2a: Evidence obtained from one well designed controlled study without randomization
In the previous post I highlighted the importance of randomization. However, in many situations randomization may not be possible or ethical (cancer drug trials, for instance).
The absence of randomization may be seen as a major weakness in such studies. This is not necessarily the case. Like all other study designs, Randomized Controlled Trials are also subject to bias (systematic error).
Trials can generate good quality evidence if bias is minimized at all stages.
This involves (the use of) the following:
Registration with one or more Clinical Trial Registries: Major methodological flaws; duplication of study; subsequent manipulation of protocol, etc. are prevented.
Trial Monitoring Committee: The purpose of a Trial Monitoring Committee is to oversee the trial. It consists of independent observers/ experts who ensure that the trial is conducted according to protocol; determine if the trial should be stopped prematurely, etc.
Randomization: This minimizes selection bias. All eligible subjects have similar chance of allocation into any arm of the trial. In non-randomized trials randomization is not undertaken.
Allocation concealment: The allocation sequence is not known to the investigators/ observers in advance. Usually, subjects are allocated using a predetermined sequence/ algorithm. The sequence would have been generated through randomization in a randomized trial. Allocation concealment minimizes selection bias- deliberate, systematic allocation to one or other arm either to influence the outcome of the trial, or because of personal preference for intervention. Even if randomization was performed, if the sequence generated by randomization is known, investigators could undermine the randomization process by selectively assigning subjects to one or other arm. Mere randomization does not minimize selection bias- it must be accompanied by allocation concealment to be effective.
Blinding/ Masking: Camouflaging the intervention(s) and control so that one cannot guess whether one is receiving the intervention(s), or placebo. This may involve one or more of the following: subject, observer, analyst. Masking the subject minimizes subject/ performance bias. Masking the observer minimizes observer bias. Masking the analyst minimizes bias during statistical analysis. In many situations masking the intervention is either very difficult, or not possible. In such situations, masking the observer is adequate.
The following article discusses the importance of allocation concealment and masking:
Prevention of attrition: Excess attrition in one arm of the trial will result in an imbalance between the arms. The resulting bias is called attrition bias. Attrition may be prevented through careful explanation of procedures, potential effects (good and bad) due to participation, importance of the trial, etc. during recruitment (before obtaining consent and randomization). Frequent follow-ups by trial staff may also help during the trial.
Contamination control: Contamination occurs when subjects in one arm begin doing/ receiving the intervention intended for another arm. Example: A study on the effect of exercise on lipid levels has an intervention (exercise) arm, and a control (no exercise) arm. If subjects in the control arm begin exercising (or those in the intervention arm stop exercising), contamination is said to have occurred. Contamination reduces the true difference between the interventions/ intervention and control. Contamination may be minimized by insisting on strict adherence to the protocol.
Intention To Treat Analysis: Once contamination has occurred, the observations from the subjects involved should not be included with another arm. Their observations should be analyzed as with other subjects from the original allocation group/ arm. Failure to do so will result in bias.
Thus, randomization, though crucial, is not the only means of addressing/ minimizing bias in a trial. It can only minimize selection bias, that too, if accompanied by allocation concealment. Other forms of bias may still occur. Minimizing those biases will improve the quality of evidence. Non-randomized trials can generate good quality evidence by doing so. Evidence from such trials constitute Level 2a.