Abstract
In this issue of SMJ, Koulaeinejad et al. 1 reminds us of importance of randomised controlled trials in assessing treatments in clinical practice. Randomisation is an accepted method of comparing estimates of treatment effects in medical research. Apparent lack of predictability using this approach ensures that biases of researchers are avoided.2–4 Inherent process of concealing allocation makes certain that participants and researchers cannot influence decision of participation into the study. In order to achieve a fair outcome, stratification of participant groups according to their characteristics (known and unknown) has to be similar and comparable. This removes any unintended influence of confounding factors on the outcomes.4 This also creates a state of not knowing which treatment is good and is termed as clinical equipoise- a position adopted under this experimental intervention both by participants and researchers. This is the only experimental research design to test new treatment methods, however there are issues with randomised controlled trial (RCT) designs as the participants are selected based on pre-defined criteria and do not necessarily be representative of the population being offered treatment on day to day basis. The rigid and inflexible design may be scientifically valid but despite spending huge resources (including cost) may not be applicable or generalizable to those who needed this most- very sick and patients with atypical conditions.5,6 The justification of assessing intervention through RCT design, however has to balanced against safety and long-term harms of treatments introduced into clinical practice through anecdotal or poor evidence.
There are some real challenges in conducting RCTs; randomisation may not be possible or refused by research participants or context of design may not allow participants to be accepted. Various methodological alternatives to RCTs have been suggested.7,8 In order to take care of confounding factors in observational studies data, propensity score matching was introduced. Propensity score matching mitigates selection bias at baseline in the data.9,10 Only drawback for this approach is that a potential exists for unknown confounding factors or less comprehensive coverage of all the known baseline characteristics. The Campbellian framework approach for observational data allows addition of design elements to test internal validity of the study designs. This includes matching and stratifying at multiple level, addition of multiple controls and nonequivalent dependent measures. This may not cover confounding factors which can be source of biases in the observational studies, however.
Quantitative assignment design is another alternative methodology to RCT design for assessing new treatment interventions especially where ethical issues are not favouring classical randomisation assignment. This type of assignment can be based on clinically relevant risk score of disease, interrupted time series or settings of neighbourhoods (access to a treatment option in one geographical area vs. another). One disadvantage of this method is unknown variables at the time of assignment which could potentially introduce biases. The issue of participation into randomised trials can also be addressed by randomisation using encouragement design. Here analysis is used to compare between those encouraged and not-encouraged rather than based on participants allocation to intervention status. This allows participants to choose intervention without commitment to adhere to protocol based allocation and compliance.