This evidence Pyramid (Figure 2) was developed to categorize clinical studies for their strength of evidence according to their risk for bias. Studies therefore are ranked from those with the least risk for bias (Level 1) and progressively down to those with the greatest risk for bias.
At the top of the hierarchy are Clinical Practice Guidelines (CPGs). CPGs represent an interpretation or translation of the highest levels of research evidence on a specific topic to provide guidance to clinicians. Many areas of practice do not have a CPG, however when one does exist, it saves time in trying to interpret the research.
Systematic Reviews and Meta-Analyses, are “Secondary” or “Filtered studies.” and provide Level 1 evidence. Secondary research provides a synthesis of the primary/individual research studies, which address the same specific and very focused question regarding a particular intervention. Pre-established rigid inclusion criteria are identified for the selection of studies. Typically, systematic reviews use only the highest levels of evidence available. They evaluate the quality of the studies as well as report on the consistency of the results between studies. When possible, a meta-analysis should be conducted which combines the data from the individual studies and conducts an analysis of this pooled data. This analysis can increase the precision of estimates of treatment effects. Thus, a meta-analysis of RCTs is considered the most valuable when testing for causality.
If no systematic review or meta-analysis is available, then an individual randomized controlled trial (RCT) is recognized as the next best evidence (Level 1) for determining causality. An RCT is the highest level of primary evidence and is a “true experiment” in which eligible individuals are randomly assigned to either the experimental or control group. The experimental group would receive the intervention being tested whereas the control group would receive a placebo or no treatment. RCT’s have the ability to safeguard against bias through “randomization” of participants and “blinding” of either the participants or the investigator (or both).
Thus, for study types used to determine causality, Level 1 evidence is preferred unless it is unethical to do studies that are experimental. For example, one could not ethically conduct a randomized controlled trial to determine if smoking causes lung cancer. For that to occur, researchers would have had to randomly assign one group of study participants to be in the intervention group (smoking) and the remainder assigned to the control group (non-smoking) and then wait for decades to see at what rate the outcome would occur (i.e., lung cancer) in each group and to see if there was a significant difference between the rate in each groups. In that particular case, Sir Bradford Hill and Dr. Doll used prospective cohort studies where 2 groups (smokers and non-smokers) were followed over a period of time and then compared the outcome (which group contracted lung cancer and which group did not). Individual cohort studies are considered Level 2 evidence; however, a systematic review of cohort studies will provide a higher level of evidence than an individual study.
There are two types of observational studies, the strongest being the Cohort Study, which provides Level 2 evidence. Cohort studies are prospective in nature and thus satisfy the Bradford Hill requirement of “temporality.” In these studies, participants are placed into groups based on their exposure to a risk factor/causal agent. One group of individuals has been exposed to a putative causal agent (e.g., tobacco), while the other group has not (no exposure to tobacco). Both groups are then followed by the researcher for a period of time to measure the “incidence” or development of the disease or outcome of interest, e.g., lung cancer, which is reported in terms of relative risk, “RR”. RR is a calculation of the incidence of the outcome in the group with the exposure, divided by the incidence of the outcome in the group without the exposure.
Case-control studies are the second type of observational studies and provide Level 3 evidence. These studies are “retrospective in nature.” In other words, they look back in time at a group of individuals who already have the disease or outcome being studied. This group is then compared with a control group who are typically matched in all possible aspects with those who have the disease/outcome except they do not have the disease/outcome. Case-control studies are at higher risk for bias for a number of reasons. One, because of their retrospective nature, recall bias can be a major problem in that investigators often have to rely on the subjects’ recall or incomplete sources of information for exposure histories. Typically, those with the disease have thought more seriously about their exposure history than those without the disease. Additionally, confounding variables are not often accurately identified by the researchers along with concerns about the suitability of the control group selected.
Laboratory and animal studies are at the bottom of the hierarchy since evidence-based practice deals with how interventions work in humans. They are both important in that testing at these levels often must occur before human trials can occur. They are useful when determining the Bradford Hill criterion of analogy.