Exam 2 Material Flashcards
Quantitative study design
Design in which the results are quantifiable (involve numbers)
Ex: Interventional and observational studies
Interventional study design
“Experimental” study design in which researcher select an intervention (usually an exposure) and forces allocation groups using randomization
Best design to prove causation, therefore is the only design that is FDA-approved
Observational study design
“Observation of naturally occurring events”
Study design in which researchers do NOT force allocation groups
Typically does not prove causation
Population vs study population
Population is NOT the same as a study population.
Population = all individuals making up a common group
Study population = a portion of the full population that representative of the group (ie. sample)
Null hypothesis (Ho)
Perspective stating that there is no true difference between comparison groups
“Innocent until proven guilty”
If not rejected - there is NO difference between groups
If rejected - there IS a difference between the groups
Statistical perspectives/questions of:
Superiority
Non-inferiority
Equivalency
Superiority: is the drug superior to ___?
Ho = the drug is NOT superior to ___.
Non-inferiority: is the drug NOT worst than ___?
Ho = the drug IS worst than __.
Comparison is typically to the “golden standard”
Equivalency: is the drug equal in effect to ___?
Ho = the drug is NOT equal to ___.
Alternative hypothesis
Perspective that there IS a difference between comparison groups
Type I error
False (+)
Null hypothesis has been inaccurately REJECTED
Type II Error
False (-)
Null hypothesis has been inaccurately ACCEPTED
Probability sampling
Equal opportunity (known, non-zero probability) of selection to be included in a sample
Simple random sampling
Completely random sampling
Where each element is assigned a random number and then numbers are randomly selected until desired sample size
Ex. Flipping a coin
Systematic random sampling
Where each element is assigned a random number, and then selection of sample is based on a predetermined sampling interval (take an element every Nth element)
Stratified random sampling
Split population into strata based on specific characteristic and then do simple random sampling in each strata (where elements are assigned random numbers and random number are selected to make sample)
Ex. Male and female strata
Stratified disproportionate random sampling
“Weighing” statrified sample to make sample proportional to population
Useful for over-sampling
Multi-stage random sampling
Random selection at different stage intervals
Using simple random sampling at multiple stages toward patient selection
Ex. Regions -> zip codes -> clinic -> patient
Cluster multi-stage random sampling
Sampling ALL elements clustered together (at any stage) of multi-stage random sampling scheme
Ex. All clinics in a zip code are included
Quasi-systematic sampling
Sampling from a fraction of the population
-may introduce selection bias
Ex. Sampling from all persons with names beginning with “M-Z”
Patient-oriented outcome (POE)
An outcome that is more directly important to a patient
Ex. Risk of heart attack (POE) vs high BP (DOE)
Disease-oriented outcome (DOE)
Outcome attributable to diseases, but not necessarily of great concern to patients
Ex. High blood pressure (DOE) vs heart attack (POE)
EQUIPOISE
Confidence that an intervention is worthwhile (risk vs benefit)in order to be tested/used in humans
4 Principles of Bioethics
Autonomy - self-rule; allow an individual to make ones own informed decision with understand of risks/benefits involved and without outside influence
Beneficence - do good for the individual (NOT society); must have the individual’s best interest in mind
Non-maleficence - do no harm to the patient
Justice - treat individuals equally and fairly regardless of any characteristics they may have
Guidelines of the Belmont Report
- Respect for persons = research is conducted on a volunteer basis
- Beneficence = risks of research are justified by potential benefits
- Justice = risk and benefits are equally distributed among the study population
Consent
Agreement to participate of a mentally-capable, fully informed 18+ yo
Assent
Agreement to participate of a minor or individual unable to otherwise give legal consent after guardians have been fully informed of the risks and benefits
Institutional Review Board (IRB)
Protects human subjects from undue risks of research
Regulated by Dept. of Human Health Services (DHHS)
Enforced by the Office of Human Research Protections (OHRP)
Full Board vs. Expedited vs. Exempt
Full board = for all interventional trials with more than a minimal risk to the human subjects; requires the most time and resources
Expedited = for trials with minimal risk and no/little patient identifiers
Exempt = for trials that have no/little risk and no patient identifiers, or use existing data; required the least time and resources
*all are for BEFORE the study begins
Data Safety and Monitoring Board (DSMB)
Board that reviews a study and provides interim analysis as the study progresses
Can stop research if findings are overly positive or overly-negative
Pre-clinical (interventional study)
Bench/animal research prior to human investigation
Phase 0 (interventional studies)
First in-human use
- Assesses drug’s target actions
