Epidemiology

EPID 801 – Introduction to Epidemiology Week 7 Summary

Clinical trials are the most well-known experimental design and the ultimate step in testing causal hypotheses
From an epidemiological viewpoint, the "gold standard" intervention study is a randomized controlled trial (or RCT).
Observational study: The investigators use the data observed in the population to make inferences on the relationship between exposure to a factor and an outcome.
Experimental study: Planned research design. Treatment and exposures occur in a “controlled” environment. Investigators can “control” the exposure. In other words, the investigator assigns study participants to the intervention or control groups
In an RCT, participants are allocated to study arms at random. "Random" has a specific meaning in this context, as we will see later. "Controlled" means that one study group receives the intervention (the intervention group), while another does not (the control group). A control group is required so outcomes in the intervention group can be compared with those in the control group.
In a randomized trial, the only difference between the intervention and control groups is whether or not they received the intervention. Therefore any difference in the outcome should be attributable to the intervention.
Key features of experimental studies: recruitment, enrollment, informed consent process, allocation concealment, randomization, masking/blinding, follow-up, outcome ascertainment, and data analysis.
Individual-level and Group-level interventions:
In some instances, individuals in the study population are randomly allocated either to receive an intervention or form part of the control group. However, some interventions are designed to have an effect on entire communities (for example, an intervention to reduce air pollution can only be introduced at the level of the community, not the individual). Some interventions may be most effectively tested by randomizing clusters (groups) of individuals.
The main problem with comparing interventions at the group level is the number of groups required. Sample sizes for cluster-randomized trials are much larger than for individually- randomized trials.
Treatment allocation is a two-step process: Randomization and allocation concealment.
Why is randomization so important?
The major advantage of randomization is that when done correctly on a large enough group, the comparison groups will usually be similar with respect to all exposures except
the intervention.
Randomization is a way of controlling for potential known and unknown confounders.
We can then check that the randomization has been successful, i.e. that the distribution of
confounders is the same in each group (Table 1 of studies).
Assigning the Intervention
Randomization means that every participant has a known chance (usually an equal
chance) of being allocated to either (or any) of the treatment groups, and neither the
investigator nor the participant can predict which group the participant will be allocated.
Randomization is usually done by assigning random numbers to the study participants and using these random numbers to indicate which group the participant is allocated.
Randomization also minimizes the possibility of bias affecting the allocation groups. When
randomization is done properly, the investigator has no control over which patient receives which intervention, and so the outcome cannot be influenced by the investigator's preferences.
Other methods of allocation may appear to be as good as randomization.
For example, we might decide to allocate patients born on an even-numbered day of the
month to the intervention group, and those born on an odd-numbered day to the control group. Alternatively we might decide to allocate alternate patients to the intervention group, according to the day of the week, or the order in which they were seen by the investigator.
Such methods are not satisfactory, because if the investigator can predict which group the patient will be assigned to before the patient is recruited to the study, the investigator could potentially manipulate the allocation system.
Special methods of randomization are sometimes useful.

o Block randomization is a way of ensuring that the number of participants allocated

to each group is equal after every block of x patients (for example, 4 or 8 etc) has entered the trial. This is a way of ensuring that the number of participants allocated to each group is similar throughout the trial.

o Stratified randomization is useful if other risk factors have a strong influence on the outcome. For example, age may be an important risk factor for the outcome. We might therefore decide do a stratified randomization by age group. In this case we would divide our study population into strata by age group, and then randomize subjects in each stratum of age group separately. This ensures that the randomized groups will be balanced with respect to the risk factor determining the strata (i.e. age).

• Allocation Concealment: \n o Allocation concealment should not be confused with blinding. Allocation concealment

concentrates on preventing selection and confounding biases, safeguards the assignment sequence before and until allocation, and can always be successfully implemented. Blinding concentrates on preventing study personnel and participants from determining the group to which participants have been assigned (which leads to ascertainment bias), safeguards the sequence after allocation, and cannot always be implemented.

o Randomization should be distant and separate from clinicians conducting the trial. o Good allocation concealment: central randomisation; numbered, opaque, sealed

envelopes; sealed envelopes from a closed bag; numbered or coded bottles or containers; drugs prepared by the pharmacy; or other descriptions that contain elements convincing of concealment.

• Concealment of allocation: \n o Procedure to protect the randomization process BEFORE participants enter the trial o Concealment of allocation is ALWAYS feasible \n o If not done, results in selection bias (randomization benefits are lost, and treatment

assignment is no longer truly random)

• Blinding: \n o Masking of the treatments AFTER randomization (once trial begins) o Failed masking of patients, investigators, outcome assessors, etc

o Blinding is not always feasible \n o If not done, can result in participants biasing their responses because of their

knowledge of treatment; can also lead to biased outcome assessment because

investigators have knowledge of treatment \n o Masking or blinding is used to increase the objectivity of the persons dealing with the

randomized study \n o Use of the terms single/double/triple blinding is discouraged (CONSORT 2010) \n o Open trials: All participants and investigators know who is getting which intervention o Double-blind: Two groups do not know - usually, it is the participants and the outcome

assessors/investigators

• What should the control group receive? \n • We need to decide what is the appropriate treatment for the comparison group. This may be: \n o A placebo, that is a treatment that resembles as closely as possible the active

treatment, but has no active constituent. Placebo controls are appropriate when we are uncertain whether the intervention is better than no treatment at all, and there is no standard treatment of proven value which is already in use. 

o Standard treatment. This would be appropriate where an existing treatment of established value for the condition in question is already in routine use.

