Studies can be broadly divided into comparative studies and non-comparative studies. Comparative studies, as the name suggestes, compare exposed participants with unexposed, or participants with and without interventions, or compare different exposures or interventions. Non-comparative studies simply report the outcome of one, or a series of cases. It is very difficult to make any clear predictions or recommendations on the basis of case studies, but they are useful in identifying the optimum rigorous research method to apply in future studies.
Comparative studies may involve investigator selection or assignment of participants (subjects), or the investigator may simply follow groups or individuals in an observational study. The best selection process involves random allocation, ideally where neither the investigator nor subject knows which group they are in (double-blind randomised controlled trial, the ‘gold standard’). It can be very difficult to arrange such trials, for example if the exposure is a surgical procedure, or carrying out a particular job task. Randomisation can be of individuals, or where this is particularly difficult it may be randomised in groups or ‘clusters’.
Where it is not possible to assign individuals to interventions, for example because the study is comparing firefighters with forestry workers, an observational study will be undertaken. This can compare groups, or it can compare individuals before and after interventions or exposures. Groups can be compared at a moment in time where both exposure and outcome are considered at the same time, in a cross-sectional study. Groups can be defined by outcome, in a case-control study. Or groups can be followed over time in a cohort study.
Sometimes a study shows very early on that one group is doing substantially better than another and it may be halted early as it is unethical to continue. Sometimes an outcome is so obvious that no actual study should be needed; the most obvious being the effect of a parachute. No test has ever been done comparing the outcome of someone falling from a plane with, and without, a parachute in order to decide whether it is worth using a parachute (Smith and Pell, 2003).
There are a number of major sources of bias in studies. The study may be funded by a pharmaceutical company or the manufacturer of a medical device. They may undertake a number of studies, some of which produce negative results and others produce positive results; they may only publish positive results. Publishers may prefer papers with positive results (who wants to read about a study that doesn’t prove anything?). The authors may have a biased view and may either deliberately or unintentionally skew the results through choice of tests, choice of subjects or selective analysis of data. Most papers will note the source of funding, and any commercial links between authors and the source of funding. Authors who have affiliations with advocacy groups are more likely to introduce bias.
The risk of introducing bias or confounders varies between the different study types noted above. The greater the risk, the less effective and reliable the study design. Studies can be graded in order of effectiveness:
A study at the bottom of this list can still be of good quality and reliable provided that there are no significant confounders or bias, the strength of the study is good and the outcome is of major significance. Parachute use has been based solely on case studies with no need to have any controls. No-one would doubt the effectiveness of a parachute. On the other hand, a high quality double blind randomised controlled trial may show that one in three patients will benefit from the new treatment, but one in three will develop substantial side effects. Further studies may be needed to see if side effects can be predicted, and whether better patient selection can make the drug useful.