Hierarchy of Evidence


Evidence-based medicine has an accepted hierarchy of evidence based on study design (see Table 1). Observational studies where patients are not randomised are susceptible to bias, albeit unintentional.

Table 1. Hierarchy of evidence for treatment studies

1 Systematic review: A summary of the evidence that uses prespecified methods to search the literature and evaluate the quality of studies and combines the results of valid studies using appropriate techniques. Systematic reviews that evaluate treatment usually include randomised controlled trials rather than other study types. Cochrane review: Uses explicit methods as laid down by the Cochrane Collaboration, published in the Cochrane Library's Database of Systematic Reviews. (Note that a Cochrane review is only as good as the quality of the studies included. In many reviews, a meta-analysis is possible, combining the results from a number of trials.) Non-Cochrane systematic review: Review using prespecified but less rigorous criteria than Cochrane to do a systematic search for all primary studies on a question and appraise and summarise them. (The abstracts of non-Cochrane systematic reviews can be found in PubMed, under Clinical Queries, and the abstracts of good-quality systematic reviews are in the Cochrane Library's Database of Abstracts of Reviews of Effectiveness.)

2 Meta-analysis: A particular form of systematic review that uses quantitative statistical techniques to combine and summarise the results of all relevant studies, almost as if it had been one huge trial; for example, a meta-analysis could be conducted of all studies of SIDS that reported sleeping position.

3 Randomised controlled trial: Subjects are randomly allocated to an experimental (treatment) group or a control (placebo or different treatment) group and the outcomes are compared.

4 Cohort study: A non-randomised study of two groups of patients. One group receives the exposure of interest (e.g. a treatment) and the other does not; an example would be a study comparing outcomes in pregnant women who took thalidomide and women who did not.

5 Case control study: Patients with the outcome being studied are matched with one or more controls without the outcome and compared regarding different exposures to look for risk factors for or predictors of the outcome. For example, a group of preterm babies with a rare outcome, say neonatal meningitis, could be compared with matched controls without meningitis with regard to various factors such as chorioamnionitis, gender, artificial ventilation, parenteral nutrition, and so on to determine risk factors for neonatal meningitis and factors that appear to protect against developing meningitis.

6 Case series: Report of a series of patients with a condition, but no controls.

7 Case report: Report of one or more patients with a condition.

The hierarchy of evidence applies to treatment studies but does not apply to studies of aetiology, diagnosis or prognosis:

1. For aetiology, the best evidence is from large cohort studies or case control studies or sometimes randomised controlled trials (RCTs).
2. For diagnosis, the best evidence is from large cross-sectional studies in a population similar to your own, because the results will be most relevant to your clinical practice. These studies compare the test or tests you are interested in to a reference test or ‘gold standard’. For example, a new test such as polymerase chain reaction (PCR) for herpes simplex virus (HSV) could be compared with viral culture, although if HSV PCR is positive but culture negative you then have to decide if the PCR is more sensitive than culture (true positive) or if PCR is ‘too sensitive’ and also detecting contamination (false positive).
3. For prognosis, the best evidence is from large cohort studies followed over time, preferably in a population similar to your own. In addition, the no-treatment or placebo groups from large RCTs can provide excellent ‘natural history’ data on prognosis.
   

You can use Medline, via PubMed, to search for systematic reviews through PubMed Clinical Queries, which can be accessed by clicking on the Clinical Queries logo under ‘PubMed Tools’ in the middle column of the PubMed home page (Fig. 2). This will take you to the Clinical Queries screen (Fig. 1). Enter your search terms into the Search box near the top of the page and click on the SEARCH logo. PubMed automatically finds systematic reviews, including Cochrane Reviews, which are displayed in the middle column, labeled ‘Systematic Reviews’ (Fig. 1). These are the most useful evidence for the clinician. If there are no systematic reviews, then column 1 (Clinical Study Categories) will give you randomised controlled trials and observational studies. Under ‘Clinical Study Categories’ are two boxes for category and scope. The default settings are ‘Therapy’ for category and ‘Broad’ for scope. The scope setting ‘Broad’ gives observational studies. Changing ‘Broad’ to ‘Narrow’ gives randomised controlled trials.


It is important to try different search terms. For example, you want to know how effective prophylactic antibiotics are in preventing recurrence of urinary tract infections (UTIs). If you type ‘(urinary tract infections) (prophylactic antibiotics)’ into the Clinical Queries Search box in PubMed, using brackets to ensure the computer searches for the correct combined term, you get 33 systematic reviews, none of which helps you. Furthermore, they do not include the most relevant Cochrane systematic review.[8]


Why? Because the term ‘prophylactic antibiotics’ is too narrow. Searching using (urinary tract infections) (antibiotics) yields 231 reviews, including the Cochrane systematic review,[8] and 1071 RCTs, but this is too many for comfort. However, if you limit the review by entering ‘(urinary tract infections) (antibiotics) (children)’ as search terms, you get 58 reviews including the most relevant[8] (Fig. 3). If you use the term ‘pyelonephritis’ (instead of ‘urinary tract infections’) plus ‘antibiotics’ plus ‘children’, you get only 19 papers, but they do include reference 8. Incidentally, the Cochrane review tells you that prophylactic antibiotics reduce the incidence of recurrent UTIs but also cause adverse effects and the emergence of resistant organisms and that you need to treat 16 children presenting with their first UTI using 1 year of prophylactic antibiotics to prevent one child having a recurrence, even if the child has mild to moderate vesico-ureteric reflux.[8]

Recommendations

The biggest barrier for many doctors is fear of trying. For those still unconvinced, many institutions run courses that can familiarise clinicians with searching the literature and help refine their searches and save time. Searching the literature is rewarding and satisfying, as well as important, as it should help improve clinical decision-making. The sort of quick search described above should take only 10 to 15 minutes.