- Used on healthy or diseased volunteers
- Small sample (<20)
- Short duration (single dose - few days)
Phase 1 (interventional study)
- Assesses safety/tolerance of pharmacokinetics
- Used on healthy or diseased volunteers
- Small sample (20-80)
- Short duration (few wks)
Phase 2 (interventional study)
- Assesses effectiveness (and safety); an expansion off phase 1
- Used on DISEASED volunteers
- Larger sample (100-300)
- Longer duration (few wks-months)
Phase 3 (interventional study)
- Assesses safety/effectiveness over longer period before FDA approval
- Used on DISEASED volunteers (may expand inclusion criteria)
- Large sample (hundreds-thousands)
- Long duration (months-years)
Phase 4 (interventional study)
- Post-marketing studies assessing long-term safety/effectiveness post FDA approval
- Used on DISEASED volunteers
- Huge sample (thousands-hundred thousands)
- Long duration (years-ongoing)
Pros/Cons of Interventional studies
Pro:
- can demonstration causation
- only study that can lead to FDA approval
Con:
- expensive / complicated / takes time
- ethical considerations (balancing risk with benefit in human trials)
- must ensure external validity can be upheld
Exploratory studies
Research study that seeks to precisely answer research questions by “exploring” the safety/usefulness/efficiency of an intervention within a non-real clinical setting
Explanatory studies
Research study that seeks to “explain” how to treat a disease in relation to a patient in a real-life clinical setting
Is less restrictive and more applicable to patients than exploratory studies
Simple study design
An interventional study design in which subjects are divided in 1 step of randomization
Useful for testing a single hypothesis
Factorial study design
An interventional study design in which subjects are divided with 2+ steps of randomization using subdivisions
Useful for looking at multiple factors and their interactions; can test multiple hypotheses at the same time based on different subdivision combinations
Note: if too complex, the study may restrict generalizability
Parallel study design
An interventional study design in which there is NO SWITCHING of intervention groups after initial randomaization
Cross-over study design
An interventional study design in which subjects may serve as their own control by crossing between intervention groups during wash-out phases
Pro:
Allows for “between”/“within” group comparisons and smaller sample size (since data can be collected from the same person)
Con:
Only for long-term conditions/diseases
Subjects are studies for longer duration (enough to collect data twice)
Beware of carry-over effects (wash-out must be done correctly)
Complex data analysis
Internal validity may be compromised (treatment may affect subjects differently at different times)
What is the use of a run-in/lead-in phase?
Acts as a wash-out phase prior to the study beginning
Allows for researchers to determine a baseline for the subjects (removes existing medication in their system and can assess their protocol compliance)
Patient oriented endpoints (POE)
Results that are more clinically relevant or important to the patient
Ex: heart attack, stroke, etc.
Disease oriented endpoint (DOE)
Study result that evaluates the risk of a patient oriented endpoint
Ex: blood pressure, cholesterol level, etc.
Randomization (3 types)
Randomization = selection of subjects such that study groups are made as equal as possible
- Simple randomization - equal probability of allocation into either study group
- Blocked randomization - randomization that ensures balance within each intervention group
- Stratified randomization - randomization that balances groups based on known confounding variables
Single blind masking
Only subjects do not known what interventional group they are in
Double blind masking
Both researchers and subjects do not know which intervention groups the subjects are in
Open-label masking
Unmasked/unblinded studies
Placebo effect
Improvement in condition by power of suggestion
What can assess adequacy of blinding?
Post hoc survey’s
Placebo (“dummy”) treatment
An inert treatment that is made to look identical in ALL aspects to the active treatment
Hawthorne effect
When study subjects change behavior solely because they know they are being studied
Intention-to-treat
The most conservative form of managing drop-outs/loss-to-follow-up
Data from the drop-out is included in the study anayway
Pros:
- preserves randomization
- preserves base-line characteristics including balance for cofounders
- maintains statistical power by keeping the original sample size
Per protocol (aka. Efficiency analysis)
Form of drop-out/loss-to-follow-up management in which they are ignored
Their data is removed from the study
Con:
- prone to type II error
- reduces generalizability
As treated
Form of managing drop-out/loss-to-follow-up in which end results are somewhat estimated
Data from the missing subjects is used as if they had stayed in the study
Assessing adherence
“Are subjects following protocol?”
Ex. Test drug levels / count pills / check bottle counter tops
Improving adherence
“How can we ensure subjects follow protocol?”
Ex. Increase follow-up visits/communication; treatment alarms; dosage containers or medication blisters for easy access
Case control studies are particularly impacted by what kind of bias?
