Advantages
- temporality is clear
- can examine multiple diseases
Disadvantages
- often expensive to run
Clinical trials
A type of cohort study in which an intervention is assessed
- the randomized controlled trial is one type
No exposure assigned via natural processes (unlike normal cohort), instead it is assigned via randomization or direct assignment
- blinded = randomized by patient groups only
- double blinded = randomized by patients and doctors (nobody knows who has what until the ending)
Randomized controlled trials
Considered the gold standard for design for evidence
- considered the best at theoretically eliminating the most bias or chances
- show causation rather than association
Drawbacks
- expensive and time consuming
- hard to administer and track
- bias can still creep up
- ethical issues may arise
What is each study best at?
Randomized control trials
- best at removing bias
- best at establishing causation
Cohort studies
- almost never unethical
- allows for long-term follow-up
- establishes weak causation
Case control
- best for rare diseases
- establishes only association
Cross sectional
- good for generating hypotheses
- best for diagnostic screening treats
- establishes only association
Case reports
- best at getting info out quickly
- establishes only association
Advantages to meta analysis
Generalize results to larger populations
Precision and accuracy can be improved when using data
Inconsistency across studies can be quantified
Hypothesis testing can be applied easy
Potential problems in meta analysis
Does not predict results of a single large study
Cannot correct for bias
Selectively publishes studies the analysis wants only
5 major types of bias
Selection
- systematic difference between comparison groups
- early on
Performance
- systematic difference in care between groups
- during the study
Exclusion
- systematic difference in study withdrawal
- during the study
Detection
- systematic difference in outcome assessment
- after the study
Interpretation
- systematic difference in drawing conclusions
- after the study
3 types of selection bias
Non random sampling
- fix by randomizing
Berkson bias
- using hospital populations is bias since they are generally les s healthy than non hospitalized patients
Non-response bias
- in some way, participants that are actually surveyed and return it differ from people who did not report back
4 types of performance bias
Recalling bias
- patients recall exposure after learning of similar cases
- decrease time from exposure -> follow up to fix this
Hawthorne effect
- behavior is changed because the participant knows someone is actively watching/measuring them
- using placebo groups fixes this
Procedure bias
- physicians treat patients in the treatment group better than patients in the placebo
- double blinding fixes this
Pygmalion effect
- if someone believes the novel treatment is going to be positive, they will falsely document more positive outcomes (can occur at both patient and physcian levels)
- double blinding fixes this
The main type of exclusion bias
Systematic differences in treatment withdrawal
- ADRs and lack of efficiency causes dropouts
- using Intent to treat analysis fixes this
Main type of detection bias
Lab differences
- each clinical site uses a different radiologist/pathologist who will assess outcomes differently
- using a central lab/imaging center that is also blinded fixes this
2 types of interpretation bias
Confounding
- drawing a conclusion from the data, without realizing there is another correlation that you didn’t take into account
- using crossover studies and matching patients fixes this
Lead-time
- interpreting the time from detection of a disease and its clinical presentation without acknowledging the patient age possibly playing an effect
- must adjust for overall survival of a patient at that age and the severity of the diagnosis
