Experimental Study Flashcards
Description & Objectives of:
- Randomized Clinical Trial
- Randomized Community Intervention
RCT = Clinical pop., like patients in a certain setting
Objective:
-test efficacy of intervention to improve disease prognosis or prevent
-Suggests the feasibility of community studies and programs
RCI = a community population
Objectives:
- identify persons at high risk/in need
-Test efficacy and effectiveness of interventions
-Suggest / justify health policies + programs
Efficacy vs Effectiveness
Efficacy - refers to whether an intervention has beneficial effect when applied properly to those intended for
Effectiveness - refers to whether an intervention has a beneficial effect under real conditions in the target pop. it is offered to.
Randomization
- Defining feature of an experimental study
- Assignment of subjects to different treatment groups involves a random process.
- Simple randomization: subjects have equal probability to be assigned to an exposure / control group.
- Helps ensure comparability among groups –i.e., assuming no treatment effect, we would expect equal distribution of the outcome in all groups
- All groups will have similar distributions of extraneous factors (confounders) that also affect the outcome = helps to control for the effect of these factors.
Unit of Randomization
Most studies - it is the individual
Sometimes the group - i.e., region, school, family– the treatment is assigned to groups
*community interventions involving environmental exposure/policy change etc
Advantages of Randomization
- Greatest control to measured/unmeasured confounders
- Enhances ability to isolate the effect of the treatment to the effects of other factors
***however compliance may not a random process, which might subvert the benefit of randomization
Disadvantages of Randomization
- Limited generalizability of results to a target population
- studies are conducted ion highly selective samples of strongly motivated/incentivized pple
- unethical to randomize if tx is too hazardous or if there is a better treatment already accepted
Cross-over design
Allows each subject to serve as their own control
each subject receives either tx, after the first period they are crossed over to the other tx.
The order in which A and B are given is randomized
Advantage + Disadvantage of Cross-over design
Advantage = Allows for smaller sample size - more efficient
Disadvantage = effects from the first f/u period may carry over into the second f/u period
Questions to Consider before Choosing a Cross-over Design
- Is the condition of the patients chronic and stable?
- Crossover trials are common for conditions such as asthma, osteoarthritis
- Period Effect: May not be appropriate for progressive diseases / acute conditions that will worsen or improve by the second period. - Does the intervention provide temporary relief, and not permanent change?
Ex, surgical interventions unsuitable for crossover trials if the surgery permanently alters the condition. - Can the outcome be repeated in 2nd f/u if it occurs in the first?
Ex, unsuitable if the primary outcome is mortality, or pregnancy in infertility studies. - Will the effect of the first intervention last into the second treatment period?
A ‘washout’ period is usually built into the trial between the two treatment periods. This is a common method to minimize ‘carry-over’ effects and ensure the participant is in the same state at the beginning of each period, - Does the trial go on long enough for drugs to have effects and outcomes to occur?
Ex may need to observe patients for long enough to make sure it’s not a particularly good or bad time in their illness.
Experimental Studies:
Selection of Study Subjects
Via Eligibility Criteria: based on scientific goals, to apply conclusions to defined pop
+
Exclusion Criteria:
Experimental Studies:
Sample Size Determination
Trial must have sufficient n to have adequate statistical power
-in order to detect existing associations
Sample size should also be small enough to be feasible
In order to:
- Increase effect size (magnitude of difference to be detected) = decrease n
- reduce type I & II error = increase n
- increase characteristics of data (variation, sd) = increase n
Type I error vs Type II error
Type I error - observed significant association when there is none
Type II error - did not observe a significant association when there was one
Experimental Study
Recruitment of Subjects
Goals:
1. Recruit a sample that
adequately represents target population
2. Recruit enough subjects to meet desired sample n
Experimental study
Follow Up and Data Collection
Follow subjects from recruitment to the end of the trial
-Assess how many patients will develop the outcome of interest
Data Collection:
-Measure baseline characteristics before randomization to avoid bias in collecting/reporting baseline characteristics.
- Outcome = factor on which the assessment of treatment efficacy is based.
Primary Outcome - must be specified in advance.
Selecting the Primary Outcome
- Typically a measure that is easy to assess in all patients.
- Objective outcome measures that are well-defined and can be observed directly are preferred. (death, survival, or biologic makers).
- Subjective outcome measures pain reduction, quality of life etc) = more subject to bias if staff is not blinded
- Outcome must be measurable according to the same criteria, regardless of tx.
All patients followed in the same way, with the same tests, and at the same intervals after treatment.