Follow-Up and Ascertainment of Outcomes
If there is incomplete follow-up, and the completeness of follow-up differs between the
intervention and the control groups, bias may be introduced.
It is important that follow-up is continued even if participants discontinue the treatment
that they are allocated, whatever the reason for discontinuation. Failure to do this may
lead to selection bias.
The outcome of interest must be clearly defined, preferably using a case definition
which was decided at the start of the study. 
We should also try to ensure that the person who assigns outcome diagnoses does not
know which treatment group the subject was allocated to.
Analysis of Intervention Studies
The main measures of effect that we use to analyze an intervention study with a binary
outcome (for example, where the main outcome is whether participants experience a specific disease event or not) are the risk ratio or the rate ratio in the intervention group compared with the control group. We should always calculate and report 95% confidence intervals around point estimates of the risk ratio or rate ratio, since these help us to judge the precision of the estimate. Alternatively, for a binary outcome, we can express the results in terms of a survival analysis: this means that we compare the time to the occurrence of the outcome of interest in the intervention and control groups. If the main outcome is not a binary variable, other methods of analysis will be required, such as the comparison of two means.
The main analysis of an intervention study should be an "intention to treat" analysis.
This means that we compare the incidence of the outcome in individuals who were
randomized to receive the intervention with the incidence of the outcome in individuals who were randomized to the control group. This is irrespective of whether some people in each group stopped taking the allocated treatment, changed to the other treatment, or were lost to follow-up.

An "intention to treat" analysis is the most important and "safest" analysis, because by doing this, we compare the intervention and control groups as they were originally randomized.
Any other analysis that we may perform, for example according to the treatment actually received, is less reliable, because the groups may have become biased.
An "intention to treat" analysis may result in an underestimation of the effect of an intervention, because some of the people who are assumed to have received the intervention may in fact not have received it (if they switched to the other treatment allocation, or were lost to follow-up).
The true effect of the intervention may therefore be diluted, because not all the people assumed to be in the intervention group actually received the intervention.
Alternative analyses can be performed, such as Per protocol: Includes only those participants who complete the trial as randomized. As treated: Participants are analyzed according to the intervention they received (which may be different from the one they were assigned to!)
Such an analysis may introduce bias, since it is likely that participants who discontinue treatment are different from those who remain on treatment.
Interpretation of Intervention Studies
If the randomization process was effective, this should have controlled for confounding.
We can assess whether the randomization was effective by comparing the baseline characteristics of the randomized groups. If the baseline characteristics of the groups are similar, it is likely that randomization has been effective, and therefore that all known and unknown confounders will have been controlled for.
In a well-conducted randomized controlled trial, in which the results are not significantly influenced by bias, confounding or random error, any difference in outcome between the treatment groups should be attributable to the intervention, because the only difference between the groups is whether they received the intervention or not.
RCTs can be classified in many ways: \n o Aspects of interventions being evaluated
§ Efficacy (can it work?) vs. effectiveness trials (does it work?) § Superiority vs. equivalence trials vs. non-inferiority \n § Phase I, II, III trials
o How participants are exposed to interventions § Parallel trials
§ Crossover trials § Factorial trials
Bias in RCTs:
Pillars of an RCT :
Minimization of:
Selection bias: by randomization and allocation concealment
Performance and detection (Information bias): by blinding; to reduce that outcome are
assessed in the same way in each group
Attrition: By intention to treat analysis
Other bias:

• Publication bias, bias due to competing interests, outcome reporting bias, etc. Ethical aspects in RCTs

• Is it ethical to randomize?

The investigator cannot allocate an intervention believed to be harmful, nor withhold an intervention believed to be beneficial. Hence a trial is only ethical when there is equipoise, i.e., there is no evidence that one of the treatment options offered is superior to the other.
Consent:
- In the majority of intervention studies, since we are doing an experiment involving
human subjects, we must obtain their informed consent to take part in the study.
- The investigators must explain to potential participants, using language that is clear to the participants, the aim of the study and what is required of them if they agree to take
part. The participants must be fully informed of the potential risks and benefits.
- In a randomized trial, study participants must understand that they may receive either
(or any) of the treatment options under study, and that they may not know which treatment they received until the end of the study. They must also be free to decline to take part, without this affecting their care. They may need time to consider whether or not they wish to participate.
- In intervention studies where the intervention is introduced at the level of the community (as we will discuss later), it is often not feasible to obtain informed consent from every member of the communities involved. In such conditions, consent is usually obtained from community leaders. Even when community consent has been obtained, the investigators should take steps to ensure that members of the community are informed about the study.
The major advantages of RCT s are:
- if properly randomized, the intervention and control groups will be similar in all respects except the intervention, and so selection bias and confounding are minimized if the participants are "blind" to the treatment allocation, reporting bias is minimized; if the investigators are "blind" to the allocation, observer bias is minimized
- because RCTs carry less risk of bias and confounding than other study designs, they can provide powerful evidence of a causal relationship between the intervention and the outcome intervention studies are similar to cohort studies in that:
- multiple outcomes can be examined
-  the incidence rate of the outcome can be measured
The major disadvantages of RCT s are:
- they are often expensive to conduct: they may require a large study team, perhaps at several sites, and may require a long follow-up period. in some situations, intervention studies are impossible to conduct for ethical reasons.