Selection Bias and Recall Bias
Sampling for Nested Case-Control studies:
Survivor sampling = samples from non-diseased individuals at the end of a previous prospective study
Base sampling = samples from non-diseased individuals from the beginning of a previous prospective study
Risk-set sampling = sampling from non-diseased individuals from within the prospective study period at the same time when case was diagnosed (control group is time dependent)
Case Control Studies
Observational studies in which group assignments are based on disease status; retrospective study
Strengths:
- useful for studying rare diseases
- can assess multiple exposures on one disease
- useful for when disease has a long induction/latent period
Weaknesses:
- cannot demonstrate causation
- impacted by selection and recall bias
- selection of controls is difficult (selection is irrespective of exposure)
Individual matching
Matching individuals based on patient-based characteristics to control for confounder
Ex. Groups will have same number of males and females
Group matching
Matching proportions of characteristics between case and control groups
Ex. 41% of cases are white and 41% of controls are white
Cohort studies
An observational study in which group assignment is based on exposure status or a shared common factor; can be conducted prospectively, retrospectively, or am-bidirectionally
Strengths:
- useful for studying rare exposures
- can represent Temporality (if conducted prospectively)
- can assess multiple outcomes/diseases from 1 exposure
Weakness:
- cannot demonstrate causation
- exposures may change over time (are hard to control)
- long induction/latency is not good for Prospective Cohorts
What biases typically impact Cohort Studies?
Healthy-worker effect and Selection Bias
Internal validity
The extent to which a study’s results accurately reflect what was being assessed
External validity
The extent to which a study’s results are applicable to an entire population; generalizability
Validity
The ability to discern between those that have disease and those that do NOT have disease
What can be “chosen” to minimize false positives and/or false negatives?
Cut-offs
Reliability
The ability of a test to give consistent results
LR+ and LR-
LR+ : the likelihood of a diseased individual receiving a (+) result compared to a non-diseased individual; useful if LR+>10
=sensitivity/(1-specificity)
LR- : the likelihood of a diseased individual receiving a. (-) result compared to a non-diseased individual; useful if LR-<0.1
=(1-sensitivity)/specificity
Likelihood Ratios
Ratio of the probability of receiving a test result (+/-) in the diseased group compared to the non-diseased group
Loss-to-follow-up increases what type of error?
Type II Error
Birth cohort vs. Inception cohort vs. Exposure cohort
Birth cohort - groups are based on time period and region of birth
Inception cohort - groups are based on a common factor
Exposure cohort - groups are based on common exposure; ex. One time events
Fixed cohort
Cohort in which sample cannot be added to (fixed max from starting sample), but can have loss-to-follow-up
Closed cohort
Cohort in which sample cannot be added to or reduced (fixed at both ends)
Open/dynamic cohort
Cohort in which sample can be added to and/or reduced
Cross-sectional study
An observational study that captures information about an exposure and disease at the same time; aka. Prevalence study
Typically are large-scale, national surveys
Strengths:
- quick / easy / relatively cheap
- estimates prevalence rates
- can use the same data for different research questions
Weaknesses:
- difficulty in studying rare diseases
- cannot determine temporal relationship for cause/effect
What cross-sectional survey combines interviews with physical exams?
National Health and Nutrition Examination Survey (NHANES)
How does the National Ambulatory Medical Care Survey sample a population?
Multi-step random sampling
What cross-sectional survey collected data with an interview by telephone?
BRFSS - Behavioral Risk Factor Surveillance System
2 questions patients should ask their physician before a medical screening:
- How accurate is the test?
2. How confident are you that the test is accurate in its prediction?
Sensitivity
How accurate a test is in detecting disease in someone who actually has the disease; proportion of true positives
=TP/(#diseased people)
=TP/(TP+FN)
Specificity
How accurate a test is in detecting the absence of disease in a non-diseased individual; proportion of true negatives
=TN/(#non-diseased)
=TN/(TN+FP)
Positive Predictive Value (PPV)
How accurate a (+) test predicts the presence of disease
=TP/(all positive results)
=TP/(TP+FP)
Negative Predictive Value (NPV)
How accurate a (-) result is in predicting the absence of disease
=TN/(all negative results)
=TN/(TN+FN)
What is the effect of prevalence on PPV and NPV?
Increasing prevalence causes an increase in PPV and NPV
This is because the more frequent a disease is in a community, the easier it is to “find” the disease
Diagnostic Accuracy
Proportion of screening in a sample that are correctly identifying the presence of absence of disease
=(TP+TN)/(total sample)
=(TP+TN)/(TP+TN+FP+FN)