Blinding
- most desirable in trials where outcomes are subjectively measured
- reduces information bias
Can Occur at 4 Levels:
- Researcher level: tx allocators do not know which tx will be assigned, reduces bias in subject selection/tamper with randomization
- Subject level: patients unaware, less likely to bias compliance/reporting of sx
- Physician level: observers taking care of patients blinded to reduce differential care
- Investigator level: cannot affect measurement or data analysis
Compliance + Adherence
-The researchers must be sure the tx group is actually receiving the intervention
- Regularly assess compliance during f/u period:
- count unused tablets
- biological methods
RCT Data Analysis
Are the 2 groups comparable?
- any difference btwn the two groups at baseline?
if no, then comparable via randomization
Analysis strategy depends on:
- Allocation method (rct or cross over)
- Randomization Unit (individual or group/cluster)
RCT Data Analysis:
Intent to Treat Analysis (ITT)
- Advantages
- Limitations
- What is required for ideal ITT Analysis?
All patients are included in analysis in the group they were originally randomized to, regardless of whether or not they actually receive the allocated treatment
Advantages:
- Retains balance in risk factors that may affect the outcome
- Gives an unbiased estimate of treatment effect
- Admits non-compliance and protocol deviations = reflects real clinical situation
Limitations:
- Conservative estimate of treatment effect due to dilution from non-compliance
- Interpretation becomes difficult if a large proportion of participants cross over
Requirements for an ideal ITT analysis:
- Full compliance
- No missing responses
- 100% retention
RCT Data Analysis:
Treatment-Received Analysis
Advantage
Disadvantages
Addresses extent to which the tx produces effect under optimal conditions by considering the outcome only in participants who completed the treatment as intended.
- Answers the research question: whether the treatment itself is better?
Disadvantages:
- Compromises balance of factors that affect the outcome
- Introduces bias into the treatment comparisons
RCT Data - Assess Baseline Characteristics
Compare baseline characteristics by tx groups
- discover possible confounders
- discover differences in outcome variables btwn tx groups
Compute sample statistics for each group
- means, sd, medians
Sometimes journals ask for stat. tests on sig differences between 2 groups, just to prove comparability.
- *Flawed thinking - bc any differences should be due to chance if randomized
- conduct the test if you suspect randomization was flawed.
RCT Data - Assess crude effect of tx
- Estimate the magnitude of the effect on the outcome measure, compute confidence interval
- A p-value can also be provided
Type of Outcome / Stat Tests:
Dichotomous outcome / chi square, Fishers, Risk ratio
Nonimal / chi squared
Continuous normal / t-test, ANOVA
Continuous non-normal / Wilcoxon test
Ordinal / chi square for trend
Time to event, censored data / Log rank test, Wilcoxon test
Single post treatment outcome measure:
Continuous Outcome Data
Parametric Test
Non Paramentric Test
Parametric statistical test
This type of tests is used with the assumption that the data are sampled from a normally distributed population.
Nonparametric statistical test
This type of tests does not make assumptions about the population distribution.
These tests use the ranks of the outcome variable from low to high, and then analyze differences in the ranks.
Parametric Test
T- Test
Analysis of Variance
Two-independent-sample t-test
- Test for a difference between two means obtained from two independent populations.
Assumptions for t-test:
- Outcome variable is continuous and normally distributed.
- Homogeneity of variance: we assume that the variances are equal for both groups in the population.
ANALYSIS OF VARIANCE:
- use when testing a difference between 3 or more means of interest.
- same assumptions
Non-Parametric Test
Wilcoxon rank test is used when assumptions for t-test are not held.
RCT Data- Adjusting for Confounders
Determine possible confounders:
- Variables with imbalance between groups
- Variables related to outcome: examine association between different variables and the outcome
Adjust
- Stratification approach (Mantel-Haenszel) : stratify data by confounder and derive adjusted measure of association
- Confounding present if difference btwn adjusted + crude estimate
- Regression Approach:
- linear regression, log binomial, poisson, cox proportional hazard model, etc
Issue of Attrition - Study Drop Out
- Are the 2 groups still comparable in terms of confounders?
- -> any differences in characteristics at the end of the trial? - If Attrition creates imbalance, tx effect will be distorted
- If no imbalance created, statistical power will still be reduced to do decreased n
Can impute missing data
Can conduct subgroup analysis if n permits
Presentation of Findings
- Describe protocol deviations
- State results in absolute numbers, not only percentages
- Present means AND SD’s
- Present detailed stats for alternative analysis and replication
- Discuss Internal validity: sources of bias and imprecision
- Discuss external validity: generalizability to other populations
